Datasets:

kilian-group
/

arxiv-classifier

Dataset card Data Studio Files Files and versions Community

Subset (4)

default · ~99.6k rows (showing the first 120k)

Split (2)

train · ~72.4k rows (showing the first 93.1k)

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
1609.05546	1	1609	2016-09-18T20:55:15	Phosphorylation-induced mechanical regulation of intrinsically disordered neurofilament protein assemblies	[ "physics.bio-ph" ]	The biological function of protein assemblies was conventionally equated with a unique three-dimensional protein structure and protein-specific interactions. However, in the past 20 years it was found that some assemblies contain long flexible regions that adopt multiple structural conformations. These include neurofilament (NF) proteins that constitute the stress-responsive supportive network of neurons. Herein, we show that NF networks macroscopic properties are tuned by enzymatic regulation of the charge found on the flexible protein regions. The results reveal an enzymatic (phosphorylation) regulation of macroscopic properties such as orientation, stress-response and expansion in flexible protein assemblies. Together with a model explaining the attractive electrostatic interactions induced by enzymatically added charges, we demonstrate that phosphorylation-regulation is far richer and versatile than previously considered.	physics.bio-ph	physics	Phosphorylation-induced mechanical regulation of intrinsically disordered neurofilament protein assemblies Eti Malka-Gibor, 1,†Micha Kornreich, 1,†Adi Laser-Azogui, 1Ofer Doron, 1Irena Zingerman-Koladko, 2Ohad Medalia, 3and Roy Beck1,∗ 1Raymond and Beverly Sackler School of Physics and Astronomy, Tel Aviv University, Israel 2Department of Life Sciences and the National Institute for Biotechnology in the Negev, Ben-Gurion University, Israel 3Department of Biochemistry, University of Zurich, Winterthurerstrasse 190, Switzerland 6 1 0 2 p e S 8 1 ] h p - o i b . s c i s y h p [ 1 v 6 4 5 5 0 . 9 0 6 1 : v i X r a Abstract The biological function of protein assemblies was conventionally equated with a unique three-dimensional protein struc- ture and protein-specific interactions. However, in the past 20 years it was found that some assemblies contain long flexible regions that adopt multiple structural conformations. These include neurofilament (NF) proteins that constitute the stress- responsive supportive network of neurons. Herein, we show that NF networks macroscopic properties are tuned by enzymatic regulation of the charge found on the flexible protein regions. The results reveal an enzymatic (phosphorylation) regulation of macroscopic properties such as orientation, stress-response and expansion in flexible protein assemblies. Together with a model explaining the attractive electrostatic interactions induced by enzymatically added charges, we demonstrate that phosphorylation-regulation is far richer and versatile than previously considered. INTRODUCTION In the past two decades it was discovered that approxi- mately 50% of human proteins are intrinsically disordered, i.e. they contain long peptide regions that do not fold into secondary or tertiary structures. The disordered regions con- tain a disproportionate number of phosphorylation sites, which are functionally crucial (1). In particular, assem- blies of phosphorylation-rich intrinsically disordered pro- teins (IDPs) are associated with neurodegenerative diseases including amyotrophic lateral sclerosis, Alzheimer's, Parkin- son's and Charcot-Marie-Tooth (2 -- 4). Despite the signifi- cance of IDP assemblies, little is known about the phos- phorylation regulation of their structural and mechanical properties. NFs make a valuable model system for the exploration of phosphorylation-driven interactions in IDP assemblies, due to their high modularity in protein content and phos- phorylation levels. In axons, NF proteins hierarchically form a filamentous network whose main roles are to provide the cell with its mechanical support and structure (5). Pro- tein subunits co-assemble into 10 nm bottlebrush-like fil- aments (Fig. 1A,B), where the filament backbone consists of the N-terminal domains. The long disordered C-terminal tail domains protrude outwards, and mediate inter-filament interactions and neuronal cytoskeletal organization (6, 7). The expression levels of NF proteins are modified dur- ing nerve growth or trauma, where changes in protein ratios are thought to accommodate the changing mechanical and structural needs of the nervous system. This modularity is enabled by three of the NF proteins, NF-L, NF-M, and NF- H, which are expressed in both the central and peripheral nervous systems (8, 9). Their disordered tails, which govern the network organization, differ considerably in sequence, length, net charge and charge distribution (Fig. 1C-F and Table S1). Additional functional versatility is achieved by enzy- matic regulation of tail charge distribution through phospho- rylation and de-phosphorylation. The majority of the phos- phorylation sites correspond to the serine residues of the Lys- Ser-Pro (KSP) repeat motifs. These repeats abound in the tail domains of NF-M and NF-H subunits, and significantly alter their charges (10). For example, de-phosphorylation of NF- H tail changes its total charge from -97 to -7 e, and charge density from -0.14 to -0.01 e/amino acid (Fig. 1 and Table S1). Given this significant charge difference, NF-M and NF- H phosphorylation is thought to govern the lateral extensions of NF tails, thereby regulating NF spacing, axonal calibre and protein transport (7, 11 -- 14). These structural roles are the focus of a recent debate, following a study that found no dependence of axonal caliber on the phosphorylation of NF-M (15). † Eti Malka-Gibor and Micha Kornreich contributed equally to this work. * Corresponding author: [email protected] 2 Figure 1: Structure of bottlebrush NF filaments. (A) A schematic of two neighboring interacting filaments. Each filament consists of three different subunit proteins, whose protruding tails are shown in colour (red, blue and green). (B) Transmission electron microscopy images of de-phosphorylated filaments (from left to right): NF-LH, NF-LM and NF-LMH. Scale bar = 100 nm. (c-f) Tail charge distributions of (C) phosphorylated NF-M, (D) phosphorylated NF-H, (E) de-phosphorylated NF- M, and (F) de-phosphorylated NF-H. The charge distributions were calculated at pH= 6.8 and averaged over a 5-amino acid window (see materials and methods). The grey boxes highlight the protein segments which are most affected by the phosphate charge removal. Two seemingly contradictory effects can be attributed to the phosphorylation-induced charge regulation. On the one hand, phosphorylation enhances the electrostatic repulsion due to its excess charge. On the other hand, phosphorylation can promote short-range attractive cross-bridging between the disordered tails (16 -- 21). Here we address the structural and mechanical aspects of these two opposing phosphory- lation effects. We show that NF assemblies have a distinc- tive phosphorylation-induced regulation of network proper- ties, such as compression resistance and orientation. The regulation stems from the structural plasticity of the dis- ordered tails, and therefore differentiates between IDP and structured-protein assemblies. MATERIALS AND METHODS Calculation of basic sequence-based properties Protein sequences are imported from UniProt database (22), and their identifiers are: P02548 (NF-L), O77788 (NF-M) and P19246 (NF-H). Since a UniProt-verified Bovine NF- H sequence is still missing, the mouse NF-H sequence is used instead. Sequence-based calculations of NF tail charge are performed at pH 6.8 using the EMBOSS amino acid pKa table (23), and the average amino acid volume (cid:0)Va = 0.134 nm3(cid:1) calculation follows Ref. (24). The phos- phoserine sites are determined using the UniProt database, and the phosphoserine pKa2 is set to 6.2 for charge calcula- tions. Average tail charges and tail fraction charges appear in Table S1. 5006007008009001000−1−0.500.5500600700800900−1−0.500.5Amino acid #Amino acid #Charge (e)deNF-MdeNF-HNF-HNF-MABCDEFdeNF-LHdeNF-LMdeNF-LMHProtein purification Filament self-assembly and hydrogel formation 3 NF subunits (NF-L, NF-M and NF-H) are purified from bovine spinal cord using a modification of an earlier protocol (25, 26). Spinal cords are homogenized in an equal volume of buffer A (0.1 M MES, 1 mM EGTA, 1 mM MgCl2, 0.02% (w/v) sodium azide, 7 mM β-mercaptoethanol, pH 6.8 with NaOH) with 1% (w/v) Triton X-100 and 1 mM phenyl- methylsulfonyl fluoride. The homogenates are centrifuged at 30k RPM (Beckman rotor type 45-Ti) for 70 min at 4 ◦C. An equal volume of glycerol is added to the supernatant and incubated overnight. A pellet of NFs is recovered from the glycerol solution by precipitation at 40k RPM (Beckman rotor type 45-Ti) for 90 min at 4 ◦C. The pellet is homog- enized in buffer A with 0.8 M sucrose and clarified by spin- ning through a step gradient of 0.8 M sucrose buffer (0.8 M sucrose in buffer A) layered on top of 1.5 M sucrose buffer (1.5 M sucrose in buffer A) for 4 h at 55k RPM (Beckman rotor type 70-Ti). The pellet is homogenized in buffer B (0.1 M potassium phosphate and 0.1% (v/v) β-mercaptoethanol in 8 M urea, pH 6.5), and applied to a DEAE sepharose column (DEAE Sepharose fast flow column, GE Health- care). The column is rinsed with buffer B containing 55 mM NaCl which elutes NF-H and protein contaminates. The next elution step, performed with buffer B at pH 7 containing 200 mM NaCl, elutes NF-L and NF-M. Using hydroxylap- atite (HT) column chromatography (hydroxylapatite bio gel HT, Bio-Rad), the contaminants are removed from the NF-H fractions. NF-L and NF-M are separated by HT column with a gradient of 0.1 to 0.4 M potassium phosphate pH 7.0. The purity and separation of the NF-H, NF-M and NF-L subunit proteins are verified by SDS-PAGE (Fig. S1). Protein de-phosphorylation For de-phosphorylated networks, purified proteins are dia- lyzed against de-phosphorylation buffer (50 mM Tris pH=8, 100 mM NaCl and 10 mM MgCl2), and then incubated with 50 units alkaline phosphatase (CIP, New England Biolabs) per 0.1 mg at 37 ◦C overnight. The process is monitored by the decreased mobility of de-phosphorylated NF-M and NF-H in SDS-PAGE (27, 28) (Fig. S1). De-phosphorylation of NF-M is further supported by liq- uid chromatography -- mass spectrometry, performed at the Biological Services Department, Weizmann institute. Tech- nical details are found in the Supporting Material. We compare our results against the listed NF-M phos- phosites in the UniProt database. We identified 7 out of the 22 experimentally verified NF-M phosphosites, as well as 7 out of 8 of the predicted phosphosites (22). None of the sites were phosphorylated (Table S2). Notably, since bovine NF-L tails contain only few (1-3) phosphorylation sites (10), networks composed of recombinant and native NF-L pro- teins do not differ in inter-filament spacing, orientation or compression response (29). Following protein purification and de-phosphorylation, pro- tein subunits are mixed in denaturing conditions at the desired composition. The protein solution is dialyzed against a MES buffer (100 mM MES, pH 6.8, 1 mM EGTA, 1 mM MgCl2, 0.02% (w/v) sodium azide, 7 mM β- mercaptoethanol and a total 150 mM of NaCl and NaOH salts) at 37 ◦C for 48 hours. For optical microscopy and small-angle X-ray scattering (SAXS), the reassembled filaments solution is centrifuged for 1 h at 50k RPM using a TLA120.1 rotor in a Beckman Coulter Optima TLX ultracentrifuge, and the supernatant is immediately removed from the pellet. The NF pellet is trans- ferred to 1.5 mm quartz capillaries, overlaid with ≈100 µl MES buffer solution and sealed with epoxy glue to pre- vent dehydration. To osmotically pressurize the network, the overlaying MES buffer solution is supplemented with 20,000 g/mol polyethylene-glycol (PEG)(30). The resul- tant PEG-induced osmotic pressure Π is determined by the PEG weight percentage (PEGwt), and follows the formula log10 Π = 1.57 + 2.75(PEGwt)0.21. To determine the protein molar ratios in the assembled hydrogel, control samples are analysed by SDS-PAGE as described in Refs. (25, 29). The protein molar ratios of the different hetero-filaments are 4:1 for NF-L:NF-H (denoted NF-LH), 7:3 for NF-L:NF-M (denoted NF-LM) and 10:3:2 for NF- L:NF-M:NF-H (denoted NF-LMH). Transmission electron microscopy For transmission electron microscopy (TEM), a sample of 10 µl is laid on a formvar coated 400 mesh grid (#3440c-FA, SPI Supplies) and then fixed and negatively stained as in Ref. (31). Images of filaments composed of de-phosphorylated subunits are shown in Fig. 1B, and resemble images of phosphorylated protein filaments (29, 32). Cross polarising microscopy (CPM) The hydrogel orientation (isotropic or birefringent nematic) is characterized by crossed polarised light microscopy (Fig. 2). Sedimented NFs are observed in 1.5 mm quartz capil- laries using a Nikon Eclipse LV 100 POL microscope fitted with 5-20X objectives. Micrographs are taken with a Nikon D90 camera. Small angle X-ray scattering (SAXS) NF hydrogels 2D diffraction data was integrated azimuthally and the intensity was plotted versus reciprocal distance q. The intensity, in arbitrary units, showed a broad peak with a maximum in the range of q = 0.1 − 0.2 nm−1. The peak location relates to the inter-filament spacing (d = 2π/q). 4 Broadening of this peak is observed due to density fluctua- tions and the semi-flexible nature of the individual filaments. Baseline background of the form A · q−B + C with B = 2 − 3 is subtracted from the integrated data, and the resul- tant peak is fitted with a Lorentzian function using Matlab (MathWorks) routines (32). Preliminary experiments were performed at our home- lab using a Pilatus 300K detector and a Xenocs GeniX Low Divergence CuKα radiation source setup with scatter- less slits (33). Subsequent measurements were performed at synchrotron facilities: I22 beamline in Diamond, Eng- land; SWING beamline in SOLEIL, France; and I911 SAXS beamline in MAX-lab, Sweden with 10 keV. Bulk modulus calculations For these calculations, we treat filament as infinitely long impenetrable cylinders of radius Rcyl = 5 nm, set in a hexagonal lattice (Fig. S3 and Ref. (34)). We define a prism- shaped unit cell whose base is an equilateral triangle with side length d, which is the inter-filament distance. The prism height is l =45 nm, which is the NF protein's rod length. As 32 tails emanate from each filament backbone every 45 nm (35), the unit cell contains a total of 32/6 · 3 = 16 tails. The surface of the equilateral triangle found in the hexago- cyl/2 , and therefore the nal model is S (d) = unit cell volume holds V (d) = S (d) · 45 nm. The volume fraction is given by φ = N Va/V , where N is the number of tail amino acids in V . The Π vs. d compression curves are fitted with smoothing splines (Fig. 3), and then used for the BT calculation: BT = −V (cid:18)dV (d) 3d2/4 − πR2 (cid:19)−1 √ dΠ dV = −S (d) d dΠ dd dd (1) RESULTS AND DISCUSSION We investigate the mechanical and structural roles of tail phosphorylation by reconstituting filaments from purified NF protein subunit, at desired subunit compositions. To pro- duce de-phosphorylated filaments, the natively phosphory- lated proteins are enzymatically treated with alkaline phos- phatase. Composite filaments include NF-L with either NF- M (NF-LM), NF-H (NF-LH) or both (NF-LMH). 10 nm wide filament formation is verified by transmission elec- tron microscopy (TEM, Fig. 1B), as reported for native NFs (17, 36). For simplicity, we refer to the de-phosphorylated filaments by a "de" prefix (i.e., deNF-LM, deNF-LH or deNF-LMH). At high concentrations, filaments condense into a hydro- gel network that phase separates from the supernatant. To characterize the mechanical response of the network, we vary the osmotic pressure (Π) using polyethylene-glycol (PEG) osmolyte. All phosphorylated filaments as well as deNF-LM filaments form large oriented (nematic) domains, Figure 2: NF network phase behaviour, as determined by CPM. (A) Bright field and CPM images of phosphorylated and de-phosphorylated networks at low (102 − 103 Pa) and high (105 − 2 · 105 Pa) osmotic pressure in quartz capil- laries. White dashed lines demark the capillary boundaries, as observed with bright field (see Fig. S2). The filaments in all networks are aligned (nematic) except for deNF-LM and deNF-LMH, which are isotropic (i.e., un-oriented) at low Π. Each capillary is approximately 1.5 mm wide. (B) Phase dia- gram showing the network phase behaviour at different Π's. Each star (cid:63) denotes a measurement point. either at low or high osmotic pressures, as determined by cross-polarised microscopy (CPM). In contrast, deNF-LH and deNF-LMH filaments form less oriented hydrogels with smaller nematic domains (Fig. 2). At low osmotic pres- sure, deNF-LH and deNF-LMH are isotropic and transition to a nematic phase at elevated osmotic pressure. Nonethe- less, in both cases, even at elevated osmotic pressures, the aligned domains still appear smaller and less illuminated NF-LMdeNF-LMNF-LHdeNF-LHNF-LMHdeNF-LMH103104105106 Nematic phase Isotropic phaseΠ (Pa)Low Π High Π Phosphorylated De-phosphorylated NF-LMNF-LHNF-LMHAB5 Figure 3: Comparison of native and de-phosphorylated networks using SAXS and osmotic pressure. (A) Intensity curves of NF-LH, NF-LM, and NF-LMH native and de-phosphorylated networks at 1% (w/w) PEG (Π = 2.2 · 103 Pa). (b-d) Semi-log plot of osmotic pressure Π vs. the inter-filament distance d of different network compositions: (B) NF-LMH, (C) NF-LH and (D) NF-LM. For calculation in Eq. 2, we integrate over the textured area in (D). Osmotic bulk modulus BT is shown in insets. The data is fitted with smoothing splines, which were then used to calculate BT . Subunit molar ratios for native and de-phosphorylated NF-LM, NF-LH and NF-LMH filaments are 7:3 (NF-L:NF-M), 4:1 (NF-L:NF-H) and 10:3:2 (NF-L:NF-M:NF-H). than those observed for native and deNF-LM hydrogels at comparable osmotic pressures. Hence, NF-H phosphoryla- tion regulates the macroscopic orientation of the hydrogel networks. Furthermore, a comparison of the CPM images at various osmotic pressures indicates that deNF-M promotes orientational order while deNF-H hinders it (Fig. 2). To study the nanoscopic structural organization and mechanics of the hydrogel, we measure the inter-filament spacing, d, using small-angle X-ray scattering (29, 32, 37). Azimuthally averaged intensity curves of phosphorylated (i.e., native) and de-phosphorylated networks at Π = 2200 Pa, are shown in Fig. 3A. For each intensity curve, the corre- lation peak position (q0) is denoted by a vertical line, and is related to the inter-filament distance by d = 2π/q0. At this given osmotic pressure, de-phosphorylation of NF-LMH and NF-LH results with a decrease of the inter-filament spac- ing from 70 to 58 nm and from 67 to 50 nm, respectively. Therefore, native phosphorylation of NF-LH and NF-LMH results with network expansion. This agrees with the conven- tionally considered roles of NF phosphorylation, where the substantial addition of charged phosphate groups is naively expected to stretch the NF tails and consequently increase the inter-filament spacing. The Π vs. d curves of NF-LH and NF-LMH exhibit the same phosphorylation-dependent effect over a wider range of osmotic pressures (Fig. 3B,C). In contrast, the correlation peak of deNF-LM shifts to lower q-values, in comparison to the NF-LM peak (Fig. 3A). This indicates that the spacing actually increases from 48 to 60 nm due to de-phosphorylation. The result is atypi- cal of charged polymers, as de-phosphorylation reduces the net charge by ∼ 50%, and thus the electrostatic repulsion between adjacent filaments is expected to decrease. This behaviour indicates an attractive interaction between phos- phorylated tails, which opposes the trend observed in NF-LH and NF-LMH networks (Fig. 3B,C). Notably, the expansion of deNF-LM in comparison to NF-LM is reversed only at high osmotic pressures, Π (cid:38) 105 Pa (Fig. 3D). The protein-dependent regulation of network expansion, alignment and osmotic stress response is schematically sum- marized in Fig. 4, together with conjectured tails micro- scopic organization (Fig. 4C,D,F). Phosphorylation of the NF-H tail aligns the network and increases the inter-filament distance (Fig. 4A). In contrast, NF-M phosphorylation has little effect on network alignment, and it surprisingly reduces the inter-filament distance (Fig. 4B). This indicates that NF-LM and NF-LH tails are organized differently. At low osmotic pressures, tails are expected to form two distinctive layers, known as the flower conformation (Fig. 4C). The inner layer corona is composed of the short NF- L tails while NF-H tails are repelled farther away from the corona into the outer layer. Since the tails within the outer layer are less dense, they assume a flower-like conformation (29, 38, 39). However, this picture does not agree with the NF-LM tails organization. The inter-filament spacing of NF- LM and NF-L networks are very similar (29, 32), suggesting that the long NF-M tails are hidden within the NF-L inner coronas (termed "truffle" regime) (29, 34). Under signifi- cant osmotic pressure, all filament types align and compress, while opposite tails increasingly inter-penetrate (Fig. 4E,F). In living cells, the osmotic pressure is induced by the crowded environment. We note that the effect of phospho- rylation on the compression response is much more pro- nounced than its effect on NF expansion. In particular, phos- phorylation at a given osmotic pressure, does not change the inter-filament spacing by more than 25%. However, in order to maintain the same spacing at the phosphorylated state, the osmotic pressure needs to increase by up to two orders of magnitudes. To further characterize this mechanical response, we cal- culate the osmotic bulk modulus, BT , which quantifies the d20406080100102103104105106Π (Pa)20406080103105107dB (Pa)20406080100102103104105106d20406080103105107B (Pa)Π (Pa)d20406080100102103104105106dΠ (Pa)20406080103105107B (Pa)DBCdq1E-30.010.11101103105Intensity (A.U)A6 Figure 4: Schematic of phosphorylation regulation of NF network expansion, alignment and osmotic stress response. (A) NF-H phosphorlyation aligns the isotropic deNF-LH network and increases the inter-filament distance, whereas (B) NF-M phosphorylation collapses the nematic deNF-LM network. Except for NF-LM networks, all protruding tails organize in two corona layers at low osmotic pressure (C). The outer layer is formed by the long tails (either NF-H or NF-M), and is denoted by a yellow background. Upon phosphorylation, deNF-LM tails transition from a (C) flower to a (D) truffle conformation. (E) Under significant osmotic compression, filaments align and compress (F) while opposite tails increasingly inter-penetrate. network's resistance to compression (Eq. 1 and Fig. 3B-D, insets). Similar to the phosphorylation dependent network expansion and collapse, we find that changes to the mechan- ical response are also protein specific. However, the effect of phosphorylation on the mechanical response is more pro- nounced, and BT is altered by as much as an order of mag- nitude. For NF-LH and NF-LMH, de-phosphorylation of the tails reduces BT at all d. In contrast, de-phosphorylation of NF-LM increases the network's resistance to compression (i.e., larger BT ) at low osmotic pressure. The latter occurs despite the reduced repulsive electrostatic forces and sug- gests that the charged phosphates also take part in attractive interactions (16, 32). Further support to the excess electro- static repulsion at high and low osmotic pressures are given by comparison to polymer scaling theories (Fig. S4). To estimate the attractive bridging energy per NF-LM phosphosite, we follow the calculation performed to quantify the attractive hydration energy between DNA double-helices (40). The free energy is derived from the Π − d diagram, under the hexagonal approximation. To evaluate the aver- age energy per phosphosite (), we integrate over the free energy difference between the two phosphorylation states, and divide by the number of phosphosites in the volume (Np): (VdeNF−LM − VNF−LM) dΠ. (2) Πint(cid:90) 0 = − kBT Np The integration is performed on the fitted smoothing-spline curves from Π = 0 to the intersection of the curves at Πint ∼ 105 Pa (See textured area in Fig. 3D), and yields approximately = −8 kBT per phosphosite. This is compa- rable to the free energy of protein salt bridges (41). Notably, the integration also includes contributions from repulsive NF-LHNF-LMPhosphorylationPhosphorylationpressureOsmoticFlowerTruffleCondensedABCDEFinteractions, and therefore provides a lower limit for the attractive average energy per phosphosite. Since these attractive interactions are sequence-dependent, they may account for the opposite phosphorylation expan- sion trends of NF-LM and NF-LH. To identify the polypep- tide segments involved in such attraction, we employ a coarse-grained "handshake" calculation, aimed at locating pairs of amino acid segments that interact via electrostatic bridges (32). We calculate the unscreened Coulomb energy of two interacting segments, where each segment is centered at a specific tail amino acid and the segment length is on the order of the polypeptide persistence length (approximately 3 nm (42)). We thus obtain a 2D matrix which points at the most electrostatically viable cross-linking pairs (Fig. 5A). Handshake analysis of the natively phosphorylated NF- H tail reveals multiple potential attractive sites located at its last 200 amino acids (Fig. 5B). This was also described by a more elaborate theoretical calculation (19) and agrees with previous experiments (43). Upon de-phosphorylation, new attractive sites are predicted near the filament backbone (Fig. 5B). These could participate in intra-filament attractive inter- actions, in agreement with the de-phosphorylated NF-H tail collapsed conformation (Figs. 3). The unexpectedly collapsed conformation of phospho- rylated NF-LM may involve interactions between NF-M segments with either NF-L or NF-M segments, as recently shown (29, 34). Here, upon de-phosphorylation, new NF-M attractive binding sites are formed further away from the fila- ment backbone (Fig. 5D,E). Consequently, the loss of excess phosphate charge actually expands the NF-M tail, as shown in Fig. 3D. CONCLUSIONS We demonstrate the roles of phosphorylation in regulating the structural and mechanical properties of NF networks. The phosphorylation-induced modifications strongly depend on the NF subunit composition, and can result with either network expansion or collapse (Fig. 4). This versatility orig- inates from the dual nature of the induced interactions, which are both repulsive and attractive and are protein-sequence dependent. The attractive interactions are clearly manifested in the deNF-LM network, where the removal of the excess phos- phate charges unexpectedly results with network expansion. This trend relates to previous studies, which suggested that phosphorylation can promote NF binding either by associ- ated proteins, by exposing hydrophobic residues, or by direct involvement in electrostatic bridging (16, 18, 19, 21). Of these possibilities, our results affirm the direct involvement of NF-M phosphosites in attractive interactions, which is justified by the stronger ionic bridging formed between the divalent phosphate group and basic amino acids (Fig. 5). 7 Figure 5: Handshake analysis of short-ranged electrostatic interactions between NF tail regions. (A) Two opposite 9 amino-acid long segments interacting. Details of the elec- trostatic interaction energy calculations are found in Refs. (29, 32). (b-e) Energy matrices for all possible segment pairs for two opposing (B) NF-H, (C) deNF-H, (D) NF-M, or (E) deNF-M tails. Interactions between two oppositely charged segments, which are more electrostatically viable, are denoted by blue colours in the interaction matrices. Solid box in (B) marks the last 200 amino-acids that are known to engage in attractive interactions. Dashed boxes in (E) denote phosphosite-rich segments (see also Fig. 1). Compar- ison of (D) with (E) reveals that de-phosphorylation forms new negative energy pairs between segments found further away from the filament backbone. The repulsive interactions govern NF-LH and NF- LMH compression response. Here, phosphorylation moder- ately increases the inter-filament distance and considerably enhances the osmotic compression resistance, characterized by the bulk modulus (Fig. 3). This questions the hypoth- esized main structural role of tail phosphorylation, i.e. to expand the NF network, and suggests a primarily mechanical role to NF phosphorylation. NF-M residue #500700900NF-M residue #500700900DNF-M residue #500700900500700900BNF-H residue #NF-H residue #60080010006008001000ENF-H residue #60080010006008001000C200-20ΔE(kBT)N-terminal++-2-+-2++--2--N-terminalCCANF-H residue #NF-M residue #8 The individual roles of NF-M and NF-H phosphoryla- tion, as demonstrated here, significantly increase opportuni- ties to regulate NF network physical properties. Therefore, the roles of NF expansion (Fig. 3), mechanics (21)(Fig. 3) and orientation (44, 45)(Fig. 2) must be considered where simultaneous changes in composition and tail phospho- rylation levels are observed. Such changes, for example, occur during neuronal growth and development, following injury, and in neurodegenerative diseases (8, 11, 46). Specif- ically, our results indicate a close relation of phosphorylation and NF compression response and orientation, which was scarcely considered before (6). Additional assemblies of hyper-phosphorylated disor- dered proteins are also involved in neurodegenerative dis- eases. These include the disordered tau and α-synuclein pro- teins, which are hyper-phosphorylated in pathological inclu- sions (2, 3). Their aggregation is commonly attributed to an indirect process, where phosphorylation-driven confor- mational changes expose new segments for attractive inter- actions. However, the electrostatic attraction observed in phosphorylated NF-M tails demonstrates that phosphates in disordered protein assemblies directly engage in signif- icant attractive interactions. This suggests that more atten- tion should be drawn to the role of phosphorylation-driven attractive electrostatics in the study of disordered assemblies. SUPPORTING MATERIAL Supplement to this article is available upon request. AUTHOR CONTRIBUTIONS R.B, A.A.Z, E.M.G, and M.K planned and initiated the project. E.M, M.K, A.A, O.D, I.K and O.M conducted the experiments. E.M and M.K prepared the samples and analysed the data. E.M, M.K and R.B wrote the paper. ACKNOWLEDGMENTS We are grateful to Dr. Geraisy Wassim of Beit Shean abat- toirs Tnuva for kindly providing us with the spinal cords. We thank Ekaterina Zhulina for useful discussions and sugges- tions. We thank the following beamlines for SAXS measure- ments: I911-SAXS at MAX IV Laboratory, Lund, Sweden; SWING at SOLEIL synchrotron, Paris, France; and I-22 beamline at Diamond, England. This work was supported by Israeli Scientific Foundation (571/11, 550/15) , the Tel Aviv University Center for Nanoscience and Nanotechnol- ogy and the Scakler Institute for Biophysics at Tel Aviv Uni- versity. Travel grants to synchrotron facilities were provided by BioStruct-X. References 1. Uversky, V. N., 2011. Intrinsically disordered proteins from A to Z. Int. J. Biochem. Cell Biol. 43:1090 -- 1103. 2. Chen, L., and M. B. Feany, 2005. Alpha-synuclein phos- phorylation controls neurotoxicity and inclusion formation in a Drosophila model of Parkinson disease. Nat. Neurosci. 8:657 -- 663. 3. Stoothoff, W. H., and G. V. W. Johnson, 2005. Tau phosphory- lation: physiological and pathological consequences. Biochim. Biophys. ACTA 1739:280 -- 297. 4. Liu, Q., F. Xie, A. Alvarado-Diaz, M. a. Smith, P. I. Mor- eira, X. Zhu, and G. Perry, 2011. Neurofilamentopathy in neurodegenerative diseases. Open Neurol. J. 5:58 -- 62. 5. Hirokawa, N., M. A. Glicksman, and M. B. Willard, 1984. Organization of mammalian neurofilament polypeptides within the neuronal cytoskeleton. J. Cell Biol. 98:1523 -- 1536. 6. Safinya, C. R., J. Deek, R. Beck, J. B. Jones, and Y. Li, 2015. Assembly of Biological Nanostructures: Isotropic and Liquid Crystalline Phases of Neurofilament Hydrogels. Annu. Rev. Condens. Matter Phys. 6:113 -- 136. 7. Sihag, R. K., M. Inagaki, T. Yamaguchi, T. B. Shea, and H. C. Pant, 2007. Role of phosphorylation on the structural dynam- ics and function of types III and IV intermediate filaments. Exp. Cell. Res. 313:2098 -- 2109. 8. Laser-Azogui, A., M. Kornreich, E. Malka-Gibor, and R. Beck, 2015. Neurofilament assembly and function during neuronal development. Curr. Opin. Cell Biol. 32:92 -- 101. 9. Yuan, A., M. V. Rao, Veeranna, and R. A. Nixon, 2012. Neurofilaments at a glance. J. Cell. Sci. 125:3257 -- 3263. 10. Trimpin, S., A. E. Mixon, M. D. Stapels, M.-Y. Y. Kim, P. S. Spencer, and M. L. Deinzer, 2004. Identification of endogenous phosphorylation sites of bovine medium and low molecular weight neurofilament proteins by tandem mass spectrometry. Biochemistry 43:2091 -- 2105. 11. Dale, J. M., and M. L. Garcia, 2012. Neurofilament Phospho- rylation during Development and Disease: Which Came First, the Phosphorylation or the Accumulation? J. Amino Acids. 2012:1 -- 10. 12. Kriz, J., Q. Zhu, J. P. Julien, and A. L. Padjen, 2000. Elec- trophysiological properties of axons in mice lacking neurofila- ment subunit genes: disparity between conduction velocity and axon diameter in absence of NF-H. Brain Res. 885:32 -- 44. 13. Ackerley, S., P. Thornhill, A. J. Grierson, J. Brownlees, B. H. Anderton, P. N. Leigh, C. E. Shaw, and C. C. Miller, 2003. Neurofilament heavy chain side arm phosphorylation regulates axonal transport of neurofilaments. J. Cell Biol. 161:489 -- 495. 14. Shea, T. B., C. Jung, and H. C. Pant, 2003. Does neurofilament phosphorylation regulate axonal transport? Trends Neurosci. 26:397 -- 400. 15. Barry, D. M., W. Stevenson, B. G. Bober, P. J. Wiese, J. M. Dale, G. S. Barry, N. S. Byers, J. D. Strope, R. Chang, D. J. Schulz, S. Shah, N. A. Calcutt, Y. Gebremichael, and M. L. Garcia, 2012. Expansion of Neurofilament Medium C Ter- minus Increases Axonal Diameter Independent of Increases J. Neurosci. in Conduction Velocity or Myelin Thickness. 32:6209 -- 6219. 16. Aranda-Espinoza, H., P. Carl, J.-F. Leterrier, P. Janmey, and D. E. Discher, 2002. Domain unfolding in neurofilament sidearms: effects of phosphorylation and ATP. FEBS Letters 9 34. Kornreich, M., E. Malka-Gibor, B. Zuker, A. Laser-Azogui, and R. Beck, 2016. Neurofilaments function as shock absorbers: compression response arising from disordered pro- teins. Phys. Rev. Lett. (accepted for publication). 35. Herrmann, H., L. Kreplak, and U. Aebi, 2004. Isolation, Char- acterization, and In Vitro Assembly of Intermediate Filaments. Elsevier, volume 78 of Methods Cell Biol., 3 -- 24. 36. Beck, R., J. Deek, M. C. Choi, T. Ikawa, O. Watanabe, E. Frey, P. Pincus, and C. R. Safinya, 2010. Unconventional salt trend from soft to stiff in single neurofilament biopolymers. Langmuir 26:18595 -- 18599. 37. Kornreich, M., R. Avinery, and R. Beck, 2013. Modern X-ray scattering studies of complex biological systems. Curr. Opin. Biotechnol. 24:716 -- 723. 38. Leermakers, F. A. M., and E. B. Zhulina, 2010. How the pro- jection domains of NF-L and alpha-internexin determine the conformations of NF-M and NF-H in neurofilaments. Eur. Biophys. J. 39:1323 -- 1334. 39. Jayanthi, L., W. Stevenson, Y. Kwak, R. Chang, and Y. Gebremichael, 2013. Conformational properties of inter- acting neurofilaments: Monte Carlo simulations of cylindri- cally grafted apposing neurofilament brushes. J. biol. phys. 39:343 -- 362. 40. Rau, D. C., and V. A. Parsegian, 1992. Direct measurement of the intermolecular forces between counterion-condensed DNA double helices - evidence for long-range attractive hydration forces. Biophys. J. 61:246 -- 259. 41. Kumar, S., and R. Nussinov, 1999. Salt bridge stability in monomeric proteins. J. Mol. Biol. 293:1241 -- 1255. 42. Bright, J. N., T. B. Woolf, and J. H. Hoh, 2001. Predict- ing properties of intrinsically unstructured proteins. Prog. Biophys. Mol. Biol. 76:131 -- 173. 43. Chen, J., T. Nakata, Z. Zhang, and N. Hirokawa, 2000. The C-terminal tail domain of neurofilament protein-H (NF-H) forms the crossbridges and regulates neurofilament bundle formation. J. Cell. Sci. 113 Pt 21:3861 -- 3869. 44. Deek, J., P. J. Chung, J. Kayser, A. R. Bausch, and C. R. Safinya, 2013. Neurofilament sidearms modulate parallel and crossed-filament orientations inducing nematic to isotropic and re-entrant birefringent hydrogels. Nat. Commun. 4:2224. 45. Storm, I. M., M. Kornreich, A. Hernandez-Garcia, I. K. Voets, R. Beck, M. A. Cohen Stuart, F. A. M. Leermakers, and R. de Vries, 2015. Liquid Crystals of Self-Assembled DNA Bottlebrushes. J. Phys. Chem. B 119:4084 -- 4092. 46. Toman, E., S. Harrisson, and T. Belli, 2016. Biomarkers in J. R. Army Med. Corps. traumatic brain injury: a review. 162:103 -- 108. 531:397 -- 401. 17. Hisanaga, S., and N. Hirokawa, 1989. The effects of dephos- phorylation on the structure of the projections of neurofila- ment. The J. Neurosci. 9:959 -- 966. 18. Gou, J. P., T. Gotow, P. A. Janmey, and J. F. Leterrier, 1998. Regulation of neurofilament interactionsin vitro by natural and synthetic polypeptides sharing Lys-Ser-Pro sequences with the heavy neurofilament subunit NF-H: Neurofilament crossbridg- ing by antiparallel sidearm overlapping. Med. Biol. Eng. Comput. 36:371 -- 387. 19. Leermakers, F. A. M., and E. B. Zhulina, 2008. Self-consistent field modeling of the neurofilament network. Biophys. Rev. Lett. 3:459 -- 489. 20. Jeong, S., X. Zhou, E. B. Zhulina, and Y. Jho, 2016. Monte Isr. J. Chem. Carlo Simulation of the Neurofilament Brush. 56:599 -- 606. 21. Eyer, J., and J. F. Leterrier, 1988. Influence of the phosphoryla- tion state of neurofilament proteins on the interactions between purified filaments in vitro. Biochem. J. 252:655 -- 660. 22. UniProt-Consortium, 2015. UniProt: a hub for protein infor- mation. Nucleic Acids Res. 43:D204 -- D212. 23. Hancock, J. M., and M. J. Bishop, 2004. EMBOSS (The Euro- pean Molecular Biology Open Software Suite). In Dictionary of Bioinformatics and Computational Biology, John Wiley & Sons, Inc., Hoboken, NJ, USA. 24. Harpaz, Y., M. Gerstein, and C. Chothia, 1994. Volume changes on protein folding. Structure 2:641 -- 649. 25. Jones, J. B., and C. R. Safinya, 2008. Interplay between Liquid Crystalline and Isotropic Gels in Self-Assembled Neurofila- ment Networks. Biophys. J. 95:823 -- 835. 26. Kas, J., 1996. Mechanical Effects of Neurofilament Cross- bridges. Modulation by phosphorylation, lipids, and interac- tions with f-actin. J. Biol. Chem. 271:15687 -- 15694. 27. Carden, M. J., W. Schlaepfer, and V. Lee, 1985. The structure, biochemical properties, and immunogenicity of neurofilament peripheral regions are determined by phosphorylation state. J. Biol. Chem. 260:9805 -- 9817. 28. Pant, H. C., 1988. Dephosphorylation of neurofilament pro- teins enhances their susceptibility to degradation by calpain. Biochem. J. 256:665 -- 668. 29. Kornreich, M., E. Malka-Gibor, A. Laser-Azogui, O. Doron, H. Herrmann, and R. Beck, 2015. Composite bottlebrush mechanics: α-internexin fine-tunes neurofilament network properties. Soft Matter 11:5839 -- 5849. 30. Parsegian, V. A., R. P. Rand, N. L. Fuller, and D. C. Rau, 1986. Osmotic stress for the direct measurement of intermolecular forces. Methods Enzymol. 127:400 -- 416. 31. Mucke, N., L. Kreplak, R. Kirmse, T. Wedig, H. Herrmann, U. Aebi, and J. Langowski, 2004. Assessing the Flexibility of Intermediate Filaments by Atomic Force Microscopy. J. Mol. Biol. 335:1241 -- 1250. 32. Beck, R., J. Deek, J. B. Jones, and C. R. Safinya, 2010. Gel- expanded to gel-condensed transition in neurofilament net- works revealed by direct force measurements. Nat. Mater. 9:40 -- 46. 33. Li, Y., R. Beck, T. Huang, M. C. Choi, and M. Divinagracia, 2008. Scatterless hybrid metal-single-crystal slit for small- angle X-ray scattering and high-resolution X-ray diffraction. J. Appl. Crystal. 41:1134 -- 1139.
1807.05100	1	1807	2018-07-13T14:17:54	The physics of large-scale food crises	[ "physics.bio-ph", "physics.soc-ph" ]	Investigating the ``physics'' of food crises consists in identifying features which are common to all large-scale food crises. One element which stands out is the fact that during a food crisis there is not only a surge in deaths but also a correlative temporary decline in conceptions and subsequent births. As a matter of fact, birth reduction may even start several months before the death surge and can therefore serve as an early warning signal of an impending crisis. This scenario is studied in three cases of large-scale food crises. Finland (1868), India (1867--1907), China (1960--1961). It turns out that between the regional amplitudes of death spikes and birth troughs there is a power law relationship. This confirms what was already observed for the epidemic of 1918 in the United States (Richmond et al. 2018b). In a second part of the paper we explain how this relationship can be used for the investigation of mass-mortality episodes in cases where direct death data are either uncertain or nonexistent.	physics.bio-ph	physics	The physics of large-scale food crises Peter Richmond1, Bertrand M. Roehner2 and Qing-hai Wang3 Abstract Investigating the "physics" of food crises consists in identifying features which are common to all large-scale food crises. One element which stands out is the fact that during a food crisis there is not only a surge in deaths but also a correlative tempo- rary decline in conceptions and subsequent births. As a matter of fact, birth reduction may even start several months before the death surge and can therefore serve as an early warning signal of an impending crisis. This scenario is studied in three cases of large-scale food crises. Finland (1868), India (1867–1907), China (1960–1961). It turns out that between the regional am- plitudes of death spikes and birth troughs there is a power law relationship. This confirms what was already observed for the epidemic of 1918 in the United States (Richmond et al. 2018b). In a second part of the paper we explain how this relationship can be used for the investigation of mass-mortality episodes in cases where direct death data are either uncertain or nonexistent. Key-words: Famine, death rate, birth rate, excess-death. Bertillon effect Version of 8 July 2018 1: School of Physics, Trinity College Dublin, Ireland. Email: peter [email protected] 2: Institute for Theoretical and High Energy Physics (LPTHE), University Pierre and Marie Curie (Sorbonne), Paris, France and CNRS (National Center for Scientific Research). Email: [email protected]. 3: Physics Department, National University of Singapore. Email: [email protected] 8 1 0 2 l u J 3 1 ] h p - o i b . s c i s y h p [ 1 v 0 0 1 5 0 . 7 0 8 1 : v i X r a 2 Introduction A plea for comparative analysis The title of the present paper1 was inspired by a book published seven years ago (Viswanathan et al. 2011) under the title: "The physics of foraging". From foraging, i.e. the collection of food by animals, to food crises there is of course a smooth tran- sition. However, in this title it is the mention of the word "physics" which was for us the most important source of inspiration for it means that the purpose of the book was to examine basic features of foraging that are shared by many species. To find common mechanisms in seemingly different phenomena has been a permanent ob- jective of physics throughout its development over past centuries. Is there a common factor in the fall of apples, rain drops, meteorites and the "fall" of the Moon toward the Earth? We now know that the common factor is the gravitational attraction. Here, we will be guided by a similar agenda. It can be obsereved that such a comparative approach was quite common in the late 19th century; see for instance the works of Louis-Alphonse Bertillon (1872), Alfred Espinas (1878), Jacques Bertillon (1892), Emile Durkheim (1895). One may deplore that in recent decades comparative studies of this kind (among which we do not include meta-analyses which is something different) have become rare. The Bertillon effect: from heat-wave mortality to large-scale food crises Here we will focus on the Bertillon effect (Bertillon 1892, Richmond et al. 2018a,b) which consists in the fact that any mass mortality is followed 9 months later by a birth rate trough2. Our previous parallel with gravitation becomes particularly relevant here for indeed, just like gravity, the Bertillon effect has a broad range of applicability. There is a long chain of cases which goes from heat-waves, to epidemics, to earthquakes, to large- scale food crises. As in any chain, it is of particular interest to consider more closely its two extremities. Heat-waves in developed countries result in excess mortality3 of the order of 24% (Rey et al. 2007, p. 536, Table 1) whereas in the food crises that we are going to study mortality rates were increased by up to 140%. The successful identification of birth rate troughs in the wake of heat-waves required a skillful and 1The paper was written by three physicists. However, we feel that our approach is very much in the Inciden- the same comparative approach was also used in our previously published papers in biodemography spirit of the methodological guidelines defined by the French sociologist Emile Durkheim (1898). tally, (http://www.lpthe.jussieu.fr/∼roehner/biodemo.html.) 2Note that the birth rate trough is much larger (usually about 10 times larger) than the reduction in births due to the fatalities. This was shown in detail in Richmond et al. (2018a) and can also be seen from the simple fact that excess fatalities would result in a one sided Heaviside fall not in a symmetrical and fairly narrow trough. This trough is mostly due to a temporary reduction in the conception rate among the survivors. 3In the sense of: [(observed deaths)-(deaths in normal years)]/(deaths in normal years). 3 pioneering analysis of daily birth data by Arnaud R´egnier-Loilier (2010 a,b). The reason why heat waves display only a faint Bertillon effect is not only due to low excess mortality but also to it being concentrated in elderly people which makes it irrelevant for birth rates4. Why did we say that heat-wave cases constitute the beginning of the chain? One could of course consider diseases (e.g. the Lyme disease) which have even lower mortality. However, as the birth rate troughs of heat-waves are already at the limit of what can be observed, the Bertillon effect for Lyme disease would be inobservable. It would be like a pulsar which is known to exist but is too far away to be seen with an optical telescope. In the cases considered in the present study the Bertillon effect is so massive that it can be identified (and studied even at regional level) with annual vital statistics. From well documented cases to uncertain situations Why did we choose to focus on the three cases selected? The main reason is that these cases are massive and statistically fairly well documented. In contrast, there are many cases of mass mortality for which there is considerable uncertainty. Thanks to the death-birth relationship (to be stated below) one is in good position to throw new light on such episodes. How? Most countries, even those which do not have a reliable statistical registration infras- tructure, conduct censuses. Because this is done periodically (e.g. once in a decade) it does not require a permanent organization. As will be shown below, the distribu- tion of the population by age measured in a census gives a fairly accurate picture of birth rate changes in the years preceding the census. For instance, the census of 1982 in China gives a good picture of the birth rate squeeze that occurred in sev- eral provinces in 1961. Then, through the relationship between birth troughs and death spikes one can get an estimate of the mortality. Even though not perfect, such estimates give at least a rough picture of what happened. Outline Our investigation will proceed through the following steps. (1) First, we present the three famine cases which will be studied. Apart from giving some social and historical background information, we will discuss the origin, reliability and accuracy of the birth and death data. (2) Then, in each case we describe the Bertillon death-birth connection. (3) In order to get a comparative view we bring together the three Bertillon rela- tionships and we compare them with the results already obtained in Richmond et al. 4Even for persons in age of having children the birth rate trough is only marginally due to those who die. Most of the effect comes from the much larger number of persons who are affected but do not die. 4 (2018b). (4) We show how to retrieve past annual birth rates from a population pyramid. (5) Finally, in our conclusion, we develop two examples in which mortality rates are derived from census data. PART 1: QUANTIFICATION OF THE BERTILLON EFFECT Background information for the crises Causes of death First of all, we should explain why we prefer to use the expression "food crises" rather than "famines". The word "famine" elicits images of people starving to death. Although, this may of course exist as documented by impressive Internet pictures of children almost reduced to their bones5, death by starvation represents only a small fraction of the global death toll of food crises. This can be illustrated by data from Finish and Indian sources. • The data given in Finland 1 (p. XXXIV) and Finland 2 (p. 421-422) tell us that during the crisis of 1868 death by starvation represented only 1.71% of the deaths. The main cause of death was typhus (43%), followed by tuberculosis (5.84%), dysen- tery (5.70%) and smallpox (3.02%). • Data for the food crisis of 1942–1944 in India are given in Maharatna (1992, p. 331). Death by starvation represented 2.1% of the deaths. The main cause of death was fever due to diseases (mainly malaria) which accounted for 34.8% followed by scabies (18.4%) and dysentery (10.9%). In short, at symptoms level, food scarcity crises have a close resemblance with epi- demics. Crisis of 1868 in Finland Immediate causes such as a rainy and cold summers in 1866 and 1867 can be men- tioned but in order to get a real understanding the crisis of 1868 should not be seen as an isolated event. In fact, there had been similar crises in 1833 and 1856. albeit of smaller magnitude (Flora et al. 1987, p, 24,51). Whereas in the 1840s the average annual death rate was around 25 0/00 it reached 46 0/00 in 1833, 34 0/00 in 1856 and 78 0/00 in 1868. The main factor in this succession of crises was certainly the rapid growth of the population. Between 1811 and 1865 it increased by 66% which is twice the 30% 5For instance the persons shown on the cover of Davis (2000). Needless to say, the inclusion of pictures which have such an emotional content also reveal something about the agenda of the author. 5 increase in France in the same time interval (Flora et al. 1987, p. 55-56). This 66% increase represented an average annual increase of 1.20% (against 0.55% in France). Certainly agricultural production did not increase at the same rate which means that any bad harvest due to adverse weather conditions would result in malnutrition or a more serious crisis. The annual death rate of 78 0/00 reached in 1868 in Finland was one of the highest ever observed anywhere. The severity of a food scarcity is best judged by the death rate peak for this is a fairly intrinsic metric. In contrast excess-deaths are very dependent on the level of the baseline death rate chosen to define normal conditions. It is true that at the level of Indian provinces the death rate reached similar values: for instance in 1900 it reached 88 in the "Central Provinces". However, if nationwide data were available they would certainly show lower peak values for in such a large land as India the crises were not completely synchronous. Food crises in India Between 1860 and 1910 there were recurrent food crises in India but most of them were limited to some parts of the country. A brief description can be found in Roehner (1995, p.5-6). Can one explain the famines in India in the same way as the famines in Finland that is to say by a rapid increase of the population. The answer is no. Assuming that the population estimate given for 1820, namely 209 millions is indeed reliable, one finds for the period 1820–1865 an average annual increase rate of only 0.27%. However, if the population stagnated the amount of food available in India may have decreased. One can mention three reasons for a shrinking food supply (Davis 2000, p. 59-66). • Between 1875 and 1900 Indian annual grain exports to Great Britain increased from 3 to 10 million tons, equivalent to the annual nutrition of 25 million people. Davis does not give trade data for the period before 1875. However, it is known that as a result of the repeal of the corn tariff around 1850 Britain's dependence on imported grain increased from 2% in the the 1830s, to 24% in the 1860s, and 65% (for wheat) in the 1880s. Thus, it is quite possible that the drain on Indian grain started before 1875. • In Berar (central India) the acreage of cotton doubled between 1875 and 1900. • Although a colony, with regard to its budget India was treated like an indepen- dent country. This meant reimbursing the stockholders of the "East India Company", paying the costs of the 1857 revolt, supporting an army of about 100,000 that was employed not only in India but also in foreign war theaters (e.g. Afghanistan, Tibet, Egypt, Ethiopia, Sudan). Military expenditures alone represented about one third of 6 the budget. As a result, little funds were available for agricultural improvement. Having said that, in India just as in Finland and in China, the immediate causes of the crises were unfavorable weather conditions, particularly drought which in Madras and Bombay provinces resulted from a reduced monsoon rain season. Food crisis in China, 1960–1961 The explanation given for Finland also applies to China and in fact to even greater degree. From 1949 to 1959 the Chinese population increased from 541 to 672 mil- lions, an increase of 24.2% which represented an average annual increase of 2.20%. This is 4 times faster than the French rate of 0.55% that we took as a yardstick in discussing the case of Finland. It is true that between 1953 and 1957 population and foodgrain production increased at the same rate of 2.0% annually: the population from 588 to 646 millions and the foodgrain production from 168 to 185 million tons. However, after a bumper crop of 212 million tons in 1958 it declined due to bad weather conditions to the point of being reduced in 1960 to its level of 1957 at which time there had been about 50 million fewer Chinese to feed (SNIE 1961, p.3). At the same time the death rate fell from 20 0/00 to about 10 0/00. It is the fall in the death rate which brought about the rapid population increase for the birth rate remained stable at around 40 0/00. Fig. 1 shows that even at its peak value in 1960 the death rate reaches only 25 0/00 (Ren kou 1988, p. 268) which is lower than the death rate in Bengal under normal conditions and only slightly higher than the death rate in 1949. In short, instead of people starving to death, one should rather think of the situation as similar to the one of 19496. It is true that the situation may have been more tragic in some specific provinces as for instance Anhui or Sichuan. However, this was not really new for, as we will see below, the best predictor for regional death rates in 1960 was the situation one decade earlier. Comparative graph Fig. 1 shows the changes in death rates in India and China. For the sake of clarity we did not represent the curve for Finland on the same graph; however, it can be noted that the baseline death rate before the Great Famine of 1868 was around 28 0/00 and that the death rate peak reached 78 0/00. In a general way, for a given peak value, the lower the baseline rate, the higher the excess mortality with respect to this baseline. The fact that for China the baseline level was 2 or 3 times lower than in India is the main reason for a substantially higher excess death number. In Fig. 1 it may appear surprising that in 1931–1940 the death rate in Bengal was about 1.5 times higher than the rate for 1874–875 in Madras. The most likely ex- 6According to the SNIE (1961. p.1) report discussed in Appendix A: "Widespread famine does not appear to be at hand but in some provinces people are now on a bare subsistence diet." ) n o i t a l u p o p , f o 0 0 0 1 r e p ( e t a r h t a e D 1874 1876 1878 1880 1882 1884 1886 1888 1890 1931 1933 1935 1937 1939 1941 1943 1945 7 50 10 5 Bengal 1941-1946 India 1874-1891 China 1950-1985 1950 1955 1960 1965 1970 1975 1980 1985 60 50 40 30 20 10 9 8 7 6 5 Fig. 1 Comparison of famines in India and China. The vertical scale is the same for the 3 curves. As there are no nationwide data for India the curve corresponds to data for the province of Madras in the south east of India. Sources: Maharatna (1992, p.55), Ren kou (1988, p.278). planation is under-reporting. The registration of deaths started around 1870 and, as is often observed, the scope and completeness of the registration increased progres- sively in the course of following decades. In the 1940s, the corrective factors used by various authors still ranged from 1.32 to 1.70 (Maharatma p. 228). This suggests that in the 1870s under-reporting may have been more serious, may be by a factor 2. The Bertillon death-birth relationship In each of the previous cases death and birth data are available not only at national level but also at provincial level. This will allow us to carry out a regression analysis in the same spirit as in Richmond et al. (2018b). Here, however we will have to work with yearly (not monthly) data. It is to maintain accuracy that we limited ourselves to massive events. As in many countries only annual data are available, it is important to 8 see whether or not in such cases the Bertillon effect can be analyzed in a meaningful way. Bertillon effect in Finland: food crisis of 1868 Fig. 2 shows the Bertillon effect for the 8 provinces which composed Finland in the 1860s. ) 8 6 8 1 ( s h g u o r t h t r i b f o e d u t i l p m A 2 1 4 2 7 3 6 2 8 1 5 3 4 Amplitude of death peaks (1868) Fig. 2 Finland, 1868: relationship between the amplitudes of death spikes and birth troughs. As the graph is a log-log plot it means that: Ab = CAα d . where α = 0.50 ± 0.26 and C = 0.89 ± 0.06 (the error bars are for a confidence level of 0.95); the coefficient of linear correlation is 0.83. Each number corresponds to one of the 8 governments (i.e. provinces) which composed Finland. Their names are as follows: 1=Uudenmaan (S), 2=Turun ja Porin (SW), 3=Hameenlinnan (S),4=Wiipurin (SE), 5=Mikkelin (SE), 6=Kuopion (E), 7=Waasan (SW), 8=Oulun (N). Sources: Finland (1871, p.34,XII), Finland (1902, p.274,397) As monthly data are available for the whole country, we know that the death rate peak occurred in May 1868; therefore, as expected, the center of the birth trough will be in December 1868–January 1869. In other words the birth trough will be split in two parts, one part in late 1868 and the other in early 1869. However, because of the rebound effect7 the annual birth number for 1869 is higher than the one for 1868. That is why in Fig. 2 the vertical scale displays births of 1868 and not 1869. Birth troughs seen as a sensitive detector of population suffering At first sight it might seem that the absence of a death spike, i.e. Ad = 1 means that 7The rebound effect occurs in the months immediately following the trough; it is a compensating rise of the birth rate above normal baseline level (see Maharatna 1992, p. 380 and Richmond et al. 2018a). In other words it is a return to equilibrium marked by an "overshooting" episode. 9 the situation is normal and should therefore be associated with Ab = 1. However, the fact that there are no excess-deaths does not mean that the population does not suffer. A simple illustration is a non fatal disease which nevertheless makes people ill. The 2003 SARS epidemic in Hong Kong discussed in Richmond et al. (2018a) was a situation of this kind. Although not strictly equal to zero, the death toll was only 40 per million. Nevertheless, the threat and disorganization due to the epidemic produced an excess birth trough of 6% (Ab = 1.06). In the same line of thought it will be seen in subsequent cases that usually the birth rate starts to fall in an early stage at a moment when no increase of the death rate can be detected. In other words, birth troughs are a detector of population suffering that is more sensitive than Ad. Bertillon effect in India: time series of food crisis Here "India" refers to the British colony before its division into Bangladesh, India and Pakistan. As already observed there are no nation-wide data for India in spite of the fact that some food crises extended to several parts of the country. Instead one has data sep- arately for several provinces: Bombay, Central Provinces, Madras, Punjab, United Provinces. Fig.3a,b displays the basic mechanism of food crises: as the death toll increases the birth rate decreases. One may wonder why here, in contrast to Finland, the birth trough occurs in the year following the death spike. Monthly data that are available in some cases (Dyson 1991a,b, 1993 and Maharatna 1992, p. 235) show that usually the maximum of the deaths occurred in August at the beginning of the wet season. The reason for that comes from the fact that the main cause of mortality is malaria whose spread is favored by humidity. With deaths spiking in August instead of May as was the case in Finland the effects of reduced conception will be mostly visible in the following year. Fig. 3a and 3b show that the analysis performed in Richmond et al. (2018a) on monthly data can be repeated in a similar way with annual data. Bertillon effect in India: global death-birth relationship Next, we wish to discover the relationship existing between the amplitudes of death spikes and birth troughs of different crises. This is summarized in Fig. 4. There is again a power-law relationship. Bertillon effect in China: time series of the food crisis of 1960-1961 Of the three countries considered in this paper it is for the case of China that we have the most complete data set. Annual data are given for over 20 provinces from 1955 to 1985. In what follows we will focus on a 10-year interval centered on the crisis of 1960-1961. Bombay, death Bombay, birth 10 ) d n a s u o h t r e p ( e t a r h t r i b / h t a e d l a u n n A 70 60 50 40 30 1895 1896 1897 1898 1899 1900 1901 1902 1903 ) d n a s u o h t r e p ( e t a r h t a e d l a u n n A 80 70 60 50 40 30 20 Bombay, death Bombay, birth (left-shifted by 1 year + inverted) 1895 1896 1897 1898 1899 1900 1901 1902 1903 -20 -25 -30 -35 -40 -45 Fig. 3a,b Bombay (1895-1903). There are two food crises in the time interval 1895-1903: first one of small magnitude in 1896-1897 and then the more serious crisis of 1899-1900. The right-hand graph (where the birth curve was shifted and inverted) makes manifest that both death spikes gave rise to a birth trough in the following year. Source: Maharatna (1992, p.55) s h g u o r t h t r i b f o e d u t i l p m A 1.7 1.6 1.5 1.4 1.3 1.2 1.1 1 1 1 8 3 5 6 9 4 7 2 2 3 Amplitude of death spikes Fig. 4 India (1877-1908). Relationship between the amplitudes of death spikes and birth troughs. The re- lationship is as follows: Ab = CAα d . where α = 0.30±0.16 and C = 1.01±0.06. The coefficient of linear cor- relation is: 0.81. The correspondence between the numbers and the provinces is as follows: 1=Madras (1877), 2=Bombay (1877), 3=Bombay (1900), 4=Punjab (1900), 5=Central Provinces (1897), 6=United Provinces (1908), 7=Bombay (1897), 8=Central Provinces (1897), 9=Berar (1897). Source: Maharatna (1992, p.55). Fig. 4 shows typical graphs for 4 different provinces . In a general way, the crisis was more serious in South China than in North China and in the south the severity increased from east to west. Apart from Beijing, the three other selected provinces are all located at the same latitude (about 500 km south of Shanghai): Fujian is on the Pacific coast, Hunan in central China and Sichuan about 700km to the west of Hunan. The graphs show very clearly two characteristics of the Bertillon effect. 11 Fujian 1956 1958 1960 1962 1964 Sichuan Beijing 1956 1958 1960 1962 1964 Hunan ) 0 0 0 , 1 r e p ( e t a r h t r i b / h t a e D ) 0 0 0 , 1 r e p ( e t a r h t r i b / h t a e D 60 50 40 30 20 10 9 8 7 60 50 40 30 20 10 9 8 7 60 50 40 30 20 10 9 8 7 60 50 40 30 20 10 9 8 7 1956 1958 1960 1962 1964 1956 1958 1960 1962 1964 Fig. 5 Birth and death rates in 4 Chinese provinces. The curve of deaths is in black with filled squares; the curve of birth is in red with filled circles. The graphs are ranked by order of increasing severity. It can be seen that the fall of the birth rates in 1957, 1958 provide so to say early warnings of an impeding crisis. The fact that the minima of the birth troughs occur in 1861 shows that the mortality rates certainly peaked in the second half of 1960. Source: Ren kou (1988) 12 • The one year time lag between death peaks and birth troughs • The birth rate rebounds in the two or three years following the troughs. In the next subsection, in order to find out the relationship between the amplitudes of death peaks and birth troughs, we extend this analysis to 22 provinces Bertillon effect in China: global death-birth relationship Fig. 6 shows the relationship between death peaks and birth troughs. As in Finland and India it can be described by a power law. The exponent is about the same as in Finland. 3 2 s h g u o r t h t r i b f o e d u t i l p m A ZHE HEB SHN SHA YUN JIA BEI MON SIC ANH HUN HEN LIA HUB GUI GUZ FUJ JIA SHG GUA JIL HEI 1 1 2 3 4 Amplitude of death peaks Fig. 6 China 1959-1961. Relationship between the amplitudes of death spikes and birth troughs. The relationship between the amplitudes of the death and birth rates is as follows: Ab = CAα d . where α = 0.42 ± 0.16 and C = 1.38 ± 0.06. The coefficient of linear correlation is: 0.76. The labels are the first three letters of the names of the provinces except for the following whose first three letters would have been identical: SHA=Shanxi, SHN=Shanghai, SHG=Shandong, GUA=Guangdong, GUI=Guanxi, GUZ=Guizhou. Source: Ren kou (1988) Structural fragility One may wonder what made the crisis more serious in some provinces than in others. We have already mentioned that globally the north was less affected than the south and the east less than the west. However, besides these fairly loose observations there is another which is both simpler and more revealing. It consists in the fact that the provinces which have the highest death rate peaks in 1960 are those which already had the highest death rate in the normal years before the crisis. This is illustrated graphically in Fig. 7. 13 SIC GUZ ANH HEN YUN HUN GUI HUB SHG JIA GUA JIA HEBSHA ) d n a s u o h t r e p ( 0 6 9 1 n i e t a r h t a e D 60 50 40 30 20 10 9 8 7 6 FUJ ZHE JIL MON BEI LIA HEI SHN 6 7 8 10 20 9 Death rate in 1956 (per thousand) Fig. 7 Relationship between the death rates in 1956 and in 1960 (food crisis). The coefficient of linear correlation is: 0.81. The labels are the same as in Fig.5. Source: Ren kou (1988) Is there a similar connection between normal and peak death rates in Finland and in India? It is impossible to say because in those cases the number of separate provinces for which data are available is too small. What interpretation can one give of the effect shown in Fig. 6? The higher death rates in normal times suggest structural fragility factors. To get better insight it is useful to compare extreme cases. On one end we consider the northern provinces of Heilongjiang and Jilin where the crisis was subdued while on the other end we consider Anhui and Sichuan where it was very severe. In Table 1 we computed the population growth between 1954 and 1959. A much faster growth in Anhui-Sichuan would suggest that, as was indeed the case for the whole country, the increase of food supply could not match the population growth. However, the data in Table 1 do not show a faster population increase in Anhui- 14 Sichuan. As a matter of fact, had it existed such a faster growth would not explain that the death rate in Anhui-Sichuan was already higher in 1956. This fact rather suggests that there was a structural fragility that was already present in 1956 but had an aggravated effect in 1960. Table 1: Regional fragility Population Population Percent increase Population Death density 1954 (per sq-km) rate 1954 (per 1,000) Death rate 1960 (per 1,000) 1954 (106) 11.7 12.7 31.7 64.4 1959 (106) 13.2 16.9 34.4 73.7 Jilin Heilongjiang Anhui Sichuan 13% 33% 8.5% 14% 60 37 224 134 10.3 10.5 16.4 15.4 10.1 10.5 50.2 47.8 Notes: Jilin and Heilongjiang were little affected by the crisis whereas Anhui and Sichuan were the two provinces where the crisis was the most severe. It seems that population density is the main factor which can explain this difference; in Anhui-Sichuan it was on average 3.7 times larger than in Jilin-Heilongjiang. This created a permanent fragility and sensitivity to weather conditions which is revealed by the high death rates even before the crisis. Source: Ren kou (1988) The data in Table 1 show that the population density was far higher in Anhui-Sichuan than in Jilin-Heilongjiang. The difference is all the more striking because in contrast to Jilin-Heilongjiang Sichuan had at that time almost no industry. Another contributing factor was the regional prevalence differential of infectious dis- eases. It is well known that malnutrition reduces the ability of the organism to fight infection. For tuberculosis in 1990 the prevalence was 147 per 100,000 in the East, 198 in the Center and 216 in the west (China Tuberculosis Control Collaboration 2004, p.421) We will not develop this analysis further for it would lead us too far away from the main purpose of our paper which to focus on general rules. In the next subsection we compare the regression lines in Finland, India and China. Comparison of the regression parameters When the three regression lines are drawn on the same graph one can compare their slopes (given by the exponent α) and their levels (given by the coefficient C). The level is given by the birth trough amplitude for a given death peak amplitude. One should realize that this depend very much upon how the birth deficit is divided be- tween the two years. Thus, for a death peak in May (as in Finland) the birth deficit will be divided almost equally between 1868 and 1869. On the contrary, if deaths 15 had peaked in November 1868 then the whole birth deficit would occur in 1869. In other words, to make this comparison significant one would need monthly data. With respect to the slopes we observe that they are fairly similar for Finland and China but 50% smaller for India. Regarding the value of α in India one can observe that in this country the data cover several food crises in successive years whereas in Finland and China they cover a single crisis. However, it is not clear why this should lead to a lower α. For the influenza epidemic of October 1918 in the United States the slope was (Rich- mond et al. 2018b): α = 0.19 ± 0.1 and C = 1.28 ± 0.22. Here, a lower value of α makes sense. To make the argument simpler, let us consider as an approximation that C = 1; this means that in all cases the regression line starts from the point (1, 1). Then the question becomes: "For a given death peak how many people will suffer to the point of reducing conceptions?" It makes sense to observe that during a food crisis more people do suffer, and suffer more severely, than during an influenza epidemic. An obvious reason is that the influenza epidemic was much shorter than any of the food-crises that we considered. It lasted only about one month, from 15 October to 15 November 1918 whereas the food crises lasted more than one year. In the last section before the conclusion we will show how the death-birth relation- ship can be used as an exploration tool for historical mass mortality events. PART 2: MASS MORTALITY EPISODES EXPLORED THROUGH THE BERTILLON EFFECT Why are mortality statistics often uncertain? The production of vital statistics comprises two fairly different tasks. • In order to register individual births and deaths one needs a network extending to hospitals and doctors of the whole country. This is a challenging task as can be seen from the fact that in a vast country like the United States it took several decades to extend the registration network to all states. The task was completed only around 1930. Even once established, registration networks may be overwhelmed during mass mortality episodes. • The second task is the organization of censuses. Although by no means an easy task, the organization of censuses does not require a permanent network extending to the whole country. In a time span of 3 or 4 months the same team of census officers can move from region to region until the whole country has been covered. The fact that the organization of a census is a much less demanding task than daily registration is demonstrated by the observation that the first US census took place 16 about one century before the national registration network was completed. How can one derive annual birth numbers from census records? In principle, a census does not give any information about births or deaths but it gives the age of all people and from these data one can derive approximate birth data. This is illustrated in Fig. 8a,b. ) s r a e y ( e g A 60 50 40 30 20 10 0 Anhui 65 64 63 62 61 60 59 58 57 56 55 54 53 52 51 50 49 48 47 46 45 44 43 42 41 40 39 38 37 36 35 34 33 32 31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 9 8 7 6 5 4 3 2 1 0 h t r i b f o r a e Y 1922 1932 1942 1952 1962 1972 1982 ) n o i t a l u p o p 0 0 0 1 , r e p ( e t a r h t r i B 70 60 50 40 30 20 10 9 Annual birth rates (registration) Annual birth rates derived from census data 1955 1960 1965 1970 1975 1980 Fig. 8a,b Derivation of birth data from a population pyramid. (a) The huge indentation is due to the birth deficit in 1960-1961. The population pyramid of Anhui province was built with the data of the census of 1982. Each step represents an age group born in a single year. Males are on the left and females on the right. The birth rate trough corresponds to people who were 1982 − 1961 = 21 year old at the moment of the census. People older than 21 were born before 1961 and people younger than 21 were born after 1961. In this way one can reconstitute approximate birth data for the whole period from 1955 to 1982. (b) Graph 7b is for males; its horizontal axis shows the years of birth as derived from the ages recorded in the census of 1982. It shows a fairly good agreement between the real birth data and the approximate birth data derived from the population pyramid. Sources: Population pyramid: IPUMS. Birth data: Ren kou (1988). In what sense are the data derived from the population pyramid approximate birth numbers? In principle the size b′(1982) of the age group 0 − 1 of the population pyramid should be equal to the number b(1982) of births in 1982. The fact that in graph 7b the two points are not identical is because we have been using a 1% sample of the census of 1982. This gives an idea of the statistical fluctuations. More generally, the individuals aged x in 1982 were born in yb = 1982 − x. For instance, the individuals born in yb = 1970 will be 12 year old in 1982. Naturally, during these 12 years some of these children died or moved from Anhui to another place (whether in China or abroad); conversely some children may have moved into Anhui. If these population movements are important the size of the 12-year age group will have little to do with the number of births in 1970. On the contrary, if 17 there were few population movements a comparison of the numbers b′(x) and b(x) will tell us how well the registration system was working. Generally speaking, the registration of births is more reliable than the registration of deaths for at least two reasons. • When funeral services are overwhelmed by the number of deceased people it may happen that they are buried by relatives or neighbors without being registered, especially in times of epidemics. On the contrary, as there are no birth rate spikes the birth registration services will not get overwhelmed. In addition the registration of newborns does not need to be done immediately after birth, it can be done at any time. • Politically, dead people are usually a more sensitive matter than newborns. The authorities may wish to publish under-estimated death numbers8. Preliminary test of the methodology on influenza deaths in Pennsylvania First of all, before using it as an exploration tool, we wish to test the accuracy of the methodology on a case in which the death toll is known. Why did we select for our test the influenza epidemic in Pennsylvania? The impact of the disease was particularly severe in Pennsylvania. The amplitude of the death peak of 1918 (with respect to the average of the 1917 and 1919 numbers) was 1.56 in Pennsylvania but only 1.35 for the whole country (or more precisely for all death registration states). The challenge now is to see if we can derive the death peak amplitude solely from the age-group data given by the census of 1930. For that purpose we need to go through the following steps. (1) First, as was already done in the previous subsection, we derive from the age- groups given by the census of 1930 proxy birth numbers. To distinguish these proxy birth numbers from the real birth numbers (given by the registration network) they will be called census birth numbers. This step is done in Fig. 9a,b. In contrast with Fig. 8a,b here there is no huge trough. Before doing the test we could not know whether or not the small trough of 1918 would be covered by the background noise. Fortunately, it turns out that it can be identified. Its amplitude is 1.09. (2) The second step is to use the death-birth relationship to derive the amplitude of the death peak. As the level of the regression line is not well defined (as was explained eralier it depends upon the month of the death spike) we take again C = 1 for the multiplicative constant. Thus, using the value of α given previously, one gets: Ab = A0.19 d ⇒ Ad = A1/0.19 b = 1.095.26 = 1.57 8For instance, according to "R´esum´e r´etrospectif (1907, p. 368), the death rate in Ireland in the period 1864–1870 was 16.6 0/00 which was the lowest rate among all European countries. It was lower than in England and Wales (22.5), Sweden (20.2), Denmark (19.9) or Prussia (27.0). Most likely such a small rate resulted from under-reporting. 18 ) s r a e y ( e g A 16 14 12 10 8 6 4 2 0 h t r i b f o r a e Y Pennsylvania 15 14 13 12 11 10 9 8 7 6 5 4 3 2 1 0 ) n o i t a l u p o p 0 0 0 1 , r e p ( e t a r h t r i b s u s n e c . s v e t a r h t r i b l a e R 1920 1930 Real births Births based on 1930 census Birth trough due to influenza epidemic 30 28 26 24 22 20 18 1914 1916 1918 1920 1922 1924 1926 1928 1930 Fig. 9a,b Derivation of birth data from the population pyramid of Pennsylvania. (a) The population pyra- mid was truncated to the ages that are required for our exploration of the birth trough of 1919 (indicated by arrows). (b) In 1930 the two data points should be identical. The small difference seen in the graph is due to the statistical fluctuations resulting from the fact that the population pyramid was derived from a 5% sample of the census of 1930. As one moves back in time it is of course natural that the distance between the two curves increases because the people born in these years appear in 1930 as older age-groups. Sources: Population pyramid: IPUMS, Birth data: Linder et al. 1947, p.666-667. (3) Now, in order to derive the death rate in 1918 from the amplitude of the death peak, we need an estimate for the death rate in normal times. Although in some cases this death rate may be known, most often it is not. In the two cases considered below the normal death rates are not known for the simple reason that there was no registration network. In such a cas one takes the death rate in a region that is similar in terms of socio-economic conditions. For instance one may take the death rate in 1915 in Sweden which is 14.7 0/00 (Flora et al. 1987, p.73). Here, as the death rate in Pennsylvania is in fact well known (for in 1915 it was already a death registration state) we can check whether the Swedish proxy is reasonable. It is indeed for in Pennsylvania in 1915 the death rate was 13.2 0/00. Thus, the predicted death rate in 1918 becomes: µ = 1.57 × 14.7 = 23.1. Compared with the real value of 18.1 0/00 there is a difference of 28%. This could seem high, but one should remember that in such matters the estimates by various authors often differ by 100% or more. In the first case considered below, no estimate whatsoever was available and in the second case the official estimate may have been highly exagerated. Georgia during the American Civil War As an illustration of the kind of situation for which the birth-death method may be useful consider the state of Georgia in 1864, the year of the Civil War during which General Sherman led Union troops through Georgia on what is called the "March to the Sea"(Nov.–Dec 1864) which resulted in great damages. 19 Georgia Virginia ) 0 0 1 = 0 7 8 1 ( x r a e y n i n r o b p u o r g e g a f o e z i S 130 120 110 100 90 80 70 Civil War 60 1854 1856 1858 1860 1862 1864 1866 1870 Year of birth (x) of age group 1868 Fig. 10 : Reduction in birth rates during the Civil War in Georgia and Virginia The horizontal axis shows the year of birth as derived from the age at the census of 1870. The fact that in 1861 and 1870 the levels are nearly the same shows that there was no permanent emigration or immigration between the year of birth and the census year. Source: 1% sample of the census of 1870 from the database of IPUMS, USA. At that time Georgia was not a registration state which means that no death data were recorded. As a result, one has no idea of the civilian death toll of the "March to the Sea". However, there was a census in 1870 through which one can measure the birth numbers in earlier years. The population pyramid shows that there was a birth trough of amplitude 1.33 and of width 3 years (Fig. 10). Applying the relationship between birth troughs and death peaks leads to a death peak amplitude Ad = (1/C)A1/α . For the sake of simplicity we take C = 1, α = 0.5 which will give a fairly conservative death estimate: Ad = 1.332 = 1.77. For the average death rate in 1861–1870 time we take the rate of Sweden namely 20 0/00 (R´esum´e r´etrospectif 1907, p. 368). Thus the peak rate in 1864 was: 1.77 × 20 = 35 0/00. In 1860 the population of Georgia was about 1 million; this leads to an excess-death toll of 15 × 1000 = 15, 000. There may be a additional excess-deaths in 1861–1863 but as we do not know the width of the death peak (usually it is more narrow than the birth trough) we will not try to estimate them. b 20 The figure of 15,000 excess-deaths should be seen as a tentative estimate because it is not obvious that the death-birth relationship derived from food crises can also be used for a war time situation. During wars, the fact that husbands and wives are separated may inflate the birth troughs to an extent which may critically depend upon how frequently servicemen can return home. If the birth troughs of Fig.8 are inflated through the separation effect, the figure of 15,000 will be an over-estimate. Measles epidemic in Fiji Islands in 1875 It is claimed that the measles epidemic of 1875 in Fiji claimed one third of the pop- ulation which was assumed to have been 150,000 before the epidemic (McArthur 1967, p.8). Note that both the population and the death toll are merely estimates made by western visitors9. ) s r a e y ( e g A 46y 50 40 30 20 10 0 Fiji 54 53 52 51 50 49 48 47 46 45 44 43 42 41 40 39 38 37 36 35 34 33 32 31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 9 8 7 6 5 4 3 2 1 0 h t r i b f o r a e Y 1875 1871 1881 1891 1901 1911 1921 Fig. 11 Was there a devastating measles epidemic in Fiji in 1875? The ages given on the left-hand side correspond to the age-groups in 1921. One expects a birth trough in 1875. In the census of 1921 this trough will be seen as a reduced size of the age-group that is 49 year old resulting in a reduced 49 year old age- group. Based on a death-birth relationship with an exponent α = 0.5, one would expect a reduction of the corresponding 5-year age group as indicated in green. Source: Fiji census of 1921, the first in which ages were recorded, cited in McArthur (1967, p.38). This sounds somewhat surprising when compared to the death toll of the influenza epidemic of 1918 which was only 5%. This last figure is probably more reliable than the one for 1875 because in the meanwhile the procedure for births and deaths registration was improved by the British administration. Fig. 11 is based on the census of 1921. This was not the first one but in the censuses of 1879,1881, 1891, 1901 and 1911 only 4 age groups were distinguished, namely: children, youths, adults and aged. 9For instance, the population estimates published by different visitors range from 100,000 to 300,000. 21 Even without doing any further calculation a comparison with Anhui (where the death toll was 5%) shows that the 33% claim seems dubious. It is true that instead of annual data here we have only 5-year averages but in Anhui the contraction would still be clearly visible on 5-year averages. More specifically, a calculation involves the following steps. (i) Computation of the amplitude of the death spike based on the 33% death toll. (ii) Computation of the amplitude of the birth trough based on the death-birth relationship; for α we selected α = 0.5 but taking another value (e.g. α = 0.3) would not make a great difference. (iii) Finally, from the amplitude of the birth trough one derives the expected contraction of the 5-year age group (shown in green in Fig. 11). One may wonder why British authorities were favorable to the thesis of a big drop in the Fijian population10. One reason which comes to mind is that in 1879 started a massive transfer of Indian indentured workers to Fiji which had the purpose of providing the manpower needed on sugar cane plantations. By 1946 the Indian com- munity had developed to the point of being larger than the native Fijian population. Conclusion Outline of the exploration of food crises In his seminal paper Jacques Bertillon used weekly and monthly birth and death data (Bertillon 1892). We wished to see to what extent his analysis can also be performed using annual data for in many countries only annual vital statistics are publicly avail- able. It was shown that the death-birth effect can be analyzed in a significant way provided one considers events marked by large-scale mortality. The annual death birth data were found to be compatible with the 9-month time lag revealed by weekly and monthly data (see Fig. 3.5). By observing regional death peaks and birth troughs we found a power law rela- tionship between their respective amplitudes in confirmation of a similar connection already observed in Richmond et al. (2018b). Agenda for future explorations of mass mortality episodes The spirit of a death registration network is to go bottom up. Starting from block level one should move up to county, state and nation level. At each step the death numbers should be aggregrated until one gets to a total number for the whole country. Any death estimate made at the macro-level without being supported by appropriate 10In British accounts the gravity of this epidemic is heavily emphasized (see McArthur 1967, p.8). The following excerpt is from a report by a commission in 1893. In 1875 "the people were estimated to number about 150,000, and it is recorded, probably with fair exactitude, that 40,000 died from measles during the epidemic which overran the whole archipelago in the space of 4 months. Whether the Fijians who survived have had their stamina permanently lowered can only be a matter for conjecture." 22 data at micro-level should be considered as suspicious. Unfortunately in many cases of interest the data that would permit a bottom up procedure are simply not available. The methodology developed in the previous section allows us to use censuses made decades after the mass mortality. For instance, in the case of the Fiji Islands we have been using a census made 4 decades after the epidemic of 1875. We believe that this procedure can bring new light in many dubious estimates. Appendix A: Specific features of the crisis in China It seems that in the course of the past two decades the factual description of the food crisis of 1960-1961 has progressively been replaced by accounts based on ill- founded stories. The Wikipedia articles entitled: "Great China famine" and "History of agriculture in the People's Republic of China" illustrate this tendency. The second one contents claims not based on any reference or outright mistakes11. Science is not only about building nice models, first and foremost it means starting from the right facts. In order to come back to a more scientific description, it is important to collect as many hard data as possible. By hard data we mean data that can be checked in some way. This is for instance the case of grain trade data because such figures are published by the two trade partners, first as exports, then as imports. It is the purpose of the present appendix to present a number of useful data. Parallel with the food crisis and meteorological situation of 1878 An article entitled "La famine en Chine" [Famine in China] was published in 1880 in a French scientific journal (Margoll´e 1880) which describes a meteorological sit- uation very similar to the one in 1959–1960, namely a severe drought in north China over 3 years together with floodings in south China. In this article, the two phenom- ena are explained by the existence of a strong high altitude air circulation from west to east which prevented the steady humid flow originating in the south from reaching the north. Instead the accumulated humidity would eventually be released in central and south China and provoke the floodings. The severity of the famine was described as follows. "Between 1876 and 1878, a lethal drought-famine struck the five northern provinces of Shanxi, Henan, Shan- dong, Hebei, and Shaanxi. By the time the rains returned late in 1878, an estimated 9 to 13 million of the affected area's total population of about 108 million people had perished (Legge 1878, Edgerton-Tarpley 2008). The publication of 1878 by the "Committee of the China Famine Relief Fund" shows that this famine triggered a mobilization in western countries. Relief operations opened a road to the action of western missionaries. The famine took place 14 years after western powers helped 11Such as the claim that China did not import grains before 1962 which, as will be seen below, is not correct. 23 the imperial government to defeat the Taiping rebellion. Excess deaths When the death rate of 1958 is taken as the baseline the peak of 1958–1961 gives an excess death toll of 14 millions, an impressive figure12. As in Finland the crisis was a two step process: the fast population growth made the country vulnerable to adverse weather conditions; then in to successive years 1959 and 1960 the country was hit by severe droughts. We will also try to assess the role of the "Great Leap" policy. Wheat imports The situation was aggravated by the difficulties and delays in importing grains. The USSR was unable to export grains to China13. A table of the SNIE (1961) reports that there were no Soviet grain exports to China. Soviet-China relations already started to sour in 1960. In July-August 1960 the USSR abruptly withdrew nearly all of the 2,500 Soviet industrial technicians present in China; moreover, a disruption in oil delivery created a shortage of petroleum products in late 1960 (SNIE 1961, p. 4). After 1950 the United States had put in place a drastic trade embargo which, through the CHICOM committee, involved also US allies. In 1959 the US had a large surplus of wheat. An article of 8 November 1959 in the New York Times is entitled "Wheat surplus is a big headache". It says that at the start of the crop year the surplus stood at 1.27 billion bushels (on the basis of one bushel of wheat weighing 30kg 12If one takes as baseline the level of 1950 the excess death toll is about 7 millions which is of the order of magnitude of the recurrent famines that occurred in the 1920s and 1930s. 13The following articles published in the "New York Times" provide some useful glimpses into the situation. • 10 May 1959: Drought perils wheat crop in North China. • 19 August 1959: China is believed to be coping successfully with one of the worst droughts in mainland China in several decades. • 27 August 1959: China announced that its 1958 production figures issued earlier this year, had been overstated. It also reduced output goals set for 1959. • 12 February 1960: Taiwan buys for $5.2 million (about 0.1 million tons) of US wheat. • 5 July 1960: The Chinese government spurs efforts to rush irrigation equipment into drought-struck north east provinces to save wheat crop. • 7 August 1960: The Chinese government is resorting to extraordinary measures to produce food. Millions of people are shifted from nonagricultural jobs to fight shortage. • 16 October 1960: Several million tons of wild plants have been collected in north China against losses caused by crop failure, the Peiping radio has reported. • 23 January 1961: In 1960 China and the Soviet Union had an "unsatisfactory agricultural output". • 4 May 1961: An agreement was signed between China and Canada for the sale of 6 million tons of grains (at a cost of $362 millions) over a period of 2.5 years. • 13 September 1961: Canada is also exporting grains to the Soviet Union and Poland. • 10 March 1977: After three years of reducing its wheat imports, China is once again diping deeply into its foreign currency reserves to feed its people. Total wheat imports reached $550 million. [At a price of $55/ton this represents 10 million tons, i.e. some 6.5% of Chinese consumption. Without such massive imports the situation might have become quite as dramatic as in 1960.] • 20 March 1977: China's economic performance last year was the worst since the Cultural Revolution in the late 1960's, with the growth rates of industrial and agricultural production well below those of 1975. [According to government data, the real GDP growth rate was 8.7% in 1975 and -1.6% in 1976.] 24 this quantity represented 38 million tons) enough to meet US domestic needs for two years. Nevertheless, US exports to China were practically nil between 1950 and 1970. Led by its fairly independent Prime Minister John Diefenbaker, Canada was willing to brave the embargo. It sold a small amount of wheat to China in 1958 and opened discussions with the US to get permission to sell more. Indeed, in 1961 Canada sold about 1 million tons of wheat and barley to China. This represented one third of total Chinese imports of foodgrains in 1961 (Lu 1997, p.22) but less than 1% of the total production of around 170 million tons according to SNIE (1961, p.3). It was both too little and too late14. In summary, it can be said that through its own embargo and its influence on other western grain exporting countries, the United States bears at least partial responsibility in the crisis of 1960. Conflicting accounts Most present-day accounts blame the "Great Leap Forward" as the main cause of the food crisis. However, a US intelligence report (SNIE 1961) which was published in April 1961 gives another perspective. It is of interest for several reasons. (1) Classified as "Secret", this joint report of the intelligence organizations of the Army, Navy, Air Force, State Department and of the Central Intelligence Agency, was not destined to be published. Therefore, one cannot claim that it was part of a public relations operation and does not reflect the opinions of its authors. As a matter of fact, it was declassified only in December 1996. (2) As it was published in April 1961 it is possible to check whether its predic- tions were confirmed by subsequent events. This test supports indeed the perspective which is presented. (3) The SNIE report contrasts with present-day mainstream accounts in several important ways. For instance, it describes in detail the measures already taken in 1960; they show that, contrary to the claim made in the Wikipedia articles mentioned above, the leadership was well informed about the reality of the situation. Was the "Great Leap Forward" an economic failure? Nowadays it is a standard and almost self-evident belief that the "Great Leap For- ward" (1957–1960) was a technical and economic failure15. It is indeed quite likely that it put additional stress on country side people, but was it an economic failure? 14A discrepancy can be noted between Lu (1997, p.22) and SNIE (1961, p.6) regarding the total imports of grain in 1960: the first source gives 5.81 million tons whereas the second gives 2.81 million tons. The reason of this difference is that the second number is based on contracts which had been concluded by April 1961 when this report was published. This means that there were additional contracts and deliveries between April and December 1961, probably from Burma and Malaysia. 15Though also held in China, this view is more based on ideological reasons developed during the Deng period than on hard facts. 25 Although only of marginal importance for the purpose of the present paper, from a scientific perspective the question certainly deserves to be raised. The following facts can be mentioned. • A short Internet search shows that at least 5 large dams were built in those years. Here is the list. (i) 1957–1960, Sanmenxia Dam, Henan/Shanxi, 106m, 16.2 cubic-km (ii) 1958–1962, Xinfengjiang Dam, Guangdong, 105 m, 13.9 cubic-km (iii) 1958–1962, Zhexi Dam, Hunan Province, 104 m, 3.65 cubic-km (iv) 1958–1981, Chengbihe Dam, Guangxi, 70 m, 1.12 cubic-km In 1964, Ren´e Dumont, a French expert in agricultural economics wrote the follow- ing (Dumont 1964, p. 393, our translation): "Between 1955 and 1964 I observed the most extraordinary transformation of the agricultural landscape. When one flies over China from Hanoi to Beijing one sees that the regions to the south of the Yangtze are now covered with canals, levees and dikes". Naturally, the other side of the coin was that for this kind of work men were often employed far away from their villages which disrupted family life and reduced conceptions. This may have amplified the Bertillon birth effect. It is a fact that birth reductions were much more pronounced south of the Yangtze than in northern provinces. • The SNIE (1961) report gives growth estimates for industrial production and GDP which are summarized in Table A1. The forecast for 1961 was made under the assumption that Soviet technicians would not come back to China; otherwise it would have been higher. Table A1: Estimates for industrial production and GDP growth during the "Great Leap Forward" 1958 1959 1960 1961 Average Industrial production GDP 33% 16% 18% 12% 8% (forecast) 12% 20% 13% Notes: It is often said that the "Great Leap Forward" was a technical failure but the present estimates made by the US intelligence community tell another story. It is true that such a rapid growth was not sustainable but the spirit of doing things notably faster than elsewhere is still present in China nowadays. The construction of the high speed rail track between Beijing and Shanghai was completed in 3-4 years, whereas in France the construction of a similar line between Lille and Montpellier (almost the same distance) took about 15 years. Source: SNIE (1961, p.2,5) [Special National Intelligence Estimate]. Quite understandably the report does not say how these figures were computed but the fact that making such estimates was one of the main duties of US intelligence agencies suggests that they had means to do that reliably. Here again, the fact that the report was not destined to be made public is important because otherwise the publication of estimates could be a way to influence the public opinion. 26 Despite the slowdown of 1960-1961 the average rates remain impressive even for a fast-growth economy like China. They justify the SNIE's observation that "Peiping recognized that it could not continue the breakneck industrialization tempo of 1958– 1959". According to government statistics released later on, there was a decline in the GDP in 1962–1963. It is difficult to separate the after-effect of the food crisis from the consequences of the departure of the Soviet technicians, the shutdown (or reduction) of Soviet oil supply and other adverse conditions connected with the end of the Soviet economic cooperation. The end of Soviet assistance was particularly critical in the face of the continued western and Japanese trade embargo at least until 1971. References Bertillon (L.-A.) 1872: Article "Mariage" in the Dictionnaire Encyclop´edique des Sciences M´edicales, [Encyclopedic Dictionary of the Medical Sciences]. 2nd series, Vol. 5, p.7-52. [In this study the author emphasizes the protective role of marriage not only with respect to suicide but also with respect to general mortality.] Bertillon (J.) 1892: La grippe `a Paris et dans quelques autres villes de France et de l'´etranger en 1889–1890 [The influenza epidemic in Paris and in some other cities in western Europe]. In : Annuaire statistique de la ville de Paris pour l'ann´ee 1890, p. 101-132. Imprimerie Municipale, Paris. [Available on Internet, for instance at the following address: http://www.biusante.parisdescartes.fr/histoire/medica/resultats/index.php?do=livre&cote=20955] China Tuberculosis Control Collaboration 2004: The effect of tuberculosis control in China. Lancet 364,417-422. Davis (M.) 2000: Late Victorian holocausts: El Nino famines and the making of the Third World. Verso, London. Dumont (R.) 1964: Les communes populaires rurales chinoises [The people's com- munes in rural China]. Politique ´Etrang`ere 29,4,380-397. [available online] Durkheim (E.) 1895: Les r`egles de la m´ethode sociologique. F´elix Alcan, Paris. Translated into English under the title: "The rules of sociological method". [The French and English versions are freely available on Internet.] Espinas (A.) 1878, 1935: Des soci´et´es animales [on animal societies]. Thesis of the university of Paris. Republished in 1935, F´elix Alcan, Paris. [To measure the novelty of this study one should remember that the author was in fact a sociologist.] Dyson (T.) 1991a: On the demography of South Asian famines. Part 1. Population 27 Studies 45,1,5-25. Dyson (T.) 1991b: On the demography of South Asian famines. Part 2. Population Studies 45,2,279-297. Dyson (T.) 1993: Demographic responses in South Asia. Institut of Development Studies 24,4,17-26. Edgerton-Tarpley (K.) 2008: Tears from iron: cultural responses to famine in nineteenth- century China. The north China famine of 1876–1879. University of California Press, Berkeley. Finland 1871. The Finnish title of this publication is: Suomenmaan Virallinen Tilasto VI [Of- ficial Statistics of Finland], Helsinki. [Available on line (April 2018): http://www.doria.fi/handle/10024/67301] Finland 1902. ´El´ements d´emographiques The French subtitle of this official publication is: principaux de la Finlande pour les ann´ees 1750-1890, II: Mouvement de la population. [Vital statistics of Finland for the years 1750–1890.] The Finnish title is: Suomen [Finland] Vaestotilastosta [demographical elements] vuosilta [years] 1750-1890, II: Vaeston [population] muutokset [changes]. Helsinki 1902. (526 p.) [Available on line (April 2018): http://www.doria.fi/handle/10024/67344] Flora (P.), Kraus (F.), Pfenning (W.) 1987: State, economy, and society in western Europe 1815-1975. A data handbook in two volumes. Volume 2: The growth of industrial societies and capitalist economies. Macmillan Press, London. IPUMS: Integrated Public Use Microdata Series. Minesota Population Center, Uni- versity of Minnesota. [The IPUMS database consists of microdata samples from the censuses of the US and a number of other countries. Most often the samples are 1% or 5% samples but for the US there are a few 100% samples.] Legge (J.) 1878: The famine in China. Illustrations by a native artist with a transla- tion of the Chinese text. Published by C. Kegan Paul for the "Committee of the China Famine Relief Fund", London. Linder (F.E.), Grove (R.D.) 1947: Vital statistics rates in the United States 1900– 1940. United States Public Health Service, Washington. Lu (F.) 1997: China's grain trade policy and its domestic grain economy. Working paper No E1997002. Peking University and Hong Kong university of Science and Technology. [The paper contains a table which gives Chinese grain imports and exports from 1953 to 1992. Grain imports were negligible until 1961 when 28 they jumped to 5.81 million tons.] Maharatna (A.) 1992: The demography of Indian famines: a historical perspective. Thesis, London School of Economics and Political Science. University of Lon- don. [In 1996 the thesis was published by Oxford University Press (Dehli) under the title: "The demography of famines: an Indian historical perspective."] Margoll´e (E.) 1880: La famine en Chine. La Nature, Revue des Sciences p. 314-315. R´egnier-Loilier (A.) 2010a: ´Evolution de la saisonnalit´e des naissances en France de 1975 `a nos jours [Changes in the seasonal birth pattern in France from 1975 to 2006]. Population, 65,1,147-189. – [This paper contains a section about the effect of heat waves.] R´egnier-Loilier (A.) 2010b: ´Evolution de la r´epartition des naissances dans l'ann´ee en France [Changes in the seasonal birth pattern in France]. Actes du XVe colloque national de d´emographie [Proceedings of the 15th national conference on demography, 24-26 May 2010] Published by the "Conf´erence Universitaire de d´emographie et d'´etude des populations". [Fig. 1 of this paper gives the pattern of monthly variations of conceptions in France from the 17th to the 20th century; it is based on Dupaquier (1976).] Ren kou 1988. The complete title has the following pinyin transcription: Ren kou tong ji zi liao hui bian. Zhong hua ren min gong he guo. [Population statistics data compilation, 1949-1985. People's Republic of China.]. Published in 1988. [This official demographic report (1030 p.) gives annual birth and death rates for the whole country and separately for each province. It contains also population data by age and province based on the three censuses of 1953, 1964 and 1982. So far, we could not find an English translation but in fact the Chinese version can be used fairly easily; one needs to know only a few key-words.] R´esum´e r´etrospectif 1907: Statistique Internationale du mouvement de la population d'apr`es les registres d'´etat civil. depuis l'origine des statistiques de l'´etat civil jusqu'en 1905. [International vital statistics from the start of official registration until 1905]. Imprimerie Nationale, Paris. Rey (G.), Fouillet (A.), Jougla (E.), H´emon (D.) 2007: Vagues de chaleur, fluctua- tions ordinaires des temp´erature et mortalit´e en France depuis 1971. Population 62,3,533-564. Richmond (P.), Roehner (B.M.) 2018a: Coupling between death spikes and birth troughs. Part 1: Evidence. Physica A 506,97-111. Richmond (P.), Roehner (B.M.) 2018b: Coupling between death spikes and birth 29 troughs. Part 2: Comparative analysis of salient features. Physica A 506,88-96. Roehner (B.M.) 1995: Theory of markets. Trade and space-time patterns of price fluctuations. Springer, Heidelberg. Roehner (B.M.) 2010: How can population pyramids be used to explore the past? Asia Pacific Center for Theoretical Physics (APCTP) Bulletin, No 25-26, p. 13-25, January–December 2010. SNIE [Special National Intelligence Estimate number 13-61] 1961: The economic situation in Communist China. Submitted by the Director of Central Intelli- gence on 4 April 1961. Declassified on 24 December 1996. [available on line] Viswanathan (G.M.), Luz (M. G. E. da), Raposo (E.P.), Stanley (H.E.) 2011: The Physics of foraging. An introduction to random searches and biological en- counters. Cambridge University Press, Cambridge (UK).
1502.05115	1	1502	2015-02-18T04:53:46	Quantitative analysis of reptation of partially extended DNA in sub-30 nm nanoslits	[ "physics.bio-ph", "cond-mat.soft" ]	We observed reptation of single DNA molecules in fused silica nanoslits of sub-30 nm height. The reptation behavior and the effect of confinement are quantitatively characterized using orientation correlation and transverse fluctuation analysis. We show tube-like polymer motion arises for a tense polymer under strong quasi-2D confinement and interaction with surface- passivating polyvinylpyrrolidone (PVP) molecules in nanoslits, while etching- induced device surface roughness, chip bonding materials and DNA-intercalated dye-surface interaction, play minor roles. These findings have strong implications for the effect of surface modification in nanofluidic systems with potential applications for single molecule DNA analysis.	physics.bio-ph	physics	Quantitative analysis of reptation of partially extended DNA in sub-30 nm nanoslits Jia-Wei Yeh 1, 2, K. K. Sriram 1, Alessandro Taloni 3, Yeng-Long Chen1, 4, 5, and Chia-Fu Chou1, 6, 7,* 1Institute of Physics, Academia Sinica, 128, Sec. 2, Academia Road, Nankang, Taipei 11529, Taiwan. 2 School of Applied and Engineering Physics, Cornell University, Ithaca, NY 14850, USA. 3 CNR-IENI, Via R. Cozzi 53, 20125 Milano, Italy. 4 Department of Chemical Engineering, National Tsing Hua University, 101, Sec. 2, Kuang- Fu Road, Hsinchu 30013, Taiwan. 5 Department of Physics, National Taiwan University, 1, Sec. 4, Roosevelt Road, Daan, Taipei 10617, Taiwan. 6 Research Center for Applied Sciences, and 7 Genomics Research Center, Academia Sinica, 128, Sec. 2, Academia Road, Nankang, Taipei 11529, Taiwan. Abstract We observed reptation of single DNA molecules in fused silica nanoslits of sub-30 nm height. The reptation behavior and the effect of confinement are quantitatively characterized using orientation correlation and transverse fluctuation analysis. We show tube-like polymer motion arises for a tense polymer under strong quasi-2D confinement and interaction with surface- passivating polyvinylpyrrolidone (PVP) molecules in nanoslits, while etching- induced device surface roughness, chip bonding materials and DNA-intercalated 1 dye-surface interaction, play minor roles. These findings have strong implications for the effect of surface modification in nanofluidic systems with potential applications for single molecule DNA analysis. Polymer reptation is one of the transport mechanisms for biological macromolecules in a crowded environment, such as an actin filament powered by molecular motors that snakes its way towards the center of the sarcomere [1], or a DNA molecule that wraps around histone proteins during nucleosome reposition [2]. A phenomenological approach to describe reptation dynamics is the classical tube model of Edward, de Gennes and Doi [3-6]. The main assumption is that the polymer is confined in a tube- like constraint along the polymer backbone, and its preferential motion in the tube resembles the slithering of a snake. Reptation motion has been directly observed for actin filament diffusing in an entangled actin network [7], single DNA molecule in a entangled polymer solution [8], and also for a tense DNA stretched by optical tweezers, for which the DNA transverse fluctuations are restricted into a tubelike region [9]. More recently, the reptation of stiff carbon nanotube filaments confined in a crowded environment was compared to the reptation of flexible and semi-flexible filaments [10]. Reptation motion has also been observed for a single polymer adhering on a surface, where the polymer motion is constrained to two dimensions [11-13]. To characterize reptation dynamics, the tube model must be tested by probing the trajectories of individual chains [6, 9, 14, 15]. Recent studies involving DNA motion driven by an electric field in 20 nm nanoslits has been described as biased-reptation behavior [16, 17]. Despite providing evidence of the polymer reptation dynamics, the chain trajectories and the ensuing slithering motion were not 2 quantitatively characterized in the former experiments. In this study, we directly observe and systematically probe the trajectories of single stretched DNA polymers in nanoslits of height between 23 to 110 nm. For the first time, we provide a scaling relationship of the tube as a function of the confinement and the slit length. Most importantly, our analysis constitutes the first quantitative study of a polymer reptation in a two dimensional confined environment. Description of the experiment- The experimental evidence of DNA reptation under strong slit-like confinement is provided in Figure 1a-b. Biased-reptation motion [16, 17] was induced by forcing DNA molecules into long nanoslits (25 nm in depth, 200 µm in length) using a square wave electric field. Time sequences of DNA molecules moving inside the slit clearly demonstrate its traces constrained in tube-like regions. This observation, however, only constitutes a qualitative indication of the constrained reptation dynamics. To quantitatively characterize the trajectories of single DNA molecule in nanoslits, DNA molecules were subjected to tug-of-war (TOW) and retraction scheme at the micro-nanofluidic interfaces [18], as illustrated in Figure 1d-f. For this, a T4-DNA polymer is transposed from the microchannel region into the nanoslit by applying an electric field (~2 V/cm), to overcome the entropic barrier. When one end of the polymer chain reaches the other side of the nanoslit, the field is turned off. At this point, most part of the DNA chain is in confined nanoslit region, with the coiled DNA ends protruding into the microchannel regions, and the TOW starts. The conformational entropy difference between a polymer chain confined in the nanoslit and in the microchannels induces equal and opposite entropic recoiling forces frec at the two micro-nano interfaces, thereby establishing a TOW between two coiled ends of the same DNA polymer across the nanoslit (Fig. 1d). This scenario could last for minutes until one end of the DNA loses the TOW, thus 3 entering the nanoslit at time t0 from one side, retracting slowly, and eventually exiting from the other side at time tf (Fig. 1e and the movies M1, M2, and M3 in the supporting information). As demonstrated in Figure 1d, the contour of the polymer in the nanoslit remains nearly constant during the TOW. Possible causes for the reptation-like behavior include reduced transverse fluctuations due to tension acting on the DNA segment, strong surface interactions, and large surface roughness. These factors are systematically examined here using nanoslits of different height and surface coating. Device fabrication: The devices were fabricated in fused silica using standard photolithography and etching methods, with the design illustrated in Figure 1c. Dry and wet etching methods were both tested for roughness influence. For dry etching of nanoslits, reactive ion etching (RIE; ANELVA DEM-451T) with a CF4/O2 mixture at 20 W was carried out; for wet etching, ten times diluted buffered hydrogen fluoride (BHF, 6:1 volume ratio of 40% NH4F in H2O and 49% HF in H2O) was used. The dimensions of each nanoslit are 10 or 30 µm in width, 2, 3, 4, 5, 10, 20, 30 or 200 µm in length and 23, 28, 40, 50, 65, 110 nm (dry etching) and 25 nm (wet etching) in height were made. Nanoslits bridge inductively coupled plasma (ICP, Samco RIE- 10iP) etched microchannels of 1.5×100 (depth×width) µm. Etched fused silica chip was bonded to a cover glass using two different methods. The first method uses polysilsesquioxane (PSQ)-coated cover glass subjected to oxygen plasma surface- treatment at room temperature [19]. PSQ solution is a mixture of xylene and Hardsil (Gelest) in 2:1 ratio, filtered using a syringe filter with 0.45 µm PTFE membrane (Basic Life). The second method is fusion-bonding [20, 21], which helps sealing fused silica chips to cover glass by thermal bonding at a temperature of 950 °C. In both cases, a Piranha solution (con. H2SO4/H2O2 in 1:1 ratio) was used to clean the 4 substrates and cover glass in the first step. After the bonding process, sample loading reservoirs were attached to the substrate by UV-curable glue (No. 108, Norland optical adhesives). Electrical contacts were made by inserting a gold electrode in each of the reservoirs. Nanoslit surface roughness was measured by AFM (Bioscope II, Veeco) with the root-mean-squared roughness (Ra) around 2 nm for dry etching and 0.7 nm for wet etching. The relative roughness is typically less than 10 percent of the slit height, and does not strongly affect the DNA trajectories. Sample preparation: T4 DNA (T4GT7 DNA, 166 kbp, Wako Japan) was stained with either SYBR-Gold (Invitrogen) or YOYO-1 (Invitrogen) fluorescent dye, with dye to base-pair ratio of 1:5. DNA samples were initially prepared at 0.1 µg/ml in 0.5×TBE buffer (Sigma) containing 2.5% (w/w) poly(n-vinylpyrrolidone) (PVP, 10 kDa, Sigma, used to suppress electro-osmotic flow), 30% (w/v) sucrose (J.T. Baker), and 10% (w/v) glucose (Sigma) used to increase solution viscosity thereby slowing down the dynamics of DNA molecules. The buffer viscosity was 4.1 cP, measured by a viscometer (Toki Sangyo). An oxygen scavenging system containing 0.5 %(v/v) β- mercaptoethanol (Sigma), 50 µg/ml glucose oxidase (Sigma), and 10 µg/ml catalase (Roche) was used to reduce photobleaching and help extend the observation for a few hours. The ionic strength (I = 1.79×10-2 M) of the buffer condition gives the Debye length of 1.6 nm and the effective DNA diameter (deff) is 10 nm [22]. Fluorescence imaging: Single DNA molecules were observed with a fluorescence microscopy system consisting of an inverted microscope (Leica DMI6000), 100× oil-immersion lens (Leica) with a numerical aperture of 1.4, and electron-multiplying charge coupled device camera (EM-CCD, Andor Technologies; IXon-888 or 897) with a pixel resolution of 0.13 or 0.16 µm, respectively. Images were captured at a rate of 17 frames/s. DNA trajectories were extracted from the 5 videos by using ImageJ software (NIH) and algorithms edited by MATLAB (The Mathworks, Natick, MA). For analysis of DNA contour trajectory, information was extracted by vectorization. The fluorescence point spread function (PSF) was determined, and DNA images were iteratively deconvolved. An algorithm was written to track the fluorescence peak and intensity area by Gaussian fitting from one end of the DNA polymer to the other end for each pixel segment. In this way, the contour trajectory inside a nanoslit was determined. Tug-of-war scenario: The present study allows characterization and subsequent determination of reptation dynamics in quasi-static TOW phase and during far-from- equilibrium retraction, respectively. In previous studies, reptation behavior has been identified by direct observation of DNA molecules [8, 9, 11], carbon nanotubes [10], or actin filament entangled dynamics [7, 15, 23], taking place in an imaginary “tube” that follows the molecule’s or stiff filament’s backbone. Our experimental setup allows quantitative determination of the “tube”, by measuring the transverse fluctuation magnitude of DNA during the TOW. As a matter of fact, the “tube” can be visualized in a simple way by superimposing a sufficient number (300~2000) of contour images where the time scale of each image is around 0.1 second, as shown in Figure 2a. This observation can be quantitatively characterized with the transverse chain fluctuations defined as the root-mean-squared standard deviations of the position of DNA segment yi(x,t;h). [8, 23, 24]. x is segmental position along the slit length ls x ∈ 0,ls ), i identifies the ith DNA molecule sample in slits with given h ( [ ] and ls (i ∈ 1, Nn,ls "# $% ), Nn,ls can be between 30 or 50 molecules according to the value of h and ls, and t is the time that the ith molecule spends within the slit in the TOW phase 6 "# ) . The magnitude of fluctuations of the confined DNA around its average $% t ∈ 0,t f i ( position is then obtained by the standard deviation σi(ls,t;h) = 1 ls ls∑ x=0 (yi(x,t;h) − yi(ls,t;h) )2 (1) where yi(ls,t;h) = ls∑ x=0 yi(x,t;h) ls . Expression (1) is then averaged with respect to the time frames collected during the TOW regime, i.e. σi(ls;h) = i t f σi(ls,t;h) t=1 ∑ t f i , and also with respect to the ensemble of i = Nh,ls molecules: σ(ls;h) = Nh,lx∑ t=1 σi(ls;h) Nh,ls . Therefore the former expression allows investigating the dependence of the average molecules transverse fluctuations upon the degree of confinement h and the slit length ls, as shown in the inset of Fig. 2b and in Figs. S3-S6 of the Supplementary material (black filled symbols). The plots display the apparent trend for which σ(ls;h) increases monotonically, with slit length ls for a given h. However, an accurate determination of the analytical dependence of σ(ls;h) on the confinement h, which is the principal aim of this analysis, is not easily inferred. This is possibly ascribable to two reasons: first, performing spatial average before temporal average could lead to different results than vice versa, although for ideal infinite chains this order is interchangeable; second, edge effects may considerably affect the overall scaling as the chain segments are more mobile close to x=0, x=ls (see movies M1, M2, M3) We test the first hypothesis by plotting the values of σ(ls;h) achieved by performing temporal average before spatial average (See supplementary material): Fig. S2 shows the same trends apart from a nearly constant offset. The second hypothesis was tested by defining a local spatial average, i.e. an average of the 7 confined DNA transverse coordinate over a spatial window of length lx, starting from x0, yi(lx;t,h) x0 = x0+lx∑ x=x0 yi(x,t;h) lx (2) lx ∈ 0,ls [ ] ] , and x0 ∈ 0,ls −lx [ where recovers yi(lx,t;h) x0 ≡ yi(ls,t;h) . We then define the local transverse fluctuation by means . When ls, one lx ≡ of the standard deviation σi(lx,t;h) = ls−lx∑ x0=0 { x0+lx∑ x=x0 " # yi(x,t;h) − yi(lx,t;h) x0 2 $ % lx } (ls −lx +1) (3) By averaging over time and Nh,ls realizations, we finally obtain the mean local transverse chains fluctuation σ(lx;h) shown in Fig. 2b and Figs. S3-S6 of the supporting information (open symbols). The above definition enjoys a largely improved statistics for spatial windows lx ≤ ls, and, on the other side, it allows a direct comparison of the local transverse fluctuations away from the edges. Fig. 2b shows an average scaling behavior of the form σi(lx,t;h) = A(h)lx ξ(h) (4) The fits of the experimental curves with larger lx are reported in Fig. 2b, while the values of the coefficient A(h) and of the scaling exponent ξ(h) are displayed in Fig. 2c. The scaling prefactor A(h) shows a monotonic increasing behavior as a function of the confinement h. On the other side, ξ(h) from fitting is reported in the inset of Fig. 2c to be between 0.31 and 0.395 with weak dependence on h: setting its average value around 0.35, the scaling coefficient A(h) seems to be fairly in accordance with the form A(h) ~ hξ/2. 8 The scaling expression (4) carries two important consequences. Firstly, it states that the longer the slit, the larger are the transverse fluctuations. Secondly, it corresponds to the scaling form typical to the roughness of growing surfaces [25]. In particular, the critical exponent ξ is also known as the roughness exponent, as it defines the rugosity of a surface around its mean spatial position [26]. ξ is 1/2 for Edward-Willkinson chains and directed 2D polymers [27]: the observed values 0.31- 0.4 indicate that the trapped DNA fluctuates, in the y direction, less than one would expect for a Rouse chain. This is certainly due to tension exerted by the recoiling forces to the free ends. As a matter of fact, smaller nanoslits augment the configurational entropy difference across the micro-nano interface, inducing higher tensile forces on the confined DNA portion. Thus, the confinement-induced tension can be reasonably considered as the only factor playing a role in suppressing transverse chain’s fluctuations. This is an important finding when compared with earlier experiments and models for reptation, where the entanglement ultimately owns to the surrounding concentrated polymer solution [7, 9, 10, 15, 23]. The recoiling forces are independent of ls [18], but strongly dependent on the confinement, being frec ~ kBT/h [28]; surprisingly, the exponent ξ does not seem to vary considerably in the range of probed h. It appears instead that the major role played by confinement is expressed by the pre-factor A(h), which indeed is absent in 1+1 dimensional models for growing interfaces. In addition, DNA-surface interaction becomes an important issue as the degree of confinement in a nanofluidic system is increased. Expected surface interactions are (1) steric trapping from a large surface roughness [29]; (2) DNA adsorption to the PSQ-coated bonding surface; (3) DNA interaction with surface passivating short chain polymer (PVP) attached on the channel surface [16, 17], which could act like a 9 mechanical obstacle in the slit. The surface roughness has been characterized by AFM and found to be small. We thus further investigate the other factors. Experiments with nanoslits sealed by PSQ-bonding [19] and fusion bonding [20, 21] were used to investigate how strongly the reptation motion depends on the DNA interaction with the surface. These experiments were performed using DNA stained with two different nucleic acid staining agents YOYO-1 and SYBR-Gold, to examine any contribution from surface interaction due to the fluorescent dye. As shown in Fig. 3a, DNA transverse fluctuations during TOW were found to be similar for YOYO-1 DNA or SYBR-Gold DNA in fusion bonded or PSQ bonded nanoslits. Thus, reptation motion does not strongly depend on either the fluorescent dyes or the bonding method. The effect of the surface-passivating PVP polymer on DNA is addressed as follows. PVP is usually added to reduce the negative surface charge of fused silica fluidic channel and the electro-osmotic flow (EOF) [17, 29-32], and earlier studies suggest that PVP may influence the movement of DNA in shallow nanoslits (~20 nm), thus causing a biased reptation [16, 17]. The 10 kDa PVP used in these experiments adsorb to the surface with an average thickness of 4 nm [33, 34]. The added concentration of 2.5 % is much higher than the saturated surface adsorption value of 0.02 %. The effective slit height is thus reduced with PVP (see schematic in the inset of Fig. 3b). The influence of PVP on the reptation motion is investigated in experiments with and without PVP, for which nanoslits of heights 28 nm and 23 nm were used respectively, thereby maintaining nearly constant effective slit height. Figure 3b shows the averaged transverse fluctuations σ = 0.065 ± 0.002 (se) µm in nanoslits with PVP coating surface is slightly smaller than in those without PVP σ = 0.07 ± 0.003 (se) µm. Based on these numbers, no firm conclusions can be drawn on 10 the effects of PVP on the lateral fluctuations of the DNA molecule, it appears instead that the size of the reptation tube does remain unaffected by the presence of PVP. The smaller number of observations in experiments without PVP is due to the difficulty in driving the DNA molecules into the nanoslits, owing to the influence of surface charges of fused silica channels in absence of surface passivation [16]. Retraction scenario: In the TOW analysis we could characterize the size of the imaginary tube constraining the dynamics of the constrained DNA. However, this observation alone, although accurate and quantitative, does not guarantee that the polymer’s dynamics corresponds to a reptating one. Thus, the reptation motion is detected and further quantified during chain retraction (when the DNA freely translocate out of the confining nanoslit), by studying the segmental tangential vector correlation of the trajectories. During the retraction process, the DNA length measured along the x axis can be defined as li(t) (see Fig. 1e). Hence x ∈ 0,li(t) i ) = 0. Taking Δ = 5 pixels (0.65 or ] in [ the reference system for which li(t0) = ls and li(t f 0.8 µm respective to IXon-888 or 897 EM-CCD), we introduce the segmental tangential vector vector whose components are vi(x,t;h) as %& where x ∈ 0,li(t) − Δ the [ Δ, yi x + Δ,t;h ( ) − yi x,t;h ( #$ ) ] . Thus, we define the tangential vector, or orientation, correlation function as Ci(x,t;h) = vi(x,t;h) vi(x,t0;h)⋅ vi(x,t0;h) vi(x,t;h) (5) and we report its behavior in Fig. 4. For Ci(x,t;h) < 1, the chain contour has changed at time t with respect to t0. For Ci(x,t;h) ≈ 1, the chain contour has fluctuated weakly at t about the t0 contour. Figure 4 shows the correlation function for a single molecule 11 (Fig. 4a) and its ensemble average (Fig. 4b) at the instances for which l (t) = 0.9 ls, 0.5 ls, and 0.3 ls. For h = 28 nm (dry etching, with PVP), the trajectory of DNA segments closely follows the backbone and the correlation function varies little with time and position (see movie M1). In contrast, there are significant fluctuations for DNA molecules in h = 50 and 110 nm (dry etching, with PVP) nanoslits (see movies M2 and M3). We tested the observed behavior by varying the value of Δ: our analysis vi(x,t;h) in the x direction, does not affect the reveals that, changing the length of qualitative trend exhibited in Fig. 4. Furthermore, Fig. 4c shows that the orientation correlation function in experiments without PVP were not as highly correlated as in experiments with PVP. This indicates that the surface-passivated PVP molecules indeed interact with the DNA molecule, and the ensuing reptation motion is strongly influenced by the presence of PVP in highly confined nanoslits. If compared with the influence that PVP has on the lateral fluctuations during TOW (Fig. 3b), we can conclude that the effects of PVP during the non-equilibrium retracting dynamics are much stronger. This can be explained by the fact that PVP acts as a surrounding entanglement medium (polymer concentration) through the entire reptation process, which is indeed a dynamical event. PVP’s presence although does not clearly affect the size of the tube, which instead is a stationary quantity that depends only on confinement-induced entropic tensile forces. In analogy with the analysis performed in the TOW phase, we study the effects of PSQ-bonding and fusion bonding on the retraction phase, achieving no experimental evidence that they hinder or promote the DNA reptation (see Fig. S7). 12 Similar results were inferred from experiments with different DNA staining agents, YOYO-1 and SYBR-Gold (Fig. S7) Finally, it may be of interest to compare the prior results of free diffusion of DNA polymer in sub-30 nm nanoslits [35-37]. The strong confinement is known to affect the DNA segmental correlation length [38, 39], and the diffusivity measurement attributed the relation to the “rod-like” chain conformation as Odijk’s argument [35, 40]. More relevant, the significant difference between the diffusivities measured along the x- and y- axis in 24 nm deep nanoslit has been dug out [37]. The origin of which could be understood due to the interaction of surface passivating PVP polymer we demonstrated here. Conclusions: In this Letter we have, for the first time, provided the quantitative characterization of DNA reptation in strong confinement conditions. This quantitative analysis, obtained by detecting the DNA transverse fluctuations in nanoslits during the equilibrium phase (TOW) and the orientation correlation of the subsequent out-of-equilibrium retraction process, shows that the strong confinement considerably suppresses the DNA freedom in the transverse direction, thus leading to reptation-like motion in the dynamical phase. Indeed, from sub-Kuhn length nanoslit confinement [18, 28] (50~100 nm) to the unconfined free solution [24, 41], the elongated DNA polymer has increasing degree of transverse fluctuations and segmental orientations. On the other hand, in strongly confining nanoslits (≤ 30 nm), DNA molecules are restricted in a tube-like region. Our study provides, for the first time, a quantitative characterization of the tube with confinement, by means of scaling relations whose critical exponents and prefactors depend on the slit height. Moreover, we have confirmed that the presence of PVP in solution significantly alters 13 the reptation dynamics, playing the active role of an entanglement medium. On the other side, the motion appears not to be influenced by the bonding surfaces or the fluorescent staining dyes. These findings help clarify the enforced reptation behavior observed in nanoslits from recent reports [16, 17, 29]. In a 3D gel environment, DNA molecules migrate through the entanglement polymer concentration by reptation and form a tube-like behavior. In this experiment, we have provided the evidence that the DNA reptation in 2D slit-like nanochannel is not only due to the confinement induced tensile forces, but also a major role of PVP by acting like a 2D gel matrix. Our work may lead to an improved device design, using simple 2D nanoslits to achieve high degree of DNA extension, and low degree of thermal fluctuation for high-resolution single molecule analysis [42-45]. Acknowledgements We thank Profs. Andreas Erbe, Cheng-Hung Chang and David Newquist for their critical review of the manuscript and insightful suggestions, Drs. Po-Keng Lin and Sheng-Qin Wang for helpful discussions, Wei-Chiao Lai for AFM measurement, and technical support from the Academia Sinica (AS) Nano Core Facilities. This work was supported by the AS Program on Nano Science and Technology and AS Thematic Projects [AS-97-FP-M02, AS-103-TP-A01] (to C.F.C.) and by the Ministry of Science and Technology (ROC) [102–2112-M-001-005-MY3, 103–2923-M-001- 007-MY3] (to C.F.C.). 14 15 Figures and Captions: Figure 1. (a) Schematic showing reptation of entangled polymer driven by an electrical field in nanoslits, with gray circles representing fictitious pillars to illustrate polymer entanglement. (b) Time sequence images and superposition of single molecule DNA trajectories in 25 nm wet-etched nanoslits. The images are separated by 0.135 sec. Applied field: 1333 V/m, 1 Hz AC; White arrow indicates the field direction. (c) Schematic of our micro/nano-fluidic device setup. (d) A DNA molecule during TOW [18]. Dotted lines represent micro-nano-interfaces. Upper and lower parts are microchannels; the middle part is the nanoslit region. (e) A DNA molecule retracting out of the nanoslit. The time interval between frames is 0.7 sec. The slit length and height are 20 µm and 28 nm, respectively. Translocation process takes place between t0 and tf, the time of complete translocation through the slit, being the retracting length l(t). (f) Schematic of DNA reptation in nanoslit. Fictitious pillars are drawn to illustrate polymer entanglement. 16 17 Figure 2. (a) Superimposed frames (separated by 0.1 s) of the DNA contour to determine the transverse fluctuation of DNA trajectories during TOW. The gray lines are 500 to 4000 trajectory traces of a DNA for h = 28, 50, and 110 nm (dry etching, with PVP) and the solid lines represent the corresponding averaged curves. (b) Average fluctuation vs. the longitudinal lengths l for h = 28, 40, 50, 65 and 110 nm (dry etching, with PVP). The solid lines represent the best fits of the transverse local fluctuation defined in Eq. (4) (l = lx, see the inset and Figs. S3-S6). Inset: the filled black symbols show the result of the global fluctuations as a function of ls for h = 110 nm (l = ls). The local roughness defined by equation (3) allows analyzing the scaling of the transverse fluctuations for values of lx ≤ ls, shown as open symbols (l = lx). The best fit of the numerical curves according to expression (4) is displayed in magenta and corresponds to the top curve in the main panel. (c) Scaling prefactor A(h) and roughness exponent ξ(h) (inset) as a function of h. The value of h seems to affect prefactor A considerably, showing an apparent reduction of the tube corresponding to the reptation dynamics, as the confinement increases. The scaling exponent ξ shows decay in correspondence of the DNA persistent length (~ 50 nm). 18 Figure 3. (a) Histogram of transverse fluctuation for YOYO-1 and SYBR-Gold dye labeled DNA in PSQ- and fusion-bonded nanoslits of 28 nm height (dry etching, with PVP) and 20 µm length. (Numbers of molecules: PSQ/YOYO-1: 19; FUSION/YOYO-1: 20; PSQ/SYBR-Gold: 31) (b) Histogram of transverse fluctuation for DNA in PSQ-bonded nanoslits, 10 µm length, of heights 28 nm (dry etching, with PVP) and 23 nm (dry etching, w/o PVP), respectively, but the effective slit height is the same in both cases. Inset shows illustration of DNA (red line) in the nanoslit. The 19 effective diameter (deff) is ≈ 10 nm; thickness of PVP layer (purple line with transparent blue) is 4 nm on each side. (Numbers of molecules: w/ PVP: 45; w/o PVP: 15) 20 21 Figure 4. (a) The tangential vector, or orientation, correlation function Ci(x,t;h), for h = 28, 50, and 110 nm (dry etching, with PVP) for slit length 20 µm for segments along the DNA backbone at times where the projected chain along the x axis is equal to 90% (black line), 50% (red line), and 30% (blue line) of the nanoslit length (ls) during retraction recoiling. (b) The ensemble average of orientation correlation C(x,t;h) for h = 28, 50, and 110 nm (dry etching, with PVP) for slit length 20 µm for segments along the DNA backbone at times where the projected chain along the x axis is equal to 90% (black line), 50% (red line), and 30% (blue line) of the nanoslit length (ls) during retraction recoiling. (c) Ensemble average of orientation correlation, C(x,t;h) , for experiments done with and without PVP (~10 molecules for each case) in PSQ bonded nanoslits at the 90% (blue line), 50% (red line), and 30% (black line) of nanoslit length during retraction recoiling. References: [1] D. Humphrey, C. Duggan, D. Saha, D. Smith, and J. Kas, Nature 416, 413 (2002). [2] H. Schiessel, J. Widom, R. F. Bruinsma, and W. M. Gelbart, Phys. Rev. Lett. 86, 4414 (2001). [3] S. F. Edwards, Proc. Phys. Soc. London 92, 9 (1967). [4] P. G. d. Gennes, J. Chem. Phys. 55, 572 (1971). [5] P. G. d. Gennes, Scaling Concepts in Polymer Physics (Cornall University Press, Ithaca, NY, 1979). [6] M. Doi and S. F. Edwards, The theory of polymer dynamics (Oxford, 1986). [7] J. Kas, H. Strey, and E. Sackmann, Nature 368, 226 (1994). [8] R. M. Robertson and D. E. Smith, Phys. Rev. Lett. 99, 126001 (2007). [9] T. T. Perkins, D. E. Smith, and S. Chu, Science 264, 819 (1994). [10] N. Fakhri, F. C. MacKintosh, B. Lounis, L. Cognet, and M. Pasquali, Science 330, 1804 (2010). [11] V. Kahl, M. Hennig, B. Maier, and J. O. Radler, Electrophoresis 30, 1276 (2009). 22 [12] S. A. Sukhishvili, Y. Chen, J. D. Muller, E. Gratton, K. S. Schweizer, and S. Granick, Nature 406, 146 (2000). [13] D. J. Olson, J. M. Johnson, P. D. Patel, E. S. G. Shaqfeh, S. G. Boxer, and G. G. Fuller, Langmuir 17, 7396 (2001). [14] T. Odijk, Macromolecules 16, 1340 (1983). [15] M. A. Dichtl and E. Sackmann, Proc. Natl. Acad. Sci. USA 99, 6533 (2002). [16] O. Castillo-‐Fernandez, G. B. Salieb-‐Beugelaar, J. W. van Nieuwkasteele, J. G. Bomer, M. Arundell, J. Samitier, A. van den Berg, and J. C. T. Eijkel, Electrophoresis 32, 2402 (2011). [17] J. D. Cross, E. A. Strychalski, and H. G. Craighead, J. Appl. Phys. 102, 024701 (2007). [18] J. W. Yeh, A. Taloni, Y. L. Chen, and C. F. Chou, Nano Lett. 12, 1597 (2012). [19] J. Gu, R. Gupta, C. F. Chou, Q. H. Wei, and F. Zenhausern, Lab Chip 7, 1198 (2007). [20] C. H. Lin, G. B. Lee, Y. H. Lin, and G. L. Chang, J. Micromech. Microeng. 11, 726 (2001). [21] P. Mao and J. Y. Han, Lab Chip 5, 837 (2005). [22] C. C. Hsieh, A. Balducci, and P. S. Doyle, Nano Lett. 8, 1683 (2008). [23] B. Wang, J. Guan, S. M. Anthony, S. C. Bae, K. S. Schweizer, and S. Granick, Phys. Rev. Lett. 104, 118301 (2010). [24] T. Baba, T. Sakaue, and Y. Murayama, Macromolecules 45, 2857 (2012). [25] A.-‐L. Barabasi and H. E. Stanley, Fractal concepts in surface growth (Cambridge University Press, Cambridge, 1995). [26] F. Family and T. Vicsek, J. Phys. A 18, L75 (1985). [27] S. F. Edwards and D. R. Wilkinson, Proc. R. Soc. London, Ser. A 381, 17 (1982). [28] A. Taloni, J. W. Yeh, and C. F. Chou, Macromolecules 46, 7989 (2013). [29] G. B. Salieb-‐Beugelaar, J. Teapal, J. van Nieuwkasteele, D. Wijnperle, J. O. Tegenfeldt, F. Lisdat, A. van den Berg, and J. C. T. Eijkel, Nano Lett. 8, 1785 (2008). [30] W. Reisner et al., Phys. Rev. Lett. 94, 196101 (2005). [31] J. T. Mannion, C. H. Reccius, J. D. Cross, and H. G. Craighead, Biophys. J. 90, 4538 (2006). [32] C. D. Zhou, W. W. Reisner, R. J. Staunton, A. Ashan, R. H. Austin, and R. Riehn, Phys. Rev. Lett. 106, 248103 (2011). [33] E. Minatti, D. P. Norwood, and W. F. Reed, Macromolecules 31, 2966 (1998). [34] Q. F. Gao and E. S. Yeung, Anal. Chem. 70, 1382 (1998). [35] Y. L. Chen, Y. H. Lin, J. F. Chang, and P. K. Lin, Macromolecules 47, 1199 (2014). [36] P. K. Lin, C. C. Hsieh, Y. L. Chen, and C. F. Chou, Macromolecules 45, 2920 (2012). [37] E. A. Strychalski, S. L. Levy, and H. G. Craighead, Macromolecules 41, 7716 (2008). [38] Y. L. Chen, P. K. Lin, and C. F. Chou, Macromolecules 43, 10204 (2010). [39] P. Cifra, Z. Benkova, and T. Bleha, J. Phys. Chem. B 112, 1367 (2008). [40] T. Odijk, Phys. Rev. E 77, 060901 (2008). [41] S. R. Quake, H. Babcock, and S. Chu, Nature 388, 151 (1997). [42] K. K. Sriram, J. W. Yeh, Y. L. Lin, Y. R. Chang, and C. F. Chou, Nucleic Acids Res. 42, e85 (2014). 23 [43] L. Lesser-‐Rojas, K. K. Sriram, K. T. Liao, S. C. Lai, P. C. Kuo, M. L. Chu, and C. F. Chou, Biomicrofluidics 8, 016501 (2014). [44] C. A. Aguilar and H. G. Craighead, Nature Nanotech. 8, 709 (2013). [45] S. K. Min, W. Y. Kim, Y. Cho, and K. S. Kim, Nature Nanotech. 6, 162 (2011). 24
1109.6065	1	1109	2011-09-28T00:12:51	Conformation changes and protein folding induced by \phi^4 interaction	[ "physics.bio-ph" ]	A model to describe the mechanism of conformational dynamics in protein based on matter interactions using lagrangian approach and imposing certain symmetry breaking is proposed. Both conformation changes of proteins and the injected non-linear sources are represented by the bosonic lagrangian with an additional \phi^4 interaction for the sources. In the model the spring tension of protein representing the internal hydrogen bonds is realized as the interactions between individual amino acids and nonlinear sources. The folding pathway is determined by the strength of nonlinear sources that propagate through the protein backbone. It is also shown that the model reproduces the results in some previous works.	physics.bio-ph	physics	November 10, 2018 3:45 WSPC - Trim Size: 9.75in x 6.5in folding 1 Conformation changes and protein folding induced by φ4 interaction Department of Physics, University of Indonesia, Kampus UI Depok, Depok 16424, Indonesia M. Januar Badan Pengkajian dan Penerapan Teknologi, BPPT Bld. II (19th floor), Jl. M.H. Thamrin 8, E-mail: [email protected], [email protected] Jakarta 10340, Indonesia A. Sulaiman Group for Theoretical and Computational Physics, Research Center for Physics, Indonesian Institute of Sciences, Kompleks Puspiptek Serpong, Tangerang, Indonesia E-mail: [email protected], [email protected], [email protected] L.T. Handoko Department of Physics, University of Indonesia, Kampus UI Depok, Depok 16424, Indonesia A model to describe the mechanism of conformational dynamics in protein based on mat- ter interactions using lagrangian approach and imposing certain symmetry breaking is proposed. Both conformation changes of proteins and the injected non-linear sources are represented by the bosonic lagrangian with an additional φ4 interaction for the sources. In the model the spring tension of protein representing the internal hydrogen bonds is realized as the interactions between individual amino acids and nonlinear sources. The folding pathway is determined by the strength of nonlinear sources that propagate through the protein backbone. It is also shown that the model reproduces the results in some previous works. Keywords: protein dynamics, protein folding, lagrangian, φ4 interaction 1. Introduction The pathway of proteins are determined by the sequences of its amino acid con- stituents. The time ordered of protein folding sequence leads from the primary to the secondary and subsequent structures. The secondary structure consists of the shape representing each segment of a polypeptide tied by hydrogen bonds, van der Walls forces, electrostatic interaction and hydrophobic effects. It is moreover formed around a group of amino acids considered as the ground state. Then it is extended to include adjacent amino acids till the blocking amino acids are reached, and the whole protein chain along the polypeptide adopted its preferred secondary structure. Our understanding on the underlying above-mentioned mechanism has unfortu- November 10, 2018 3:45 WSPC - Trim Size: 9.75in x 6.5in folding 2 nately not been at the satisfactory level. For instance, the studies based on statistical analysis of identifying the probabilities of locating amino acids in each secondary structure are still at the level of less than 75% accuracy. Moreover, the main mech- anism responsible for a structured folding pathway have not yet been identified at all. On the other hand, it is known that the protein misfolding has been identified as the main cause of several diseases like cancers and so on.1 Recently, Mingaleev et.al. have shown that the nonlinear excitations play an im- portant role in conformational dynamics by decreasing the effective bending rigidity of a biopolymer chain leading to a buckling instability of the chain.2 Following this understanding, a model to explain the transition of a protein from a metastable to its ground conformation induced by solitons has been proposed.3 In the model the mediator of protein transition is the Davydov solitons propagating through the protein backbone. At present, the most reliable theoretical explanation for this kind of the con- formational dynamics of biomolecules is the so-called ab initio quantum chemistry approach. This however requires astronomical computational power to deal with realistic biological systems.4,5 In contrary, there are some phenomenological model describing the folding pathway as a result of the interplay between the energy trans- fer from a solitary solution that travels along the protein backbone and string ten- sion.6 There are also some attempts to describe the dynamics in term of elementary biomatter using field theory approach7 and open quantum system.8,9 This paper follows the later approach, but starting from the first principle us- ing the lagrangian method to derive the responsible interactions and to clarify its origins. The paper is organized as follows. First, the model and the underlying assumptions are explained in detail. It is then followed by the derivation of rele- vant equation of motions (EOMs). Summary and conclusion based on the numerical analysis are given at the end of the paper. 2. The models The model is an extension of the toy model proposed in.10 More than considering a self-interaction mechanism as proposed in10 and subsequently developed in,3,6 more realistic model is introduced. In the model, the dynamics of amino acids forming proteins is initially considered as a free and linear system of bosonic matters. Further, external nonlinear sources, like laser or light bunch, are introduced. The sources which propagate through the protein backbone interact each other with the amino acids to induce conformation changes. The model describes the conformation changes as the dynamics of amino acids using a free and massive (relativistic) bosonic lagrangian as below, lc = (∂µφc)† (∂µφc) + 1 2 m2 φc φ† cφc , (1) where φc represents the conformation field. The hermite conjugate is φ† ≡ (φ∗)T for a general complex field φ. On the other hand, the nonlinear sources represented November 10, 2018 3:45 WSPC - Trim Size: 9.75in x 6.5in folding by the field φs are also governed by a massless bosonic lagrangian, ls = (∂µφs)† (∂µφs) + V (φs) , with an additional potential V (φs) taking the typical φ4− self-interaction, V (φs) = 1 4 λ (φ† sφs)2 , 3 (2) (3) where λ is the coupling constant. It should be noted that both scalar fields, φc = φc(t, x) denotes the local curvature of the conformation at position x with φc(x) = 1 or 0 for α or β−helix. The choice of interactions in Eqs. (1) and (2) are justified by the following considerations, • The conformation changes are assumed to be linear. It is actually not nec- essarily massive. Although one can put by hand the mass term m2 cφc φc in the lagrangian as written above, the massive conformational field could also be generated dynamically through certain symmetry breaking as shown later. φ† • The source is assumed to be massless concerning the laser or light source injected to the protein chains to induce the foldings. • Its non-linearity is realized by introducing the φs self-interaction which leads to the non-linear EOM. • For the sake of simplicity, the lagrangian is imposed to be symmetry under certain transformations, for instance in the present case is time and parity symmetry, i.e. φ(t, x) → −φ(−t, −x) for one-dimensional space. We should remark here that the model is although written in a relativistic form, after deriving relevant EOMs one can take its non-relativistic limits to obtain final EOMs describing the desired dynamics. Secondly, instead of using the vector elec- tromagnetic field Aµ to represent the nonlinear sources, like laser for instance, it is more convenient to consider the nonlinear source as a bunch of light or laser such that one might represent it in a 'macrosocopic' scalar field φs. Considering the dimensional counting and the invariance on time-parity sym- metry, the most general interaction between the conformation field and nonlinear sources is, lint = −Λ (φ† cφc)(φ† sφs) , (4) with Λ denotes the strength of the interaction. Eqs. (3) and (4) lead to the total potential in the model, Vtot = 1 4 λ (φ† sφs)2 − Λ (φ† cφc)(φ† sφs) . Eqs. (1), (2) and (5) provide the underlying interactions in the model. Concerning the minima of the total potential in term of source field, that is ∂Vtot ∂φs (cid:12)(cid:12)(cid:12)(cid:12)hφsi,hφci = 0 , (5) (6) November 10, 2018 3:45 WSPC - Trim Size: 9.75in x 6.5in folding 4 at the vacuum expectation values (VEV) of the fields yields the non-trivial solution, hφsi = r 2Λ λ hφci . (7) Imposing certain local symmetry, namely the phase or U(1) symmetry to the above total lagrangian, the VEV in Eq. (7) obviously breaks the symmetry. The symmetry breaking at the same time shifts the mass term for φc as follow, m2 φc → m2 φc ≡ m2 φc − 2Λ2 λ hφci2 , (8) from Eq. (4). On the other hand, Eq. (7) induces the 'tension force' which plays an important role to enable folded pathway in the present model. This will be discussed in the following section. 3. EOMs and its behaviours Having the total lagrangian at hand, one can derive the EOM's using the Euler- Lagrange equation, ∂ltot ∂φ − ∂µ ∂ltot ∂(∂µφ) = 0 , (9) where ltot = lc + ls + lint. Substituting Eqs. (1), (2) and (4) into Eq. (9) in term of φc and φs, one imme- diately obtains a set of EOMs, 1 c2 (cid:18) ∂2 ∂x2 − (cid:18) ∂2 ∂x2 − 1 2 m2 ∂2 ∂t2 − ∂2 1 ∂t2 + 2Λ φ2 c2 s(cid:19) φc = 0 , s(cid:19) φs = 0 . φcc2 + 2Λ φ2 c − 3λ φ2 (10) (11) Here the natural unit is restored to make the light velocity c and reappear in the equation. The last term in Eq. (11) determines the non-linearity of the EOM of source. One should also put an attention in the last term of Eq. (10), i.e. ∼ k φc with k ∼ 2Λhφsi2. This actually induces the tension force in the dynamics of conformational field enabling the folded pathway as expected. Hence, solving both EOMs in Eqs. (10) and (11) simultaneously would provide the contour of conformational changes in term of time and one-dimensional space components. 4. Numerical analysis Since the EOMs under consideration involves non-linear term, one should solve them numerically. The numerical analysis and simulation in the present paper are done using the finite difference method.11 Throughout numerical works, non-relativistic November 10, 2018 3:45 WSPC - Trim Size: 9.75in x 6.5in folding 5 Fig. 1. The discretized grid for solving the EOMs over the coordinate space R. limit v = ∂x/∂t ≪ c and the following boundary conditions for both fields are deployed, φs(0, t) = φs(L, t) = 0 and φc(0, t) = φc(L, t) = 0 for 0 ≤ t ≤ b , φs(x, 0) = f (x) and φc(x, 0) = p(x) for 0 ≤ x ≤ L , ∂φs(x, 0) ∂φc(x, 0) = q(x) for 0 < x < L , = g(x) and ∂t ∂t (12) with f (x), p(x), g(x) and q(x) are newly introduced auxiliary functions. In finite difference scheme, it is more convenient to replace φs and φc with u and w respec- tively, and rewrite them in discrete forms. Then, let us consider the coordinate space R = {(x, t) : 0 ≤ x ≤ L, 0 ≤ t ≤ b} discretized on a grid consisting of (N − 1) × (M − 1) rectangles with side length ∆x = δ and ∆t = ǫ shown in Fig. 1. Solving the equations over the grid gives us the desired numerical solutions. Both coupled EOMs in Eqs. (10) and (11) are rewritten in explicit discrete forms November 10, 2018 3:45 WSPC - Trim Size: 9.75in x 6.5in folding 6 as follows, ui,j+1 = 2ui,i − ui,j−1 + c2ǫ2(cid:18) ui+1,j − 2ui,j + ui−1,j wi,j+1 = 2wi,i − wi,j−1 + c2ǫ2(cid:18) wi+1,j − 2wi,j + wi−1,j δ2 δ2 + 2Λw2 i,jui,j − 3λu3 i,j(cid:19) ,(13) + 2Λu2 i,jwi,j (14) − c2 2 m2 φcwi,j(cid:19) , for i = 2, 3, · · · , N − 1 and j = 2, 3, · · · , M − 1. In order to calculate all values of Eqs. (13) and (14), the initial values for two lowest rows in Fig. 1 must be given. On the other hand, the value at t1 is fixed by the boundary conditions in Eq. (12). The second order of Taylor expansion can also be used to determine the values in the second row. Therefore, the values at t2 are determined by, ui,2 = fi − ǫgi + wi,2 = pi − ǫqi + for i = 2, 3, · · · , N − 1. c2ǫ2 2 (cid:18) fi+1 − 2fi + fi−1 2 (cid:18) pi+1 − 2pi + pi−1 δ2 δ2 c2ǫ2 + 2Λp2 i fi − 3λf 3 + 2Λf 2 i pi − i (cid:19) , c2 φcpi(cid:19) , 2 m2 (15) (16) For the initial stage, suppose the nonlinear sources has a particular form f (x) = 2sech(2x) ei2x and g(x) = 1 to generate the α-helix, while g(x) = q(x) = 0 for the sake of simplicity. Then, one can obtain the initial values in this case using Eqs. (15) and (16). The subsequent values are generated by substituting the preceeding values into Eqs. (13) and (14). The higher order values can be obtained using iterative procedure. The result is given in Fig. 2. The left figure in each box describes the propagation of nonlinear sources in protein backbone, while the right one shows how the protein is folded. As can be seen in the figure, the protein backbone is initially linear before the nonlinear source injection. As the soliton started propagating over the backbone, the conformational changes appear. It should be remarked that the result is obtained up to the second order accuracy in Taylor expansion. In order to guarantee that the numerical solutions do not contain large amount of truncation errors, the step sizes δ and ǫ are kept small enough. Nevertheless, this should be good approximation to describe visually the mechanism of protein folding. 5. Conclusion An extension of phenomenological model describing the conformational dynamics of proteins is proposed. The model based on the matter interactions among the relevant constituents, namely the conformational field and the nonlinear sources represented as the bosonic fields φc and φs. It has been shown that from the relativistic bosonic lagrangian with φ4 s self-interaction, the nonlinear and tension force terms appear naturally as expected in some previous works.6 November 10, 2018 3:45 WSPC - Trim Size: 9.75in x 6.5in folding 7 Fig. 2. The soliton propagations and conformational changes on the protein backbone inducing protein folding. The vertical axis in soliton evolution denotes time in second, while the horizontal axis denotes its amplitude. The conformational changes are on the (x, y, z) plane. However, the present model has different contour since the EOMs governing the whole dynamics are the linear and nonlinear Klein-Gordon equations. Note that the original model by Berloff deployed the linear Klein-Gordon and nonlinear Schrodinger equations. Moreover, the present model has inhomogenous tension force, in contrast with the homogeneous tension force in the Berloff's model, due to simultaneous solutions of Eqs. (10) and (11). These lead to wrigling folded pathways as shown in Fig. 2 which should be more natural than the homogeneous one. Acknowledgments The authors greatly appreciate fruitful discussion with T.P. Djun throughout the work. AS thanks the Group for Theoretical and Computational Physics LIPI for warm hospitality during the work. This work is partially funded by the Indonesia Ministry of Research and Technology and the Riset Kompetitif LIPI in fiscal year 2010 under Contract no. 11.04/SK/KPPI/II/2010. November 10, 2018 3:45 WSPC - Trim Size: 9.75in x 6.5in folding 8 References 1. C. M. Dobson, Nature 426, p. 884 (2003). 2. S. F. Mingaleev, Y. B. Gaididei, P. L. Christiansen and Y. Kivshar, Europhys. Lett. 59, p. 403 (2002). 3. S. Caspi and E. Ben-Jacob, Phys. Lett. A 272, p. 124 (2000). 4. A. Garcia and J. Onuchi, Proc. Natl. Acad. Sci. USA 100, p. 13898 (2003). 5. J. N. Onuchi and P. G. Wolynes, Curr. Opin. Struct. Biol. 14, p. 70 (2004). 6. N. G. Berloff, Phys. Lett. A 337, p. 391 (2005). 7. A. Sulaiman and L. T. Handoko, J. Compt. Theor. Nanoscience in press (2010). 8. A. Sulaiman, F. P. Zen, H. Alatas and L. T. Handoko, Phys. Rev. E 81, p. 061907 (2010). 9. A. Sulaiman, F. P. Zen, H. Alatas and L. T. Handoko, Int. J. Mod. Phys. A in press (2010). 10. S. Caspi and E. Ben-Jacob, Europhys. Lett. 47, p. 522 (1999). 11. J. H. Mathews and K. D. Fink, Numerical Methods using Matlab 4th Ed. (Prentice- Hall, 2004).
1909.02665	1	1909	2019-09-05T22:58:42	Nanoscale structural alterations in cancer cells to assess anti-cancerous drug effectiveness in cancer treatment using TEM imaging	[ "physics.bio-ph", "physics.med-ph" ]	Understanding the nanoscale structural changes can provide the physical state of cells/tissues. It has been now shown that increases in nanoscale structural alterations are associated with the progress of carcinogenesis in most of the cancer cases, including early carcinogenesis. Anti-cancerous therapies are intended for the growth inhibition of cancer cells; however, it is challenging to detect the efficacy of such drugs in early stages of treatment. A unique method to assess the impact of anti-cancerous drugs on cancerous cells/tissues is to probe the nanoscale structural alterations. In this paper, we study the effect of different anti-cancerous drugs on ovarian tumorigenic cells, using their nanoscale structural alterations as a biomarker. Transmission electron microscopy (TEM) imaging on thin cell sections is performed to obtain their nanoscale structures. The degree of nanoscale structural alterations of tumorigenic cells and anti-cancerous drug treated tumorigenic cells are quantified by using the recently developed inverse participation ratio (IPR) technique. Results show an increase in the degree of nanoscale fluctuations in tumorigenic cells relative to non-tumorigenic cells; then a nearly reverse of the degree of fluctuation of tumorigenic cells to that of non-tumorigenic cells, after the anti-cancerous drugs treatment. These results support that the effect of anti-cancerous drugs in cancer treatment can be quantified by using the degree of nanoscale fluctuations of the cells via TEM imaging. Potential applications of the technique for cancer treatment are also discussed.	physics.bio-ph	physics	Nanoscale structural alterations in cancer cells to assess anti-cancerous drug effectiveness in cancer treatment using TEM imaging Prakash Adhikari,1 Mehedi Hasan,1 Vijayalakshmi Sridhar,2 Debarshi Roy,3† and Prabhakar Pradhan1* 1Department of Physics and Astronomy, Mississippi State University, Mississippi State, Mississippi 39762, USA 2Mayo Clinic College of Medicine, Department of Experimental Pathology, 3Department of Biology, Alcorn State University, Lorman, Mississippi 55905, USA Minnesota, 39096 USA Emails: *PPradhan: [email protected] ; and †DRoy: [email protected] Abstract: Understanding the nanoscale structural changes can provide the physical state of cells/tissues. It has been now shown that increases in nanoscale structural alterations are associated with the progress of carcinogenesis in most of the cancer cases, including early carcinogenesis. Anti-cancerous therapies are intended for the growth inhibition of cancer cells; however, it is challenging to detect the efficacy of such drugs in early stages of treatment. A unique method to assess the impact of anti-cancerous drugs on cancerous cells/tissues is to probe the nanoscale structural alterations. In this paper, we study the effect of different anti-cancerous drugs on ovarian tumorigenic cells, using their nanoscale structural alterations as a biomarker. Transmission electron microscopy (TEM) imaging on thin cell sections is performed to obtain their nanoscale structures. The degree of nanoscale structural alterations of tumorigenic cells and anti-cancerous drug treated tumorigenic cells are quantified by using the recently developed inverse participation ratio (IPR) technique. Results show an increase in the degree of nanoscale fluctuations in tumorigenic cells relative to non-tumorigenic cells; then a nearly reverse of the degree of fluctuation of tumorigenic cells to that of non-tumorigenic cells, after the anti- cancerous drugs treatment. These results support that the effect of anti-cancerous drugs in cancer treatment can be quantified by using the degree of nanoscale fluctuations of the cells via TEM imaging. Potential applications of the technique for cancer treatment are also discussed. Keywords: mesoscopic physics, inverse participation ratio, ovarian cancer, anti-cancerous drugs, tight binding model, structural disorder I. Introduction: TEM imaging and probing nanoscale changes in cancer: TEM imaging is a method where we can probe ~1 nm resolution within the sample, and this has been used for imaging of cells at the nanoscale to see the inner structures of the cells, in general, qualitatively. It has been established that the cancer progression is associated with the nanoscale structural alteration in a cell due to the rearrangements of the building blocks 1 of the cell such as DNA, RNA, and lipids. Therefore, TEM imaging is a good modality to probe cancerous changes in cells at the nanoscale level. Furthermore, the recently developed light wave localization technique, 'inverse participation ratio' (IPR), has shown success in quantifying the degree of structural changes in one parameter, known as the degree of structural disorder [1,2]. In particular, using the IPR method, a TEM image is used to construct a disordered 2D mass matrix, and from this we generate a 2D refractive index matrix. Optical waves are then solved for their eigenvalues and eigenfunctions using the refractive index matrix with closed boundary conditions. The light localization properties are measured by the average inverse participation ratio, <IPR>, and standard deviation of the IPR, σ(IPR), of the eigenfunctions of the light waves in these samples. It is shown that the degree of structural disorder is proportional to the <IPR> or σ(IPR) [3,4]. Therefore, the <IPR> can be used as a measure of the degree of nanoscale structural disorder, and to monitor structural change in cells under diseases condition. The IPR method is a very versatile approach. The IPR method, using TEM imaging, has been recently generalized to study the structural changes in brain and colon cells in chronic alcoholism [5,6]. Furthermore, IPR method also extended to study the molecular specific (DNA, histone, etc.) structural changes in cells by using molecular specific fluorophores and confocal microscopy imaging [7,8]. Ovarian cancer: Ovarian cancer (OC) ranks the 5th in cancer related deaths among women and accounts for more deaths than any other cancer of female reproductive system. The American Cancer Society (ACS), estimated new cases of OC in USA in 2019 would be 22,530, whereas estimated deaths would be 14,000. Most OC cases are diagnosed at a very late stage, of which 51% are diagnosed as stage III and 29% are diagnosed as stage IV [9]. The exact cause that triggers OC is not clearly understood but there are several risk factors such as fertility therapy, late pregnancy, family history, hormone therapy after menopause etc. are associated with the development of OC. Metabolic alterations, suppression of tumor suppressor genes, oncogenic activations are also considered as triggering factors for OC initiation and progression of disease [10,11]. Although initially sensitive to chemotherapy treatment, however majority of the OC patients develop chemo resistant. 10 years survival rate for most patients of all stages of OC is ~30%. Development of chemoresistance, widespread disease during the time of diagnosis and tumor recurrence are the major challenges in the therapeutics of ovarian cancer [12]. In this study, we are focusing on analyzing the impact of novel anti-cancerous drug treatment in the tumor forming OC cell line in vitro. HSulf-1 knockdown OV202 cells are selected for this study for 2 their aggressive tumor forming ability and high proliferation rate [13,14]. This method of analysis is aimed to understand the effect of anti-cancerous drugs on cells in the early phases of treatment. Here we propose a novel approach to assess the impact of anti-cancerous drugs in cancerous cells by quantifying the degree of nanoscale structural disorder. II. Method: Analytical formulation for the inverse participation ratio (IPR) technique from TEM images: TEM experiment has a resolution of ~1nm and can identify the nanoscale architectural alterations inside the cells which take place in normal cells when affected by cancer. These nano-alterations happen in the cells due to the rearrangement of the basic building blocks of the cells, such as DNA, RNA, lipids, macromolecules, etc. This results in mass density fluctuations in the cells. Using a thin slice of cell (~100nm), the mass density variations can be probed by TEM imaging. The IPR calculation is an efficient technique to measure and quantify the cancerous level of aggressiveness in a cell through its mass density fluctuations. A higher <IPR> or σ(IPR) value indicates an increasing amount of the nanoscale mass density fluctuations in cells. The IPR technique is described in detail in earlier publications [1,2,4,5]. However, for a self-sufficient and completeness of this paper, we will describe the IPR technique in brief. The refractive index of a thin cell slice at a point n(x,y) with constant width dz has a voxel of volume dV=dxdydz which can be written as n(x,y) = no+dn(x,y), where no is the average refractive index and dn(x,y) is the fluctuation of refractive index at (x,y) indicated voxel. TEM image intensity at any voxel point (x,y) for a thin cell sample is represented as ITEM(x,y) and can be expressed as ITEM(x,y) = I0TEM + dITEM(x,y), where I0TEM is the average pixel intensity and dITEM(x,y) is the fluctuation part of the pixel intensity. Here, the intensity fluctuation ITEM(x,y) is less than the average intensity I0TEM and similarly, the fluctuation of refractive index dn(x,y) is less than the average refractive index n0. Optical parameter refractive index n(x,y) of the scattering substances is linearly proportional to the mass density of a biological cell for the thin samples [1-2]. Therefore, the intensity of a TEM images is linearly proportional to the mass, M, and refractive index of the voxel: 𝐼𝑇𝐸𝑀(𝑥, 𝑦) ∝ 𝑀(𝑥, 𝑦) ∝ 𝑛(𝑥, 𝑦) (1a) 𝐼0𝑇𝐸𝑀+ 𝑑𝐼𝑇𝐸𝑀(𝑥, 𝑦) ∝ 𝑀0 + 𝑑𝑀(𝑥, 𝑦) ∝ 𝑛0 + 𝑑𝑛(𝑥, 𝑦) (1b) 3 From this, we can calculate optical potential of the voxel point as εi(x,y) to generate an optical lattice: 𝜀𝑖 = 𝑑𝑛(𝑥, 𝑦) 𝑛0⁄ ∝ 𝑑𝐼𝑇𝐸𝑀/𝐼0 (2) Knowing the optical potential at every point, an Anderson disorder tight binding model TBM Hamiltonian [15-17] can be generated as follows: 𝐻 = ∑ 𝜀𝑖𝑖 >< 𝑗 𝑖 + 𝑡 ∑ (𝑖 >< 𝑗 + ⟨𝑖𝑗⟩ 𝑗 >< 𝑖). (3) Where i> and j> are the eigenvectors of i-th and j-th lattice sites, εi(x,y) or simply εi, is the i-th lattice site optical potential energy and t is the overlap integral between sites i and j. The eigenfunctions (Ei's) can now be generated from the above Hamiltonian by its diagonalization. Finally, the average IPR value of the whole samples we can define as [1-4, 18,19]: < 𝐼𝑃𝑅 >𝑁 = 1 𝑁 ∑ 𝑁 𝑖=1 𝐿 ∫ ∫ 𝐸𝑖 0 𝐿 0 4(𝑥, 𝑦)𝑑𝑥𝑑𝑦, (4) where Ei denotes the i-th eigenfunction of the Hamiltonian, N is the total number of potential points on the refractive index matrix (i.e., N=(L/dx)2). It has been shown that the average IPR value, <IPR>, is proportional to the degree of structural disorder Ld=dn×lc, where dn is the std of the all n(x,y) point and lc is the spatial correlation decay length of the n(x,y) over the sample [1,2]. Then, (5a) ⟨𝐼𝑃𝑅⟩ ≡< 𝐼𝑃𝑅 >𝑒𝑛𝑠𝑒𝑚𝑏𝑙𝑒 ~ 𝐿𝑑 = ⅆ𝑛 × 𝑙𝑐 , (𝐼𝑃𝑅) ≡ (𝐼𝑃𝑅)𝑒𝑛𝑠𝑒𝑚𝑏𝑙𝑒 ~ 𝐿𝑑 = ⅆ𝑛 × 𝑙𝑐 . (5b) Our statistical analysis involves calculating the average and standard deviation of the disorder strength of IPR values, i.e. Ld values over the samples. Using this structural disorder strength <IPR> or (IPR) or Ld as a biomarker, we study the structural properties of the ovarian cancer cells with anti-cancerous drug treatments. In particular, we expect an increase in the structural disorder with the growth of cancer, and the reversibility of the structural disorder when the cancerous cells are treated with anti-cancerous drug, if the nanoscale structural disorder is a good cancer stage biomarker. III. Sample Preparation and TEM imaging Ovarian normal and cancer cell lines: OV202 cell line is a low-passage primary ovarian cancer cell line established at the Mayo Clinic [13]. OV202 NTC (expressing HSulf-1) and Sh1 cells (HSulf-1 deficient) 4 are developed by Dr. Shridhar's group at Mayo Clinic and is described earlier elsewhere [14]. Subcutaneous injection of OV202 Sh1 cells resulted in tumor formation in nude mice, whereas HSulf-1 expressing OV202 NTC cells did not form tumor [13]. Both cells were grown in minimum essential medium alpha 1X (Cellgro) supplemented with 20% fetal bovine serum (Biowest) and 1% penicillin- streptomycin (Cellgro). All cells were grown in the presence of 1 μg/ml puromycin as a selection marker for the HSulf-1 shRNA cells were treated with 10 µl of AACOCF3 or MAFP (cPLA2 inhibitors; Cayman chemicals) for 24 hours. Following this treatment, cells were washed twice with PBS and then fixed in Trump's fixative containing 4% formaldehyde and 1% glutaraldehyde in a phosphate buffer pH ~7.3, post-fixed in 1.0% OsO4, dehydrated with ethanol gradation, and transitioned into propylene oxide for infiltration and embedding into super epoxy resin. TEM imaging: Cell samples were fixed in Trump's fixative (pH 7.2) at 4°C overnight, spun down and the supernatant removed. They were re-suspended in agarose which was cooled and solidified. The cells in agarose were then post-fixed in 1% OsO4, dehydrated through a graded series of ethanol and embedded in Spur resin. 100nm (or 0.1 μm) ultra-thin sections were mounted on 200-mesh copper grids, post-stained with lead citrate, and observed under a JEOL JEM-1400 transmission electron microscope at 80kV. IV. Results and discussions: The TEM images of the ovarian cancer cells are obtained as described in the above section. IPR analyses were performed for the samples on 165nm165nm TEM images. IPR averaging over the single cells, as well as over the different cells, were performed for ensemble averaging. As discussed above, the average <IPR> and σ(IPR) value for each TEM image was calculated and provides the degree of the structural disorder strength at a defined length scale. Figure 1(a)-(d) are the representative grayscale TEM images of a thin section (~100nm) of cell from the following ovarian cell lines: (i) non-tumorous NTC, (ii) tumorous Sh1, (iii) Sh1 treated with drug AACOCF3 (Sh1-AACOCF3), and (iv) Sh1 treated with drug MAPF (Sh1-MAFP). For each case study, ~ 8-10 different cells TEM images were taken from the cell line for averaging. Figure 1(a')-(d') are the corresponding (IPR) images of Figure 1(a)-(d), at a length scale of 165nm and sample size of 165165nm2. As can be seen from Figure 1, σ(IPR) images represent different intensities of disorder pattern in the cell line distinctly than conventional grayscale TEM images. In the IPR images, intensities 5 patterns of higher fluctuations in the cells are represented by the red spots and lower intensities with blue. In the figure, it can be seen that NTC Sh1 Sh1-AACOCF3 Sh1-MAFP (a) (b) 2µm (c) 2µm (d) 2µm 2µm (a') (b') (c') (d') TEM Images IPR Images Fig 1: (a)-(d) are the representative TEM images and (a')-(d') are their respective IPR images from ovarian cells of the following: non-tumorous (OV202 NTC); tumorous (OV202 Sh1); AACOCF3 treated tumorous Sh1, Sh- AACOCF3; and MAFP treated tumorous Sh1 Sh1-MAFP. IPR images are distinct from the TEM images. Fig. 2: Variations of the standard deviation σ(IPR(L)) with the increase of length scale L, for cell lines non-tumorigenic (OV202 NTC); tumorigenic (OV202 Sh1); AACOCF3 treated tumorigenic Sh1 (Sh1-AACOCF3); and MAFP treated tumorous Sh1 (Sh1-MAFP). The ensemble averagins were performed over TEM images of ~8-10 different cells. It can be seen that the deviation between NTC and Sh1 started becoming prominent around the length scale ~100 nm. Interestingly, the drug treated Sh1 tumorigenic cells fluctuation degrees reverse to the non-tumorigenic cells. 6 the increasing fluctuation or (IPR) value increases from the less proliferating NTC cells to highly proliferating Sh1cells, and decreasing of the fluctuations or (IPR) values decreases with the treatment of two different anti-cancerous drugs, AACOCF3 and MAFP. The drug effect can be distinctly visualized in the IPR images. Figure 2 shows length (L) dependent fluctuations with the sample size (L×L). We plotted, variations of the standard deviation σ(L) with the increase of sample lengths: L = 41, 82, 123, 165, 206, 247, 288 nm. These lengths are for the cells from the following cell lines: non-tumorigenic (OV202 NTC); tumorigenic (OV202 Sh1); AACOCF3 treated tumorigenic Sh1 (Sh1-AACOCF3); and MAFP treated tumorous Sh1 (Sh1-MAFP). As can be seen from the figure that the deviation in the degree of nano- fluctuations between non-tumorigenic cells NCT and tumorigenic cells Sh1 started becoming prominent around the length scale ~100 nm. Interestingly, the degree of nano-fluctuations of anti-cancerous drugs 7 treated Sh1 tumorigenic cell reverse to that of the non-tumorigenic NTC cells. This confirms the efficacy of these two anti-cancerous drugs. Figure 3 presents the bar graphs of the standard deviation of calculated (IPR) or (Ld) value, of the ovarian cells at the fixed length scale 165nm. The variations are similar at lower sample length scales >165nm, however we have chosen 165nm to show a prominent difference. Statistically, the standard deviation is the more reliable marker than the average, as it only depends on the width of the distribution, irrespective to the mean position. The result shows the standard deviation of the degree of structural disorder (IPR) value increased by 70% from NTC to Sh1 cells. Furthermore, when Sh1 cells were treated with 2 different anti-cancerous drugs, AACOCF3 and MAFP, the (IPR) values decreased by around 60% for AACOCF3 and 50% for MAFP, relative to the (IPR) value of the Sh1 tumorous cells. In particular, with the treatment of the anti-cancerous drug, the structural biomarker parameter (IPR) or Ld value decreased nearly back to the normal value. The normalcy detection of these anti-cancerous drug treated cancerous cells may require further investigations using different modalities. It has been earlier shown that AACOCF3 is a better anti-cancerous agent producing more anti-cancerous effects in OC cells compared to MAFP [20]. It can be seen in Fig. 3 that similar trend of bar graphs which show a reduction in the degree of structural disorder (IPR) value for AACOCF3 (60%) > MAFP(50%), consistent with the known qualities of the drugs, in this length scale. Hence, the quantitative analysis technique, called IPR, quantifies the nanoscale structural disorder (IPR) or Ld, as an important biomarker to study the structural alterations at nanoscale level and has potential to detect the effect of anti-cancerous drugs in carcinogenesis ovarian cancer. IV. Conclusions The nanoscale mass-density fluctuations are quantified with the progression of carcinogenesis, as wells as the effects of two anti-cancerous drugs on non-tumor forming OV202NTC and tumor forming OV202Sh1 cells are studied using TEM imaging and IPR technique. The nanoscale fluctuations are quantified by the std value of the IPR, (IPR), performed over an ensemble of samples. Results show an increase in the nanoscale fluctuations or (IPR) value from non-tumorous NTC to tumorous Sh1 cells. The (IPR) values for two different drugs treated tumorous cells, Sh1-AACOCF3 and Sh1-MAFP, have reduced value of the (IPR) from tumorous cells Sh1 and the reduced values are nearly same to the NTC non-tumorous cells. Earlier IPR analysis of a different cell line has verified the increase of nanoscale structural disorder with 8 the progression of cancer [1,2]. Based on the results presented, we investigate the potential applications of the IPR technique in measuring and quantifying the effectiveness of different anti-cancerous drugs on ovarian cancer treatment. This quantification of effectiveness of anti-cancerous drugs in ovarian cancer treatment could enhance better drugs treatment modalities at its earliest and helps to control the deadly ovarian cancer. Although this study is based on ovarian cancer, however, the technique can be applied to the varieties of cancers to assess the effectiveness of different anti-cancerous drugs in treatment. Acknowledgements This work was partially supported by: NIH R01EB016983 and Mississippi State University (PP); Mississippi INBRE, an Institutional Development Award (IDeA), NIH P20GM103476 (DR); NIH CA106954 and the Department of Experimental Pathology and Laboratory Medicine and the Mayo Clinic (VS). Special thanks to Scott Gamb for helping in TEM imaging in Microscopy and Cell Analysis core at Mayo clinic. References 1. 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). 2. 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). 3. Pradhan, P. & Sridhar, S. Correlations due to Localization in Quantum Eigenfunctions of Disordered Microwave Cavities. Phys. Rev. Lett. 85, 2360 -- 2363 (2000). 4. Pradhan, P. & Sridhar, S. From chaos to disorder: Statistics of the eigenfunctions of microwave cavities. Pramana 58, 333 -- 341 (2002). 5. Sahay P, Shukla PK, Ghimire HM, Almabadi HM, Tripathi V, Mohanty SK, Rao R, Pradhan P. Quantitative analysis of nanoscale intranuclear structural alterations in hippocampal cells in chronic alcoholism via transmission electron microscopy imaging. Physical biology. 2017 Mar 1;14(2):026001. 9 6. Ghimire HM, Shukla P, Sahay P, Almabadi HM, Tripathi V, Nanoscale intracellular mass- density alteration as a signature of the effect of alcohol on early carcinogenesis: A transmission electron microscopy (TEM) study, arXiv:1512.08593, 2015 (https://arxiv.org/abs/1512.08593) 7. Sahay P, Ganju A, Almabadi HM, Ghimire HM, Yallapu MM, Skalli O, Jaggi M, Chauhan SC, Pradhan P. Quantification of photonic localization properties of targeted nuclear mass density variations: Application in cancer‐stage detection. Journal of biophotonics. 2018 May;11(5):e201700257. 8. Sahay P, Almabadi HM, Ghimire HM, Skalli O, Pradhan P. Light localization properties of weakly disordered optical media using confocal microscopy: application to cancer detection. Optics express. 2017 Jun 26;25(13):15428-40. 9. Torre LA, Trabert B, DeSantis CE, Miller KD, Samimi G, Runowicz CD, Gaudet MM, Jemal A, Siegel RL. Ovarian cancer statistics, 2018. CA: a cancer journal for clinicians. 2018 Jul;68(4):284-96. 10. Galluzzi L, Senovilla L, Vitale I, Michels J, Martins I, Kepp O, Castedo M, Kroemer G. Molecular mechanisms of cisplatin resistance. Oncogene. 2012 Apr;31(15):1869. 11. Richardson RB. p53 mutations associated with aging-related rise in cancer incidence rates. Cell cycle. 2013 Aug 1;12(15):2468-78. 12. Li SS, Ma J, Wong AS. Chemoresistance in ovarian cancer: exploiting cancer stem cell metabolism. Journal of gynecologic oncology. 2017 Dec 11;29(2). 13. Staub J, Chien J, Pan Y, Qian X, Narita K, Aletti G, Scheerer M, Roberts LR, Molina J, Shridhar V. Epigenetic silencing of HSulf-1 in ovarian cancer: implications in chemoresistance. Oncogene. 2007 Jul;26(34):4969. 14. Roy D, Mondal S, Wang C, He X, Khurana A, Giri S, Hoffmann R, Jung DB, Kim SH, Chini EN, Periera JC. Loss of HSulf-1 promotes altered lipid metabolism in ovarian cancer. Cancer & metabolism. 2014 Dec;2(1):13. 15. Lee, P.A. & Ramakrishnan, T.V. Disordered electronic systems. Rev. Mod. Phys. 57, 287 -- 337 (1985). 16. 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). 10 17. Kramer, B & Mackinnon, A. Localization -- theory and experiment. Rep. Prog. Phys. 56, 1469 -- 1564 (1993). 18. Prigodin, V.N. & Altshuler, B.L. Long-range spatial correlations of eigenfunctions in quantum disordered systems. Phys. Rev. Lett. 80:9, 1944 (1998). 19. Murphy NC, Wortis R, Atkinson WA. Generalized inverse participation ratio as a possible measure of localization for interacting systems. Physical Review B. 2011 May 31;83(18):184206. 20. Roy D, Mondal S, Khurana A, Jung DB, Hoffmann R, He X, Kalogera E, Dierks T, Hammond E, Dredge K, Shridhar V. Loss of HSulf-1: The missing link between autophagy and lipid droplets in ovarian cancer. Scientific reports. 2017 Feb 7:41977. 11
1806.01463	1	1806	2018-06-05T02:09:54	Femtosecond Photonic Viral Inactivation Probed Using Solid-State Nanopores	[ "physics.bio-ph" ]	We report on the detection of inactivation of virus particles using femtosecond laser radiation by measuring the conductance of a solid state nanopore designed for detecting single virus particles. Conventional methods of assaying for viral inactivation based on plaque forming assays require 24-48 hours for bacterial growth. Nanopore conductance measurements provide information on morphological changes at a single virion level. We show that analysis of a time series of nanopore conductance can quantify the detection of inactivation, requiring only a few minutes from collection to analysis. Morphological changes were verified by Dynamic Light Scattering (DLS). Statistical analysis maximizing the information entropy provides a measure of the Log-reduction value. Taken together, our work provides a rapid method for assaying viral inactivation with femtosecond lasers using solid-state nanopores.	physics.bio-ph	physics	Femtosecond Photonic Viral Inactivation Probed Using Solid-State Nanopores Mina Nazari, 1,4 Xiaoqing Li, 9 Mohammad Amin Alibakhshi, 7 Haojie Yang, 10 Kathleen Souza, 8 Christopher Gillespie,^ 8 Suryaram Gummuluru,6 Björn M. Reinhard, 3,4 Kirill S. Korolev,2,5 Lawrence D. Ziegler,3,4 Qing Zhao, 9 Meni Wanunu,7* Shyamsunder Erramilli, 2,4* Departments of 1Electrical and Computer Engineering, 2Physics, 3Chemistry and 4The Photonics Center, 5Bioinformatics 6Department of Microbiology, Boston University School of Medicine, Boston, MA 02118, United States Program, Boston University, Boston, MA 02115, United States 7 Department of Physics, Northeastern University, Boston, MA 02115, United States 8 Next Generation Bioprocessing, MilliporeSigma, Bedford, MA 01730, United States 9 School of Physics, Peking University, Beijing, P. R. China 10 Department of Mechanics, Southeast University, Nanjing, China. We report on the detection of inactivation of virus particles using femtosecond laser radiation by measuring the conductance of a solid state nanopore designed for detecting single virus particles. Conventional methods of assaying for viral inactivation based on plaque forming assays require 24-48 hours for bacterial growth. Nanopore conductance measurements provide information on morphological changes at a single virion level. We show that analysis of a time series of nanopore conductance can quantify the detection of inactivation, requiring only a few minutes from collection to analysis. Morphological changes were verified by Dynamic Light Scattering (DLS). Statistical analysis maximizing the information entropy provides a measure of the Log-reduction value. Taken together, our work provides a rapid method for assaying viral inactivation with femtosecond lasers using solid-state nanopores. I. INTRODUCTION Existing and emerging viruses are a major threat to human and veterinary public health. The need for safe and reliable inactivation or removal of viruses is universal in antiviral therapies, pharmaceuticals, and viral vaccine development. Conventional pharmaceutical pathogen inactivation methods are quite effective, but they involve substantial collateral damage and have undesirable side effects. [1-3] Chemical-free viral inactivation methods such as ultraviolet (UV) and gamma-irradiation have been used to minimize some of the side-effects. Unfortunately, these methods still adversely affect thermolabile compounds and denature biomolecules of interest in the medium containing the virus. Ultrashort Pulsed lasers (USP) provide new opportunities for chemical-free pathogen disinfection in solution. Photonic methods have the potential to provide an attractive alternative to existing biocides and ionizing radiation techniques. [4-8] Photonic inactivation has been successfully achieved with the focused femtosecond (fs) laser pulses for exposure times of ≥ 1h on sample volumes typically of ≤2 ml. [4-7,9,10] Although the ultrafast laser inactivation method for viral inactivation is fairly well established, a systematic understanding of the inactivation mechanism, which can contribute to the design and optimization of protocols, is currently lacking. Viral inactivation is a complicated process and its outcome highly depends on the specific treatment method. A wide range of biological assays could detect and quantify intact viruses in an ensemble manner which is extremely laborious, time-consuming, and low-sensitive. [11] A different approach is to explore viruses at the single virion level. Different optical methods have been developed to characterize single virus; [12-14] but still there is a need for a technique that is fast, sensitive and uses low sample volumes. Although imaging techniques, such as AFM and TEM, are capable of characterizing viruses with high sensitivity, results will be inevitably affected by the tedious and costly sample preparation steps. Nanometer-sized pores in a membrane offer the capability of electrically detecting molecules in a label-free manner at single-molecule level in a volume as small as a few microliters and detection times as short as a few seconds. Passage of molecules and particles through a nanopore causes transient disruption in the ion current flux through it, from which the size, concentration, and distribution of analytes can be deduced. [15,16] The electrical signal characteristics in a given analyte sample ^ Current address of C. Gillespie: Immunogen, 830 Winter St Waltham MA * [email protected] and [email protected] strongly depend on the analyte passing through the pore, the pore geometry, and the experimental conditions such as pH, ionic strength, applied voltage, temperature, etc. This single-particle electrical sensor has been used for quantifying the conformational properties of proteins, [17- 23] understanding DNA transport [24-26] and detecting small molecules, [27,28] among many other applications. Previous studies have reported the ability of nanopore sensors to detect spherical and icosahedral viruses,[29] virus capsids, [30] the masses and zeta potentials of viruses, [31,32] and to explore the translocation of a stiff, rod-shaped virus. [33] This ability of nanopores to detect the translocation of nanometer size particles motivated us to study the effect of an optical viral therapy on a single virus level, crucial for the preparation of very safe biotherapeutics. In this paper, high incident fs laser intensities of >100 GW/cm2, which are more than 105 times greater than the previous studies, have been used to inactivate viruses, leading to 4-log reduction in viral activity reduction in 1 min irradiation of ~ 2 ml sample volume. This result shows nearly more than four orders of magnitude improvement in treatment time compared to conventional pulsed laser viral inactivation methods.[34] Furthermore, we demonstrate the capability of the nanopore technique to precisely characterize individual viruses, explore how vital viral function is affected by treatment, and quantify how effective this label-free viral inactivation technique is. In light of these points, we investigate the effects of fs laser on inactivated ΦX174 bacteriophage, which has the first sequenced DNA-based phage genome widely used standard for viral clearance, as well as a surrogate for enteric human viruses. [35] By electrically counting virus particles in a small volume sample, we monitor here changes in the physical properties of treated virus samples. Further, we develop a statistics-based method, to monitor the reduction value of viruses using sequential measurements, and compare it with a plaque-forming assay. Moreover, we compare the effect of inactivation on viruses in an ensemble manner and at the single-virus level using Dynamic Light Scattering (DLS) measurement and nanopores, respectively. II. METHODS Femtosecond laser irradiation. A femtosecond laser based upon a Legend Elite Duo (Coherent Inc.) Ti-sapphire regenerative amplifier has been used as the excitation source in this study. The laser produces a continuous train of 35 fs pulses at a repetition rate of 1 kHz. The output of the second harmonic generation system of the laser centered at ~400 nm with energies up to 2.5 mJ was used to irradiate the virus samples. Figure 1a depicts the experimental setup. The laser beam with spot size ~1 cm2 was incident upon a typically 1 cm quartz cuvette containing 2 ml of virus sample while a stirrer was used to homogenize the virus's interaction with the laser beam. For ΦX174 sample with 1012 pfu/ml concentration, the laser treatment is made by exposing 250 µl of viruses in 2 mm cuvette. Typical sample exposure times to the laser beam were 15 min. All experiments were carried out at 22 °C, and all samples were immediately stored at 4°C after irradiation. All experiments have been done in triplicate. Virus sample preparation. ΦX174 samples with 2 × 1012 plaque forming unit (pfu) per ml concentration (Promega TiterMax ΦX174 Bacteriophage) in 0.05 M Sodium Tetraborate were stored at -80°C. Before the experiment, samples were thawed to room temperature, aliquoted, and kept at 4°C. For diluted samples, ΦX174 spiked feed solutions were prepared by serial dilution in Sorenson's buffer to the final concentration of approximately 106 pfu/ml. Infectious plaque assay. To count the ΦX174 in the solution, samples were diluted with Sorensen's buffer dilution blanks, to bring plaque to within a statistically valid range of 30-300 plaques per plate. Samples were assayed in triplicate by adding 0.1 mL of diluted sample and 0.1 mL of host cell suspension to a test tube containing 3 mL of molten (46-48 °C) ΦX174 overlay agar consist of 10 g of tryptone peptone (Difco), 8.5 g of agar (Difco) and 5 g of NaCl per liter of reagent water. Then the solution containing host cell and bacteriophage was vortexed and to the ΦX174 bottom plate agar (2.5 g of NaCl, 2.5 g of KCl, 10 g of tryptone peptone from Difco, 10 g of agar from Difco, and 1 mL of 1 M CaCl2 per liter of reagent grade water) and incubated overnight at 37 °C. Then plaques were counted and the corresponding bacteriophage concentrations were reported as pfu/mL. Sorensen's phosphate buffer (pH 7.3), ΦX174 bottom plate agar, and ΦX174 overlay agar were purchased from Northeast Laboratory Services (Winslow, ME). transferred Nanopore device fabrication and measurement. Our electrical detection system is composed of a nanopore formed in a 50-nm-thick insulating silicon nitride (SiN) membrane. The silicon nitride is deposited on a silicon substrate with 2-micron silicon dioxide previously grown on, which is chemically etched by potassium hydroxide (KOH) to obtain a freestanding membrane. The electron beam of a JEOL 2010F transmission electron microscope (TEM) was finely focused on the membrane in order to make a pore with controlled size. [26] The nanopore chip is then mounted in a fluoropolymer cell that allows electrical measurement of ionic current through the nanopore. The cell is filled with 0.1 M KCl solution (16.1 mS/cm conductivity), buffered to pH 7 using 10 mM tris. The silver-silver chloride (Ag/AgCl) electrodes are inserted in both cis- and trans- chambers, and a DC voltage is applied to flow current and drive charged molecules across the pore. (1) ö ÷ ø LRV = log 10 ae ç è C C U T Here UC is the concentration of the untreated sample and TC the concentration of the treated samples exposed to the laser irradiation as described. Control samples consist of a sample with no laser exposure which was held under refrigeration during the experiment and another sample which experienced the same pipetting and stirring condition as treated sample but without the laser exposure. These control samples never differed significantly and were taken to check for loss of titer in the treated suspension. As represented in Fig. 1b more than 104 reduction values of ΦX174 with 106 pfu/ml concentration is achieved by irradiating 2 ml of virus suspension with 2.5 W femtosecond laser for different treatment times ranging from 1 min to 15 min. The combination of the large laser beam diameter with intensified beam results in fast viral treatment which can overcome the need for long irradiation times, and remove constraints on the corresponding practical implementation. Data acquisition and analysis. A Chimera VC100 (Chimera Instruments LLC) was used for recording the ion current through the nanopore. Data was digitized at 4.17 MS/s, and streamed/saved to the computer at a 1 MHz bandwidth. Prior to analysis, recordings were further filtered using a 100 kHz digital low-pass filter. To verify the pore's stability, before the introduction of a virus sample to the nanopore, several seconds of current were collected to ensure that no event is detected and the baseline current is stable. Three key independent parameters are extracted from the nanopore data: the dwell time of viruses at the pore, td, the fractional current blockade, FI, and the inter-event waiting time, δt, from which virus capture rates can be extracted. Dynamic Light Scattering. In order to obtain information on the size distribution of viruses in an ensemble manner, we performed DLS measurement using Zetasizer Nano S90 from Malvern Corp. This DLS measurement is based on the Brownian motion of spherical particles; using the Stokes- Einstein relation the size is calculated based on measured diffusion constant of particles. the size measurement, ΦX174 virus with 1012 pfu/ml concentration and ΦX174 Virion DNA (New England Biolabs) with 1,000 µg/ml concentration, are diluted by 8-fold and 30-fold respectively in 10mM Tris. An aliquot of about 80 µl of the diluted samples is transferred to the cuvette for DLS analysis with the measurements done at 23 °C. [36] For III. RESULTS AND DISCUSSION in Obtaining extremely high levels of viral clearance is a substantial step the purification of protein-based therapeutics. The presence of even a single virus in the final drug product could be harmful to the consumer's health. To prevent this, implementation of an effective viral inactivation strategy is crucial. USP viral inactivation with greater than 4-log reduction in viral infectivity would enable new chemical free pathogen clearance technology. [34,37,38] The first objective of this study is to extend the work done on USP inactivation of viruses, using a regeneratively amplified laser system. We expedite the reported USP photonic inactivation (>1 hr) [34] to 1 min. In this study, 35 femtosecond pulsed laser irradiation working at ~ 400 nm is used to irradiate 2 ml of ΦX174 bacteriophage for different irradiation times. Inactivation of viruses is measured by virus infectivity assay and the concentrations of the original samples were calculated by multiplying the plate count by the dilution factor, reported as pfu/mL. [39] The final results for the viral inactivation experiments is reported as the Log Reduction Value (LRV) which provides a direct measure of viral inactivation. The LRV was calculated according to: FIG. 1. a) Schematic of experimental setup for photonic viral inactivation. The following abbreviation are used: fs: femtosecond, QC: quartz cuvette, BBO: Beta barium borate. b) Log Reduction Value (LRV) measured for ΦX174 virus with 106 pfu/ml (patterned) and six- orders of magnitude increased concentration (solid) after USP laser irradiation for different exposure time. The next goal is to precisely monitor changes implied on the treated virus on the single virus level. A 20 µl aliquot of ΦX174 with ~1012 pfu/ml is treated for 15 min exposure with the same laser setup. For this low volume of sample, we used a micro quartz cuvette with no stirring. Again, the strong reduction in viral infectivity (LRV>3) was achieved for six- orders of magnitude increased virus concentration (Fig. 1b). As shown in Fig. 2, the ΦX174 viruses are being voltage- driven through a ~38 nm pores made of silicon nitride (SiN). A TEM image of the pore is shown as an inset. The pore size is intentionally chosen close to the virus size to slow down the translocations and allow for accurate measurement of the events. Viruses are electroosmotically driven through the nanopores upon application of a negative bias to the trans chamber. The application of voltage results in a steady-state countercurrent of K+ and Cl- ions across the pore, which produces a stable baseline open pore current, 𝐼". When 0.5 µL of virus sample is added to 50 µL of buffer in the cis chamber, passage of individual viruses through the pore reduces the ionic current, which results in a spike in the current. This volume was sufficient to generate > 2×103 events in 10 s for good statistics in untreated sample. The spike contains information about single virus, which can be extracted using statistical analysis methods. The current is measured by a Chimera VC100 amplifier which streams 1 MHz bandwidth data to a computer at a sampling rate of 4.17 MHz, followed by application of a 100 kHz low-pass filter in software to reduce the high frequency noise dominated by the chip capacitance. Figure 3a shows continuous current traces obtained when untreated ΦX174 viruses were added to the cis chamber at 40 mV applied voltage. Each spike corresponds to transport of a single virus through the ~38 nm pore. The electric signal represents two predominant current levels, the higher one corresponds to open pore current when viruses do not translocate through the pore, 𝐼", and the lower corresponds to virus-occupied level. Based on the measured open pore current which is 1.13 nA, the pore conductance was calculated as G=28.2 nS. The inset shows a magnified view of a randomly selected event corresponding to individual ID , which is the intact ΦX174. The translocation current, difference between the baseline current and the pulse minimum, depends on pore geometry and virus size. Knowing the nanopore diameter as 38 nm, the nanopore length was estimated from the open pore conductance to be ℎ$%%=35 𝑛𝑚. This value, thinner than the overall initial 50 thickness of ℎ$%%=ℎ/3 previously found for the very nm membrane thickness, h, was still thicker than the expected narrow pores. [40] As shown in Fig. 3d, the mean fractional current blockade for untreated ΦX174 virus is measured to be 27%. This number is consistent with the 29% blockade predicted by the following theoretical equation, derived in Supplemental material, for a 30 nm diameter spherical virus: F I = I D I o R b = R o - R b (2a) In which R o = 1 d s p + 4 h eff d sp 2 p is the open pore resistance bR is the nanopore resistance in the blocked state: and R b = 1 d s p + 4 h - eff d sp d 2 p ae ç ç ç è (2b) ö ÷ ÷ ÷ ø 4 2 d p 1 - tan +³ 2 d - sp Here 𝑑 is the virus diameter, d 2 - p ( h eff d ) d 2 d pd is the pore diameter, and 𝜎 is the salt conductivity which for the 100 mM KCl buffer was measured as 16.7 mS/cm using the conductivity meter. FIG. 2. Schematic of the nanopore setup for virus detection. Pores are fabricated in freestanding SiN membranes and an external bias is applied across the membrane to drive ΦX174 viruses, potassium ions (yellow dots) and chloride (red dots) ions through the pore. Inset: TEM image of a fabricated nanopore. (scale bar: 20 nm) The agreement between theoretical and experimental translocation ratio verifies that the untreated viruses maintain their shape integrity during transport through the pore. Additionally, a 1-D drift-diffusion model can be used to describe the dwell time distribution of virus translocation through the nanopores. Fittings the probability density function (PDF) of dwell time distribution ( dt ) with the following equation yields important parameters: poreD , and diffusion constant of viruses inside the nanopore, their drift velocity, two dv . ( ) P t = h eff D tp 4 3 pore d exp v t d 2 ) d h - eff t D 4 d pore (3) ö ÷ ÷ ø - ( ae ç ç è Application of this equation, which is based on the assumption of barrier-free transport, [41] for the untreated 0.5 2 a drastic change in the time trace of treated viruses compared to untreated one which will be explored in more detail. To explore the effect of laser treatment, the scatter plot of fractional current blockade versus dwell-time of treated and untreated viruses at 40 mv voltage is shown in Fig. 3c. Two clear groupings of events are visible noted, corresponding to the treated and untreated viruses which can be visually distinguished by drawing a line as shown in Fig. 3c. A histogram of fractional current blockades and dwell-time distributions for both treated and untreated samples at 40 mv along with generalized extreme value distribution fits to the distributions are shown in Fig. 3 d-e respectively. The untreated sample is centered at FI = 25.04+/-3.1 % with log10(td) = 2.05+/-0.36 µs and the treated sample is centered at FI = 10.4+/-2.7 % with log10(td) =1.3+/-0.12 µs. These two sets of independent parameters (FI and td) clearly show the effect implied on viruses by laser treatment. The FI is related to the size of the particle and hence a strong decrease in the fractional blockade yields the first important information on the effect of fs laser treatment, i.e., the global appearance of 3e) and poreD yields m sµ= / (Fig. m s / . viruses dv 410 -< To gain insight into the virus transport kinetics through this nanopore, we compare the in-pore diffusion coefficient of viruses with coefficient, using where kB is the Stokes−Einstein equation a bulk diffusion Bk T dph 3 D = 2 14 Boltzmann constant, T the absolute temperature, h the viscosity of the solution, and d the hydrodynamic diameter of the virion. Using DLS we measured average diameters for bulkD was ΦX174 of ~30 nm at 23°C, and accordingly, /m sµ . The result shows small reduction in calculated as poreD which can be the consequence of virus-pore hydrodynamic interaction, as previously observed with protein transport through smaller pores. [42] Upon successful detection of untreated viruses using nanopores, we probed the effect of laser treatment on the virus sample. A time trace for translocation of laser treated ΦX174 with LRV>3 (Fig. 1b) is shown in Fig. 3b. There is FIG. 3. Comparison of pre-treatment and post-treatment of ΦX174 virus with a pulsed (35 fs) 800&400 nm laser with an average power of 2.5 W for15 min. a,b) Continuous ion current traces collected using a 40 mv applied voltage low-pass filtered to 100 kHz, Inset: Zoomed=in portion of selected representative events. c) Scatter plot of fractional current blockade percentage, FI (%) at 40 mV (0.1 M KCl, 20°C) vs dwell time, td. d) Histograms of fractional current blockade at 40 mv showing decrease in FI (%) after laser treatment. e) Histograms of dwell time at 40 mv voltage demonstrates faster translocation of treated viruses through the pore. the non-enveloped ΦX174 virus does not remains intact after treatment which is in agreement with previously established data. [6] Interestingly using the capability of our label-free resistive pulse technique, one can monitor the inactivation of viruses by looking for presence of intact viruses in the solution with high sensitivity. One simple model is to consider an ellipse in the scatter plot of untreated events centered at the average value of fractional current in y direction and the average of log10(td) in x direction. The semi- minor and semi-major axis radius of this ellipse is equal to three times the standard deviation of the FI and log10(td) of untreated events, respectively. (Fig. 4a). The standard deviations are obtained based on the Gaussian fits of the corresponding data. Counting the number of the treated (red) data points in the black ellipse gives us a rough estimate of the number of intact viruses in the treated sample, n. The number of points in untreated and treated scatter plot is 2,723 and 2,564, respectively. The ratio of the n to the total number of red data points of the treated sample, N , gives us an estimation of the survival fraction (r) (4) r = f n I f N T where fI and fT are the capture probabilities of intact and treated virus. In our case by assuming these probabilities to be equal, the survival fraction is approximately 0.02. This quantity can provide a simple method for rapidly estimating the efficiency of the inactivation technique. A more precise way to estimate the r using nanopore data is to determine an upper bound for the number of survived viruses after treatment. To find the upper bound we developed a statistical formula that made use of a probability distribution function derived from Fig. 3d. The first step is to estimate the probability of finding a translocation event of the Tx inside the treated sample with a distinguishing feature untreated sample region with Ux . In a simplest case Tx and Ux can be any feature which separates the two populations; in our case it can be fractional current blockade, dwell time or any other combination of these two. For example if we consider these one dimensional scaler as the fractional current blockade, in our case there is a big separation between the mean of untreated events, Ux , and treated ones, Tx , so that x . Some of the viruses in the treated sample could be T intact. We can quantify this possibility by considering the ) that an untreated virus has a value of the probability TP x I Tx , which is the cumulative distribution feature as low as ) function for the untreated sample: . Thus, for the treated data points that are far away from the untreated events, the result would be zero while for the ones ( CDF x T ( P x I T x< U = ( ) U FIG. 4. a) Scatter plot of fractional blockade of untreated (green) and treated (red) viruses versus event duration, under a driving voltage of 40 mV b) upper bound determination of the survival fraction, inset shows the zoom in portion of the graph. The minimum occurs at ε= 0.2294. which has overlap, the result is more than zero. By assuming all untreated viruses are intact, we can consider a threshold, ε, on the probability PI, and using a Heaviside step function, θ, we decide if an event can be counted as an intact virus (if IP x e³ ) or not (if IP x e< ). So for applying this threshold on all events, we can define a function ρε as equation 5 which is a sum on all treated data points normalized by the number of treated events, N. )T )T ( ( ( q r e N = å i 1 = ( P x I i N ) - ) e (5) So after treatment, there is ρε×N number of events in the treated sample which have the possibility bigger than ε to be counted as the intact viruses. This count can be written as r e ) ´=´ -+- ´ e N n (1 ( N n ) q e (6) Where n in the first part is the actual number of intact viruses in the treated sample, and qeis the possibility of counting a broken virus as an intact one in the treated sample. In this £ , the r which is n/N, will be calculation, since 0 1 qe£ bounded by r £ er 1 - e . By minimizing the over the ε er 1 e- ( ) TP x I one can determine the upper bound for survival fraction, but still there is another point that we need to explore. This value of r has a large error bar due to the shot noise, so that the error in estimating n is proportional to the n . In practice for large enough epsilon there will be no data points satisfying the e³ and the estimated value of ρε would condition of be zero. Thus as a practical approach, we consider a cutoff and vary epsilon from 0 to the maximal value at which 5Ner ´ ³ , where 5 is an arbitrary threshold that controls how accurate this method is. As shown in Fig. 4b in our case, rmax can be estimated as 2.5´10-3 which is similar to the inverse of the reduction value ~103 reported using infectivity assays (Fig. 1b). So these data show that the proposed formulation makes it possible to determine the inactivation efficiency using the nanopore method; while in contrast to the conventional infectivity assays, nanopore technique is capable of probing small damages to the viruses in linear base with high precision. the force Generally electrophoretic driving and electroosmotic flow are the main effects that govern a particle's motion in a nanopore [43]. The first one originates from the force exerted to the charged nanoparticle in an electric field and the second one is due to the viscous drag by the fluid flowing through the charged nanopore in response to the applied voltage. Combining the equations for electrophoretic and electroosmotic transport yields an effective virus velocity v in an external electric field E inside the pore as [44] Where z is zeta potential, η and ε are the viscosity and permittivity of the electrolyte. For our negatively charged viruses attempting to translocate through the nanopore with negative surface charge, these two forces oppose each other and the relative magnitudes of the forces determines the direction and duration of translocations [45]. In our study, electroosmotic flow is a dominant factor, and therefore, the difference in membrane and viruses surface charge plays a dominant role on capture and transport. As shown in Fig. 3e, the dwell time of most of the events in the treated sample are <90 µs, while untreated viruses have much longer dwell times. In addition, upon treatment, the dwell time distribution narrows substantially and the fractional current blockades are reduced, which can be attributed to a morphological change in the virus sample. To further examine the impact of fs laser on virus conformation, we employed DLS to characterize changes in the radius of ΦX174 viruses and its DNA caused by laser treatment. The DLS measured the average diameter of the untreated viruses 30.9±2.3 nm (Fig. 5a), which is in agreement with the ΦX174 size.[46] DLS data entails two useful pieces of information: first it shows that the laser- treated sample no longer possesses a monomodal size distribution and a second peak is observed at 458±56 nm which will be discussed shortly. Another important piece of information can be obtained from the intersection of the correlation curve on the y-axis of the correlogram. This y- intercept is related to the signal-to-noise ratio (SNR) of the measured sample. As shown in Fig. 5b the correlation coefficient for the treated ΦX174 viruses has lower y- intercept of 0.4 compared to the untreated one which is ~0.85. The lower y-intercept in the treated sample can be attributed to the concentration variations in this sample and presence of 458 nm particles. This concentration variation can also be Ev e ( =- z h virus z pore ) (7) FIG. 5. a) DLS size distribution obtained from b) Quasi-elastic light scattering intensity autocorrelation function or correlogram of ΦX174 virus before (untreated) and after irradiation (treated) with IR& blue 35 fs laser pulses for 30 min (average laser power 2.5 W) c) inter-event waiting time obtained from nanopore measurement of the data in Fig. 3 at 40 mv applied bias, fitted by an exponential function. d) Size distribution e) Correlogram of ΦX174 viral DNA. Laser irradiation conditions were identical to (a,b). explored by nanopore measurements. In case of the nanopore measurements, the waiting time between two successive events (δt) is inversely proportional to the concentration of the detectable particles. An exponential fit to the waiting time data (Fig. 5c), indicates that capture rate of the viruses has decreased by more than one order of magnitude after laser treatment at the same voltage bias (40 mV). Capture rates for untreated and treated samples were 52±0.87 s-1 and 3.49±0.13 s-1 respectively. Particles of the second mode in the DLS data of ΦX174 viruses have a hydrodynamic radius that is too large to be detected with our 38 nm nanopore. To investigate the second peak, we performed DLS measurement of ΦX174 virion DNA, the single-stranded viral DNA isolated from purified phage by phenol extraction (New England Biolabs). This measurement indicates weather the second peak is a signature of free DNA ejected from virus under photonic exposure. The DLS measured average diameter of the untreated phenol- extracted DNA was 113.5±30 nm. A part of this variation may be attributed to ~ 15% of the phenol-extracted DNA molecules not being in circular form. The radius of gyration, Rg and hydrodynamic radius, Rh. of free ΦX174 DNA polymer can be estimated from the known persistence length 2 R = g P L c 6 (8) where P is persistence length and Lc is contour length. [47- 49] For ΦX174 DNA with 5386 nucleotides and the persistence length ~4.6 nm for DNA, [50] the Rg and Rh can be estimated as 72 nm and 45.6 nm, respectively. [41,51] So the corresponding hydrodynamic diameter is approximately 91 nm which is consistent with the peak in DLS measurement of untreated DNA. Interestingly, DLS data of laser exposed DNA (Fig. 5d), still contains this peak which argues against complete agglomeration of single-stranded DNA due to laser exposure. Also, as the correlogram of the DNA (Fig. 5e) shows there is little difference in the time autocorrelation function between exposed and unexposed DNA samples. Our observations indicate that the 458 nm peak observed in laser- treated ΦX174 viruses sample is not from the presence of free single-stranded DNA. Based on our data after laser treatment there are three possible morphology changes that can happen to ΦX174 viruses: (i) fragmentation to small pieces, (ii) perforation of viruses that leads to DNA expulsion with a small associated change in the virus shell, and (iii) agglomeration of viral particles (Fig. 6). Nanopore data can be used to evaluate each of the possibilities. The fact that the fractional current blockade decreases after treatment, suggests that the particle diameter gets smaller upon laser exposure. Also if we neglect possible changes in the zeta potential of the viruses, the smaller viruses will translocate faster consistent with the observation of a shorter dwell times in the treated sample. Therefore, the current blockade and the dwell time both point FIG. 6. Schematic of the possible effect of the fs laser treatment on ΦX174 virus. to detection of entities smaller than the original viruses, and support the hypothesis of virus perforation. At the same time, the nanopore capture rate suggests that viruses for the most part have been fragmented to small pieces not detectable by the nanopore conductance. Neither DLS nor nanopore data don't suggest extensive virus agglomeration. Since the differential scattering cross- section for elastic scattering in the Rayleigh regime is proportional to the 6th power of the particle radius with given dielectric properties, DLS may be expected to be sensitive to agglomeration or presence of larger particle species. Additionally, significant agglomeration is expected to lead to a higher intensity at second diffraction mode which is not observed. Also in nanopore data, significant agglomeration of viruses can block the pore which leads to long lasting dwell times in contrast to what is observed in our experiments. So taken together, our data suggest that a small fraction of viruses (7%) get perforated, while a negligible fraction (< 10- 3% based on Rayleigh scattering assumption) get agglomerated, and the remaining largest fraction of the viruses (93%) are fragmented. While DLS provides an ensemble averaged estimate of the particle size distribution, it does not provide information at the single particle level. Our experiments suggest nanopore single-particle conductance measurements to be helpful in detecting intact viruses in small-volume samples (∼0.5 µl) to characterize morphological changes caused by treatment and inactivation. laser IV. CONCLUSION The ultrashort pulsed laser technology presented here can be readily used for the rapid and effective disinfection of viruses in a label free manner. The application of this technology to the disinfection of viruses can lead to some changes in the viruses which needs to be monitored with high precision. Nanopore technique allowed us to detect intact viruses and monitor the effect of fs laser on a model virus in a single molecule level. Analysis of changes in the ionic current through the nanopore provides information about size and physical properties of viruses before and after laser treatment. To the best of our knowledge, this is the first time P. Hellstern, Current opinion in hematology 11, K. T. Tsen, S.-W. D. Tsen, Q. Fu, S. M. Lindsay, that nanopores have been used to monitor changes induced by an inactivation method in viruses. Our data is a promising step towards developing a label-free detection technique that can also be used as an effective method for monitoring the survival fraction of viruses using low sample volume, high precision and fast assay time. ACKNOWLEDGMENTS Grants: Kirill S. Korolev was partially supported by Cottrell Scholar Award (#24010) by the Research Corporation for [1] 346 (2004). G. Rock, Vox sanguinis 100, 169 (2011). [2] J. AuBuchon, ISBT Science series 6, 181 (2011). [3] [4] K.-T. Tsen, S.-W. D. Tsen, C.-L. Chang, C.-F. Hung, T.-C. Wu, and J. G. Kiang, Journal of biomedical optics 12, 064030 (2007). [5] S.-W. D. Tsen, T. Chapa, W. Beatty, K.-T. Tsen, D. Yu, and S. Achilefu, Journal of biomedical optics 17, 128002 (2012). [6] S.-W. D. Tsen, D. H. Kingsley, C. Poweleit, S. Achilefu, D. S. Soroka, T. Wu, and K.-T. Tsen, Virology journal 11, 20 (2014). [7] Z. Li, S. Cope, S. Vaiana, and J. G. Kiang, Journal of biomedical optics 16, 078003 (2011). [8] (2017). [9] Journal of Healthcare Engineering 1, 185 (2010). [10] G. Kiang, Journal of Physics: Condensed Matter 20, 252205 (2008). [11] manual (Academic Press, 1996). [12] G. Daaboul, A. Yurt, X. Zhang, G. Hwang, B. Goldberg, and M. Unlu, Nano letters 10, 4727 (2010). [13] L. Novotny, ACS nano 4, 1305 (2010). [14] Cytometry Part A 65, 140 (2005). [15] Bohn, Chemical Society Reviews 39, 1060 (2010). [16] Selzer, A. Zlotnick, and S. C. Jacobson, Analytical chemistry 87, 699 (2014). [17] Nanopores (Springer, 2011), pp. 129. [18] Protein and peptide letters 21, 256 (2014). [19] [20] Chemical Society 131, 9287 (2009). A. Oukhaled et al., ACS nano 5, 3628 (2011). D. S. Talaga and J. Li, Journal of the American S.-W. D. Tsen, Y.-S. D. Tsen, K. Tsen, and T. Wu, A. Mitra, B. Deutsch, F. Ignatovich, C. Dykes, and K. Tsen, S.-W. D. Tsen, C.-F. Hung, T. Wu, and J. H. O. Kangro and B. W. Mahy, Virology methods B. Ledden, D. Fologea, D. S. Talaga, and J. Li, in A. Piruska, M. Gong, J. V. Sweedler, and P. W. Z. D. Harms, D. G. Haywood, A. R. Kneller, L. J. Li, D. Fologea, R. Rollings, and B. Ledden, C. L. Stoffel, R. F. Kathy, and K. L. Rowlen, M. Nazari et al., Scientific reports 7, 11951 Science Advancement, a grant from the Simons Foundation (#409704) and an award from Gordon and Betty Moore foundation (#6790.08). X.L. and H.Y. were supported by the China Scholarship Council (CSC), and M.A.A. was supported by the National Institutes of Health (R01- HG009186). B. Cressiot, A. Oukhaled, G. Patriarche, M. K. J. Freedman, M. Jürgens, A. Prabhu, C. W. O. Braha, L.-Q. Gu, L. Zhou, X. Lu, S. Cheley, K. J. Freedman, S. R. Haq, J. B. Edel, P. Jemth, S. Carson and M. Wanunu, Nanotechnology 26, D. Branton et al., Nature biotechnology 26, 1146 [21] and M. J. Kim, Scientific reports 3 (2013). [22] Ahn, P. Jemth, J. B. Edel, and M. J. Kim, Analytical chemistry 83, 5137 (2011). [23] Pastoriza-Gallego, J.-M. Betton, L. c. Auvray, M. Muthukumar, L. Bacri, and J. Pelta, ACS nano 6, 6236 (2012). [24] (2008). [25] 074004 (2015). [26] M. A. Alibakhshi, J. R. Halman, J. Wilson, A. Aksimentiev, K. A. Afonin, and M. Wanunu, ACS nano 11, 9701 (2017). [27] Biophysical journal 85, 897 (2003). [28] and H. Bayley, Nature biotechnology 18, 1005 (2000). [29] K. Zhou, L. Li, Z. Tan, A. Zlotnick, and S. C. Jacobson, Journal of the American Chemical Society 133, 1618 (2011). [30] (2011). [31] Analytical chemistry 86, 4637 (2014). [32] Borghs, Analytical chemistry 84, 8490 (2012). [33] Ma, K. Luo, and Q. Liu, Anal. Chem 88, 2502 (2016). [34] 064042 (2009). [35] environmental microbiology 65, 1186 (1999). [36] J. Zou, H. Chen, and Q. Huo, Journal of the American Chemical Society 130, 2780 (2008). [37] Hung, T. Wu, and J. G. Kiang, Virology journal 4, 50 (2007). [38] Tsen, Journal of biomedical science 19, 62 (2012). [39] H. Brough, C. Antoniou, J. Carter, J. Jakubik, Y. Xu, and H. Lutz, Biotechnology progress 18, 782 (2002). K.-T. Tsen et al., Journal of biomedical optics 14, Z. D. Harms et al., Analytical chemistry 83, 9573 X. Liu, Q. Dai, L. Austin, J. Coutts, G. Knowles, S.-W. D. Tsen, T. C. Wu, J. G. Kiang, and K.-T. H. Wu, Y. Chen, Q. Zhou, R. Wang, B. Xia, D. S. S. Thompson and M. V. Yates, Applied and K.-T. Tsen, S.-W. D. Tsen, C.-L. Chang, C.-F. L. Movileanu, S. Cheley, and H. Bayley, N. Arjmandi, W. Van Roy, and L. Lagae, N. Arjmandi, W. Van Roy, L. Lagae, and G. J. Larkin, R. Y. Henley, M. Muthukumar, J. K. P. Waduge, R. Hu, P. Bandarkar, H. Yamazaki, B. G. Huang, K. Willems, M. Soskine, C. Wloka, and [40] M. J. Kim, M. Wanunu, D. C. Bell, and A. Meller, Advanced materials 18, 3149 (2006). [41] Cressiot, Q. Zhao, P. C. Whitford, and M. Wanunu, ACS nano 11, 5706 (2017). [42] Rosenstein, and M. Wanunu, Biophysical journal 106, 696 (2014). [43] G. Maglia, Nature Communications 8, 935 (2017). [44] M. Firnkes, D. Pedone, J. Knezevic, M. Döblinger, and U. Rant, Nano letters 10, 2162 (2010). [45] D. V. Melnikov, Z. K. Hulings, and M. E. Gracheva, Physical Review E 95, 063105 (2017). [46] M. Bayer and R. DeBlois, Journal of virology 14, 975 (1974). [47] (Cambridge University Press, 2012). [48] Doniach, Physical Review E 86, 021901 (2012). [49] K. Rechendorff, G. Witz, J. Adamcik, and G. Dietler, The Journal of chemical physics 131, 09B604 (2009). [50] (2006). [51] Dorfman, Macromolecules 46, 8369 (2013). D. Boal and D. H. Boal, Mechanics of the Cell A. Y. Sim, J. Lipfert, D. Herschlag, and S. G. S. Manning, Biophysical journal 91, 3607 D. R. Tree, A. Muralidhar, P. S. Doyle, and K. D.
1303.6517	1	1303	2013-03-26T14:56:27	Phase Transition and dissipation driven budding in lipid vesicles	[ "physics.bio-ph", "cond-mat.soft", "q-bio.SC" ]	Membrane budding has been extensively studied as an equilibrium process attributed to the formation of coexisting domains or changes in the vesicle area to volume ratio (reduced volume). In contrast, non-equilibrium budding remains experimentally widely unexplored especially when time scales fall well below the characteristic diffusion time of lipids{\tau} . We show that localized mechanical perturbations, initiated by driving giant unilamellar vesicles (GUVs) through their lipid phase transition, leads to the immediate formation of rapidly growing, multiply localized, non-equilibrium buds, when the transition takes place at short timescales (<{\tau}). We show that these buds arise from small fluid-like perturbations and grow as spherical caps in the third dimension, since in plane spreading is obstructed by the continuous rigid gel-like matrix. Accounting for both three and two dimensional viscosity, we demonstrate that dissipation decreases the size scale of the system and therefore favours the formation of multiple buds as long as the perturbation takes place above a certain critical rate. This rate depends on membrane and media viscosity and is qualitatively and quantitatively correctly predicted by our theoretical description.	physics.bio-ph	physics	Phase Transition and dissipation driven budding in lipid vesicles Thomas Franke,1§ Christian T. Leirer1§, Achim Wixforth1, Nily Dan2, Matthias F. Schneider1,3* 1University of Augsburg, Experimental Physics I, D-86159 Augsburg, Germany 2Drexler University, Department of Chemical and Biological Engineering, Philadelphia, USA 3Boston University, Department of Mechanical Engineering, Boston, USA § contributed equally * Corresponding author: Matthias F. Schneider Phone: +49-821-5983311, Fax: +49-821-5983227 [email protected] University of Augsburg, Experimental Physics I, Biological Physics Group Universitaetstr. 1 D-86159 Augsburg, Germany New Address (September 1, 2009): [email protected] Boston University, Dept. of Mechanical Engineering 110 Cummington St Boston, Ma 02215 USA 1 Abstract Membrane budding has been extensively studied as an equilibrium process attributed to the formation of coexisting domains or changes in the vesicle area to volume ratio (reduced volume). In contrast, non-equilibrium budding remains experimentally widely unexplored especially when time scales fall well below the characteristic diffusion time of lipidsτ . We show that localized mechanical perturbations, initiated by driving giant unilamellar vesicles (GUVs) through their lipid phase transition, leads to the immediate formation of rapidly growing, multiply localized, non-equilibrium buds, when the transition takes place at short timescales (<τ ). We show that these buds arise from small fluid-like perturbations and grow as spherical caps in the third dimension, since in plane spreading is obstructed by the continuous rigid gel-like matrix. Accounting for both three and two dimensional viscosity, we demonstrate that dissipation decreases the size scale of the system and therefore favours the formation of multiple buds as long as the perturbation takes place above a certain critical rate. This rate depends on membrane and media viscosity and is qualitatively and quantitatively correctly predicted by our theoretical description. 2 Introduction Lipid bilayer membranes are “soft” two dimensional objects with a width of ~ 4nm and area in the order of 106-1010 nm2. Upon heating, many lipid membranes undergo a transition from a rigid gel phase to a flexible fluid phase with a bending modulus of the order of κ ∼ 15 kbT (1). Due to their high flexibility, fluid lipid membranes display intensive shape fluctuations, which have been studied thoroughly from both a theoretical and an experimental perspective (2) (3) (1). The formation of membrane ‘buds’, which are spherical protrusions extending out of the membrane plane, is a key step in cellular endo- and exocytosis. Qualitatively, similar transitions have been described in model systems as a global shape transformation (3-5) as well as a consequence of intermembrane domain formation in a multi component, fluid-fluid phase separated system (6) (7). The driving force for bud formation in these systems is thought to be the formation of minority domains in a continuous phase matrix, giving rise to a line tension penalty. Escape of the domain into the third dimension (out of the plane) reduces the contact line, and thus the associated line tension penalty (8). Morphological transitions in these systems were induced using different experimental methods, such as osmotically (3), by detergents (9), as well as by temperature (3, 10) and the resulting buds grew on in the range of seconds (10) to minutes (4). In biology, however, small membrane deformations are expected to result from localized perturbations in the lipid bilayer properties and are therefore more conveniently described as a non equilibrium process (11, 12). As a consequence, budding transitions have been studied theoretically in terms of dynamic processes (12-14). Experimentally, however, only one single paper reports on the dynamics of budded vesicles following a temperature jump (10). In contrast to our work, the time scale of the observed process is in the order of seconds, the typical relaxation time of the system. Here, we demonstrate that rapid heating of giant unilamellar lipid vesicles (GUVs) from their gel to fluid phase leads to the formation of dynamic, non-equilibrium buds. The observed growth rate exceeds the typical time scale (diffusion time) of the system by at least an order of magnitude emphasizing that this is a force driven process. The subsequent relaxation into the global equilibrium shape takes place on the same time scale as reported earlier (10). A theoretical approach following Sens (12) reveals that, while the number of buds increases with increasing heating rate and viscosity, the size of the transient buds follows exactly the opposite relation. In other words, while dissipation favours multiple small buds, elastic processes favour single bud formation. This is in excellent agreement with our observations. Materials and Methods 1,2-Dipalmitoyl-sn-Glycero-3-Phosphocholine 1,2-Dimyristoyl-sn-Glycero-3- (DPPC), Phosphocholine (DMPC), 1,2-Dipentadecanoyl-sn-Glycero-3-Phosphocholine (D15PC) and 1,2-Dilauroyl-sn-Glycero-3-Phosphocholine (DLPC) were purchased from Avanti Polar Lipids (Alabaster, Alabama, USA) and used without further purification. Fluorescently labeled 1,2-dihexadecanoyl-sn-glycero-3- Red (Texas DHPE T-Red phosphoethanolamine,triethylammonium salt, was obtained from Invitrogen/Molecular- 3 Probes (Carlsbad, Ca) and Dil18 from Sigma Aldrich (Germany). For all experiments, ultrapure water (18.2MΩ, pure Aqua, Germany) was used. Vesicles were prepared by a standard electroformation method (20). Briefly, lipid solutions, containing 0.2% fluorescent dye, were spread on Indium Tin oxide (ITO) coated glass slides and dried from organic solvent (Chlorophorm). The remaining solvent was removed in a vacuum desiccator for at least 3 h. Two ITO plates were assembled in parallel an separated by a Teflon spacer of 2mm thickness. To initiate lipid swelling medium was added to the films and after subsequent application of an alternating electric field of a frequency of 10Hz and an amplitude 1V/mm, Giant Unilamellar Vesicles (GUVs) formed within 6 hours. GUVs were carefully transferred with a pipette into to experimental chamber containing the same isoosmotic solution as the preparation chamber. Both fluorescent and phase contrast images were collected with a standard CCD camera (Hamamatsu Photonics Deutschland, Herrsching am Ammersee, Germany) coupled to an Axiovert 200M microscope (Zeiss, Oberkochen, Germany). Our experiments were performed in a closed, temperature controlled chamber with optical access from the bottom and the top. The temperature during the experiments was controlled with the aid of a standard heat bath (Julabo, Seelbach, Germany) and temperature measurement was performed with an thermocouple within the experimental chamber. Results We observed the phase transition induced budding of a DPPC vesicle (GUV) immersed in a narrow gap between two heated glass slides as shown in Fig. 1. The initial, wrinkled structure is typical for vesicle membranes in the gel-like phase where lipid mobility is suppressed. Increasing the temperature above the transition temperature leads to some flattening of the membrane surface and the immediate appearance of small domains, or buds, which area grows as shown in Fig. 2. The average growth rate of area increase is estimated to be ΔA n ∼3⋅10-10m²/s from measuring the bud radius dependence on time. Generally, the transition into the fluid phase is associated with an area increase of about 20- 25% (15). Considering the size of the vesicle in our experiment, this corresponds to a total change in membrane area during melting of approximately ΔAtot≈3⋅10-9m² (calculated from the vesicle in the fluid phase because there wrinkles are absent) over a temperature interval of ΔTm≈1°C. At an experimentally controlled heating rate of 5°C/s, we end up with a rate of change in area of λ ≈ 10-8m²/s for the entire vesicle shown in Fig. 1. Dividing this externally “forced” increase in area λ by the observed area growth rate ΔA n of an individual bud (Fig. 2) demands that the total area increase ΔAtot is accommodated by the formation of roughly 30 buds, which is in good agreement with our experiments were we typically found 20-40 buds (Figure 1). It shows that most of the area increase due to the melting process is consumed by formation of the microscopically visible buds. During the melting process the surrounding of the buds begins to melt (an therefore expand) as well. For our estimate, however we assume that the entire increase in membrane area is contained within the buds, which is supported by the observation that buds initially formed to not show any diffusion indicating a “rigid” matrix. This indicates that the process is controlled by the dynamic properties of the system. Furthermore, all buds appear as individual objects and are triggered by local rather than global changes in the membrane properties, in contrast to earlier studies on GUV vesicles from SOPC and bovine brain lipids (3, 4). The experiments were fairly reproducible. However, it is important to note, that not all vesicles exhibit a transition when heated. Approximately 30% of all (freshly prepared) vesicles showing a transition did exhibit a budding transition of the same time and size scale as presented. However, recalling that this is a dynamic process depending on local curvature, local rate of expansion, etc. it is not at all surprising that there is more than just one budding scenario. We believe, that the reason that only 30% but not all vesicles undergo such a transition is a fingerprint of uncontrollable asymmetry in lipid 4 distribution between the inner and outer monolayer, which in turn changes the spontaneous curvature. In multiple component systems, variations in lipid distribution becomes increasingly hard to be experimentally controllable. In the next section, we will clarify the nucleation process of the growing buds which is necessary in order to analyse the dynamics of the process. Fluid Domains and Bud Nucleation It is well know, that small fluid-like domains nucleate in a gel-like matrix during lipid phase transition (16). The fact that we do not observe significant diffusion along bud nucleation (Fig. 1d-f) supports the constellation of a gel-like majority phase with nucleating fluid-like domains. To determine whether the buds are indeed fluid, we added a small (0.1 mol%) amount of the fluorescently labelled lipid DiL18, which preferentially incorporates into the gel-phase. In Figure 3 a series of fluorescence images taken during the heating process of a DLPC-D15PC (1:1) mixture is shown. The darker areas in the images indicate lower dye solubility (note, however, that the dye solubility in the fluid phase is not zero) and therefore a predominately fluid phase, providing support, that indeed growing fluid like domains in a gel- like matrix are the origin of the observed buds in Figure 1 and 2. After nucleation, the melting process and accompanied area increase continues. As noted earlier, the membrane area in the fluid phase increases by ~25% (15). However, no water permeation takes place over the short time scales studied, constraining the overall vesicle volume. Therefore, the increase in area results from the rapidly growing fluid-like domain, which is surrounded by a mechanically rigid gel-like matrix, and leads to a projection out of the vesicle plane, as sketched in Figure 4. Dynamics of Bud Formation Previous experimental studies examined budding on time scales of a few seconds to minutes (3) (4, 10). Here, however the change in area takes place very rapidly, leading to the formation of single buds of a diameter ∼1µm during ∼ 60ms. Recalling that typical time scales of bulk and membrane diffusion are in the range of seconds on this size scales (1, 4), we expect dissipation to be a major contribution to the total energy of the morphological change. A thorough theoretical model of the dynamics of the budding process is beyond the scope of this work. In principle our theory does not distinguish between intra and extravesicular budding. A local asymmetry between the inner and outer monolayer may create a localized change in spontaneous curvature which will favor either one of the two budding scenarios. Lipid composition, anchoring to the cytoskeleton or differences in the inner and outer bath renter this asymmetry quite likely (21). However, we briefly discuss the role of dissipative terms, following the work of Sens (12) who studied non-equilibrium bud formation due to introduction of a local change in lipid density on one side of a lipid bilayer. Although, in our experiments budding is not induced by a local perturbation in lipid symmetry between the two leaflets of the bilayer membrane as in Sens work, the proposed analysis of energy dissipation still holds for our experimental situation. The change in energy Etot due to the overall increase in area Δ totA during the lipid phase transition is given by the sum of the elastic (bending) and dissipative terms E Pn ∫+ where n is the number of individual buds formed and P the dissipated power. The elastic contributions εB of each individual bud are simply described by the membrane bending energy according to Helfrich (17) n = ε B (2) tot Dis dt , 5 ε B = κ 2 ∫ ⎛ ⎜ ⎝ 22 ⎞ ⎟ R ⎠ dA (3) where κ is the bending modulus, A the area and Rn the radius of the formed spherical cap (Fig. 4). The dissipative term has two contributions: One arising from the flux of volume (12) Pbulk and a second one from the flux of membrane area 22 η 5  V ( 2) a 3 π (4) = =  A ( 2) µ a 3 π . (5) Pmem Here, η represents the bulk viscosity, V and A the change in volume and area respectively. a denotes the neck radius of the bud and µ the lateral membrane viscosity. The fact that only the budding, fluid membranes contribute to the dissipation allows us to tread these terms like a homogenous system Using a Lagrangian description, Sens derived the dynamic equation for bud formation using Equ. (3) – (4). However, it will turn out sufficient for our discussion to focus on the impact of tA /)(=λ t n, the number of buds, and the rate of increase in area on the total energy Etot (Equ. 2), which can be calculated as follows: Assuming all n buds end up in the same shaped spherical shape with radius Rn (Figure 4), the E n 8n elastic contribution (Equ. 3) becomes simply . In order to calculate the = = ε πκ B B dissipation and its dependence on the rate of area-increase, we need to describe the volume of the spherical cap as a function of the constant radius Rn and the momentary area of the cap An(t) (Fig. 4) V n = with the time derivative  V n Inserting this result into Eq. (4) results in the dissipated power of the bud P bulk A nA Rn 4 2 where a≈Rn and have been used. In reality, as a ≤ Rn ,this term Δ = π= tot n n represents a lower limit of the dissipated energy per time, but should produce at least the order of magnitude correctly. Finally, integrating Eq. 8 over the entire time of bud evolution ttot= ΔAtot/λ, and accounting for the fact that volume flux takes place on the inner and outer side of the bud, we arrive at the energy dissipated by volume flux  tAR )( n n tAR )( n n λ . n 2/1 2/1 π 44 45 A Δ tot (7) , ) 1 − (8) (6) 1 3 2 ηλ ( n 2/3 1 3 1 3 R , = = = 6 (9) 2/1 Atot Δ n 2/3 2/1 π = 88 ηλ 45 Dis ε bulk or for all n buds 2/1 A 88 Δ E n (10). Dis Dis tot = = ε ηλ bulk bulk 45 n 2/1 2/1 π Following the same path and using the same approximations, the energy contribution due to area flux results in E or, by using Eq. 2: A = µλ Δ tot (11) Dis mem 88 ηλ 45 2/1 A Δ tot n 2/1 2/1 π tot = + + , (12) n 8 πκ A Δ µλ tot E A A = 25.0 where Δ was applied. ⋅ tot gel In Fig 5, we plot the result of Eq. 12 for different numbers n of buds and bulk viscosities η as a function of growth rate λ. For small velocities, the dissipative terms do not contribute significantly and a single bud is the most likely state. Upon increasing λ, and therefore increasing dissipation, the formation of multiple buds becomes more likely. This effect is even more pronounced when the media viscosity is higher, because bulk dissipation is increased as can be seen from Eq. 8. The model presented here suggests that under dynamic, non-equilibrium conditions, increasing media viscosity supports the formation of multiple, small buds. This is experimentally confirmed and can be seen when comparing bud formation in η≈0,001 Pa⋅s (Figure 1) with bud formation when η≈0,05 Pa⋅s (Figure 6). Clearly, despite the low heating rate of 5°C/min, which favours small bud formation, more buds are formed under high media viscosity conditions. Quantitatively, a transition width of ΔTm ≈ 5°C (DMPC/DPPC) at a heating rate of 5°C/min leads to a rate of expansion λ ≈ 5⋅10−11m²/s as opposed to λ ≈ 10-8m²/s for the low viscosity experiment (Fig. 1). Comparing λ for the high and low viscosity case to the model predictions (see Fig. 5), we find that our simple non-equilibrium description correctly predicts the formation of multiple individual buds as in fact observed during our experiments. In this context, it is important to note that in biology, intracellular trafficking takes place under higher media viscosity. Kuimova et al. very recently reported a media viscosity around η≈0,08 Pa⋅s in human ovarian cells (18), a value very close to the viscosity used in our experiments. Relaxation We consider fluid bud formation in a gel matrix as a non-equilibrium phenomenon, since a local perturbation can not relax within the membrane plane (eq. 1): The excess area due to the transition from gel to fluid phase must extend out of the vesicle’s plane. However, as the melting progresses, the entire membrane becomes more fluid, and the vesicle is free to relax into its equilibrium state defined by the minimal bending energy (5) at a fixed area to volume ratio (reduced volume ≈ 0.7). A typical relaxation process following non-equilibrium budding is shown in Figure 7. After a temperature jump, many buds arise on the surface of the vesicle. When melting is complete, the buds are re-absorbed into the membrane plane (see also Fig. 7 1d-g) leading to a global change in the vesicles morphology defined by its minimal bending energy. Here, the vesicle relaxes into a spherocylindric shape as expected from a reduced volume of vred≈0,7 before leaving the focus of the objective. In conclusion, we present experimental evidence for a thermodynamically driven non- equilibrium budding transition in single and multi component GUVs undergoing a transition from the gel to the fluid phase. We were able to demonstrate that the immediate evolution of localized fluid-like buds in a gel-like matrix arise as a consequence of minimal dissipation and the local mechanical properties, and are not due to global changes in the lipid membrane. The corresponding time scales exceed the diffusion limit underlining the fact that this is a force driven process. By increasing the media viscosity to at least more relevant intracellular media conditions (18), we demonstrate that our theoretical description is in good qualitative agreement with the rate of area perturbation necessary for non-equilibrium bud nucleation. Finally, we like to point out that the interpretation of our findings being caused by the physical properties of localized microdomains and their surrounding matrix, makes it not less likely to be of biological relevance: Biological membranes are highly heterogeneous (19), dynamic systems with local changes in composition and order and therefore also in their mechanical properties, which can trigger not only budding of lipid vesicles, as described but also fission of lipid membranes as has been shown recently (22). However, in biology the appropriate trigger may rather be protein adsorption (or binding) or pH changes then temperature. Acknowledgement Financial support is acknowledged from the DFG (SFB 486), the Cluster of Excellence via NIM. MFS and C.L. acknowledges their support by the Bayrische Forschungsstiftung. References 1. 2. 3. 4. 5. 6. 7. 8. Sackmann, E. 1996. Structure and Dynamics of Membranes. In From Vesicles to Cells. E. Sackmann, and R. Lipowsky, editors, Elsevier: Amsterdam. Lipowsky, R. 1991. The conformation of membranes. Nature 349:475 - 481. Döbereiner, H. G., J. Käs, D. Noppl, I. Sprenger, and E. Sackmann. 1993. Budding and fission of vesicles. Biophys J 65:1396-1403. Döbereiner, H.-G., E. Evans, U. Seifert, and M. Wortis. 1995. Spinodal Fluctuations of Budding Vesicles. Physical Review Letters 75:3360. Miao, L., U. Seifert, M. Wortis, and H.-G. Doebereiner. 1994. Budding transitions of fluid-bilayer vesicles: The effect of area-difference elasticity. Phys Rev E Stat Nonlin Soft Matter Phys 49:5389. Julicher, F., and R. Lipowsky. 1993. Domain induced budding of vesicles. Phys Rev Lett 70:2964-2967. Baumgart, T., S. T. Hess, and W. W. Webb. 2003. Imaging coexisting fluid domains in biomembrane models coupling curvature and line tension. Nature 425:821-824. Towles, K., and N. Dan. 2008. Coupling between line tension and domain contact angle in heterogeneous membranes. BBA 1778:1190-1195. 8 9. 14. 15. 12. 13. 11. 10. Staneva, G., M. Seigneuret, K. Koumanov, G. Trugnan, and M. I. Angelova. 2005. Detergents induce raft-like domains budding and fission from giant unilamellar heterogeneous vesicles: a direct microscopy observation. Chem Phys Lipids 136:55- 66. Li, X. Liang, M. Lin, F. Qiu, and Y. Yang. 2005. Budding Dynamics of Multicomponent Tubular Vesicles. Journal of the American Chemical Society 127:17996-17997. Chernomordik, L. V., and M. M. Kozlov. 2008. Mechanics of membrane fusion. Nature Structural Molecular Biology 15:675. Sens, P. 2004. Dynamics of Nonequilibrium Membrane Bud Formation. Phys Rev Lett 93:108103-108101. Hong, B., F. Qiu, H. Zhang, and Y. Yang. 2007. Budding Dynamics of Individual Domains in Multicomponent Membranes Simulated by N-Varied Dissipative Particle Dynamics. The Journal of Physical Chemistry B 111:5837. Sunil Kumar, P. B., G. Gompper, and R. Lipowsky. 2001. Budding Dynamics of Multicomponent Membranes. Physical Review Letters 86:3911. Heimburg, T. 1998. Mechanical aspects of membrane thermodynamics. Estimation of the mechanical properties of lipid membranes close to the chain melting transition from calorimetry. Biochim Biophys Acta 1415:147-162. 16. Mouritsen, O. G. 1991. Theoretical models of phospholipid phase transitions. Chem Phys Lipids. 57:179-194. Helfrich, W. 1973. Elastic properties of lipid bilayers: theory and possible experiments. Z Naturforsch [C] 28:693-703. Kuimova, M. K., G. Yahioglu, J. A. Levitt, and K. Suhling. 2008. Molecular Rotor Measures Viscosity of Live Cells via Fluorescence Lifetime Imaging. Journal of the American Chemical Society 130:6672-6673. Ole G. Mouritsen, K. J. 1992. Problems and paradigms: Dynamic lipid-bilayer heterogeneity: A mesoscopic vehicle for membrane function? BioEssays 14:129-136. 20. M. I. Angelova and D. S. Dimitrov 1986. Liposome electroformation. Faraday Discuss. Chem. Soc., 81: 303. fission in lipid vesicles BiophysChem 143:106–109. Iglic, A. and Hagerstrand, H. 1999. Amphiphile induced spherical microexovesicle corresponds to an extreme local area difference between two monolayers of the membrane bilayer.Med. Biol. Eng. Comp., 37:125. C. Leirer, B. Wunderlich, V.M. Myles, M.F. Schneider. 2009 Phase transition induced 21. 17. 18. 19. 22. 9 Figure Captions Figure 1: Phase transition induced extravesicluar budding of pure DPPC vesicles: as the temperature is increased from T=35°C to 45°C, the vesicle undergoes a phase transition from gel (a) to the fluid state (b – d). Clearly, small buds of a final size ranging between R≈1-4µm (R=2.8 ± 1.4µm) depart from the mother vesicle and disappear again. The marks in d) - f) point out that originally there is no diffusion of the buds on the vesicle surface (see supporting information: Movie_Fig_1_Budding). Repeating experiments revealed, that more than 30% of all (freshly prepared) vesicles undergoing a morphological transition in the first place, did exhibit a budding transition of the same time and size scale as presented. Figure 2: The dynamics of bud growth. Analyzing 50buds, an average growth takes rate of (1- 3)⋅10-10m²/s has been calculated. Scale bar 7µm. Figure 3: Fluorescence image of bud formation. DiL18 preferentially incorporates into the gel phase revealing that the bud does in fact origin from a fluid-like domain. To follow the process, experiments where performed by slowly increasing the temperature in a DLPC- D15PC (1:1) mixture. The rather broad phase transition regime allows to maintain the coexistence over a longer time period. Figure 4: Domain growth under lateral confinement. The surrounding gel-like matrix does not expand during melting of the fluid domain. The increase in area due to gel phase melting requires an escape in the third dimension, thus forming a spherical cap. The height of the cap, and therefore the contact angle, is set by the projected area of the domain (namely, the area of the gel phase that transitioned into the fluid phase) and the area ratio between the fluid and gel phase, as shown in Eq. 1. Figure 5: Plot of Eq. 12 at different media viscosities and for different number of buds n. Changing the area per molecule at a rate above ∼ 10-10m²/s leads to the formation of multiple buds in order to minimize energy dissipation. When the media viscosity is increased from 1mPa⋅s to 50mPa⋅s, the rate at which dissipation dominates the system is drastically decreased to ∼ 10-12m²/s . Figure 6: Bud formation is increased in a GUV of DMPC/DPPC under high media viscosity (50mPa⋅s) condition. The rate of area increase was λ ≈ 5⋅10−11m²/s. Apart from the formation of more individual buds, the bud size is decreased roughly to the limit of optical resolution. Figure 7: After melting is completed (∼35s), the buds start to reintegrate. Since the volume of the vesicle is constant, we arrive at a new area to volume ratio. The reduced volume in this case is vred≈0,7 (DPPC + T-Red at λ ≈ 10-8m²/s ). 10 Figure 1 11 Figure 2 Figure 3 12 Figure 4 Figure 5 13 Figure 6 Figure 7 14
1812.08248	1	1812	2018-12-19T21:17:21	Piezoelectric Barium Titanate Nanostimulators for the Treatment of Glioblastoma Multiforme	[ "physics.bio-ph", "q-bio.TO" ]	Major obstacles to the successful treatment of gliolastoma multiforme are mostly related to the acquired resistance to chemotherapy drugs and, after surgery, to the cancer recurrence in correspondence of residual microscopic foci. As innovative anticancer approach, low-intensity electric stimulation represents a physical treatment able to reduce multidrug resistance of cancer and to induce remarkable anti-proliferative effects by interfering with Ca2+ and K+ homeostasis and by affecting the organization of the mitotic spindles. However, to preserve healthy cells, it is utterly important to direct the electric stimuli only to malignant cells. In this work, we propose a nanotechnological approach based on ultrasound-sensitive piezoelectric nanoparticles to remotely deliver electric stimulations to glioblastoma cells. Barium titanate nanoparticles (BTNPs) have been functionalized with an antibody against the transferrin receptor (TfR) in order to obtain the dual targeting of blood-brain barrier and of glioblastoma cells. The remote ultrasound-mediated piezo-stimulation allowed to significantly reduce in vitro the proliferation of glioblastoma cells and, when combined with a sub-toxic concentration of temozolomide, induced an increased sensitivity to the chemotherapy treatment and remarkable anti-proliferative and pro-apoptotic effects.	physics.bio-ph	physics	Piezoelectric Barium Titanate Nanostimulators for the Treatment of Glioblastoma Multiforme Attilio Marino1,†,, Enrico Almici2,†, Simone Migliorin2, Christos Tapeinos1, Matteo Battaglini1,3, Valentina Cappello4, Marco Marchetti5,6, Giuseppe de Vito5,7, Riccardo Cicchi5,7, Francesco Saverio Pavone5,6,7, Gianni Ciofani1,2, 1Istituto Italiano di Tecnologia, Smart Bio-Interfaces, Viale Rinaldo Piaggio 34, 56025 Pontedera, Italy 2Politecnico di Torino, Department of Mechanical and Aerospace Engineering, Corso Duca degli Abruzzi 24, 10129 Torino, Italy 3Scuola Superiore Sant'Anna, The Biorobotics Institute, Viale Rinaldo Piaggio 34, 56025 Pontedera, Italy 4Istituto Italiano di Tecnologia, Center for Nanotechnology Innovation, Piazza San Silvestro 12, 56127 Pisa, Italy 5European Laboratory for Nonlinear Spectroscopy (LENS), Via Nello Carrara 1, 50019 Sesto Fiorentino, Italy 6Università di Firenze, Department of Physics and Astronomy, Via Giovanni Sansone 1, 50019 Sesto Fiorentino, Italy 7National Institute of Optics, National Research Council (INO-CNR), Largo Enrico Fermi 6, 50125 Firenze, Italy †These authors contributed equally to this work CORRESPONDING AUTHORS: [email protected]; [email protected] 1 ABSTRACT Major obstacles to the successful treatment of gliolastoma multiforme are mostly related to the acquired resistance to chemotherapy drugs and, after surgery, to the cancer recurrence in correspondence of residual microscopic foci. As innovative anticancer approach, low-intensity electric stimulation represents a physical treatment able to reduce multidrug resistance of cancer and to induce remarkable anti-proliferative effects by interfering with Ca2+ and K+ homeostasis and by affecting the organization of the mitotic spindles. However, to preserve healthy cells, it is utterly important to direct the electric stimuli only to malignant cells. In this work, we propose a nanotechnological approach based on ultrasound-sensitive piezoelectric nanoparticles to remotely deliver electric stimulations to glioblastoma cells. Barium titanate nanoparticles (BTNPs) have been functionalized with an antibody against the transferrin receptor (TfR) in order to obtain the dual targeting of blood-brain barrier and of glioblastoma cells. The remote ultrasound-mediated piezo-stimulation allowed to significantly reduce in vitro the proliferation of glioblastoma cells and, when combined with a sub-toxic concentration of temozolomide, induced an increased sensitivity to the chemotherapy treatment and remarkable anti-proliferative and pro-apoptotic effects. KEYWORDS Barium titanate nanoparticles; piezoelectricity; wireless stimulation; glioblastoma multiforme; blood-brain barrier. 2 INTRODUCTION Despite the dramatic efforts to develop diagnostic and therapeutic tools, the treatment of brain cancer remains a huge challenge in oncology, and successful treatments are still far from being attained. The main obstacles to the successful treatment of brain tumors include i) the structural complexity of the central nervous system, ii) the recurrence of the tumors, and iii) the acquired drug resistance during chemotherapy.1 The most common and detrimental primary brain tumor among adults is represented by glioblastoma multiforme (GBM), a particularly aggressive malignant astrocytoma. Although various treatments are available for GBM, including surgical resection, chemotherapy, and radiation, prognosis remains extremely poor.2 The average survival time following diagnosis of GBM patients is only fourteen months, while the five-year survival rate is about 5%. As alternative anticancer approaches, effective physical treatments based on low-intensity alternating currents (AC) demonstrated great potential for inhibiting the proliferation of different kind of cancer cells without the use of any drug/chemical.3,4 Specifically, AC is known to inhibit cell division by interfering with Ca2+ and K+ homeostasis and with the cytoskeletal components involved in cell division. Low-intensity AC resulted able to enhance the efficacy of a standard chemotherapy drug, temozolomide (TMZ), by reducing multidrug resistance,5 and have been recently tested in combination with TMZ for the treatment of glioblastoma multiforme in clinical trials.6,7 The involved mechanism seems to be mediated by a AC-dependent translocation of the drug transporter P-glycoprotein (P-gp) from the plasma membrane to the cytosol.5 However, healthy brain cells (i.e., human astrocytes) are also sensitive to AC-dependent antiproliferative effects3 and, in this context, the local delivery of electrical stimuli to cancer cells is highly desirable. 3 The rapid development of innovative nanotechnological tools is allowing for the targeting of remote physical stimulations (e.g., thermal, electrical, oxidative, ionic, etc.) in deep tissues.8 In the field of nano-oncology, different nanotransducers have been designed to mediate photothermal, photodynamic, or magnetothermal conversion, and to locally deliver anticancer stimuli at tumor level.9 These nanotechnology-assisted remote stimulation approaches exploit a non-invasive source of energy, such as, for example, alternated magnetic fields and near-infrared radiations, which penetrates the biological tissues and is finally transduced by the nanomaterial into another potential toxic form of energy (e.g., heat). In this context, our group proposed, for the first time in the literature, the remote electric stimulation of living cells mediated by piezoelectric nanoparticles,10,11 an extremely interesting approach for the modulation of cell behavior and activities.12,13 Taking advantage of the direct piezoelectric effect, these nanomaterials have been exploited to convert mechanical into electrical energy.14,15 Electric potentials can be generated by piezoelectric nanoparticles in remote modality by using ultrasounds (US),16 mechanical pressure waves that can be safely and efficiently conveyed into deep tissues. Electro-elastic mathematical models11 allowed to estimate, at nanoparticle level, the magnitude of the output voltage (φoutput ~ 0.5 mV) evoked in response to US intensity IUS = 0.8 W/cm2, while electrophysiological recordings17 and real-time Ca2+/Na+ imaging11 of electrically excitable cells experimentally demonstrated the efficacy of nanoparticle-assisted piezo-stimulation. Recently, our group successfully exploited the antiproliferative effects of nanoparticle-assisted remote electric stimulation as non-invasive "wireless" therapy suitable for inhibiting proliferation of SK-BR3 breast cancer cells.18 Similarly to low-intensity AC, chronic piezo-stimulations resulted able to inhibit cancer cell cycle progression by interfering with Ca2+ homeostasis, by upregulating the gene expression of inward 4 rectifier potassium channels, and by affecting the developing of the mitotic spindles during cell division.18 Associated to the difficulties of treatment of pathologies at the level of the central nervous system, we find the problem of blood-brain barrier (BBB) crossing. The recent growth of nanotechnology promises to revolutionize the delivery of nanomaterials across BBB to brain cancers.19 At first instance, the delivery of different nanomaterials through the BBB at the tumor site can be efficiently obtained by taking advantage of the enhanced permeability and retention (EPR) effect.20 This phenomenon is associated to a highly fenestrated and permeabilized BBB in correspondence of newly formed tumor vessels. A complementary strategy, that appears to be particularly relevant for diagnostic and therapeutic purposes, is the functionalization of nanomaterials with specific ligands to promote their BBB crossing and their targeting to specific cell types or anatomical districts.21 Typical receptors on cancer cell membrane, as the folate, the transferrin, or the epidermal growth factor (EGF) receptors, can be targeted for an efficient delivery of nanostructures to cancer cells. Particular attention has been dedicated to the antibody against the transferrin receptor (anti-TfR Ab), since it can be successfully exploited as a dual- targeting ligand for both enabling the BBB-crossing and the uptake by cancer cells.22-25 In this work, we report the preparation of functionalized piezoelectric nanoparticles for in vitro BBB crossing, active glioblastoma cell targeting, imaging, and remote electric treatment. To this aim, tetragonal crystalline barium titanate nanoparticles (BTNPs) have been chosen as lead-free piezoelectric nanotransducers26 because of their excellent level of biocompatibility,27 high piezoelectric coefficient (d33 ~ 30 pm/V),28 peculiar optical properties,29-31 and possibility to finely control their morphology.32 Finally, the synergic effects of the chronic piezoelectric stimulation combined with sub-toxic TMZ treatment have been in vitro investigated. 5 MATERIALS AND METHODS Nanoparticle functionalization with anti-TfR antibody (AbBTNPs) Non-centrosymmetric piezoelectric barium titanate nanoparticles were purchased by Nanostructured & Amorphous Materials, Inc (nominal nanoparticle size 300 nm in diameter, as indicated by the provider, purity > 99.9%). In the literature, many different dispersing agents have been adopted to obtain a stable dispersion of these nanoparticles, like poly(vinylpyrrolidone) (PVP),33 hexamethylenetetramine (HMT),34 ascorbic acid,35 and ethanolamine.35 In this work, a wrapping with the amphiphilic 1,2-distearoyl-sn-glycero-3- phosphoethanolamine-N-[methoxy(polyethylene glycol)-5000] (DSPE-PEG, Nanocs, purity > 99%) was carried out, as this copolymer allows for easy and straightforward functionalization with many kinds of targeting moieties18. DSPE-PEG was mixed with BTNPs (1:1 w/w) in ddH2O; the mixture underwent sonication with a tip sonicator (8 W for 150 s, Mini 20 Mandelin Sonoplus) and, after a centrifugation step (20 min at 900 rcf, Hettich®Universal 320/320R centrifuge), supernatant containing free DSPE-PEG was discharged. The wrapped nanoparticles were thereafter washed twice in ddH2O and finally re-dispersed at a 5 mg/ml concentration in ddH2O (for electron microscopy imaging), in PBS (for estimation of functionalization efficiency), or in complete cell medium (for stability studies and for biological experiments). Concerning the nanoparticle functionalization with the antibody against the transferrin receptor, BTNPs were firstly coated with biotin-DSPE-PEG (20 % w/w, Nanocs, purity > 95%) and DSPE-PEG (80 % w/w), and subsequently conjugated to streptavidin-Ab anti-TfR (2.5 µg of Ab / mg of BTNPs, Abcore), similarly as described in a previous work.18 Ab-functionalized BTNPs will be indicated in the text as AbBTNPs. DPSE-PEG-coated BTNPs have been used as control and will be indicated in the following as BTNPs for easiness of reading. The non- 6 functionalized plain BTNP powder will be indicated as plain BTNPs. The quantification of Ab functionalization efficiency was carried out through the bicinchoninic acid (BCA) assay following the manufacturers' procedures (enhanced test tube protocol, Thermo Fisher). Nanoparticle size, Z-potential, and polydispersity of AbBTNP and BTNP suspension (100 μg/ml) were characterized by using dynamic light scattering (DLS, Nano Z-Sizer 90, Malvern Instrument); the dynamic measurements of size and polydispersity were performed every ten minutes for two hours. Fourier-transformed infrared spectroscopy (FT-IR) was performed using a Shimadzu Miracle 10 as previously described.36 Multimodal imaging of BTNPs Imaging of BTNPs was performed by using scanning electron microscopy (SEM), second harmonic generation (SHG) microscopy, and confocal laser scanning microscopy (CLSM). A drop of the diluted BTNP dispersion (100 μg/ml) was deposited and let dry on a glass coverslip. SHG imaging of tetragonal crystal lattice of piezoelectric BTNPs was carried out with a multimodal custom-made non-linear microscope using a femtosecond pulsed laser source (Discovery, Coherent Inc.) for excitation. Images were acquired using an excitation wavelength of 800 nm and a 20X water immersion objective lens (XLUM 20X 0.95 NA, Olympus Corporation). SHG signal at 400 nm was collected in the epi-direction using a dichroic filter. Emission spectrum was obtained exciting with a pump-and-probe beam at 810 nm and a Stokes beam at 1060 nm. BTNPs were also detected by CLSM (C2s system, Nikon) with a 642 nm laser (emission collected at 670 nm < λem < 750 nm), as showed elsewhere.11,18,31 BTNP signal from SHG and CLSM images related to the same region of the glass coverslip were obtained and then merged with ImageJ software (https://imagej.nih.gov/ij/). 7 For SEM, the coverslip with the deposited nanoparticles was gold-sputtered at 60 nA for 25 s, and imaging was carried out by using a Helios NanoLab 600i FIB/SEM, FEI. Characterization of the blood-brain barrier model Cultures of immortalized brain-derived endothelioma bEnd.3 cell line (ATCC® CRL-2299™) were seeded at high confluence (seeding density 8·104 cells/cm2) and maintained in proliferative conditions on 3 µm porous transwells (Corning Incorporated) in order to obtain a functional endothelial barrier mimicking the BBB.37 In this configuration, endothelial layer separates the luminal compartment (on the top) from the abluminal compartment (on the bottom). Both the abluminal and luminal compartments were incubated with complete cell medium, composed by Dulbecco's Modified Eagle's Medium (DMEM, Thermo Fisher Scientific), 10% fetal bovine serum (FBS, Gibco), 100 IU/ml penicillin (Gibco), 100 μg/ml streptomycin (Gibco). The development of a functional biological barrier was assessed at day 1, 3 and 6 of proliferation by measuring both FITC-dextran permeability and transendothelial electric resistance (TEER). BBB model permeability was analyzed by incubating the luminal compartment with 200 μg/ml of FITC-dextran (Sigma, molecular weight 70 KDa) and measuring the fluorescence emission (λex = 485 nm, λem = 535 nm, Perkin Elmer Victor X3 UV-Vis spectrophotometer) of the abluminal compartment at different time points (10, 20, 30, 60, 120 min). TEER was assessed with a Millipore Millicell ERS-2 Volt-Ohmmeter device. Resistance across the plain transwell (blank) was subtracted to all the TEER measurements. After the quantitative BBB model characterizations, all the subsequent experiments reported in the text were performed on BBB models at day 3. The qualitative morphological integrity of the BBB models at day 3 and the expression of a specific marker of tight junctions (zonula occludens-1) were respectively verified by Coomassie® Brilliant Blue Staining (BioRad, 0.2% for 5 minutes) 8 and by immunocytochemistry (please refer to the Materials and methods "Immunofluorescence staining"). Investigations of nanoparticle-cell interactions and BBB model crossing BTNPs associated to bEnd.3 cells were observed with SEM imaging combined with energy- dispersive X-ray spectroscopy (EDX). Samples incubated for 30 min with 100 µg/ml BTNPs were washed twice in PBS and fixed with paraformaldehyde (PFA, 4 % in PBS). Subsequently, cells were washed twice with ddH2O and treated with glutaraldehyde solution (2.5 % in ddH2O for 30 min at 4°C) and dehydrated by using progressive ethanol gradients (0 %, 25 %, 50 %, 75 %, and 100 % in ddH2O). Before SEM/EDX imaging (Helios NanoLab 600i FIB/SEM), samples were gold-sputtered as described above. Concerning TEM imaging, samples incubated for 30 min with BTNPs or AbBTNPs were washed twice with PBS, fixed with a solution of 1.5 % glutaraldehyde in sodium cacodylate buffer (0.1 M, pH 7.4) and the pellet treated for epoxy resin embedding. Briefly, cells were post- fixed in 1% osmium tetroxide plus 1% K3Fe(CN)6 at room temperature; then cells were en bloc stained with 3 % solution of uranyl acetate in 20 % ethanol; finally, they were dehydrated and embedded in epoxy resin (Epon 812, Electron Microscopy Science). Polymerization has been performed for 48 h at 60°C. Samples were then sectioned with a UC7 Leica ultramicrotome equipped with a 45° diamond knife (DiATOME), and the slices of 80-90 nm were collected on 300 mesh copper grids. The ultrastructural analysis was performed by using a Zeiss Libra 120 Plus instrument operating at 120 kV equipped with an in-column omega filter. Fluorescence staining of plasma membranes and acidic organelles in living bEnd.3 cells was carried out. For these experiments, bEnd.3 cells were seeded on 35 mm µ-dish (Ibidi) at 8·104 cells/cm2 density for 3 days and then incubated with 100 µg/ml BTNPs / AbBTNPs for 24 and 72 9 h. After nanoparticle treatment, cells were washed in PBS and stained with CellMask Green Plasma Membrane Stain (1:1000 dilution; Invitrogen) or with Lysotracker (50 nM; Invitrogen) following the manufacturers' procedures. Nuclear staining was performed with Hoechst 33342 (1 μg/ml, Invitrogen) in all samples. Finally, cells were washed and incubated with HEPES- supplemented (25 mM) phenol red-free DMEM (Thermo Fisher) supplemented with 10% of FBS for CLSM imaging (C2s system, Nikon). Images were acquired by using the same acquisition parameters for the different experimental classes and were subsequently analyzed with NIS- Elements software (Nikon). Concerning the analysis of nanoparticle internalization, signals of plasma membranes and nanoparticles were selected and then measured upon intensity thresholding. Intersections between the areas of BTNPs / AbBTNPs and plasma membranes or of nanoparticles and intracellular regions were then obtained and expressed as percentages of the total nanoparticle area. Co-localization between acidic organelles and nanoparticles was investigated by assessing Mander's overlap coefficient. 3D reconstruction of z-stack images was carried out by using NIS-Elements software (Nikon). Investigations of nanoparticle internalization were also performed on U-87 cells (ATCC ® HTB-14), a cell line derived from a human primary glioblastoma that is well characterized and commonly used in brain cancer research.38 The composition of the medium used for culturing U- 87 cells was the same of that for bEnd.3 cells (U-87 seeding density 2·104 cells/cm2). Internalization studies were performed by incubating 100 µg/ml of nanoparticles directly on U- 87 cells seeded on 35 mm µ-dish (Ibidi). Alternatively, U-87 cells were seeded in the abluminal compartment of the transwell and 100 µg/ml of nanoparticles were dispersed in cell medium of the luminal compartment. Staining, CLSM imaging, and image analysis were carried out as described above for bEnd.3 cells. SHG imaging of nanoparticle internalization was carried out 10 with the multimodal custom-made non-linear microscope described above, exploiting a pump- and-probe beam at 800 nm and a Stokes beam at 1040 nm. Coherent anti-Stokes Raman spectroscopy (CARS) signal at 650 nm and SHG signal from pump beam at 400 nm were acquired simultaneously in epi-direction. BBB model-crossing was investigated through flow cytometry (CytoFLEX, Beckman Coulter). 100 µg/ml of nanoparticles were incubated in the luminal compartments of a BBB model. At 4 h, 24 h and 72 h of nanoparticle treatment, concentrations of BTNPs / AbBTNPs were assessed in the abluminal compartments. The number of events measured by flow cytometry was then converted to nanoparticle concentrations thanks to a calibration curve obtained at different known concentrations of BTNPs (R2 = 0.997, Figure S1). Chronic ultrasound (US) stimulations and temozolomide (TMZ) treatment US were generated by a KTAC-4000 device (Sonidel) through a tip transducer (S-PW 3 mm diameter tip). Chronic US stimulations were applied with 1 W/cm2 intensity and 1 MHz frequency. Single US stimuli lasted 200 ms and were delivered every 2 s, 1 h per day, for 4 days. This protocol of US treatment was chosen since was not able to detectably increase the temperature of the cell medium neither to affect cell behavior / proliferation.17,18 Concerning TMZ treatment, different concentrations of the drug (0-400 μg/ml) were assessed at 24 and 72 h in order to evaluate TMZ effects. The highest non-toxic concentration (50 μg/ml) was then tested in combination with US stimulations. Cell viability assays Metabolism of cell cultures after the treatment with temozolomide (TMZ, Sigma-Aldrich) and after chronic US stimulation was assessed with WST-1 Assay Reagent ((2-(4-iodophenyl)-3-(4- nitrophenyl)-5- (2,4-disulfophenyl)-2H-tetrazolium sodium salt, BioVision), as previously 11 described.39 Samples were washed twice with PBS and then incubated with the WST-1 reagent (1:10 dilution in complete medium with phenol red-free DMEM, 50 minutes at 37°C). The absorbance of the collected supernatants was measured with a multiplate reader (Perkin Elmer Victor X3 UV-Vis spectrophotometer); the blank, corresponding to the non-specific absorbance of the WST-1 dilution in phenol red-free DMEM, was subtracted from all measurements. Finally, all data were normalized with respect to the non-treated controls. Immunofluorescence staining Immunofluorescence was carried out to detect the expression of the tight junction marker zonula occludens-1 (ZO-1) in the BBB model. PFA-fixed cells were incubated with a 0.1% Triton X-100 solution in PBS (25 min at room temperature) for membrane permeabilization and with 10% goat serum in PBS (1 h at room temperature) as a blocking solution. Subsequently, samples were treated with rabbit IgG primary antibody against ZO-1 (Invitrogen, 1:100 dilution in PBS supplemented with 10% goat serum, 3 h at room temperature) and, after 5 washes with PBS supplemented by 10% goat serum, were incubated with goat Alexa Fluor 488-IgG anti- rabbit secondary antibody (Invitrogen, 1:200 dilution in PBS supplemented with 10% goat serum, 2 h at room temperature). TRITC-conjugated phalloidin (100 μM, Millipore) and Hoechst 33342 (1 μg/ml, Invitrogen) were also included in solution with the secondary antibody in order to stain f-actin and nuclei, respectively. Double immunofluorescence was performed to analyze the expression of the Ki-67 proliferation marker and of the p53 tumor suppressor marker on U-87 cells after 4 days of remote chronic piezoelectric stimulation and TMZ treatment. Immunocytochemistry was performed as described above with a primary mouse monoclonal anti-p53 antibody (Abcam, 1:200), a primary rabbit IgG anti-Ki-67 antibody (Millipore, 1:150), a TRITC-conjugated secondary anti-rabbit 12 antibody (1:200, Millipore), and a FITC-conjugated secondary anti-mouse antibody (1:75, Millipore). Ca2+ imaging Ca2+ imaging was performed during US stimulation, with or without piezoelectric AbBTNPs, taking advantage of Fluo-4 AM Ca2+-sensitive fluorescence dye, similarly as in a previous work.11 Before US stimulation, U-87 cells were stained with Fluo-4 AM (Invitrogen, 1 µM in DMEM for 30 min at 37°C), washed twice with PBS and incubated with HEPES-supplemented (25 mM) phenol red-free DMEM (Thermo Fisher). Fluorescence time-lapse imaging was performed with CLSM (C2s system, Nikon), and obtained images were processed by using ImageJ (http://rsbweb.nih.gov/ij/). The average intracellular fluorescence intensity was defined as F0 at time t = 0 s, and as F for t > 0 s. F/F0 values were calculated for both US and AbBTNPs+US experimental groups and reported in the graph. Statistics For multiple sample comparisons, ANOVA followed by Tukey's HSD post-hoc test was performed by using R software (https://www.r-project.org/); regarding the analysis of nanoparticle internalization in bEnd.3 and U-87 cells, independent two-sample t-tests were carried out by using Excel software. Statistically significant differences among distributions were indicated for p < 0.05. Finally, data were plotted in histograms as average ± standard error by using Excel software. RESULTS The scheme of the experimental design is represented in Figure 1. In Figure 1a the strategy of nanoparticle functionalization is depicted: piezoelectric BTNPs (showed in red) are wrapped with DSPE-PEG and DSPE-PEG-biotin, and subsequently conjugated with strepatavidin-Ab 13 against human TfR to finally obtain AbBTNPs. In Figure 1b the luminal (in red) and abluminal (in light blue) compartments of the in vitro BBB model are showed, where bEnd.3 and U-87 cells are respectively seeded (cell membranes are shown in green, nuclei in blue, and AbBTNPs in red). After 72 h of nanoparticle incubation in the luminal compartment, U-87 cells exposed to nanoparticles that crossed the BBB model have been piezoelectrically stimulated with chronic US treatments, as schematically indicated in Figure 1c. Characterization and imaging of piezoelectric BTNPs Imaging of piezoelectric BTNPs is presented in Figure S2. Figure S2a reports a representative SEM image of the sample. Figure S2b represents the emission spectrum obtained by illuminating the tetragonal crystal lattice of piezoelectric BTNPs with a pair of spatially- and temporally- overlapped laser beams at 810 nm and 1060 nm (pump-and-probe beam and Stokes beam, respectively). Figure S2c shows the multi-modal imaging of piezoelectric BTNPs; signal of BTNPs observed by SHG of the pump beam (in red) co-localizes with that one detected by CLSM (in green). Fourier transformed infrared spectroscopy (FT-IR) was performed in order to verify the successful functionalization of the BTNPs. Starting from the low wavelengths, the peaks in the range 530-600 cm-1 (Figure S3) can be attributed to the Ti-O stretching bond and they are characteristic of the BaTiO3 compound.40 Shifting to higher wavelengths, the peak at 1450 cm-1 that can be seen in spectrum i) of plain BTNPs can be attributed to an impurity of BaCO3 as it has been reported elsewhere.40 The peaks between 1000-1100 cm-1 (spectrum ii)) are attributed to the C-O-C and C-O-H stretching41 vibration of the aliphatic chain of poly(ethylene glycol) (PEG), while the peaks in the range 1600-1670 and 1300-1460 cm-1 (spectrum iii)) can be attributed to the Amide I (C=O stretching)42 and Amide III40 vibrations of the attached anti- 14 transferrin antibody. The peaks at 2280-2400 and 2850-3000 cm-1 (spectra ii) and iii)) are attributed to the C-H stretching bond of the DSPE-PEG while the peak at 3320 cm-1 (spectrum iii)) can be attributed both to the O-H stretching vibration of the DSPE-PEG/TfR antibody as well as to the Amide A (N-H stretching)43 of the TfR antibody. The corresponding vibrations and wavelengths are summarized in Table S1. A small yet significant difference in Z-potential was observed between BTNPs (-29.6 ± 0.8 mV) and AbBTNPs (-22.0 ± 0.6 mV), thus further supporting the hypothesis of the successful functionalization of BTNPs with the Ab. Quantitative measurements of functionalization efficiency indicated an amount of 1.1 ± 0.4 µg of Ab per mg of BTNPs (~ 44% of the Ab used for the reaction successfully linked to BTNPs). Considering the molecular weight of the Ab (~ 90 KDa) and the number of BTNPs per mg of powder (1.2·1010 particles / mg), about 624 ± 227 molecules of Ab were conjugated to each BTNP. Polydyspersity index (PDI) and hydrodynamic diameter (Rd) of BTNPs and AbBTNPs were investigated (Figure S4). The PDI was found stable over time for both BTNPs and AbBTNPs (Figure S4a; 1 measurement / 10 min for 110 min total; for t = 0 min, 0.29 ± 0.05 for BTNPs and 0.25 ± 0.02 for AbBTNPs; for t = 110 min, 0.37 ± 0.04 for BTNPs and 0.37 ± 0.04 for AbBTNPs); in both cases 0.2 < PDI < 0.4, thus indicating a moderate dispersivity.44 Furthermore, hydrodynamic diameter Rd of both samples were stable along the experiment (Figure S4b; 1 measurement / 10 min for 110 min total; for t = 0 min, Rd = 274 ± 1 nm for BTNPs and Rd = 252 ± 11 nm for AbBTNPs; for t = 110 min, Rd = 304 ± 13 nm for BTNPs and Rd = 280 ± 2 nm for AbBTNPs). AbBTNPs efficiently target endothelial-like cells and cross BBB model 15 In order to obtain a functional biological barrier mimicking the BBB, confluent cultures of bEnd.3 cells were maintained in proliferative conditions on 3 µm porous transwell for 1, 3 and 6 days (BBB model characterization is reported in Figure S5). Transendothelial electric resistance (TEER) was measured to assess the performances of the barrier at the different time points (Figure S5a). After 3 and 6 days of maturation, BBB model showed similar TEER levels (41.9 ± 8.9 Ω·cm2 and 48.5 ± 7.4 Ω·cm2 at day 3 and day 6 of culture, respectively), in both cases significantly higher with respect to the 1 day culture (20.1 ± 0.9 Ω·cm2; p < 0.05). The crossing of FITC-dextran through the BBB model (at day 1, 3 and 6 of maturation) is shown in Figure S5b and has been expressed as % of the maximum theoretically achievable fluorescence intensity in the abluminal compartment at different time points (10, 20, 30, 60 and 120 min). BBB model at day 3 and day 6 showed similar permeability to FITC-dextran after 20 min of incubation (19.8 ± 0.6% at day 3 and 17.8 ± 0.4% at day 6), while a significant lower permeability at day 6 of maturation (35.1 ± 1.7%) was observed after 120 min of dextran treatment with respect to both day 1 and day 3 (101.2 ± 2.7% at day 1 and 52.3 ± 2.4% at day 3; p < 0.05). The developing of cell multilayers was also observed at day 6. Considering the good performances of the BBB model at day 3 as well as the scarce mechanical stability and resistance to the shear forces of BBB immediately after day 6 (delamination and cell layer detachments were observed in different cultures starting from day 7-8), nanoparticle-crossing through barrier was tested on BBB starting from day 3. In Figure S5c the Coomassie (left image) and the immunofluorescence staining (right image, ZO-1 in green and nuclei in blue) of the 3-day BBB model are reported; it is possible to appreciate the complete maturation of functional junctions among bEnd.3 cells, that develop a endothelial layer separating the luminal from abluminal compartment of the transwell. 16 Analysis of BTNP / AbBTNPs interacting with bEnd.3 cells and assessment of BBB model crossing are shown in Figure 2. In Figure 2a the SEM imaging and the energy dispersive X-ray analysis (EDX) of BTNPs associated to the plasma membranes of bEnd.3 cells are reported. Qualitatively, TEM observations (Figure 2b) highlighted a higher amount of AbBTNPs associated to plasma membranes and up-taken by bEnd.3 cells with respect to the non- functionalized BTNPs. CLSM of immunofluorescence staining against the ZO-1 marker of tight junctions after 72 h of BTNP / AbBTNP treatment is showed in Figure 2c (nuclei in blue, f-actin in red, ZO-1 in green, nanoparticles in white). CLSM imaging revealed that both BTNPs and AbBTNPs were internalized in bEnd.3 cells; however, increased nanoparticle internalization can be appreciated in samples treated with AbBTNPs. Plasma membrane imaging was carried out at 72 h of nanoparticle treatment (Figure 2d) and showed a higher amount of nanoparticles internalized in cell body with respect to those associated to the plasma membranes (this was observed for both BTNPs and AbBTNPs). Histograms of Figure 2e and 2f show the cell membrane and intracellular areas (%) of bEnd.3 cells co-localizing with BTNPs / AbBTNPs at 24 and 72 h of nanoparticle incubation, respectively. The quantitative analysis demonstrates a significantly higher amount of AbBTNPs, both associated to membranes (1.06 ± 0.27%) and internalized by bEnd.3 cells (2.04 ± 0.30%), with respect to BTNPs (0.30 ± 0.14% associated to plasma membranes and 0.55 ± 0.31% internalized in cells; p < 0.05) at 24 h. Furthermore, the amount of intracellular nanoparticles (both AbBTNPs and BTNPs) decreased from 24 to 72 h (AbBTNPs and BTNPs internalized in cells for 72 h correspond, respectively, to 1.02 ± 0.14% and 0.24 ± 0.05%), likely due to the active transport mechanisms through the bEnd.3 cells (e.g., transcytosis and exocytosis). Despite this decrement, the amount of functionalized nanoparticles internalized in bEnd.3 cells remained significantly higher with respect to BTNPs at 72 h (p < 17 0.05). 3D reconstructions of nanoparticles (BTNPs / AbBTNPs in red) and bEnd.3 plasma membranes (in green) are available in Figure S6 (72 h of incubation). Moreover, co-localization analysis of nanoparticles and acidic cell compartments (i.e., late endosomes and lysosomes) at 4, 24 and 72 h of BTNP / AbBTNP incubation is reported in Figure S7, and showed a progressive accumulation in the acidic organelles of the cells. Higher AbBTNPs co-localization with acidic cell compartment was found with respect to BTNPs at both 24 h (Mander's coefficients were 0.56 ± 0.06 for AbBTNPs and 0.36 ± 0.03 for BTNPs; p < 0.05) and 72 h (Mander's coefficients were 0.78 ± 0.14 for AbBTNPs and 0.36 ± 0.07 for BTNPs; p < 0.05), presumably as a consequence of the higher internalization level with respect to the non-functionalized ones. In order to measure BBB model-crossing, BTNP / AbBTNP were incubated in the luminal compartment of the BBB model and nanoparticle concentrations in the abluminal compartment were measured at 4, 24 and 72 h of nanoparticle treatment. Results reported a progressive BBB- crossing of BTNPs / AbBTNPs at the different time points. Similar BTNP and AbBTNP concentrations were found at 4 h (3.25 ± 0.01 µg/ml and 2.96 ± 0.26 µg/ml respectively for BTNP and AbBTNP) and 24 h (6.26 ± 0.83 µg/ml and 6.72 ± 0.10 µg/ml respectively for BTNP and AbBTNP). Instead, a significantly higher BBB-crossing ability of AbBTNPs was observed at 72 h with respect to non-functionalized nanoparticles (~34% increase: 8.01 ± 0.03 µg/ml and 10.69 ± 0.17 µg/ml respectively for BTNP and AbBTNP; p < 0.05). All together, these results indicated a higher BBB-targeting and BBB-crossing efficiency of the functionalized nanosystem. Dual targeting of AbBTNPs Additionally to the measurements of nanoparticle concentration in the abluminal compartment, the ability of AbBTNPs to efficiently bind glioblastoma cells was tested (Figure 3). CLSM analysis of U-87 cells that were incubated for 24 h with 100 μg/ml BTNPs or AbBTNPs is 18 shown in Figures 3a-b. Interestingly, a higher level of AbBTNPs (1.00 ± 0.23% intracellular nanoparticles and 1.32 ± 0.42% associated to membranes) were found with respect to BTNPs (0.16 ± 0.03% intracellular nanoparticles and 0.26 ± 0.11% associated to membranes; p < 0.05). Qualitative observations with CARS / SHG scans confirmed the higher level of AbBTNP internalization (Figure 3c). Moreover, the CLSM analysis of U-87 cells exposed to nanoparticles that crossed the BBB model was carried out; Figure 3d shows representative CLSM images of U-87 cells cultured in the abluminal compartment after a 72 h treatment with BTNPs or AbBTNPs in the luminal compartment. In Figure 3e, the quantitative co-localization analysis revealed a higher amount of AbBTNPs in the abluminal compartment that are associated to the plasma membranes (1.11 ± 0.35%) and internalized by U-87 cells (0.96 ± 0.25%) compared to BTNPs (0.38 ± 0.15% associated to plasma membranes and 0.16 ± 0.03%; internalized by U-87 cells; p < 0.05), thus demonstrating as the AbBTNPs resulted a successful nanosystem able to cross the in vitro BBB model and to target U-87 cells with higher efficiency with respect to the non-functionalized BTNPs. For this reason, the following experiments have been performed just by using AbBTNPs. Chronic piezoelectric stimulation inhibits proliferation of human glioblastoma cells Inhibitory effects of chronic piezoelectric stimulation on U-87 proliferation are shown in Figure 4. Two concentrations of AbBTNPs have been investigated: 100 µg/ml, already successfully tested on SK-BR3 breast cancer cells,18 and 10 µg/ml, the concentration of nanoparticles able to cross the BBB model after 72 h. The expression of the nuclear proliferation marker Ki-67 has been analyzed through immunofluorescence assays combined with CLSM imaging (Figure 4a). Qualitatively, it is possible to appreciate a lower Ki-67 expression in piezoelectrically-stimulated cells (AbBTNPs+US) with respect to the control cultures (non- 19 stimulated and non-incubated controls, cells incubated with AbBTNPs but non-stimulated with US, cultures stimulated with US without the presence of AbBTNPs); quantitative analysis of Ki- 67+ nuclei (%) are presented in Figure 4b and reported as the lowest proliferation rate was found for AbBTNPs+US cultures incubated with 100 µg/ml of nanoparticles (28.7 ± 2.5%), followed by AbBTNPs+US cultures incubated with 10 µg/ml of nanoparticles (51.1 ± 4.5%). Both these 2 experimental conditions resulted characterized by a significantly lower proliferation rate with respect to all the other control groups (72.3 ± 3.7% for non-stimulated and non-incubated controls, 77.4 ± 3.2% for cells incubated with AbBTNPs but non-stimulated with US, 86.8 ± 1.9% for cultures stimulated with US without the presence of AbBTNPs; p < 0.05). Time-lapse Ca2+ imaging on AbBTNPs+US (10 µg/ml) cultures demonstrated the successful remote activation of the cells (Figure 4c): remarkable long-term Ca2+ waves are observed in response to the US stimulation in the presence of the nanoparticles. The peak of the Ca2+ wave was detected ~ 5 min after starting the US stimulations, and the Ca2+ concentrations remain higher than the basal levels even after 25 min of stimulation. The time lapses video of Ca2+ imaging performed on US and AbBTNPs+US cultures are available as Supplementary Information (Video S1 and S2, respectively). The stability of Ca2+ levels and the regular proliferation rate observed in response to the plain US stimulation (i.e., without the presence of AbBTNPs) support the safeness of the proposed stimulation method. Synergic efficacy of remote piezoelectric stimulation with temozolomide treatment The ability of nanoparticle-assisted piezoelectric stimulation to improve the anticancer efficacy of temozolomide (TMZ) treatment was investigated (Figure 5). Toxic effects of TMZ were assessed by testing different concentrations of the chemotherapy treatment (0-400 μg/ml) at two different time points (24 and 72 h) through WST-1 assay (data are normalized and expressed as 20 percentage of WST-1 absorbance values measured at 24 h on control cultures). Metabolism of U- 87 cultures at 72 h of treatment was significantly affected when treating with concentrations at least of 200 μg/ml (Figure 5a): the best anti-proliferative effects were observed with the highest tested TMZ concentration (400 μg/ml), while first significant effects were observed by using 200 μg/ml. The hypothesis that piezoelectric stimulation could increase the sensitivity of TMZ was investigated by using 50 μg/ml of this chemotherapy drug, the highest concentration that was not effective in our testing conditions (Figure 5b-e). Experimental classes we represented by control cultures, cultures incubated with 10 μg/ml AbBTNPs, cultures incubated with 50 μg/ml TMZ, cultures incubated with 50 μg/ml TMZ and 10 μg/ml AbBTNPs, cultures chronically stimulated with US, cultures stimulated with US in the presence of 10 μg/ml AbBTNPs, cultures stimulated with US in the presence of 50 μg/ml TMZ, and, finally, cultures stimulated with US in the presence of 10 μg/ml AbBTNPs and of 50 μg/ml TMZ. WST-1 assay (Figure 5b) was performed at day 4 on control cultures (100.0 ± 7.2%), cultures incubated with AbBTNPs (101.3 ± 1.7%), cultures incubated with TMZ (97.4 ± 2.4%), cultures incubated with TMZ and 10 μg/ml AbBTNPs (95.3 ± 0.9%), cultures chronically stimulated with US (94.9 ± 4.5%), cultures stimulated with 10 μg/ml AbBTNPs+US (87.8 ± 1.3%), cultures stimulated with US and TMZ (94.7 ± 4.1%), and, finally, cultures stimulated with 10 μg/ml AbBTNPs+US in the presence of TMZ (TMZ+AbBTNPs+US; 72.1 ± 1.7%). Results confirmed the anti-proliferative effects of nanoparticle-assisted piezoelectric stimulation (AbBTNPs+US), that was able to significantly decrease the metabolic activity without the presence of TMZ with respect to the other control conditions (control, AbBTNPs, TMZ, AbBTNPs+TMZ, US, US+TMZ; p < 0.05). However, the major effects were observed by synergistically combining piezo-stimulation with TMZ (TMZ+AbBTNP+US; p < 0.05). 21 The expression of the Ki-67 proliferation marker and of the p53 tumor suppressor marker in response to 50 μg/ml TMZ, 10 μg/ml AbBTNPs+US, and of 10 μg/ml AbBTNPs+US with 50 μg/ml TMZ (TMZ+AbBTNPs+US) were compared with control cultures and are showed in Figure 5c. Qualitatively, a decreased number of cells and a lower Ki-67 expression were found in both AbBTNP+US and TMZ+AbBTNP+US experimental classes, compared to both control and TMZ. This observation is in line with the lowest metabolism levels reported in response to these treatments. Moreover, a higher amount of p53+ nuclei was detected in response to TMZ+AbBTNP+US treatment with respect to the other experimental groups. Quantitatively, Ki- 67+ nuclei in control (72.4 ± 2.8%) and in TMZ-treated (65.8 ± 5.7%) cultures were significantly higher with respect to the cultures treated with AbBTNP+US (49.2 ± 3.7%; p < 0.05) and with TMZ+AbBTNP+US (27.7 ± 2.5%; p < 0.05), the last of which resulted the strongest antiproliferative treatment (p < 0.05; Figure 5d). Higher levels of p53+ nuclei were found in response to the combined TMZ+AbBTNP+US therapy (28.3 ± 6.6%; p < 0.05; Figure 5e) with respect to all the other treatments (1.2 ± 1.3% for AbBTNP+US; 3.4 ± 1.1% for TMZ) and control cultures (1.0 ± 0.7%). Overall, these results indicate that piezoelectric stimulation affects proliferation of U-87 cells and increases their sensitivity to TMZ. Indeed, TMZ therapy at non-toxic concentrations, when combined with chronic piezoelectric treatment, was able to promote cell apoptosis and reducing cell proliferation. DISCUSSION Recent advances in nanobiotechnology are directed to the development of smart and biocompatible sensors / actuators that are able to detect and respond to specific physicochemical conditions in the human body.45-47 Piezoelectric nanomaterials are a promising class of 22 nanostructures, that have been successfully exploited both as mechanical sensors for energy- harvesting and mechanobiology studies, and as nanostimulators for indirect electrical activation of excitable cells.48,49 In this work, we report for the first time the successful crossing of a piezoelectric nanomaterial through a BBB model. Piezoelectric barium titanate nanoparticles used in this study are characterized by a 300 nm diameter size, and resulted able to cross a BBB model with a quite good efficiency; crossing was however improved of ~30% by promoting nanoparticle targeting to BBB cells thanks to surface functionalization with anti-TfR Ab. These results are in line with observations of Wohlfart et al., that reviewed various nanoparticles adopted for the delivery of different drugs into the brain and reported as most of the successfully ones are characterized by a size ranging from 150 to 300 nm.50 Moreover, nanoparticles of 300 nm size are still small enough to passively cross the large defenestrations of the tumor-associated vessels developed during aberrant angiogenesis.11,51 Indeed, the cutoff size of porous blood vessels in most of tumors is 380-780 nm, and 400 nm size nanoparticles are known to efficiently accumulate in the brain tumors.52 These considerations are extremely important in view of exploiting piezoelectric BTNPs for in vivo and preclinical studies, especially considering the potential impact of these nanomaterials in nanomedicine, not only for brain cancer treatment, yet also for the non-invasive electric deep brain treatment of different neurodegenerative pathologies that are characterized by a defenestrated vasculature, such as Parkinson's and Alzheimer's diseases.53 The higher levels of AbBTNPs associated to plasma membranes and internalized in cell body with respect to BTNPs confirm the efficacy of the dual targeting strategy mediated by the antibody against TfR, a receptor highly expressed by the endothelial cells of the neurovascolature54 and by different cancer cells (i.e., glioma, lymphoma, leukemia, breast, lung, 23 bladder).55,56 In agreement with our observations, Cui et al. and Chang et al. exploited TfR targeting to promote the targeting of poly(lactic-co-glycolic acid) (PLGA) nanoparticles to glioblastoma cells, both in vitro and in vivo. In these cited works, PLGA nanoparticles were functionalized with Tf. However, recent researches reported a decrease of specificity of Tf- functionalized nanosystems in biological environment due to the high levels of endogenous free Tf.57 Therefore, following an approach adopted also by other groups,58,59 we performed nanoparticle functionalization with anti-TfR Ab, that does not compete with endogenous Tf for TfR binding. The cell-targeting efficiency of our nanoplatform was investigated by exploiting different imaging techniques, as SEM/EDX, TEM, CLSM and SHG, the last of which represents an advanced imaging technique allowing detecting the crystal asymmetry of BTNP tetragonal lattice. Taking advantage of these imaging approaches, nanomaterial was detected in biological samples without the need of any kind of surface modification (e.g., with fluorophore functionalization or quantum-dot decoration) that can potentially interfere with nanomaterial-cell interaction and with its internalization fate. In this regard, thanks to their peculiar optical properties, non-centrosymmetric BTNPs display a potential impact for cancer theranostics.31 Concerning piezoelectric stimulation, AbBTNP+US treatment resulted able to affect the proliferation of different types of cancer cells, thus suggesting a high versatility of this anticancer approach. Particularly, we observed a remarkable decrease of proliferative U-87 cells after 4 days of chronic piezo-stimulation (from 86.8 ± 1.9% of Ki-67+ nuclei, observed in control cultures, to 28.7 ± 2.5% of Ki-67+ nuclei, when stimulating with 100 µg/ml AbBTNPs+US); the decrease in U-87 cell proliferation in response to piezoelectric stimulation was even more pronounced with respect to that observed on SK-BR-3 cells (from 80 ± 8% of Ki-67+ nuclei observed in control cultures to 56 ± 13% of Ki-67+ nuclei when stimulating with 100 µg/ml AbBTNPs+US). 24 Moreover, the anti-proliferative effects of piezoelectric stimulation resulted preserved, albeit to a lesser extent, when reducing nanoparticle concentration to 10 µg/ml (corresponding to the concentration of nanoparticles that crossed the BBB model after 72 h). No significant increases of apoptotic glioblastoma cells were observed when treating cells only with the piezoelectric stimulation. Instead, the piezo-stimulation approach, when combined with sub-toxic TMZ treatment, was able to significantly increase the percentage of apoptotic cells of about 25% and to further reduce the proliferation rate of the cells with respect to the piezo-stimulation alone. These results demonstrated as the nanoparticle-assisted remote piezoelectric stimulation increases the sensitivity of glioblastoma cells to TMZ treatment. The synergic attack (chemical, thanks to the chemotherapy drug, and physical, thanks to the remote electric stimulation) remarkably reduced the cell number and the metabolic activity of glioblastoma cultures. The remote piezo-stimulation has the potential to improve the therapeutic success by overcoming the main obstacles for brain tumor treatment indicated in the Introduction, and will be tested in more complex in vivo models and in preclinical studies. Particularly intriguing is the future perspective to target also small microscopic foci of the GBM, that are the main cause of the recurrence of the disease. A further point worth of investigation will be the analysis of the effects of nanoparticle size / morphology on anticancer effects; indeed, the size / shape also affect the values of piezoelectric and dielectric susceptibility coefficients,60 thus not allowing to easily and independently control the nanomaterial morphology and its piezoelectric behavior. Summarizing, the main novelties of this research consist in the preparation of piezoelectric nanoceramics able to cross a BBB model, to target glioblastoma cells, and to provide remote electric stimulations for increasing GBM sensitivity to TMZ-based chemotherapy; on the other hand, it is also necessary to underline the limits characterizing the present work, where in vitro 25 models of BBB and GBM were adopted, thus highlighting once more the needing for future in vivo experiments. CONCLUSIONS We presented for the first time the preparation of functionalized piezoelectric BTNPs for BBB- crossing, active cancer cell targeting, imaging, and remote US-driven electric treatment. Moreover, we demonstrated the versatility of this nanotechnological approach, that allows the successful delivery of antiproliferative stimuli to glioblastoma cells. Furthermore, the chronic piezoelectric stimulation, in synergic combination with a sub-toxic concentration of TMZ, induced an increased sensitivity to chemotherapy treatment and remarkable anticancer effects. All together, these findings open new interesting perspectives in nanomedicine, with a potential positive impact for the remote therapy of brain cancer and neurodegenerative conditions. Future works will be focused on investigating the efficacy of nanoparticle-assisted piezo-stimulation in xenograft models, in order to explore the realistic translation of these nanomaterials in the future clinical practice. The possibility to fabricate piezoelectric BTNPs with different size and higher piezoelectric coefficient by maintaining the same level of biocompatibility will be assessed, and the effects of nanoparticle morphology on BBB crossing and on piezo-stimulation efficiency will be evaluated. Moreover, the anticancer performances of remote piezo-stimulation approach will be tested in combination with TMZ for the treatment of TMZ-resistant glioblastoma cells, analyzing the molecular mechanisms at the base of TMZ resistance and sensitivity. Finally, the combination of piezo-stimulation with different anticancer drugs, radiotherapy and hyperthermia is envisaged in order to develop an efficient anticancer protocol for pre-clinical studies. 26 AUTHOR STATEMENT OF CONTRIBUTIONS A.M. performed the nanomaterial functionalization, the piezoelectric stimulation experiments and the Ca2+ imaging and contributed to write the manuscript. E.A. carried out the tests concerning the BBB crossing, and performed immunochemistry and statistical analysis. S.M. performed viability studies and CLSM imaging. C.T. carried out nanomaterial characterization (analysis of nanoparticle size, stability and BCA assay to evaluate the efficiency of nanomaterial functionalization). M.B. developed and characterized the BBB model (TEER analysis, ZO-1 expression and Coomassie® Brilliant Blue Staining). V.C. performed TEM analysis. F.S.P. and R.C. supervised the SHG acquisition. R.C. and M.M. built the SHG setup. M.M. and G.d.V. performed the SHG acquisition and analyzed SHG data. G.C. supervised and planned the whole work and contributed to write the manuscript. All authors have given approval to the final version of the manuscript. ACKNOWLEDGEMENTS This work has received funding from the European Research Council (ERC) under the European Union's Horizon 2020 research and innovation program (grant agreement N°709613, SLaMM). 27 Figure 1. Experimental scheme of a) BTNP functionalization with antibody against transferrin receptor (TfR), b) nanoparticle crossing through a static 2D model of the BBB (nuclei in blue, cell membranes in green and AbBTNPs in red), and c) chronic piezoelectric stimulation of glioblastoma cells. 28 Figure 2. Analysis of BTNPs and AbBTNPs interaction with bEnd.3 cells and assessment of the BBB model crossing. a) SEM imaging and EDX analysis of BTNPs associated to the plasma 29 membranes of bEnd.3 cells (Ba in green and Ti in red). b) TEM image highlighting a higher amount of AbBTNPs associated to plasma membranes and up-taken by bEnd.3 cells with respect to the non-functionalized BTNPs. c) CLSM of immunofluorescence staining of bEnd.3 cells against the ZO-1 marker after 72 h of BTNP / ABTNP treatment (nuclei in blue, f-actin in red, ZO-1 in green and nanoparticles in white). d) CLSM imaging of bEnd.3 plasma membranes (in green), nanoparticles (in red) and nuclei (in blue), after 72 h of nanoparticle treatment. e-f) Histograms reporting intracellular and cell membrane areas (%) co-localizing with BTNPs / AbBTNPs after 24 and 72 h of nanoparticle incubation, respectively. g) Concentrations of BTNPs / AbBTNPs measured in the abluminal compartment after BBB crossing at different time points (4, 24 and 72 h). p < 0.05. 30 Figure 3. BTNP / AbBTNP targeting to glioblastoma cells. a-c) U-87 cells incubated for 24 h with 100 μg/ml BTNPs or AbBTNPs. a) CLSM imaging (plasma membranes in green, nanoparticles in red and nuclei in blue), b) histogram of nanoparticle localization, and c) SHG signal from nanoparticles (in red) overlaid on the outlines of the cells generated from the CARS images. d-e) CLSM analysis of U-87 cells exposed to nanoparticles after BBB model crossing, d) representative CLSM images of U-87 cells cultured in the abluminal compartment after 72 h of BTNP or AbBTNP treatment in the luminal compartment; e) quantitative analysis of experiment depicted in d). * p < 0.05. 31 Figure 4. Inhibitory effects of chronic piezoelectric stimulation on U-87 proliferation by testing different AbBTNP concentrations (100 µg/ml and of 10 µg/ml, the latter corresponding to the concentration of nanoparticles crossing the BBB model after a 72 h treatment). a) CLSM analysis of Ki-67 proliferation marker on control cultures, AbBTNPs-treated cells, US-stimulated cells, and on AbBTNPs+US treated cultures. b) Histogram reporting Ki-67+ nuclei (%). c) Time- lapse Ca2+ imaging in response to plain US and to AbBTNPs+US (10 µg/ml). Images at the top show F/F0 signal of cells after 5 min of US (top left) and AbBTNP+US (top right) stimulations. At the bottom, the graph reports F/F0 traces of cultures stimulated with US (in red) and with US+AbBTNP (in black). * p < 0.05. 32 Figure 5. Nanoparticle-assisted piezoelectric stimulation (AbBTNPs+US) improves anticancer efficacy of temozolomide (TMZ). a) WST-1 assay on U-87 cells incubated for 24 h and 72 h with different concentrations of drug (0-400 μg/ml; data are normalized and expressed as percentage of WST-1 absorbance values measured at 24 h on control cultures). b) WST-1 assay respectively performed on control cultures, cultures incubated with 10 μg/ml AbBTNPs, cultures incubated with 50 μg/ml TMZ, cultures incubated with 50 μg/ml TMZ and 10 μg/ml AbBTNPs, cultures chronically stimulated with US, cultures stimulated with US in the presence of 10 μg/ml AbBTNPs, cultures stimulated with US in the presence of 50 μg/ml TMZ, and, finally, cultures stimulated with US in the presence of 10 μg/ml AbBTNPs and of 50 μg/ml TMZ. c) CLSM imaging of Ki-67 and p53 expression in the different experimental conditions. The histograms reporting the Ki-67+ and p53+ nuclei are respectively showed in d) and e). * p < 0.05. 33 REFERENCES (1) S. S. Kim, J. B. Harford, K.F. Pirollo, E. H. Chang, Biochem. Biophys. Res. Commun., 2015, 468, 485-489. (2) C. Adamson, O. O. Kanu, A. I. Mehta, C. Di, N. Lin, A. K. Mattox, D. D. Bigner, Expert Opin. Invest. Drugs, 2009, 18, 1061-1083. (3) L. Cucullo, G. Dini, K. L. Hallene, V. Fazio, E. V. Ilkanich, C. Igboechi, K. M. Kight, M. K. Agarwal, M. Garrity-Moses, D. Janigro, Glia, 2005, 51, 65 -- 72. (4) E. D. Kirson, Z. Gurvich, R. Schneiderman, E. Dekel, A. Itzhaki, Y. Wasserman, R. Schatzberger, Y. Palti, Cancer Res., 2004, 64, 3288 -- 3295. (5) D. Janigro, C. Perju, V. Fazio, K. Hallene, G. Dini, M. K. Agarwal, L. Cucullo, BMC Cancer, 2006, 6, 72. (6) R. Stupp, S. Taillibert, A. A. Kanner, S. Kesari, D. M. Steinberg, S. A. Toms, L. Taylor, P. F. Lieberman, A. Silvani, K. L. Fink, G. H. Barnett, J. J. Zhu, J. W. Henson, H. H. Engelhard, T. C. Chen, D. D. Tran, J. Sroubek, N. D. Tran, A. F. Hottinger, J. Landolfi, R. Desai, M. Caroli, Y. Kew, J. Honnorat, A. Idbaih, E. D. Kirson, U. Weinberg, Y. Palti, M. E. Hegi, Z. Ram, JAMA, 2015, 314, 2535 -- 2543. (7) R. Stupp, S. Taillibert, A. Kanner, W. Read, D. M. Steinberg, B. Lhermitte, S. Toms, A. Idbaih, M. S. Ahluwalia, K. Fink, F. Di Meco, F. Lieberman, J- Zhu, G. Stragliotto, D. D. Tran, S. Brem, A. F. Hottinger, E. D. Kirson, G. Lavy-Shahaf, U. Weinberg, C- Kim, S- Paek, G. Nicholas, J. Burna, H. Hirte, M. Weller, Y. Palti, M. E. Hegi, Z. Ram, JAMA, 2017, 318, 2306-2316. (8) G. G. Genchi, A. Marino, A. Grillone, I. Pezzini, G. Ciofani, Heathcare Mater., 2017, 6, 1700002. (9) A. Sneider, D. VanDyke, S. Paliwal, P. Rai, Nanotheranostics, 2017, 1, 1-22. 34 (10) G. Ciofani, S. Danti, D. D'Alessandro, L. Ricotti, S. Moscato, G. Bertoni, A. Falqui, S. Berrettini, M. Petrini, V. Mattoli, A. Menciassi, ACS Nano, 2010, 4, 6267 -- 6277. (11) A. Marino, S. Arai, Y. Hou, E. Sinibaldi, M. Pellegrino, Y.-T Chang, B. Mazzolai, V. Mattoli, M. Suzuki, G. Ciofani, ACS Nano, 2015, 9, 7678 -- 7689. (12) A. Marino, G. G. Genchi, V. Mattoli, G. Ciofani, Nanotoday, 2017, 14, 10-13. (13) A. Marino, G. G. Genchi, E. Sinibaldi, G. Ciofani, ACS Appl. Mater. Interfaces, 2017, 9, 17663 -- 17680. (14) X. Wang, J. Liu, J. Song, Z. L. Wang, Nano Lett., 2007, 7, 2475 -- 2479. (15) Y. Zhao, Q. Liao, G. Zhang, Z. Zhang, Q. Liang, X. Liao, Y. Zhang, Nano Energy, 2015, 11, 719 -- 727. (16) X. Wang, J. Song, J. Liu, Z. L. Wang, Science, 2007, 316, 102 -- 105. (17) C. A. Rojas Cifuentes, M. Tedesco, P. Massobrio, A. Marino, G. Ciofani, S. Martinoia, R. Raiteri, J. Neural Eng., 2018, 15, 036016. (18) A. Marino, M. Battaglini, D. De Pasquale, A. Degl'Innocenti, G. Ciofani, Sci. Rep., 2018, 8, 6257. (19) M. Pinto, M. Arce, B. Yameen, C. Vilos, Nanomedicine (Lond)., 2017, 12, 59-72. (20) N. Bertrand, J. Wu, X. Xu, N. Kamaly, O. C. Farokhzad, Adv. Drug Deliv. Rev., 2014, 66, 2-25. (21) M. I. Alam, S. Beg, A. Samad, S. Baboota, K. Kohli, J. Ali, A. Ahuja, M. Akbar, Eur. J. Pharm. Sci., 2010, 40, 385-403. (22) J. Chang, A. Paillard, C. Passirani, M. Morille, J- Benoit, D. Betbeder, E. Garcion, Pharm. Res., 2012, 29, 1495-505. (23) Y. Cui, Q. Xu, P. K. Chow, D. Wang, C. H. Wang, Biomaterials, 2013, 34, 8511-8520. 35 (24) T. Kang, M. Jiang, D. Jiang, X. Feng, J. Yao, Q. Song, H. Chen, X. Gao, J. Chen, Mol. Pharm., 2015, 12, 2947-2961. (25) Z. Pang, W. Lu, H. Gao, K. Hu, J. Chen, C. Zhang, X. Gao, X. Jiang, C. Zhu, J. Control. Release, 2008, 128, 120-127. (26) H. Wei, H. Wang, Y. Xia, D. Cui, Y. Shi, M. Dong, C. Liu, T. Ding, J. Zhang, Y. Ma, N. Wang, Z. Wang, Y. Sun, R. Wei, Z. Guo, J. Mater. Chem. C, 2018, doi: 10.1039/C8TC04515A. (27) G. G. Genchi, A. Marino, A. Rocca, V. Mattoli, G. Ciofani, Nanotechnology, 2016, 27, 232001. (28) Z. Deng, Y. Dai, W. Chen, X. Pei, J. Liao, J. Nanoscale Res. Lett., 2010, 5, 1217-1221. (29) C- Hsieh, R. Grange, Y. Pu, D. Psaltis, Biomaterials, 2010, 31, 2272-2277. (30) J. Čulić-Viskota, W. P. Dempsey, S. E. Fraser, P. Pantazis, Nat. Protoc., 2012, 7, 1618- 1633. (31) A. Marino, J. Barsotti, G. de Vito, C. Filippeschi, B. Mazzolai, V. Piazza, M. Labardi, V. Mattoli, G. Ciofani, ACS Appl. Mater. Interfaces, 2015, 7, 25574 -- 25579. (32) Z. Sun, L. Zhang, F. Dang, Y. Liu, Z. Fei, Q. Shao, H. Lin, J. Guo, L. Xiang, N. Yerrad, Z. Guo, CrystEngComm, 2017, 19, 3288-3298. (33) L. Zhang, W. Yu, C. Han, J. Guo, Q. Zhang, H. Xie, Q. Shao, Z. Sun, Z. Guo, J. Electrochem. Soc., 2017, 164, H651-H656. (34) L. Zhang, M. Qin, W. Yu, Q. Zhang, H. Xie, Z. Sun, Q. Shao, X. Guo, L. Hao, Y. Zheng, Z. Guo, J. Electrochem. Soc., 2017, 164, H1086-H1090. (35) B. Song, T. Wang, H. Sun, Q. Shao, J. Zhao, K. Song, L. Hao, L. Wang, Z. Guo, Dalton Trans., 2017, 46, 15769-15777. 36 (36) C. Tapeinos, A. Marino, M. Battaglini, S. Migliorin, R. Brescia, A. Scarpellini, C. De Julián Fernández, M. Prato, F. Drago, G. Ciofani, Nanoscale, 2018, doi: 10.1039/c8nr05520c. (37) Y. He, Y. Yao, S. E. Tsirka, Y. Cao, Stroke, 2014, 45, 2514-2526.(38) M. J. Clark, N. Homer, B. D. O'Connor, Z. Chen, A. Eskin, H. Lee, B. Merriman, S. F. Nelson, PLoS Genet., 2010, 29; 6, e1000832. (39) A. Marino, S. Arai, Y. Hou, A. Degl'Innocenti, V. Cappello, B. Mazzolai, Y.-T. Chang, V. Mattoli, M. Suzuki, G. Ciofani, ACS Nano, 2017, 11, 2494 -- 2508. (40) N. Emelianova, Eur. Phys. J. Appl. Phys., 2015, 69, 10401. (41) F. Chemat, M, A. Vian, G. Cravotto, Int. J. Mol. Sci., 2012, 13, 8615-8627. (42) F. Mallamace, C. Corsaro, D. Mallamace, S. Vasi, C. Vasi, G. Dugo, Comput Struct Biotechnol J., 2014, 15, 33-7. (43) J. Kong, S. Yu, Acta Biochim. Biophys. Sin., 2007, 39, 549 -- 559. (44) S. Bhattacharjee, J. Control. Release, 2016, 235, 337-351. (45) C. Hu, Z. Li, Y. Wang, J. Gao, K. Dai, G. Zheng, C. Liu, C. Shen, H. Song, Z, Guo, J. Mater. Chem. C, 2017, 5, 2318-2328. (46) Y. Li, B. Zhou, G. Zheng, X. Liu, T. Li, C. Yan, C. Cheng, K. Dai, C. Liu, C. Shena, Z. Guo, J. Mater. Chem. C, 2018, 6, 2258-2269. (47) A. Marino, M. Battaglini, I. Pezzini, G. Ciofani, Smart Inorganic Nanoparticles for Wireless Cell Stimulation. In "Smart Nanoparticles for Biomedicine", pp. 201-218, edited by Ciofani G. (Elsevier, UK, 2016). (48) A. Marino, S. Arai, Y. Hou, M. Pellegrino, B. Mazzolai, V. Mattoli, M. Suzuki, G. Ciofani, Assessment of the effects of a wireless neural stimulation mediated by piezoelectric 37 nanoparticles. In "Use of Nanoparticles in Neuroscience", pp. 109-120, edited by Santamaria F., Peralta X.G. (Humana Press, 2018). (49) G. G. Genchi, A. Rocca, A. Grillone, A. Marino, G. Ciofani, Boron nitride nanotubes in nanomedicine: Historical and future perspectives. In "Boron Nitride Nanotubes in Nanomedicine", pp. 201-218, edited by Ciofani G., Mattoli V. (Elsevier, UK, 2016). (50) S. Wohlfart, S. Gelperina, J. Kreuter, J. Control. Release, 2012, 161, 264 -- 273. (51) F. Yuan, M. Dellian, D. Fukumura, M. Leunig, D. A. Berk, V. P. Torchilin, R. K. Jain, Cancer Res., 1995, 55, 3752-3756. (52) S. K. Hobbs, W. L. Monsky, F. Yuan, W. G. Roberts, L. Griffith, V. P. Torchilin, R. K. Jain, Proc. Natl. Acad. Sci. USA, 1998, 95, 4607-4612. (53) C. Saraiva, C. Praça, R. Ferreira, T. Santos, L. Ferreira, L. Bernardino, J. Control. Release, 2016, 235, 34-47. (54) N. Bien-Ly, Y. J. Yu, D. Bumbaca, J. Elstrott, C. A. Boswell, Y. Zhang, W. Luk, Y. Lu, M. S. Dennis, R. M. Weimer, I. Chung, J. Ryan, R. J. Watts, J. Exp. Med., 2014, 211, 233 -- 244. (55) T. R. Daniels, T. Delgado, J. A. Rodriguez, G. Helguera, M. L. Penichet, Clin Immunol., 2006, 121, 144-258. (56) D. L. Schonberg, T. E. Miller, Q. Wu, W. A. Flavahan, N. K. Das, J. S. Hale, C. G. Hubert, S. C. Mack, A. M. Jarrar, R. T. Karl, A. M. Rosager, A. M. Nixon, P. J. Tesar, P. Hamerlik, B. W. Kristensen, C. Horbinski, J. R. Connor, P. L. Fox, J. D. Lathia, J. N. Rich, Cancer Cell, 2015, 28, 441 -- 455. (57) T. Kang, M. Jiang, D. Jiang, X. Feng, J. Yao, Q. Song, H. Chen, X. Gao, J. Chen, Mol. Pharm., 2015, 12, 2947-2961. 38 (58) K. B. Johnsen, M. Bak, P. J. Kempen, F. Melander, A. Burkhart, M. S. Thomsen, M. S. Nielsen, T. Moos, T. L. Andresen, Theranostics, 2018, 8, 3416-3436. (59) K. B. Johnsen, A. Burkhart, F. Melander, P. J. Kempen, J. B. Vejlebo, P. Siupka, M. S. Nielsen, T. L. Andresen, T. Moos, Sci. Rep., 2017, 7, 10396. (60) P. Zheng, J. L. Zhang, Y. Q. Tan, C. L. Wang, Acta Mater., 60, 5022-5030. 39 Graphical abstract 40
1711.04999	1	1711	2017-11-14T08:40:42	The 2015 super-resolution microscopy roadmap	[ "physics.bio-ph" ]	Far-field optical microscopy using focused light is an important tool in a number of scientific disciplines including chemical, (bio)physical and biomedical research, particularly with respect to the study of living cells and organisms. Unfortunately, the applicability of the optical microscope is limited, since the diffraction of light imposes limitations on the spatial resolution of the image. Consequently the details of, for example, cellular protein distributions, can be visualized only to a certain extent. Fortunately, recent years have witnessed the development of 'super-resolution' far-field optical microscopy (nanoscopy) techniques such as stimulated emission depletion (STED), ground state depletion (GSD), reversible saturated optical (fluorescence) transitions (RESOLFT), photoactivation localization microscopy (PALM), stochastic optical reconstruction microscopy (STORM), structured illumination microscopy (SIM) or saturated structured illumination microscopy (SSIM), all in one way or another addressing the problem of the limited spatial resolution of far-field optical microscopy. While SIM achieves a two-fold improvement in spatial resolution compared to conventional optical microscopy, STED, RESOLFT, PALM/STORM, or SSIM have all gone beyond, pushing the limits of optical image resolution to the nanometer scale. Consequently, all super-resolution techniques open new avenues of biomedical research. Because the field is so young, the potential capabilities of different super-resolution microscopy approaches have yet to be fully explored, and uncertainties remain when considering the best choice of methodology. Thus, even for experts, the road to the future is sometimes shrouded in mist. The super-resolution optical microscopy roadmap of Journal of Physics D: Applied Physics addresses this need for clarity. It provides guidance to the outstanding questions through a collection of short review articles from experts in the field, giving a thorough discussion on the concepts underlying super-resolution optical microscopy, the potential of different approaches, the importance of label optimization (such as reversible photoswitchable proteins) and applications in which these methods will have a significant impact.	physics.bio-ph	physics	and plasticity of the a of fast large and that basic microscopy fraction long-term Super-resolution in Neurosciences: zoom in on synapses – Laurent COGNET & Brahim LOUNIS, Univ. Bordeaux, Institut d'Optique & CNRS, Talence, France Status The remarkably efficient functioning of the brain largely mirrors its multiscale complexity. In particular, it became clear since the emergence of modern neuroscience investigations of brain organization at the subcellular scales are key for the understanding of brain processes. Since then, our knowledge of the cellular mechanisms involved in neuronal communication and their evolution during life has literally exploded. Most of our comprehension comes from the discovery of molecules, genes and signaling cascades that play central roles in the maintenance neuronal communication. Notably, the identification of the synapse has attracted much attention. It contains key molecules required to induce plastic changes and mediates neuronal communication adaptations. The dimensions of brain synapses are however small, at the submicron scale and therefore they cannot be resolved with conventional optical microscopes. Electron microscopy has been providing ultra-structural information of synapse architectures and to some extent, knowledge about local molecular densities in these structures. However, this knowledge is out-of-reach in living cells and in intact tissues with this imaging modality. In this context, light microscopy constitutes the imaging modality of choice to study synapses in live cell. Continuous developments and techniques delivered an impressive amount of information about synaptic molecule organizations at the nanoscale. A first decisive achievement came from the possibility to detect single molecules in living cells. It not only allows localizing molecules with nanometer accuracies, far below the optical resolution, but also provides the ability to probe the intimate variety of molecule dynamic environments from nano- to micro-scales levels. In the early 2000's the presence of mobile receptors for neurotransmitters was revealed for the first time2,3 in the postsynaptic membrane of dissociated neuronal cultures and the role of neurotransmitter receptor diffusion in fast synaptic transmission was demonstrated4. The advent of several super-resolution methods that followed these early achievements raised great expectations in neurosciences by providing optical structures with of unprecedented resolutions. refinements of optical the content and neuronal imaging images or (u)PAINT)5. imaging volumes. Two major Figure 1 – High-density single-molecule trajectories (top) and super-resolution imaging (bottom) of endogenous glutamate receptors measured on a live neuron with the uPAINT method The circle highlights the presence of a synapse (from1). Scale bar = 500 nm. Current and Future Challenges Super-resolution microscopy methods mostly rely on the control of the number of emitting molecules in specific types of techniques found major applications in neurosciences: those based on STED where highly localized fluorescence emission volumes are optically shaped and those based on single-molecule localizations ((f)PALM, (d)STORM Single-molecule localization based methods have proven to be powerful to study nanoscale molecular organizations such as that of postsynaptic receptors and scaffold proteins (e.g., 1,6- 8) but also that of actin and actin binding proteins in axons9. On the other hand, STED microscopy allows visualizing dendritic spine shapes in living neurons with remarkable resolution and revealed that spine neck plasticity regulates compartmentalization of synapses10 and further deciphered the dynamic organization of actin. As exemplified in the aforementioned achievements, imaging synaptic structures and their molecular contents with nanometer resolutions constitute already a breakthrough for the understanding of synaptic functioning. Because these demonstrations have predominantly been performed on dissociated neuronal cultures or thin fixed brain slices, one of the main challenge will be transfer such cutting edge technologies to more integrated and ultimately intact living sample. This will allow major fields of neurosciences, such as development, aging and neurodegenerative disorders, the unprecedented degree of details provided by these microscopies. to benefit from to in limit in in for depletion living brain is conditioned the penetration depth favourable photophysics still need in Science and Technology to Meet Advances Challenges The possibility to perform super-resolution imaging in tissues intact address major to neurosciences questions to several light scattering and requirements. Because both absorption thick the visible domain, super- biological samples is only emerging. imaging resolution tissues Scattering is particularly detrimental for STED microscopy, since the realisation of a zero intensity region is mandatory. Two-photon microscopy, which uses near-infrared laser excitations, is a common strategy to excite molecules deep in brain tissues. However, due to the high intensities required, photobleaching rates are drastically enhanced and single molecule based super-resolution with two-photon excitation remains difficult. To fully exploit the biological transparency window, near-infrared probes with to be developed, as current red-shifted dyes are not strong emitters. Another key development will consist in achieving fast wide field super-resolved imaging at rates compatible with the inherent movements of living samples in 3D. Current super-resolution methods are indeed restricted to rather small imaging areas (sub-millimetric), which is constitutive to the novel opportunity to obtain images at superior resolutions. Indeed, by increasing the image resolution by typically a factor 10 to 50 per dimension, the size of a 3D image will be increased by a factor 103 up to ~105. This imposes constrains on the imaging speed as well as data handling. A promising strategy would consist in performing multi-scale imaging where only "relevant" brain sub-areas are imaged at the nanometer scales while the other regions are imaged at lower imaging "relevance" within the brain might depend on time and/or the physio-pathological local state of the sample as well as the nature of the information provided by the low resolution imaging modality. In this context, correlative imaging with other modalities, which do not have to be limited to optical techniques, are interesting routes. super-resolution approaches can also provide brand novel opportunities. Finally, a full understanding of the brain function cannot be obtained solely by images. Manipulation of the brain physio-pathological state will be needed using methodologies super- resolution imaging. Optogenetic methods provide promising tools in this context. Concluding Remarks For widespread application of super-resolution imaging in neuroscience, the main challenge will be to link resolutions. The definition of that are compatible with Combining multiple subcellular information gathered at the molecular scale (e.g. about synaptic processes) to the global organ function obtained at a macroscopic scales, in a well- defined functional state. To this aim, a multidisciplinary effort will be needed where physicist, chemists, computer and neuropathologists will have to work together. The task is vertiginous, but is at the level of the complexity of the brain. neurophysiologists scientist, long term potentiation on Figure 2 – Effect of the morphology of synaptic spines and heads revealed by STED microscopy (bottom panels are zoom of top ones. (from10). Scale bar = 500 nm. References 1 G. Giannone, E. Hosy, F. Levet, A. Constals, K. Schulze, A. I. Sobolevsky, M. P. Rosconi, E. Gouaux, R. Tampé, D. Choquet, and L. Cognet, Biophys. J. 99, 1303 (2010). 2 C. Tardin, L. Cognet, C. Bats, B. Lounis, and D. Choquet, EMBO J. 22, 4656 (2003). 3 M. Dahan, S. Levi, C. Luccardini, P. Rostaing, B. Riveau, and A. Triller, Science 302, 442 (2003). 4 M. Heine, L. Groc, R. Frischknecht, J.-C. Beique, B. Lounis, G. Rumbaugh, R. L. Huganir, L. Cognet, and D. Choquet, Science 320, 201 (2008). 5 A. G. Godin, B. Lounis, and L. Cognet, Biophys J 107, 1777 (2014). 6 A. Dani, B. Huang, J. Bergan, C. Dulac, and X. Zhuang, Neuron 68, 843 (2010). 7 H. D. MacGillavry, Y. Song, S. Raghavachari, and T. A. Blanpied, Neuron 78, 615 (2013). 8 D. Nair, E. Hosy, J. D. Petersen, A. Constals, G. Giannone, D. Choquet, and J. B. Sibarita, J Neurosci 33, 13204 (2013). 9 K. Xu, G. Zhong, and X. Zhuang, Science 339, 452 (2013). 10 J. Tonnesen, G. Katona, B. Rozsa, and U. V. Nagerl, Nat Neurosci 17, 678 (2014).
1905.12329	1	1905	2019-05-29T11:17:28	A practical review on the measurement tools for cellular adhesion force	[ "physics.bio-ph" ]	Cell cell and cell matrix adhesions are fundamental in all multicellular organisms. They play a key role in cellular growth, differentiation, pattern formation and migration. Cell-cell adhesion is substantial in the immune response, pathogen host interactions, and tumor development. The success of tissue engineering and stem cell implantations strongly depends on the fine control of live cell adhesion on the surface of natural or biomimetic scaffolds. Therefore, the quantitative and precise measurement of the adhesion strength of living cells is critical, not only in basic research but in modern technologies, too. Several techniques have been developed or are under development to quantify cell adhesion. All of them have their pros and cons, which has to be carefully considered before the experiments and interpretation of the recorded data. Current review provides a guide to choose the appropriate technique to answer a specific biological question or to complete a biomedical test by measuring cell adhesion.	physics.bio-ph	physics	A practical review on the measurement tools for cellular adhesion force Rita Ungai-Salánki 1,2,3 , Beatrix Peter 2, Tamás Gerecsei 1,2, Norbert Orgovan 1,2, Robert Horvath 2, Bálint Szabó 1,2 1Department of Biological Physics, Eötvös University, Pázmány Péter sétány 1A, Budapest, H-1117 Hungary 2Nanobiosensorics Group, Research Centre for Natural Sciences, Institute for Technical Physics and Materials Science, Konkoly Thege M. út 29-33. 1121, Budapest, Hungary 3 CellSorter Company for Innovations, Erdőalja út 174, Budapest, H-1037 Hungary Abstract Cell -- cell and cell -- matrix adhesions are fundamental in all multicellular organisms. They play a key role in cellular growth, differentiation, pattern formation and migration. Cell-cell adhesion is substantial in the immune response, pathogen-host interactions, and tumor development. The success of tissue engineering and stem cell implantations strongly depends on the fine control of live cell adhesion on the surface of natural or biomimetic scaffolds. Therefore, the quantitative and precise measurement of the adhesion strength of living cells is critical, not only in basic research but in modern technologies, too. Several techniques have been developed or are under development to quantify cell adhesion. All of them have their pros and cons, which has to be carefully considered before the experiments and interpretation of the recorded data. Current review provides a guide to choose the appropriate technique to answer a specific biological question or to complete a biomedical test by measuring cell adhesion. Keywords (maximum 6): cell adhesion, single cell, adhesion force, force measurement techniques Commonly used units: Shear stress: 1 dyn/cm2 = 0.1 Pa, 1 pN/µm2 = 1 Pa Force: 1 dyn = 10-5 N 1 Abbreviations: AFM: atomic force microscopy BFP: biomembrane force probe CTFM: cell traction force microscopy DDS: dimethyldichlorosilane ECM: extracellular matrix FluidFM: fluidic force microscopy FRET: fluorescence resonance energy transfer GETS: genetically encoded molecular tension sensor ICAM-1: intercellular adhesion molecule-1 MAT: micropipette aspiration technique NTA: nitrilotriacetic acid OT: optical tweezers PA: polyacrilamide PCR: polymerase chain reaction PDMS: poly-dymethylsiloxane PLL-g-PEG: poly (L-lysine)-grafted-poly(ethylene glycol) PMMA: polymethylmethacrylate PPFC: parallel plate flow chamber QCM: quartz crystal microbalance QPD: quadrant photodiode RBC: red blood cell 2 RFC: radial flow chamber RGD: arginyl-glycyl-aspartic acid RPM: revolutions per minute SAM: self-assembled monolayer Sc: critical separation pressure SCFS: single cell force spectroscopy SPR: surface plasmon resonance SPT: step-pressure technique Corresponding author: [email protected] 3 Table of contents Introduction ............................................................................................................................................. 5 1. Centrifugal and shear flow assays ............................................................................................. 11 1.1 Centrifugal assay ..................................................................................................................... 11 1.2 Spinning disk ........................................................................................................................... 13 1.3 Radial flow chamber................................................................................................................ 15 1.4 Parallel plate flow chamber ..................................................................................................... 18 1.5 Microfluidic rectangular channels ........................................................................................... 20 2. Micropipette manipulations ....................................................................................................... 22 2.1 Step-pressure technique (SPT) ................................................................................................ 23 2.2 Biomembrane force probe (BFP) ............................................................................................ 25 2.3 Micropipette aspiration technique (MAT) ............................................................................... 26 2.4 Computer-controlled micropipette .......................................................................................... 29 3. Optical tweezers (OT) ................................................................................................................... 32 4. Atomic force and fluidic force microscopy ................................................................................... 36 4.1 AFM to probe force normal to the surface .............................................................................. 38 4.2 AFM to probe tangential/lateral force ..................................................................................... 41 4.3 Fluidic force microscopy (FluidFM) ....................................................................................... 42 5. Cell traction force microscopy and FRET force sensors ........................................................... 46 5.1 Cell traction force microscopy ................................................................................................ 46 5.2 FRET force sensors ................................................................................................................. 49 Summary ............................................................................................................................................... 53 Acknowledgment................................................................................................................................... 60 References ............................................................................................................................................. 61 4 Introduction Cell adhesion [1] is the ability of a cell to stick to another cell or to the extracellular matrix (ECM). Most mammalian cell types are anchorage-dependent and attach firmly to their environment [2]. Cell adhesion mediated by cell surface receptor molecules, such as integrins [3],[4], cadherins [5],[6], selectins, and members of the immunoglobulin superfamily is a fundamental phenomenon [7] vital for both multi and single cellular organisms (Fig.1). Upon binding their extracellular ligand (such as fibronectin, vitronectin or collagen), integrin molecules cluster to form focal adhesions, complexes containing structural and signalling molecules crucial to the adhesion process [8]. Best-characterized adhesions are the 'classical' focal adhesions (also termed focal contacts) and variants including fibrillar adhesions, focal complexes and podosomes [9]. Focal adhesions mediate strong adhesion to the substrate, and they anchor bundles of actin microfilaments through a plaque that consists of many different proteins [9]. Fibrillar adhesions are associated with ECM fibrils. Focal complexes are present mainly at the edges of the lamellipodium. Podosomes contain typical focal contact proteins -- such as vinculin and paxillin and they are found in various malignant cells [9]. Communication between cells and the ECM is critically influenced by the mechanical properties of cell surface receptor-ligand interactions. Integrin binding forces were measured in intact cells by atomic force microscopy (AFM) for several RGD containing (Arg-Gly-Asp) ligands and ranged from 32 to 97 pN. The context of the RGD sequence within the ligand protein has considerable influence upon the final binding force [10]. Lee and Marchant reported that the binding force of the single molecular interaction between the RGD-ligand and the integrin was 90 pN [11]. Mechanical forces are central to the functioning of living cells. Cell adhesion is known to be closely related to the actin cytoskeleton, of which organization is crucial in determining the structural and mechanical properties of cells. While significant experimental progress has been made to measure the forces generated by cells, interpretation of these experimental results poses a challenge. Measurements can only be fully interpreted in the light of biophysical models of cell mechanics. However, even a single cell is so complex that theoretical models can usually describe only a few aspects of its behaviour, e.g., contractility considering some cellular components such as stress fibers [12]. In cell cultures on stiff substrates, the actin cytoskeleton tends to be organized into stress fibers, bundles of actin filaments tensed by myosin II molecular motors. Stress fibers usually end in focal adhesions: integrin-based adhesion contacts at the cell 5 membrane. The coupling between mechanics and biochemistry should also be considered in the model [13]. Some models are inspired by architecture like the the so-called tensegrity model [14] of the cytoskeleton [15]. Adhesion-dependent cells sense the mechanical properties of their environment by mechanotransductory processes through focal adhesions. Simplified models applying springs replacing the complex architecture and biochemistry of the cell can be successful to explain certain experimental results [16] A two-spring model [17] showed that the stiffer the environment, the stronger force is built up at focal adhesions by molecular motors interacting with the actin filaments. Dynamic control of cell adhesion is indispensable to the developing embryo [7], to the immune system [18], and critical in the metastasis of tumors [19],[20],[21] or in the successful implantation of a prosthesis. In case of inflammation, several successive steps of the immune cell adhesion are known [18]. Although the molecular background of this particular process is partially explored there is a lot more to discover about cellular adhesion in general. Fig.1 Basic processes of cell adhesion: the artificially tailored surface contains ligands that bind to the integrin receptors found in the membrane of cells. Throughout the process of adhesion, the actin filament structure of the cell is reorganized and a traction force is generated in the substrate. The cytoskeletal reorganization is also regulated by external stimuli. After adhesion to the surface, the cell can interact with other cells through membrane 6 proteins such as cadherins, selectins and members of the Ig superfamily. In tissues a variety of multiprotein complexes called cell junctions (e.g. tight junctions), can form between cells to promote intercellular communication and mechanical stability. In vitro environments to study cell adhesion. Both biochemical techniques for tailoring large surfaces on the molecular level and methods capable of the high-throughput high sensitivity measurement of cell adhesion [22],[23],[24],[25] have emerged in the past few years. Appropriate masking (blocking) of the surface area not covered by specific biomolecules has central importance to get rid of non-specific adhesion of cells on, e.g., glass or plastic substrates. For this the protein repellent PLL-g-PEG synthetic polymer [26],[27],[28],[29] proved to be exceptionally useful. Grafting the RGD motif of three amino acid residues into the PEG side chains results in a passivated surface designed for cell adhesion exclusively via RGD-binding cell surface receptors [30],[31]. For investigating cell adhesion mediated by functional proteins (engineered with a His tag) NTA-functionalized PLL-g-PEG [32] offers a versatile possibility. Recently, a 10-fold increase in surface passivation efficiency could be achieved over PEG using the Tween-20 molecule on DDS surfaces [33]. Recent studies demonstrated that a compact and oriented flagellin protein layer - mimicking the surface of bacterial flagellum - has excellent cell repellent capabilities[34],[35],[36],[37]. By genetically engineering the molecule, peptid sequences (for example RGD motif) can be incorporated into this cell repellent layer facilitating receptor specific cellular adhesion. Flagellins can be produced in bacteria and are available in large quantities, opening up an interesting new direction in developing biomimetic engineered surfaces. Tools to measure cell adherence. A number of different techniques can be applied to measure cell adhesion force [38] (Fig.2). Many of them, including the simple washing assay [39], the spinning disk technique [40] and flow chambers [41] are based on the hydrodynamic shear flow removing cells from the surface [42]. However, these techniques do not enable single cell targeting, and cell shape has a strong impact on the shear force making difficult to calculate the exact adhesion force. Furthermore, only weakly adhered cells can be probed due to the technically limited magnitude of shear stress. 7 Fig.2 Classification of cell adhesion measurement techniques. Population based assays rely on removing the cells from a surface by hydrodynamic sheer flow. The adhesion force can be calculated from the applied flow rate, 8 however the extracted value depends on other factors as well, such as cell shape. Single cell based techniques can be further divided into categories depending on their principle of operation. Micropipette manipulation assays apply a glass micropipette to aspirate cells from the surface and calculate the adhesion force using hydrodynamic simulations. Atomic force microscopy combined with chemical tip modifications or a nanofluidic channel can also be used to investigate cell adhesion. Other methods include FRET sensors that apply genetically modified cells expressing proteins whose fluorescent signal depends on mechanical tension, as well as force traction measurements capable of detecting forces generated in the substrate by the adhering cell. Data originating from population and single cell based methods differ fundamentally as the population methods are measuring an average over thousands of cells, while single cell measurements can explore differences in heterogeneous populations. To directly probe the adhesion force of single cells, cytodetachment with an AFM tip [2], [43],[44] or micropipette aspiration [45],[46],[47] can be chosen as a tool. Both of them are inherently low throughput methods: typically 5-10 cells can be measured in a day. A modified AFM applying vacuum on cells with a fluidic micro-channel (FluidFM) [48] eliminates the painful AFM cantilever chemistry. In a FluidFM, a cantilever can be used for about 10 cells, thus throughput is increased by a factor of 10 compared to conventional AFM. Additionally, the force range is enlarged to reach the µN regime. Automated glass micropipette equipped onto an inverted microscope can probe the adhesion force of single cells with a relatively high throughput: hundreds of cells can be measured in a ~30 min experiment. Cells can be automatically selected on the basis of computer vision. Single viable cells picked up by the micropipette can be further analyzed by, e.g., DNA/RNA sequencing. Hydrodynamic lifting force acting on cells positioned under the micropipette is calculated from off-line computer simulations. Cell traction force microscopy (CTFM) [49] enables the force (mechanical stress) field generated by single cells on an elastic substrate to be reconstructed with a resolution of ~1-10 μm. Since its introduction in 1996, CTFM has undergone impressive developments. However, the computation of the traction filed is challenging and strongly depends on the mathematical model applied. Thus the derivation of the cellular traction force is not straightforward. Molecular force sensors [50] inside cells based on Förster resonance energy transfer (FRET) offer a brilliant tool for the high resolution mapping of force distribution inside live cells. The FRET sensor is incorporated into a specific protein of the cytoskeleton in a transgene cell type. The extra gene is constructed to express the two fluorescent proteins separated by an elastic 9 linker region. When the elastic linker region expands as a spring pulled by a mechanic force, the distance of the fluorescent proteins, i.e., the FRET donor and acceptor changes, and can be detected in the fluorescent signal. An interesting alternative to measure cellular adhesion with high sensitivity and high temporal resolution is the application of novel evanescent field based optical biosensors [22],[51],[52],[53]. Here, the biosensor signal is directly proportional to the cell-substratum contact area and also correlates with the number of biomolecules in the adhesion contact zone, presumably proportional to the strength of adhesion [51]. These label free methods can readily monitor not only the strength but the real-time kinetics of cellular adhesion [22],[54]. But they are indirect, and thus the calculation of the adhesion force is not straightforward; an open problem not yet solved. However, they have clear advantages when the biomolecular adsorption, prior to the adhesion of living cells, has to be also recorded in a single experiment [55]. Surface plasmon resonance (SPR) microscopy can be used as an alternative method for visualizing and quantifying cell/substrate contacts of living cells [56]. SPR, the most well- known evanescent field based optical method is usually used in in biosensor setups to monitor molecular interactions such as receptor/ligand binding in real time. However, SPR has not been demonstrated to measure adhesion forces directly or indirectly. Quartz crystal microbalance (QCM) with dissipation [57],[58],[59] is a simple, high-resolution mass sensing technique allowing in situ real-time measurements of mass and viscoelasticity changes through the resonance frequency (f) and energy dissipation (D) during the cell adhesion process. It can monitor morphological and cytoskeletal changes of surface adherent cells in a non-invasive way [60]. Nevertheless, QCM cannot measure the adhesion force. Sensitive monitoring of the adhesion process of a large number of single cells on surfaces precisely decorated with biomolecules is expected to produce information on ligand binding, receptor function and signaling pathways with a quality and magnitude used to be unattainable with former cell adhesion assays. Exploring the many faces of molecular scale (nanoscale) direct physical interactions between cells and their environment is a fundamental goal of biophysics, which now seems to be accessible. 10 Below we divided the available techniques into two categories: population methods and single- cell approaches (Fig.2). Whereas population methods provide cell adhesion data with good statistics, targeting and probing single cells offer a direct and more sensitive measurement of cell adhesion [61]. Inherent heterogeneity of cell populations can be also explored by single cell measurements. Centrifugal and shear flow assays 1. Adhesion force of a cell on a substrate can be probed with simple cell detachment assays like centrifugal or shear flow chambers based on a centrifugal or hydrodynamic shear force to remove weakly attached cells from the substrate [62],[63]. While the applied centrifugal force can act both in a direction normal and parallel to the surface onto cells are attached, shear force is always parallel to the surface. Cells adhered to the surface weaker than the centrifugal or shear force are removed. The exact value of the shear force is difficult to control and therefore ill-defined [2] as it sensitively depends on parameters such as cell size and shape [43]. Although simple washing assays allow the identification of key adhesion components, they provide only qualitative adhesion data. Hydrodynamic shear flow assay applying a well-controlled shear stress to adherent cells is the most common way to quantify cell adhesion. Most widespread methods are the spinning disk, radial flow chambers and parallel plate flow chambers (Table 1). 1.1 Centrifugal assay Cell adhesion strength can be determined using a centrifugal cell detachment assay as described by Chu et. al.(1994) [64] introduced earlier by McClay et al. (1981) [65] and Lotz et al. (1989) [66]. Centrifugal assay employs standard laboratory centrifuges to apply forces on cells adhered to a substrate. Briefly, substrates for cell adhesion are glued to the bottom of a 24-well plate. At the end of the incubation, the wells are filled with medium and covered with sealing tape to avoid medium loss and air bubbles. Then the plates are inverted and spun in a swing-bucket centrifuge for 10 minutes (at 25°C, 800g) [31],[67] (Fig.3). After centrifugation, the number of cells is quantified manually. This assay measures the average response of a cell population and typically examines the fraction of cells that remain adhered to the surface after centrifugation. 11 Fig. 3 Steps followed in the centrifugal assay in the study of Chu and co-workers [64]. A substrate for cell-adhesion is prepared in a 24-well plate then filled with cell suspension. Cells are let to adhere in an incubator, then the well is sealed airtight and placed in a centrifuge with an opposite orientation. As the centrifugation begins, the cells are affected by a detaching force in the range of 1-100 nN depending on the used acceleration. The fraction of cells detached from the surface can be quantified using optical microscopy. The presented method is simple and cost-efficient, however the force range is limited by the centrifuge's properties. Even with ultracentrifuges reaching 100,000 g in acceleration. Only weakly adhering cells can be studied. Furthermore, the cell shape is altered significantly by the high acceleration endured by the cells during centrifugation. a. Number of cells in an experiment Typical cell number is 105 cell [68]. b. Typical range of centrifugal force This assay applies controlled detachment forces to a large population of adherent cells [69].To determine the adhesion strength, the number of cells is quantified before and after applying the centrifugal force [42]. The force in the centrifuge is calculated as follows: F = G x Vcell x (ρcell- ρmedium), (1) 12 where F is the force exerted on the cell, G is the centrifugal acceleration, Vcell is the cell volume, ρcell is the density of the cell (typically 1.07 g/cm3), ρmedium is the density of the medium (typically 1.00 g/cm3) [31],[42]. Typical G values fall in the range of 20-1,000g and the forces on an individual cell is 1-2,000 pN When using an ultracentrifuge, adhesion force can be measured up to 100 nN [70],[71] with a maximum acceleration of 110,000 g. However, before cells detach due to such a high acceleration, cell shape changes dramatically similarly to shear force induced shape alteration. c. Experiment duration The duration of load application in centrifuge tests typically range from 5 to 10 min [31],[42], [69],[72]. d. Advantages Simplicity. It does not require specialized equipment: the method is widely accessible [42]. e. Disadvantages Normally restricted to the investigation of weakly-adhered cells [42]. f. Main applications Adhesion of MC3T3-E1 cells on self-assembled monolayers (SAMs) as a function of fibronectin density [42],[69] . Cell-cell adhesion mediated by E-cadherin [68]. Cell adhesion to specific RGD-modified substrates over a wide range of forces [31]. Strength of bacterium- receptor interactions [73]. 1.2 Spinning disk The apparatus applies a disk spinning in a large volume of fluid. The cell adhesive surface is mounted onto a rotating circular stage, which produces a fluid flow in the chamber [74] . Fluid flow over the cells on the disk creates a detachment force [75], which is calculated from the properties of the buffer and the rotational speed (Fig.4). At the axis of rotation the detachment force is zero and it increases linearly with the distance from the axis. This fewer cells will 13 remain near the edge of the disk than closer to the axis. After spinning, the remaining adherent cells are fixed, stained, and counted. Fig.4 Spinning disk arrangement in the experiment of Wertheim et al [74]. Cells are allowed to settle onto a fibronectin-coated matrix atop the stage. As the stage is rotated, the spinning stage inside the fluid is generating a sheer flow on the surface. The stress caused by this flow is linearly increasing from the center of the disk to the edge, causing a number of cells to detach in the function of distance from the center. The distribution of cells that remain on the stage after spinning can be quantified by optical microscopy. Because of the linearly changing shear stress on the surface, a range of forces can be studied at the same time. The disadvantage of this method is common in all shear flow based techniques namely the dependence of the measured detachment force on cell shape. This particular construction also makes it impossible to optically observe the cells during measurement. a. Number of cells in an experiment 400 cells/mm2 [76] are uniformly seeded onto coverslips. b. Typical range of shear stress Detachment force is proportional to the hydrodynamic wall shear stress, τ (shear force/area) 14 τ=0.800 r √𝝆µ𝝎𝟑 (2) where r is the radial position from the disk center, ρ and µ are fluid density and viscosity, respectively and ω is the angular speed (rad/s). Typical frequency range: 500-3000 RPM [42].The shear force increases linearly with the radius from the center of the disk, thus it can generate a wide range of forces at the same time. A value of 200 Pa shear stress was reported [77]. The range of detachment forces is (0-10 Pa2) [74]. c. Experiment duration Disks were usually spun for 5- 10 min at constant speed [76],[77]. d. Advantages The key advantage of the method is the shear stress increases linearly with the radial position on the disk allowing a wide range of forces to be applied at the same time [42],[74]. Spinning disk method can generate high shear stresses and detach relatively strongly-adhered cells. e. Disadvantages Hydrodynamic shear force depends on cell shape [75]. It is not compatible with simultaneous microscopic imaging [61]. f. Main applications Binding strength between cells and Fn-coated micropattern islands [40]. Modell system: erythroleukemia cell line, expressing a single fibronectin receptor, integrin α5β1. Cell detachment profile and mean detachment force measurements [75],[77]. Bacterial spore adhesion [78]. 1.3 Radial flow chamber In the axisymmetric flow geometry of radial flow chambers (RFC), a wide range of radially dependent shear stress is applied to the adherent cells in a single experiment [79]. A design of 15 RFC by Stone (1993) [80] employs a chamber on an inverted microscope, allowing in situ observation of adherent cells exposed to shear stress. The surface is pre-incubated, then the cells are introduced into the RFC by a syringe and allowed to adhere to the surfaces for a given time (~30-60 min), prior to applying the shear. Fluid [79],[81] flows radially between the surface with the cells and a glass slide at a constant volumetric flow rate for ~5 min (Fig.5). The chamber has a 200 μm gap between the surfaces. Cells are removed, where the shear force exceeds the adhesive force of cells [80] . The mean velocity of the fluid and hence the shear stress, decreases with increasing radial position, r. For the radial flow, the Reynolds number is inversely proportional to radial position: Re= 𝟐𝝆𝑸 𝝅𝝁𝒓 (3) where Q is the flow rate, ρ and μ are the density and viscosity of the fluid, respectively. To ensure laminar flow, the critical upper limit to Re is 2000. Fig.5 Schematic illustration of the radial flow chamber based on the publication of Goldstein et al. (1998) [81]. The cells are seeded on the glass support, exposed to a laminar flow coming from above. The flow rate determines the shear stress on the cells up to the point where the flow would become turbulent. Much like in the spinning disk case, the stress is linearly changing with the position relative to the center of the dish but in this case the maximum 16 falls to the center and decreases as we approach to the edge. This construction can be directly mounted on an inverted microscope thus enabling the optical monitoring of the detachment process. a. Number of cells in an experiment 105 cells / ml added to a 60-mm-diameter dish. Seeding density: ~40 cells/ mm2 [79],[80]. Cells are plated at an approximate density of 10.25 × 104 cells /mL (∼8.34 × 104 cells/cm2, ~2.64 × 106 cells/surface) [82]. b. Typical range of shear stress Cells were allowed to attach to fibronectin-coated substrates for 30 min and then subjected to a spatially dependent range of shear stress for 5 min (28 -- 220 dyn/cm2) [83]. After 75 to 90 min, at a shear stress of 350 dynes/cm2, more than 50% of the spread cells are detached from the surface. Cells with higher spreading areas stay longer at the glass surface [84]. c. Experiment duration Cells are incubated for 30-60 min. Shear stress is applied for ~ 5 min [79],[81],[83],[84]. d. Advantages Broad range of the shear stress simultaneously [81]. If the RFC is mounted on an automated microscope, in situ observation of cell detachment can be monitored under the hydrodynamic shear stress [79],[83]. It is allowed the application of uniform stress fields to cells while enabling visualization of the cells via microscopy [61]. e. Disadvantages These chambers are unable to generate stresses high enough to detach well-spread cells under laminar flow conditions [61]. Shear stress limitations due to the upper limit of Reynolds number to avoid turbulence [81]. Complex fluidics needed [85]. f. Main applications In situ detachment of 3T3 murine fibroblasts from substrata of SAMs of dodecanethiolate [79],[81] and fibronectin. Cell deformations (elongation) after applying the shear stress [79],[86]. Adhesive nature of the modified substrates [82]. Bacterial adhesion experiments [87]. 17 1.4 Parallel plate flow chamber In the parallel plate flow chamber (PPFC) (Fig.6) the flow is precisely defined to produce a uniform shear stress [84],[88],[89],[90]. Shear stress τ (dyne/cm2) is calculated from the following equation: 𝝉 = 𝟔𝑸 𝝁 𝒘⁄ 𝒉𝟐 (4) where Q is the volumetric flow rate (cm3/s), μ is the viscosity of the medium, w is the width of the chamber and h is the height of the chamber. Usually, a peristaltic pump is used to adjust the required flow rate [90]. 18 Fig.6 Schematic illustrations of a parallel plate flow chamber used to characterize the effect of nanotopology on cell adhesion based on the publication of Martines et al.(2004) [91]. A polymethylmethacrylate (PMMA) sample (turquoise) was patterned with 100 nm nanopits, then a glass slide (light blue) was attached to it with a thermoplasitc gasket (dark blue) in between. The channels in the PMMA were designed to create a laminar flow within the chamber. The inlet and outlet for the solution is provided by circular holes in the glass slide on the top of the device. The adhesion strength can be measured by monitoring the attached cell's density as a function of the flow in the channels. The authors have successfully shown the adhesion reducing property of the nanopatterned surface topology. a. Number of cells in an experiment This technique can be used to monitor the removal of a population of cells from a surface. Seeding concentration: 104 -106 cells /ml resulting in 100-1,000 cells/mm2 [88],[91],[92]. However, normally only few hundred cells are analyzed. b. Typical range of shear stress 1-25 Pa [89],[90],[92],[93]. Cells can be exposed to either a single shear stress or incremental shear stress levels [89]. Also the effect of increasing pulsatile flow on cell adhesion can be monitored [93]. c. Experiment duration Time elapsing after seeding cells on the surface, and before applying the shear stress: 1-24 hours [88]. Duration of shear stress can be several hours [90]. In the incremental shear studies shear stress can be increased in the time scale of minutes (4-15 min) [89],[93]. In ref [92] cells were exposed to a range of shear stress levels each applied for only 2-10 sec. d. Advantages Parallel plate flow chamber can be utilized widely due to its simple well-characterized flow regime [89]. The chamber can be mounted onto a microscope, thus the behavior of the cells as a response to the shear stress can be observed in situ. Parallel flow chambers can provide a precisely controlled, uniform shear stress with dynamic, well-defined shear regulation [94]. In this chamber the motion of a sphere can be easily calculated. Cell trajectories, speed and adhesion process can be monitored [91]. It is a frequently applied design because of its simplicity. It can be employed on a broad range of surfaces, such as silicone rubber, dental enamel and metals [95]. 19 e. Disadvantages The most parallel plate flow chamber have a height >200 μm [61]. This sets a practical limit on the shear stress value considering the normally available range of flow rate or overpressure. f. Main applications Channels can be used to determine the strength of adhesion of fibroblast cells to various substrates, with different wettabilities [89]. Cell retention, morphology and migration as a function of flow rate and the influence of adhesion time were also studied [88]. Endothelial cell adherence onto polymer surfaces with different hydrophilicity [90]. Investigation of nano- patterned surfaces [91]. Development of flow circuits, in which the behavior of cells can be continuously monitored on a microscope [88]. The flow circuit usually consists of three parts: a flow loop, a heating system, and an image analysis component [88],[89]. Adhesion of bacterial and yeast strains to a PEO (Poly(ethylene oxide))-brush covalently attached to glass [96]. Measurement of the attachment and detachment rates of Escherichia coli to and from a glass surface [97]. Nonspecific surface attachment of hydrophobic yeast cells [98]. 1.5 Microfluidic rectangular channels This technique is similar to the parallel plate flow chamber, but the key advantage is the small characteristic dimension of the flow channel (h < 100 μm) that allows substantially higher shear stresses to be generated under laminar flow conditions [61] (Fig.7). The channels can be readily mounted onto a microscope, and thus cell detachment can be monitored in situ. The device is usually constructed from PDMS, because it is quick and simple to use, cheap, transparent, and biocompatible [99],[100]. 20 Fig.7 Parallel microfluidic shear device pattern, consisting of eight microchannels used in the study of Young and co-workers [101]. The flow is directed into the device through the common inlet on the left. The 516 µm wide and 59 µm high channels are designed to ensure identical flow in all branches. The main advantage of this arrangement is the parallelization of measurements: a fraction of the channels were coated with proteins, while others were used as control during the same experiment. Cell spreading and attachment can be monitored by optical observation of the channels by an inverted microscope. a. Number of cells in an experiment Cell suspension density: 106 -107 cells/ ml Number of analyzed cells: 100 - 1,000. Usually 1- 10 microscopic fields of view with 10-100 cells/ field of view [61],[100],[101],[102]. b. Typical range of shear stress 50-200 Pa [61]. When the shear stress is time dependent, its rate is usually in the 1-10 Pa/sec range [61]. 1 - 20 Pa [101]. c. Experiment duration Typical experiment duration: 10-15 min [61]. Adhesion strength of cells was investigated in 12 min, while shear stress level was increased step by step in 4 min intervals in ref [101]. d. Advantages This method offers a wide range of forces and relatively high-throughput. The experimental setup is simple, and it can be integrated with other microanalytic modules. Using time-lapse 21 microscopy, cell detachment can be monitored in situ. The microfluidic device offers advantages with its small dimensions and ensures laminar flow even at high fluid velocities. The small channel height (h<100 μm) enables higher shear stress than the similar parallel plate flow channel. Thus also the strongly adherent cells (cannot be studied in most conventional assays) can be investigated in microfluidic rectangular channels [61]. The device is usually fabricated from poly-dymethylsiloxane (PDMS) bonded to glass using the rapid technique of soft photolithography [100]. The technique has small sample consumption and it provides a stable and well characterized flow [101],[102]. A parallel microfluidic network of the channels permits the parallel analysis of multiple samples or conditions [101]. e. Disadvantages At the micron scale, channel deformation effect becomes important and must be quantified for predictable assay performance [99]. Cell morphology is affected by the flow: cells become elongated due to the applied shear stress. It shows that the flow can significantly alter the cytoskeleton several minutes before cells detach from the surface [101]. This "side-effect" needs to be thoroughly considered when evaluating experimental results. f. Main applications Measurement of the adhesion strength of well-spread cells on different surfaces [61]. Investigating the adhesion difference between normal and cancerous epithelial cells on nanostructured polymer surfaces. Adherence of vascular and valvular endothelial cells on different extracellular matrix proteins (fibronectin, collagen I) [101]. Observation of bacterial adhesion [103],[104]. 2. Micropipette manipulations The use of micropipette manipulation studies can be traced back to Mitchison and Swann (1954). They first developed and applied a micropipette-based elastimeter to determine the membrane elastic modulus and internal pressure of the unfertilized sea urchin eggs. The method can measure the deformation of a single cell attached to the tip of a micropipette by a precisely controlled vacuum in the micropipette [104]. Ten years later, Rand and Burton (1964) refined this method. They measured the stiffness of the erythrocyte (RBC) membrane and the pressure inside the cell [105],[106]. In the 1970-80s, the micropipette manipulation method was further 22 improved and a number of studies applied it to explore the mechanics of RBCs, white blood cells, endothelial cells, lipid vesicles, and liposomes. These studies imposed either a local force or a suction underpressure on individual cells or liposome surface. In the 1990s, the technique was extended to studies including tumor cell metastasis and the kinetics of cell-cell contact interactions [46]. The common approach is to quantify the adhesion strength by imposing tension to rupture the adhesive contact between two opposite faces. An increasing force is employed to measure the magnitude of the rupture force. Alternatively, a constant force is applied to determine the adhesion lifetime [46]. Micropipette manipulation techniques are summarized in Table 2. 2.1 Step-pressure technique (SPT) Sung et al. (1986) developed the first micropipette based technique, which was used to quantify cell adhesion strength [47]. The system consists of a cylindrical glass micropipette with an internal diameter of a few μm-s and a manometer to control the pressure inside the micropipette with a resolution of 0,01pN/ μm2 [43],[107],[108]. One cell is tightly held by micropipette 1 with high suction vacuum, while the second is held by micropipette 2 with lower suction force. First, the two cells are brought into contact, so they can adhere to each other. Then micropipette 2 is pulled away (Fig.8). The suction vacuum in micropipette 2 is increased until the cells are separated from each other. The minimum suction vacuum which leads to the separation of the cells is the critical separation pressure (Sc). One of the cells can be replaced with a protein coated substrate to determine the cell- substrate adhesion strength [46]. 23 Fig.8 Schematic illustration of step-pressure technique (based on the publication of Shao et al.) [46]. Two micropipettes each with a cell sucked into their tips are facing each other as the cell's surfaces are pushed together. After a certain adhesion time, the cell on the left is pulled away. If the moving cell slips out of the micropipette, the suction force was smaller than the adhesion force between the cells. By increasing the suction force, an upper estimate can be determined for the inter-cell adhesion force and energy. This method was the first to be used to measure single cell adhesion by glass micropipettes. This technique allows a precise control of cell contact area and time, however the nonspecific adhesion between the glass tips and the cell membrane must be eliminated. Another disadvantage of any similar method is the dependence of the measured adhesion properties on the speed of the separation of the micropipette tips. a. Number of cells in an experiment ~10 pairs of cells [109]. b. Typical force range The applied vacuum is in the 10-100 Pa range [109],[110]. Sc =1,5 nN/μm2, However, it is difficult to measure the exact contact area between the cells. Separation force is in the 1-10 mN range [46],[47]. c. Experiment duration Cells are brought into contact for ~10 sec [110]. In ref [47] Sung et al. measured the adhesion force between two cells for 120 min. The first measurement started at 12 min and subsequent measurements were taken in every 10 min. d. Advantages Enables the measurement of the adhesion force between two similar or different single cells in vitro [47]. Cell-cell contact area and contact time can be controlled [46],[110]. e. Disadvantages Nonspecific adhesion between the cell and pipette wall has to be much weaker than the overall adhesion to the other cell or to the substrate [46]. Low throughput, manual measurement. 24 f. Main applications Enables the real time observation of two cells assembling and then losing the contact zone in between them. Cell adhesion force can be measured directly [46]. 2.2 Biomembrane force probe (BFP) Evans et al. (1995) described a simple apparatus and procedure for probing weak forces at biological surfaces [111]. The technique applies a force transducer (these are cell-sized membrane capsules such as a vesicle, liposome or an erythrocyte) which is held by a micropipette with a suction force while a micro-bead is attached to the force transducer [46],[47]. The microscopic bead is glued biochemically to the transducer. The bead is coated with the protein of interest in order to interact with the cell (Fig.9). Pulling or pushing force applied to the bead result in membrane deformation of the force transducer [46]. Fig.9 Schematic illustration of biomembrane force probe (Based on the publication of Shao et al. [46]). A liposome is sucked into a micropipette that is biochemically glued to a latex bead. By pushing or pulling the liposome, a force can be exerted on the bond between the cell and the bead. As the latter can be coated with a protein of interest, receptor-ligand bonds can be studied under a controllable loading rate and force. Using this arrangement Evans et al. showed that the rupture force of a ligand-receptor bond depends on the speed at which the force is applied. a. Number of cells in an experiment 1 cell/experiment. 25 b. Typical force range Evans et al. (1998) employed two modes of BFP; vertical and horizontal modes. Vertical mode was used to test weak bonds under slow loading, when the force was 0.2-0.5 pN; while the horizontal mode was utilized to study stronger bonds under fast loading (1-10 pN) [112]. The transducer sensitivity is tuned to measure forces from 0.01 to 1,000 pN with a range of loading rates from 0.1pN/s to 100,000 pN/s. Evans et al. have shown that, rupture force of a receptor- ligand bond depends on the loading rate of the force, i.e., how fast the bond is pulled [46],[113], [114],[115],[116],[117],[118]. c. Experiment duration No information. d. Advantages The loading rate of the force is adjustable [46]. The interference to the cell can be minimized. Gentle measurement minimally altering the cytoskeleton. Weak adhesive bonds can be readily detected. Sensitivity of the technique enables the detection of local activation of cytoskeletal structures [111]. The method has sub-pN force and nanometer scale displacement resolution [46]. e. Disadvantages The resolution of probe movement is not as good as in AFM [111]. f. Main applications It allows the quantification of single molecular bonds [38]. Receptor-ligand binding [46]. 2.3 Micropipette aspiration technique (MAT) 26 The technique was developed by Shao and Hochmuth (1996). They applied the method to measure the magnitude of pN forces [46],[107]. Their aim was to create a simple method based on micropipette suction, allowing the measurement of detachment force of one cell from another cell or a solid surface [107] (Fig.10). Fig.10 Schematic illustration of micropipette aspiration technique (Based on the publication of Shao et al.) [46]. In the micropipette on the left a bead is aspired that can move freely within the glass walls. As the micropipette is pushed to the cell, the functionalized bead surface is brought into contact with the cell membrane using a gentle pressure. After the ligand-receptor bonds are formed, the bead is sucked into the micropipette with a controllable force. The point at which the bead detaches is the adhesion force as measured by this technique. Alternatively, the bead can be replaced with a spherical cell (Spillmann et al.) [119] to directly study cell-cell interaction. a. Number of cells in an experiment One cell/experiment. b. Typical force range A spherical object either a cell (e.g. human neutrophil) or a bead is used as a MAT transducer and it can exert forces lower than 10-20 pN [46]. Between the transducer and the pipette wall, a small clearance is necessary to allow free movement of the transducer inside the pipette. A freely sliding cell or bead is used as a force transducer. A positive pressure allows the bead (which is coated with the protein of interest) to contact and adhere to the cell [107]. Then, a constant suction underpressure provides the tensile force to break the adhesive bond. According to the diameter of the pipettes (1-10 μm), the force range is 10 pN-1 nN [45]. The force exerted by the static cell (or bead) can be determined as follows: 27 F=πR2 p Δp (5) where Rp is the radius of the pipette, Δp is the suction underpressure [80]. A bead was partially aspirated by a micropipette as described by Shao and Hochmut (1996) [107]. The diameter of the pipette was smaller than the bead, thus the bead was attached to the opening of the pipette. Another micropipette was used to aspirate a neutrophil cell. Diameter of this pipette was slightly smaller or equal to the diameter of the neutrophil. Adhesion strength is defined as the minimum force needed to detach a single cell from its substrate [38]. The technique was used to determine the minimum force (45 pN) necessary to form a membrane tether from neutrophils [107]. The smallest bead (3.2 μm) applied, resulted in a force of a few pN. As a high force limit, hundreds of nN could be exerted [46]. c. Experiment duration Lomakina (2004) hold a bead and a neutrophil in contact for a user-specified length of time (2s and 1 min ) and then separated them [120]. d. Advantages The strength of the MAT is in its simplicity. It can measure the force between cells without attaching cells onto a solid surface. Cell-cell interactions can be studied directly. Interestingly, a spherical cell can be used as a force transducer [46]. A constant localized force can be imposed to monitor the effect of the force on single-bond kinetics [46]. While the low force sensitivity of the MAT is similar to that of optical tweezers, it can exert much higher forces as well [46]. e. Disadvantages One of the cells (or probe) has to fit snugly inside the pipette. Forces smaller than 10-20 pN cannot be measured precisely when the diameter of the pipette is ~10 μm [107]. As the adhesion between the bead-transducer and the pipette wall would affect the measurement, it is important to insure that the bead does not adhere to the pipette wall [46]. Evaporation in the chamber can be a significant technical issue [107]. 28 f. Main applications Viscoelastic properties of soft cells e.g. red blood cells, white blood cells; and more rigid cells, such as endothelial cells can be measured with it [45]. Micropipette suction is a versatile method to study mechanical properties of living cells and to examine the viscous response of solid cells e.g. endothelial cells, chondrocytes [45]. It can measure cell-cell interactions, directly Lomakina et al. [46],[119],[120] investigated interactions between neutrophils and ICAM-1 coated substrates using 2 micropipettes. One of them (stationary pipette) hold the bead and the other hold the neutrophil to manipulate the cell [120]. 2.4 Computer-controlled micropipette Computer-controlled micropipettes can manipulate and sort cells in a Petri dish, individually [121],[122] (Fig.11). Cells are selected on the basis of their phase contrast and/or fluorescent images). Sorting is performed by a micropipette with an aperture of 10-70 μm with a sorting speed of 3-4 cell/min. After sorting, single cells are deposited into another Petri dish/multiwall plate/PCR tube or glass cover slip in a minute volume of liquid in the nanoliter-microliter range. Individual cells inside the drops on the glass cover slip can be studied with high resolution, immediately after sorting [123]. The technique is suitable for high-throughput single cell adhesion force measurements by repeating the pick-up process with an increasing vacuum. Adhesion force between individual human white blood cells and specific macromolecules were studied with the technique [124],[125],[126]. Hundreds of cells adhered to specific macromolecules can be measured one by one in a relatively short period of time (~30 min). 29 Fig.11 Schematic illustration of the computer controlled micropipette. The cells are approached by a micropipette from above, then a negative pressure is applied generating an upward fluid flow. If the lifting force is greater than the adhesion force between the cell and the surface, the cell is aspired. By increasing the vacuum in steps, the adhesion force distribution can be measured on a relatively large population of cells as the computer guides the micropipette through the Petri dish. The bottom surface can be functionalized by specific natural or artificial ligands. a. Number of cells in an experiment Single cell measurements in a cell culture dish containing hundreds of cells can be performed. In our experiments the total number of human immune cells probed by the micropipette was 200-600 for each of the three cell types [23]. b. Typical force range Experimental vacuum value in the syringe can be converted to hydrodynamic lifting force acting on single cells on the basis of computer simulations of the flow in the micropipette. To estimate the lifting force acting on a real cell, the total force on a model cell (e.g., a hemisphere with a diameter of 20 μm) can be determined in 3D simulations. 30 In our experiments with human monocytes, adhesion force of most cells fell into the [0, 2] µN interval on the fibrinogen surface [23]. c. Experiment duration Duration of the adhesion force measurements is typically 30 min [23]. d. Advantages It can measure the adhesion force of individual cells with relatively high throughput: hundreds of cells in ~30 min, especially when compared to AFM or FluidFM. Measurements can be carried out on cells incubated on the specific surface for several hours or days to investigate physiological, potentially strong cell adhesion. The device can be mounted onto a normal inverted microscope. Both biologically and medically relevant results gained with the automated micropipette were reinforced in standard microfluidic shear stress channels. Automated micropipette offers a higher sensitivity than the measurement using the shear stress of a microfluidic channel [23] and it has less experimental side-effect than the shear stress channel. Cells usually become elongated and aligned to the direction of the flow in a shear stress channel. Similar effect was not observed in the experiments with the micropipette. e. Disadvantages A drawback of the technique as compared to AFM: to calculate the value of the adhesion force, hydrodynamic simulations depending on cell size (and less sensitively on cell shape) have to be carried out. However, computer simulations are not needed when different cell types or different treatments have to be compared without a need for a scaled value of the adhesion force. 31 f. Main applications The adhesion force of cells attached to specific molecular surfaces can be accurately probed. Measurements can be carried out on cells incubated on the specific surface for several hours or days to investigate physiological cell adhesion. Cell-cell interactions can be also studied with this method [124]. 3. Optical tweezers (OT) Laser tweezers were developed for the microscopic manipulation of cells and organelles. Conventional manipulation techniques - including optical tweezers/optical traps, magnetic tweezers, acoustic traps and hydrodynamic flows - cannot achieve high sensitivity and high throughput at the same time. Conventional OT (Table 3) (Fig.12) use a strongly focused Gaussian laser beam to trap and manipulate microscopic objects such as small dielectric spherical particles (bead) [127]. Trapping lasers operating in the near infrared regime (800-1100 nm) minimize optically induced damage in biological specimens [128]. High NA (typically 1.2-1.4 NA) microscopic objective lens is used to focus the trapping laser [127]. A small bead is captured in an optical trap. The bead is positioned to touch the surface of a cell, and then the bead is pulled away from the cell until the chemical bond breaks. The force between the bead and the cell is determined on the basis of the displacement of the bead from the focus perpendicular to the optical axis. Optical tweezers allow fine control of positioning (~10 nm for trap beam stability) and of forces (~ 0.1 pN resolution) on a wide range of particle sizes (25 nm to 25 µm) in a non-invasive manner [129],[130]. Owing to their precisely controlled force-exerting characteristics, OT are often used for a variety of mechanical force measurements in the pN range for single cells. Although most cells cannot be directly grabbed by the OT due to their size or shape, a small number of cell types such as yeast cells, RBCs and spermatozoa are readily tweezed and provide model systems for such force studies [131]. For those cells that cannot be directly tweezed, the use of microspheres as handles for force probes has allowed the measurement of cellular properties such as membrane tension [127]. 32 Fig. 12 An optical trap is created by focusing a near infrared laser light to a diffraction-limited spot with a high numerical aperture (NA) microscope objective. The laser beam used for trapping has a 2-D Gaussian intensity distribution with an intensity gradient in the x- and y-plane (perpendicular to the beam axis). A particle (bead) is captured in the optical trap and positioned to touch the surface of a cell. Then the particle is moved away from the immobilized cell. The force between the bead and the cell is determined on the basis of the displacement (x) of the bead from the focus perpendicular to the optical axis. The particle displacement changes the intensity distribution of the transmitted infrared light. The shift of the intensity maximum can be detected with a quadrant photodiode (QPD). The signal of the QPD can be converted into units of displacement (nm) and force (pN). a. Number of cells in an experiment 1 cell/experiment. b. Typical force range 0.1-100 pN [127]. The trapping force depends on the intensity of laser power, the shape of laser focus, the size and shape of the trapped particle and the index of refraction of the trapped particle relative to the surrounding medium. It is difficult to measure the trapping force directly but there are several ways to calibrate it [127]. The external forces applied to a single particle within an optical trap can push a single particle away from the focus, and the force from the OT 33 draws the particle back to the center of the trap. In the equilibrium position, the external force equals the force from the OT. For a small displacement, the force from the OT, termed as restoring force, can be estimated by: F=−kX (6) where k is the trap stiffness and X is the displacement of the particle from the center of the trap [127],[130]. c. Experiment duration A bead carrying ligands on its surface is brought in contact with the cell. After an incubation time of a few seconds, it is pulled away from the cell applying stress to the chemical bonds [132]. When the cell itself is trapped for several minutes at high power (P > 300 mW), membrane stiffening can be observed, and the cell elasticity is affected. Thus the manipulations on the same cell should take normally less than 15 min [133]. d. Advantages OT offer exceptionally high force sensitivity [106]. The technique can be applied to measure weak forces of single molecules that cannot be achieved by traditional AFM or most of the alternative tools [127]. High spatial resolution [130]. Non-contact force measurement [127],[131]. e. Disadvantages Optical manipulation is limited in space due to focusing requirements [106]. Choice of the laser can be critical depending on the application. Laser absorption by the sample can lead to damage as the highly focused spot has an intensity in the range of MW/cm2 [131]. Thus wavelength is a central parameter when cells are trapped: wavelengths below 800 nm can easily damage cells [134]. Thermal effect of the high laser intensity has to be considered [127],[135]. Trapping in cell extract or in a medium containing impurities is generally precluded as trapped impurities can distort or mask the position signal. Still the optical trapping of lipid vesicles 34 within eukaryotic cells [136], and organelles in yeast cells [137] have been successfully implemented. The applied force is limited to a maximum of 100 pN set by the maximum laser intensity in the specimen plane [130]. f. Main applications Various cell types, including mammalian cells (red blood cells, nerve cells, gametes and stem cells), yeast cells [138],[139],[140] and bacteria, such as Escherichia coli have been studied [127]. In the binding experiments, the 3D manipulation capability is exploited to impose a specific interaction between the trapped object and a fixed partner, and to measure the force and displacement resulting from the interaction [130],[141]. A particle is decorated with ligand molecules binding to the cell surface. The optical trap pulls the particle away from the cell until the chemical bond breaks [132]. Measurement of the force and displacement of optically trapped kinesin coated beads moving along fixed microtubules was pioneered by Block et al. [142],[143]. The binding probability and unbinding force was measured between virus coated beads and erythrocytes. The binding force of single fibrinogen/fibronectin-integrin pairs could also be quantified in living cells [143],[144]. OT can capture the force characteristics of intermolecular bonds on the cell surface [132]. In the so-called tether-pulling experiments, OT have been applied to study the mechanical properties of cell membranes. Tether extraction is an accurate method to quantitatively characterize the plasma membrane. To form a membrane tether, micrometer-sized particles (beads) are used to grab the cell membrane. A bead trapped by the OT is held on the cell plasma membrane for a few seconds, and then moved away from the cell to pull out a membrane tether, a thin cylindrical strand of plasma membrane between the bead and the cell [145],[146]. The force needed to manipulate the bead can be measured by OT [127]. Thus the bending rigidity of the membrane can be determined [132]. OT can probe the viscoelastic properties of whole cells under physiological condition. For this two optical traps are applied to attach beads to two opposite points of the cell. It can give an insight into the internal structure and organization of the cell [132]. 35 4. Atomic force and fluidic force microscopy Atomic force microscopy (AFM) developed by Binnig et al. (1986) [147] can be applied to capture single cells with its functionalized cantilever gently pressed to the cell (Fig.13 A). This in turn converts the living cell to a probe brought into contact with a functionalized surface or another cell (Fig. 13 C). Subsequently, the cantilever is pulled back at a constant speed, detaching the cells from its binding side. Deflection of the cantilever is proportional to the force acting between the cell and the substrate. It is recorded as a force-distance curve [43]. Deflection of the cantilever is detected with a laser beam focused on the top side of the cantilever and reflected into a quadrant photodiode [148]. In the force spectroscopy mode, AFM (Table 4) acts as a versatile tool to probe nanomechanical properties and to extract quantitative parameters of biological systems, including tissues, cells, proteins and nucleic acids, and of biomimetic systems, such as functionalized surfaces or matrices. AFM-based force spectroscopy [43],[149] involves an atomic force microscope tip to locally record interactions with the sample or its nanomechanical properties. Approach and retraction force -- distance curves characterize the sample deformation and tip -- sample adhesion, respectively [150]. Analysis of the approach force -- distance curve, and in particular the region describing indentation, allows properties including deformation, elasticity and dissipation to be determined. The retraction curve quantifies the adhesion force between the tip and sample. Single- molecule force spectroscopy (SMFS) [151],[152] is frequently applied to detect the binding strength of ligand receptor pairs. It was applied to measure the force required to unbind streptavidin and biotin [153], it was quickly recognized that measuring rupture forces provides information about the kinetic properties of a bond. Single-cell force spectroscopy (SCFS) [154] measures the adhesion of a single cell to a biointerface, which can be tissue, another cell or a surface functionalized with ligands. A single cell is attached to the free end of a tipless cantilever [150]. A number of methods have been developed to attach the cell onto the AFM cantilever and to allow probing cell-cell or cell- substrate interactions [148]. Specific receptor-ligand interactions [155],[156] electrostatic interactions [157], glue [158] or chemical fixation [159] can all be applied. An important issue when applying these protocols is to make sure that the native surface of cells is not altered or denatured. In this respect, an appropriate approach is to attach individual cells to the AFM cantilever with lectin. This method allowed the measurement of the adhesion force between 36 two neighboring cells of D. discoideum on the single-molecular level [148],[155]. After that, the functionalized cantilever is lowered into contact with a trypsinized cell, which readily attaches to the cantilever. Then, the probe cell is brought into contact with a biointerface for a given contact time and force, and then withdrawn while a force -- distance curve is recorded. Analysis of the force curve provides the maximum adhesion force of the cell. This approach allows investigate how cells strengthen adhesion to ECM proteins or other substrates [150]. AFM-based force spectroscopic modes enable the characterization of single receptor -- ligand bonds, protein unfolding and refolding, and the mechanoelastic properties of peptides, nucleic acids, sugars and polymers. Cell adhesion to the interface of substrates, other cells or tissues can also be quantified using such modes. AFM can be used in 2 ways: it can either measure the adhesion of the cantilever-attached cell to the substrate surface or the adhesion of the cell immobilized on the substrate to the cantilever [149] . Fig.13 Schematic illustration of atomic force microscopy applied for cell adhesion measurements. The interaction between the cell and the tip is measured through the displacement of the laser beam on the photodiode caused by the deflection of the cantilever. The latter acts as a spring probing the force acting between the tip and the cell. The tip surface can be functionalized with a molecule of interest then it is pushed to the cell for a predetermined time that allows the receptor-ligand bonds to form (panel A). As the tip is pulled back with a controllable speed, the distance-force curve can be registered by measuring the signal on the photodiodes. As an alternative to tip functionalization with biomolecules, a living cell can be attached to the tip (panel B). Then the cell attached to the cantilever is pushed to the functionalized substrate surface or to the other cell on the substrate, thus allowing the direct measurement of cell-cell adhesion (panel C). Since both the spatial resolution and force sensitivity are high, the AFM was the first method capable of measuring cell adhesion on the single receptor level. 37 4.1 AFM to probe force normal to the surface a. Number of cells in an experiment Each cell requires a separate cantilever that must be functionalized (typically ~30 min) and calibrated, which impedes the ability to obtain high-throughput measurements [48]. At least three force curves have to be recorded at different locations in each cell. Although it is beneficial to take multiple curves on each cell for reliable statistical data, taking too many force curves can lead to changes in cell stiffness due to the mechanical stress caused by the AFM probe. AFM yields only a few measured cells (from 1 to 5) per condition in most studies [160]. b. Typical force range The AFM (Binnig et al., 1986) can be used to measure nano-newton to pico-newton forces and micrometer to 0.1 nm displacements [161]. Imaging in aqueous solution allows observation of biomolecules under physiological conditions and estimate strong, capillary forces between the sample and probe when the imaging is done in air [162],[163]. In order to prevent deformation of weak biological sample, a vertical force is used (50-100 pN=minimum force). At 100-1000 pN force, the biological sample is reversibly deformed, and above this value irreversibly [163],[164],[165]. AFM can be used to analyze accurately and then remove genetic material from the chromosome [162],[166]. Antibodies that specifically bind to the cytoplasmatic membrane surface can be removed with smaller than 0.8 nN [162],[163],[167]. In the future it is expected that the use of smaller cantilever improves the resolution, allowing measurement of the smaller non-binding forces. AFM-based SCFS (single cell force spectroscopy) enables measurement of wider force range (5 pN-100 nN) [149]. An individual WM115 cell was bound to a lectin functionalized cantilever with a given force (500 pN) by Puech et al. and then the cantilever with the bound cell was slowly retracted 100 μm from the substrate. This allowed a strong adhesion of the cell to the cantilever (5-15-min). After that, adhesion of an individual, captured melanoma cell and a model endothelial layer of HUVEC cells were tested. A given force (few 100 pN) was applied on the cell layer for a certain time (range of sec). Finally, probe cell was separated from the surface while detachment event 38 was recorded [44]. This work described that, AFM is a sensitive and quantitative method to measure long-distance cell-adhesion forces. A D. discoideum cell was bound to the AFM cantilever by Benoit et al [155]. Then a target cell was placed at the bottom of the Petri dish and it was approached while a defined repulsive contact force was established. This force was held constant to enable establishment of cell adhesion. While the cantilever was retracted, it was recorded the force as a function of the distance until connection between the cells was broken. It was allowed the interaction for 20 s at the force of 150 pN. Adhesion between these cells produced unbinding forces of the order of 1 nN [155]. SCFS (single-cell force spectroscopy) was used by Taubenberger et al. [168] to investigate the α2β1-mediated cell adhesion to collagen I.; they compared adhesion of CHO-WT and CHO- A2 cells to collagen. It was found that, CHO-A2 cells adhered to collagen (40-600 pN) strongly, than another cell (<50 pN). When contact time was < 120 s, usually 2-5 force cycles/cell were carried out, whereas for longer contact time, 1-2 force curves/cell were recorded [168]. Lehenkari and Horton determined integrin-ligand interaction of osteoclasts. Integrin-binding forces were measured in intact cells by AFM for several RGD-containing ligands with a range of 32-97 pN. It was reported that, AFM can be applied in cell biology studies and it provides an opportunity to analyses receptor-ligand interactions in the cell membrane [10]. c. Experiment duration First, a chemical functionalization of the cantilever is needed allowing irreversibly ''glue'' the cell of interest to the cantilever; where the lever is coated with wheatgerm agglutinin for at least 1 h. Consequently, each cell requires a separate cantilever that must be functionalized (typically ~30 min) and calibrated, which impedes the ability to obtain high-throughput measurements [48]. The AFM cantilever is positioned using a piezoelectric crystal, and its deflection is measured by laser reflection onto a split photodiode. Positioning precision in the z-direction is 1 Å, and force sensitivity is within 3 pN. The cantilever is moved with a velocity of 2.5 ± 0.5 µm s -- 1 [155]. Cells are immobilized on a substrate or on the force sensor itself [155] (Fig.13 A,B). The sensor surface is functionalized with an adhesive molecule [169] to attach the cell to it. Cells immobilized on the sensor were used to investigate cell-cell interaction [161] (Fig. 13 C). To avoid scattering of the laser beam of the detection system, non- or weakly-adherent cells are 39 removed by gently rinsing the dish. An attached cell is slightly loosened by pushing its flank with the side of the cantilever. The extreme end of the lever is then lowered onto the cell at a force of a few nN and hold in contact for ~ 30 s to allow the lectin on the lever to bind; the cell is then lifted off the bottom of the dish [155]. Using AFM, it needs ~10 min for proper cell -- cantilever interaction. In this technique, the force range is limited by the force with which the cell is bound to the cantilever, which usually results in the application of only relatively short contact times (msec-20min) [43],[170] between the cell and the substrate of less than one hour before the adhesion force exceeds the detectable range [48],[161]. d. Advantages AFM can measure adhesion from a single molecular event in intact cells under physiological conditions. Both spatial and force resolutions are high. AFM has force sensitivity in the pN range and nm positioning accuracy, therefore atomic force microscopy is a powerful device to explore dynamics and strength of interactions between individual ligands and receptors. These studies require adhesion of specific biomolecules or cells onto the AFM tip or to a solid surface. Currently, AFM is the only force- measurement technique which can map and analyze individual receptors with nanoscale lateral resolution [148]. AFM has several unique properties: it can be operated in solution, allowing observation of biological structures in native environment [171]; individual proteins can be observed at a resolution >1 nm [165],[172] conformational changes of single biomolecules can be directly monitored [162],[165],[170],[173]. e. Disadvantages State- of- the- art AFM measurements have limitations, for example adhesion measurements, which use single cells are time-consuming, costly methods. Only one cell can be characterized at a time. Each cell requires a separate cantilever which must be calibrated and functionalized [43],[48],[149] impeding to obtain high-throughput measurements [170]. Chemical attachment of the cell to the cantilever is cumbersome (~30 min) and it can alter the physiology of the cell [161],[174]. High number of detachment force-distance curves need to be measured to gain reliable statistics which restricts the length of the contact time. Usually only a relatively short contact time is established between the cell and the substrate (msec-20min) [43],[170]. Another 40 challenge is to develop a fast scanning AFM as the temporal resolution of the method is limited [148]. Thermal drift is almost unavoidable in AFM experiments which complicates long- contact-time experiments (>20min). Strong adhesion of the cells falls out of the force range of the technique after a longer contact time (>1h). f. Main applications Due to its wide force detection range, it can measure both the adhesion of a whole cell and interaction [175] in a ligand-receptor system [43],[149]. It can study the dynamic formation of cellular adhesion. It has been adapted to measure cell-cell and cell-substrate adhesions. AFM can be combined with modern optical imaging techniques [43]. It can quantify forces guiding microbial cell adhesion [78],[175],[176],[177],[178]. 4.2 AFM to probe tangential/lateral force Manipulation force microscopy [2] (Table 4), an atomic force microscopy can measure the force that dislodges micro-sized objects attached to the surface. The technique applies laser beam deflection and an inclined AFM cantilever to measure the force. The cell is brought into contact with the cantilever, and a tangential force is applied gradually. Finally, the cell is released from the surface as all adhesion bonds are broken. a. Number of cells in an experiment Adhesion strength of cervical carcinoma cells were examined by this technique. In an experiment, force measurement on ~200-300 cells were carried out [2]. b. Typical force range Adhesion force of several hundred nN was measured on different surfaces [2]. A force of 19 nN was required to detach E. faecalis cells from hydrophobic materials, and 6 nN to detach from hydrophilic materials [176]. Chang and Hammer in their numerical simulations of functionalized microbeads found that the lateral detachment force is several times lower than the perpendicular detachment force. Thus a tangential force is considered to be more effective 41 to detach a particle than normal forces [177]. The reason is thought to be the low ratio of the bond length to the bead radius. c. Experiment duration Adhesion time is usually between 10 and 90 min [2]. d. Advantages Studies have shown that the lateral force required to displace a bacterial cell is considerably (up to 10 times) smaller than the perpendicular force [177],[178]. Enables the real-time imaging of bacterial adhesion and aggregation under physiologic conditions or affected by antibiotics [176]. e. Disadvantages Interpretation of the experimental data is not straightforward [2]. The measured force depends strongly on the details of the experiment: the shape of the tip, the scanning speed, the torsional spring constant of the cantilever, the exact direction of the applied force, and the nature of the binding between the cell (or bead) and the substrate [78],[179]. f. Main applications Binding force of protein-covered silica spheres adsorbed to polystyrene surfaces [180]. Adhesion force of individual cervix cells to various substrates [2]. Characterization of medical- grade polymers and their resistance to microbial adhesion and biofilm formation [176]. 4.3 Fluidic force microscopy (FluidFM) FluidFM (fluidic force microscope) is a unique micromechanical and microfluidic device that combines the force-controlled high spatial precision of AFM with the capability of direct liquid delivery by microfluidics. The needle-like AFM probe or the bare cantilever of the FluidFM contains a fluidic microchannel inside (Fig.14). It is directly connected to an external fluid circuit controlling the pressure and thus the flow in the microchannel. This novel technique can both mechanically manipulate living cells and microinject biochemical reagents into them 42 [181]. The whole FluidFM BOT system involved an inverted optical microscope [182]. Individual microbial (S.cerevisiae, C.albicans) and mammalian cells (HEK, HeLa) were used to measure the cell adhesion force [48]. FluidFM (Table 4) reuse the same probe for the measurement of multiple cells to record single cell force spectroscopy (SCFS) curves reducing the time required to obtain statistically relevant data compared with conventional AFM [48]. Trapping the cell to the cantilever occurs within a few seconds before the SCFS measurement. Then the cells can interact with the substrate or the other cell for a longer time. During the measurement, first the cells are selected, and then it is approached with a set point (10 nN for yeast cells and 50 nN for mammalian cells), after that it is followed a pause (5 s for yeast cell and 3 s for mammalian cell) with force feedback using underpressure and the cell is immobilized to the cantilever. While using AFM, it requires ~10 min for cell-cantilever interaction but with FluidFM needs only 5 sec underpressure for cell-cantilever contact. In the conventional AFM, cells are immobilized to the cantilever, while FluidFM to fix the cell to the cantilever, apply underpressure. Here chemical fixation of the cell to the cantilever is replaced so that cells are sucked into the aperture of a hollow cantilever (few sec). The cell release from the substrate so that it is applied an overpressure pulse makes the immediate reutilization of the cantilever possible, thereby resulting in the ability to perform serial measurements. Potthoff et al. have shown that FluidFM can increase the amount of data that can be recorded in a single day. Compared to the conventional SCFS, FluidFM based SCFS can be performed up to tenfold as many experiments / day and 200 experiments can be carried out with this technique compared to the conventional cantilevers [99]. This approach is similar to the microinjection with glass pipette, but there are differences. Microinjection requires optical microscopy. Cells are often damaged, however using AFM precise force feedback reduces potential damage to the cell [181]. 43 Fig.14 Schematic illustration of fluidic force microscopy. The nanofluidic channel inside the cantilever allows the grabbing of the cell by the cantilever or tip using a negative pressure. When the upper side of the cell is immobilized this way, the tip is moved upwards picking the cell up from the surface. This method can be used to directly measure the complete adhesion force and energy between the adhered cell and the surface in a less difficult way than using the traditional AFM method. Another advantage is the rapidity of the measurements, as the tip can be reused after a ~ 5 min washing step. The upper limit of the measured force is set by the aperture of the fluidic microchannel and the spring constant of the cantilever. Since there are multiple tip designs available, the same device can be used for other applications, such as cell injection, lithography, etc. a. Number of cells in an experiment It was compared the maximum adhesion strength of 2 types of yeast cells (S.cerevisiae, C. albicans) to hydrophobic and hydrophilic surface, after an adhesion time of 15 min. Then HeLa (n=11-12 cells were examined serially) and HEK cells (n=8-9cells) were compared on glass and fibronectin-coated surfaces. Up to 200 yeast and 20 mammalian cells per probe can be performed with this technique [48]. In half an hour, ~10 cells can be studied using this technique. b. Typical force range The measured adhesion forces were between 500 pN-1.6 µN [48]. 44 It has been shown that, adhesion force is increased as a function of incubation time and dependent on the temperature. It has been demonstrated rapid and serial yeast cell adhesion which can available with FluidFM so it can be investigated adhesion strength to the specific substrates and it can be determined long-term adhesion interactions. It was studied the adhesion of C. albicans to hydrophobic surface that was 39 nN and to hydrophilic surface was 10 nN. Adhesion strength of S.cerevisiae hydrophilic and hydrophobic surfaces was 2 and 5 nN, respectively. In both cases the same behavior was occurred; adhesion to the hydrophobic surface was stronger. It was studied whether FluidFM is suitable to detach mammalian cells from standard surfaces while long-contact time is established between cell and substrate but only a minimum time is required to fix the cell to the cantilever. HeLa cells on fibronectin surface showed 40-fold greater adhesion strength as compared to yeast cells. HeLa was compared to another cell line, HEK, and adhesion of this cell to fibronectin was 53 nN which is much smaller than in case of HeLa [48]. c. Experiment duration Compared with the conventional SCFS, using FluidFM-based SCFS allows to carry out 10- times more experiments (using FluidFM can be performed up to 200 attempts while using AFM only 20 attempts). Only a few minutes is needed to target, immobilize, and release the cell as well as to change the cantilever position to the next cell [48]. d. Advantages It can be measured the adhesion force directly. It has been demonstrated a fast and serial yeast cell adhesion measurements. It can be monitored adhesion of single cells to specific substrates and can be quantified the long-term adhesion interactions. The technique enables long contact times (orders of magnitude) between the cells and substrates. It has been demonstrated the universality and broad applicability of this method for different cell types. It can be measured higher cell-substrate adhesion forces (one order of magnitude), than using AFM [43]. Much shorter measurement time and many more adhesion measurements can be performed with it. Instead of chemically immobilizing the cells to the cantilever as in conventional AFM experiments, it is applied underpressure to fix the cell to the cantilever aperture [183]. 45 e. Disadvantages Using mammalian cells, it was unable to obtain the same quantify of serial measurements which were monitored with yeast cells, but it could be recorded 10 times more force curves than with conventional SCFS [48]. Throughput is lower [181] than that of the automated micropipette. Microfabricated cantilevers come with a high measurement cost. Cells come into direct contact with the cantilever potentially perturbing or damaging cells [181]. f. Main applications The universality and versatility of the FluidFM opens the way for original experiments in physics, materials, sciences, chemistry and molecular electronics. The method reduces the time required to obtain statistically relevant data compared with conventional SCFS. The use of FluidFM based SCFS enables AFM based adhesion force measurements [48],[184]. This is a universally applicable method for living cells. It can measure the adhesion force of those cells that were difficult or impossible to measure with conventional AFM [48]. Microinjection of cells [185]. It can be used for printing of 2D surface patterns with various formations [186] or for precisely controlled template-free 3D micro- and nanoprinting [187],[188]. Micropatterning of living mammalian cells on carboxymethyl-dextran (CMD) hydrogel layers was presented using the FluidFM BOT technology [182]. 5. Cell traction force microscopy and FRET force sensors 5.1 Cell traction force microscopy Cell traction force microscopy (CTFM) [49] (Table 5) enables the force (mechanical stress) field generated by single cells on an elastic substrate to be reconstructed with a resolution of ~1-10 μm. The workflow of the technique is built up of four subsequent procedures, namely: 1) preparation and characterization of the substrate; 2) microscopic imaging of cells and deformation markers inside the elastic substrate; 3) computational recovery of the displacement field of the substrate; 4) computational determination of the traction stress field from the displacement field, based on an elastic model. 46 Most often, the substrate is a polyacrilamide (PA) gel coated with proteins that specifically promote cell adhesion (Fig. 15). PA has the advantage that it is linearly elastic within a wide range of mechanical stresses and thus completely recovers after stress removal [189]. Furthermore, by varying the degree of crosslinking in the PA gel, its rigidity can straightforwardly be tuned in the physiologically relevant range of 0.1-100 kPa (1𝑘𝑃𝑎 = 1𝑛𝑁𝜇𝑚−2). The PA gel is embedded with randomly distributed fluorescent beads (usually 200- 500 nm in size) which are then used as markers of the substrate deformation. Fig.15 Schematic illustration of the principle of cell traction force microscopy. The substrate (in purple) is randomly embedded with fluorescent beads with a diameter of around 500 nm. Then the cells are placed on the surface and let to adhere for a predetermined time. As the cells interact with the adhesion proteins on the polyacrilamide substrate, a force is exerted that rearranges the previously recorded bead distribution. The fluorescent patterns before and after cell adhesion are captured by wide field microscopy. From this displacement field, the traction field can be reconstructed by an elastic computational model with a spatial resolution of ~1 µm. Using the 3D version of the method, the traction field can be measured in 3D cell cultures. CTFM has been successfully applied for the mapping of individual cell's traction fields as well as for the examination of the migration of cell groups. b. Typical force range Traction force microscopy relies on culturing cells on soft compliant polymer films that deform under tension [190]. These deformations are then deconstructed using computational finite 47 element analysis to calculate the lateral force vectors applied to micrometer elements. Using this method, the literature estimate for integrin tension within focal adhesions is approximately 2-3 pN per receptor [191]. The typical force range is nN in magnitude [191]. c. Experiment duration Up to several hours when combined with time-lapse imaging. After pre-incubating the cells on the substrate for an arbitrary time, comes the imaging stage of the workflow when at least a set of three photomicrographs are taken. First the cell-substrate contact is imaged, often in simple phase-contrast mode. Next, the microbead distribution near the surface of the substrate is captured in fluorescent mode for the cases when i) the cells exert forces onto the substrate (force-loaded image) and when ii) they have been removed from it via trypsinization (force- free image) [192],[193]. Fluorescent imaging is generally done with a wide field epifluorescent microscope. d. Advantages Since its introduction in 1996, CTFM has undergone impressive developments. Innovations at the substrate level gave rise to CTFM variants which characterize the substrate deformations in a more or less different manner as described above [194]. For instance, a CTFM variant utilizes a PDMS substrate with a flexible array of single micropillars which act as independent tiny strings. Unlike the original CTFM, this variant has the advantage that traction forces can be easily calculated from the deflections of the micropillars [194]. On the other hand, cell adhesion is artificially constrained to discrete islands, and the spatial resolution of the mapping of traction forces is limited by the number of the micropillars. Recent technological and/or computational advancements have led to the development of i) high resolution CTFM offering a spatial resolution of ~1 μm [190] or even super-resolution [195]; ii) 2.5D CTFM, which enables all components (x,y,z) of the traction generated by a cell on a 2D substrate to be determined [192]; and iii) 3D CTFM, which permits traction force measurements for the case of cells embedded in a 3D matrix [196],[197]. e. Disadvantages 48 In order to recover the displacement field, the pixels of the force-loaded and force-free images have to be one by one appropriately matched (image registration problem). The most straightforward method would be to directly monitor the movements of the microbeads but given the high number (few thousands) of microbeads, this task is computationally demanding and is thus rarely used. Albeit alternatives to this method have been proposed, image registration methods continue to share some common problems, such as the deterioration of the displacement field when the beads become more sparsely distributed, or the difficulty of error estimation (for reviews see [193] and [194] ). Nevertheless, once the displacement field is calculated, the deconvolution of the traction field remains the last step in CTFM to be carried out. It is critical to deliberately choose an elastic model to formulate the relation between the displacement and traction fields. Generally, the substrate is considered as a semi-infinite half space, thus only the substrate surface enters analysis and the simple Boussinesq equation [198] has to be solved . However, it has not got a unique solution and therefore regularization terms have to be added to exclude non-physical solutions. Furthermore, the inversion of the elastic equations, and thus the computed force field are very sensitive to noises. To circumvent this disadvantage, more sophisticated approaches have been proposed [193],[194]. f. Main applications CTFM has hitherto been applied to map the traction field of individual cells [199],[200] and that of cell aggregates [201]; to detect phenotypic changes that are accompanied by a change in traction forces during cell differentiation [202]; and to investigate the migration of individual cells [203], as well as the collective migration of a sheet of cells [189],[204]. 5.2 FRET force sensors FRET (Förster/Fluorescence resonance energy transfer) (Table 5) is a form of non-radiative energy transfer between two fluorophores (donor and acceptor) [205] due to dipole-dipole interaction. Fluorescent organic molecules have been widely used as donors and acceptors and they offer advantages such as small size, compatibility with numerous and simple covalent 49 coupling strategies, and a relatively strong optical signal. Careful selection of an appropriate donor−acceptor pair ensures high transfer efficiency and provides two measurable parameters: quenched donor photoemission and enhanced acceptor fluorescence [206]. FRET probability depends non-linearly on the distance between the donor and acceptor. This technique, capable of measuring distances on the 2-8 nm scale, relies on the distance-dependent energy transfer between the donor and acceptor fluorophores [207]. It is widely applied for mapping large scale protein structures as a ''molecular ruler''[205],[208],[209]. FRET measurements can follow receptor−ligand interactions, changes in protein conformation upon binding a target analyte or can be utilized to reveal the response to changes in the solution conditions (e.g., temperature or pH). It can also monitor the nanometer scale displacements of cell adhesion ligands. A molecular force sensor can be constructed by attaching the donor and acceptor to the ends of a molecular spring: a flexible chain of atoms, usually a polymer. The distance of the donor and acceptor changes, when the molecular spring is under tension [50], and thus the tension can be read out from the FRET signal (Fig.16). Tension across the cellular receptor of interest leads to an elongation of the force-sensitive unit, which then can be microscopically detected [210]. FRET-based biosensors need high sensitivity microscopy and appropriate data analysis algorithms to determine the force in cells [211]. The most frequently used method is based on intensity measurements, in which the donor fluorophore is excited and the emission intensities of donor and acceptor fluorophore are used to calculate the FRET ratio. However, these measurements do not readily yield quantitative information on FRET efficiency, they are sensitive to the experimental settings and require careful image analysis. Alternatively, fluorescence lifetime imaging microscopy (FLIM) can be used to calculate FRET efficiencies from the donor lifetime in the presence or absence of the acceptor [211]. The molecular tension sensors can be divided into two categories, those that are genetically engineered and expressed within living cells (genetically encoded molecular tension sensors, GETS) and those that are anchored to a surface to probe cellular receptor forces at the interface between living cells and their external ligands (immobilized tension sensors) [212]. a. Number of cells in an experiment 50 Analysis was performed by using measurements from a minimum of 10 cells at each condition [209]. b. Typical force range As FRET force sensors measure the tension built up in single molecules, the overall adhesion force of the cell cannot be simply determined on the basis of FRET microscopy. FRET-based tension sensors provide piconewton (pN) sensitivity within cells [210]. GETS are limited to the detection of forces within the range of 1 to 7 pN [212]. In 2010, Grashoff et al. designed a tension sensor module (TSMod) resolving forces lower than 6 pN. The talin, a molecule that connects integrin receptors with the actin cytoskeleton through multiple interaction sites experiences mechanical forces of ~ 7 -- 10 pN [210]. The tension across vinculin in stationary focal adhesions (FAs) was measured to be ~ 2.5 pN [50],[212]. In the last years, TSMod-s gained popularity [50] and have been used to determine tension across the adhesion protein vinculin [50],[213], cadherins [214],[215],[216], PECAM-1 [216], spectrin [217] or the glycoprotein MUC-1 [217]. Immobilization of molecular tension sensors to a solid support allows forces between cell membrane receptors and their extracellular ligands to be investigated. Integrin receptors exert an adhesion force of 1-5 pN to their ligands [218]. c. Experiment duration ~ 20 min [209]. d. Advantages Nanoscale interactions can be detected on the basis of the FRET [219] signal. FRET can be applied as a molecular force sensor built into single biomolecules. Genetic insertion of such a tension sensor module into the protein of interest and the expression of the resulting construct in cells allows the analysis of molecular forces in living cells [211]. It allows the measurement of single molecule reaction trajectories from about 1 millisecond to several minutes [220], [221]. FRET does not require mechanical perturbation of cells, which could alter cellular traction and adhesion forces. FRET can study how cells manipulate the ligands to which they adhere, and simultaneously determine cellular traction forces without perturbing the adhesion events [209]. 51 A fundamental advantage of the technique is to measure the internal distance in the molecular frame rather than in the laboratory frame and hence it is largely immune to instrumental noise and drift [220]. The technique is adaptable to a wide variety of instrumentations, including fluorescence spectroscopes, conventional, total internal reflection (TIRF), confocal microscopes, and FACS [219]. e. Disadvantages Artificial molecular force sensors are needed to be constructed and built into the cellular system, in most cases into a specific protein. Practically only one (or very few) protein type(s) can be measured in an experiment. Thus the overall cellular traction or adhesion force exerted by a large number of different proteins cannot be determined. The selection of an appropriate elastic molecular element of the force sensor is critical. The elastic linker has to be short or the increase in linker length has to be sufficiently large so that the applied tension can be observed as a change in FRET efficiency [211],[212]. FRET requires spectrally matched fluorophores. An ideal fluorophore for single molecule studies must be bright, photostable, small and water soluble. A problem inherent to any molecular force sensor is that its insertion can interfere with the properties of the host protein, and the host protein can interfere with the function of the force sensor [50]. f. Main applications FRET can be used as a molecular ruler to monitor the nanometer scale displacements between adhesion ligands, and the corresponding traction force between integrin receptors and adhesion ligands [209]. It has been applied to quantitatively analyze the parameters of cell interactions with both 2- and 3-dimensional adhesion substrates [209],[219]. It is widely used to study intermolecular interactions [205],[222],[223]. The number of bonds between RGD and integrin receptors can be estimated in a 3D synthetic ECM [224]. Functionalization of the flagelliform peptide with organic dyes and RGD-ligands allows the estimation of force across single integrin receptors [218]. A FRET -- based biosensor can follow the distribution of RhoA and Rac1 activities during cell -- cell adhesion [225]. FRET can be used to explore the role of mechanical forces across proteins including actin-binding proteins and cell adhesion molecules like 52 cadherin, PECAM-1 [216] and vinculin [50] in cellular systems. This tool may help to develop appropriate synthetic matrices useful for tissue engineering or cell-based therapies [209]. Fig.16 Schematic illustration of the operation of FRET force sensors. The method relies on the non-radiative energy transfer between two fluorophores. The emitted fluorescence of the donor molecule is quenched by the vicinity of the acceptor. Thus their distance can be calculated from the fluorescent signal. Binding a donor and an acceptor to both ends of an elastic biological polymer (such as a protein) allows to monitor the end to end distance of the polymer, which in turn characterizes the force exerted on the molecule. The technique is especially suited for the investigation of adhesion molecules such as integrins and actin binding cytoskeletal proteins. Summary Current review provides a guide to choose the appropriate technique for the measurement of cell adhesion in vitro. Most important parameters to be considered are the i) number of cells to be measured, ii) range of the adhesion force, iii) duration of the experiment. We identify the major advantages and disadvantages of each method. We propose that the combination of a 53 simple cost effective method to gain data of the average behavior of a larger cell population and an advanced technique for single cell targeting can be a good strategy to obtain powerful data for biological and biophysical model development. We also pointed out that the preparation of the adhesive surface for subsequent cellular studies has a crucial importance [22],[23]. Masking (blocking) the surface area not covered by specific biomolecules is a prerequisite for relevant results cleansed from the contribution of non-specific cell adhesion on glass, plastic or other artificial substrates [26],[30],[32],[33]. Thus we collected relevant examples and emphasized the importance of carefully controlling and pre-testing the surface chemistry. 54 Method a) Number of cells in an experiment b) Typical range of shear stress c) Experiment duration 1.1 Centrifugal assay large number of adherent cells [68] 1-2000 pN shear stress [31],[42], [69],[70],[71] 5-10 min [31],[42], [69],[72] 1.2 Spinning disk 1.3 Radial flow chamber 1.4 Parallel plate flow chamber 1.5 Microfluidic rectangular channels population of cells in a single experiment [76] population of cells in a single experiment [42],[79],[80] * 300 cells/ experimental group/ experiment [92] * 100-1,000 cells/mm2 [88],[91],[92] * ~100 cells/ experiment [102] * Number of analyzed cells: 100 - 1,000 [61], [100],[101],[1 02] 1-200 Pa shear stress [74],[77] 5-10 min [76],[77] < 20 Pa shear stress [42] 5 min [79],[81], [83],[84] * < 20 Pa, shear stress [42] * 1-25 Pa [89],[90], [92],[93] A few sec- 210 min [89], [90],[92], [93],[94] Several hundred Pa [61],[101] 10-15 min [61],[101] d) Advantages * Simplicity * Widely available setup * No special equipment required, only common lab instruments [42] * Applying a wide range of stresses in a single experiment [74] * High stresses * Flexible * Direct visualization of cells * In situ observation of cell detachment [79],[83] * Simplicity [95] * Well-characterized fluid flow field [89] * Direct visualization of cells [94] * It can monitor cell trajectories, speed and adhesion. Motion of spherical cells in the flow can be calculated. [91] * Direct visualization of cells [100] * It can capture the cell detachment * Wide range of adhesion forces can be measured * High-throughput [61] * Simple experimental setup e) Disadvantages f) Main applications * Sensitive, quantitative measurement of weak binding forces [68] * Quantifying cell adhesion on SAMs [69] * Quantifying cell adhesion to specific RGD-modified substrates [31] * Employing in more fundamental experiments [40],[75], [77] * In situ detachment of fibroblasts [79],[81] * Determination of the adhesive nature of modified substrates [79],[82],[86] * Determining the adhesion strength of fibroblast cells to various substrates [89] * Investigation of cell retention, morphology and migration as a function of shear stress and of adhesion time [88] * Investigating the relationship between adhesion strength and cell geometry [101] * Not possible to precisely determine the force acting on the cell * Limited to weakly adherent cells * Limited to short- term adhesion studies [42] * Not compatible with in situ microscopy, nor with the direct visualization of cells [61] * Limited to weak adhesion forces [81] * >200 μm channel height in most PPFC limiting the cellular adhesion force that can be measured [61] * Sustained flow affects the cytoskeleton and the morphology of cells attached to the surface. Cells become elongated in the direction of the flow being an apparent side effect of the measurement [101] Table 1. Centrifugal and shear flow assays. 55 Method a) Number of cells in an experiment b) c) Typical Experiment d) Advantages e) Disadvantages force range duration f) Main applications * Cells are brought into contact for ~10 sec [110] * Force measurement between two cells for 120 min [47] * Cell-cell contact area and contact time can be controlled [46],[110] * Measures cell adhesion force directly [46] * Low throughput, manual measurement * Measuring the adhesion force between two similar or different single cells in vitro [47] 2.1 Step pressure technique ~10 pairs of cells [109] 2.2 Biomembrane force probe 1 cell/ experiment * Sc represents the adhesion strength; Sc =1.5 nN/μm2; 744 nN separation force [46],[47] * Difficult to measure the exact contact area between the cells  Vertical mode: 0.2-0.5 pN  Horizon NA tal mode: 1-10 pN [112] * Adjustable loading rate of the force [46] * Sub-pN force and nanometer scale displacement resolution [46] * Gentle measurement minimally altering the cytoskeleton [111] * Can detect weak adhesive bonds * Simplicity * Measures the force without attaching cells onto a solid surface * Can study cell-cell interactions directly * Higher forces than with optical tweezers * Versatility [46] * Relatively high throughput [23], especially when compared to AFM or FluidFM * Direct visualization of cells [23] * Higher sensitivity and less side-effect than in microfluidic shear stress channel [23] * The resolution of probe movement is not as good as in AFM [111] * Studying receptor- ligand binding [46] * Alignment of probe and cell: one of the cells (or probe) has to fit snugly inside the pipette * Cannot measure <10-20 pN forces precisely when the diameter of the pipette is ~10μm [107] * Low force sensitivity * Evaporation in the chamber is technically challenging [118] * Severe cell deformation prior to detachment * Measuring viscoelastic properties of soft cells * Studying mechanical properties of living cells [118] * Hydrodynamic simulations are needed to convert the experimental vacuum value to hydrodynamic lifting force [23] * Probing single cell interactions with specific macromolecules [23] * Studying cell-cell interactions [124] 2.3 Micropipette aspiration technique 1 cell/ experiment * A force lower than 10-20 pN [46] * Hundreds of nN forces [46] User- specified length of time (2s and 1 min) [120] 2.4 Computer controlled micropipette Hundreds of cells: total number of human immune cells probed by the micropipette was 200-600 [23] 0-2 µN [23] ~30 min [23] Table 2. Micropipette manipulation assays. 56 Method a) Number of cells in an experiment b) Typical force range c) Experiment duration d) e) f) Advantages Disadvantages Main applications 3. Optical tweezers 1 cell/ experiment 0.1-100 pN [127] Manipulations of a cell takes less than 15 min [133] * Extreme force sensitivity down to 0.1 pN [127],[130] * High spatial resolution (0.1-2 nm) [130] * The range of applied forces is limited to max. 100 pN [130] * Photodamage and thermal damage [127], [135] * Contactless force for manipulations [127] * Interaction and binding assays [130], [132], [141],[142], [143],[144],[226] * Tethered assay [127],[132], [145],[146] Table 3. Optical tweezers. 57 Method a) Number of cells in an experiment b) c) Typical force range Experiment duration d) e) f) Advantages Disadvantages Main applications 4.1 Atomic force microscopy 4.1.1 normal force * a few (1-5) cell measured per condition [160] * nN- pN forces [161] * Minimum force: 50-100 pN * At 100- 1,000 pN force, the biological sample is reversibly deformed, above this value irreversibly deformed [163],[164], [165] AFM-based SCFS: 5 pN- 100 nN [149] Chemical functionalization of the cantilever: 1h * ~10 min needed for proper cell -- cantilever interaction * Relatively short contact times (msec- 20min) [43],[170] * Spatial and force resolution is high [148] * pN force sensitivity nm positioning accuracy Can map and analyze individual receptors with nanoscale lateral resolution [148] 4.1 Atomic force microscopy 4.1.2 Tangential/ lateral force 200-300 cells [2] 10-100 nN [2],[176] Adhesion time:10- 90 min [2] * Much lower force is sufficient to detach a particle than a normal force [177],[178] 4.2 Fluidic force microscopy * Up to 200 yeast and 20 mammalian cells/probe * Studying ~10 cells in half an hour 500 pN-1.6 µN [48] * Carrying out ~10- times more experiments than conventional AFM * Only a few minutes to target, immobilize, and release the cell as well as to change the cantilever position to the next cell [48] Directly measures the adhesion force ~10 times higher throughput than conventional AFM can provide [48] * Increased maximum force as compared to AFM [43] * No need for chemical functionalization of the cantilever [183] * Costly method [43],[48],[149] * Time-consuming [161],[174] * Requires a separate calibrated and functionalized cantilever for each cell [43],[48],[149] * Chemical attachment of the cell to the cantilever can change the physiology of the cell [161],[174] * High number of detachment force- distance curves need to be measured to gain reliable statistics [43],[170] * Limited to relatively short contact times [43] * Interpretation of the data is not straightforward [2] * The measured force depends on experimental details (shape of the tip, scanning speed, torsional cantilever spring constant) [78],[179] * Throughput is lower than that of the automated micropipette Microfabricated cantilevers come with a high measurement cost Cells come into direct contact with the cantilever potentially perturbing or damaging cells * Investigation of cell adhesion and the interactions between specific ligand- receptor pairs [149] * Studying the dynamic formation of cellular adhesion * Measuring cell-cell and cell-substrate adhesion [43] * Extends its use from quantitatively characterizing whole- cell adhesion down to single receptor-ligand interactions [43] * Measuring the adhesion force of protein-covered silica spheres adsorbed to polystyrene surfaces [180] * Measuring the adhesion force of individual cervix cells to various substrates [2] * Monitoring the adhesion of single cells to specific substrates [48] Microinjection of cells [185] Microprinting [186],[187],[188] Table 4. Atomic force and fluidic force microscopy. 58 Method a) Number of cells in an experiment b) c) Typical Experiment force range duration d) e) f) Advantages Disadvantages Main applications 5.1 Cell traction force microscopy 1-5 in single cell experiments, several hundreds of cells in collective migration studies * Depending on the elasticity of the substrate, which can be tuned in the physiological range [189] * Not possible to resolve small forces if the overall magnitude of traction is high * The imaging stage is only a few minutes * Image analysis and deconvolution of the traction field is time consuming [193] 5.2 FRET force sensors Minimum of 10 cells at each condition [209] pN sensitivity [210] ~20 min [209] * Single cell measurements * Remaining cells unperturbed * High resolution mapping of cell- exerted forces * Following the evolution of traction through time, although practically with a poor temporal resolution * Using a confocal microscope, all components (x,y,z) of forces exerted on a 2D substrate can be attained (2.5D CTFM [192] * Using confocal microscopy, the traction forces of a cell embedded in a 3D matrix can be determined (3D CTFM) [196] * Can detect protein- protein nanoscale interactions [219] * Does not require mechanical perturbation or stimulation of cells [209] * Measures the force exerted by individual molecules [220] * Can be tracked in time [221] * Mapping the 2D and 2.5D traction field of [189] - individual cells [199],[200] - cell aggregates [201] - migrating single cells [203] - a layer of collectively migrating cells [204] * Mapping the traction field of cells embedded in a 3D matrix [196], [197] * Correlating adhesion force with local actin density or with the size/molecular composition of adhesion complexes [191] * Exploring the role of mechanical forces built up in proteins [50],[216] * Estimating the force across single integrin receptors [218] * Quantifying conformational dynamics in single molecules [212] * Measuring the parameters of cell interactions with 2-3D adhesion substrates [209],[219] * Extremely low throughput * Image analysis and the deconvolution of the traction field is time consuming [193],[194] * The traction field may be very sensitive to noises [193],[194] * 2.5D and 3D CTFM requires a confocal microscope [196] * 3D CTFM is in its infancy * Artificial molecular force sensors are needed to be built into the cellular system (a specific protein) * The selection of an appropriate elastic molecular element of the force sensor is critical [211],[212] * Requires spectrally matched fluorophores in the force sensor [212], [220] * High sensitivity camera is needed for imaging * Requires data analysis algorithm to determine FRET [211] Table 5. Cell traction force microscopy and FRET force sensors. 59 Acknowledgment Funding sources This work was supported by the National Research, Development and Innovation Office (grant numbers: PD 124559 for R. U. S., KH_17, KKP 129936 for R. H. ), "Lendület" Program of the Hung. Acad. Sci., ERC_HU for R. H., MedInProt grant of the Hung. Acad. Sci., Bolyai Scholarship for B. S. Declaration of interest B.S. is a founder of CellSorter Company for Innovations, a startup company that developed the computer-controlled micropipette device. 60 References [1] Bachir AI, Horwitz AR, Nelson WJ, Bianchini JM. Cell adhesions: Actin- Based Modules that Mediate Cell- Extracellular Matrix and Cell-Cell Interactions. Cold Spring Harb Perspect Biol 2017;9:a023234. doi:10.1101/cshperspect.a023234. [2] Sagvolden G, Giaever I, Pettersen EO, Feder J. Cell adhesion force microscopy. Proc Natl Acad Sci U S A 1999;96:471 -- 6. doi:10.1073/pnas.96.2.471. [3] Iwamoto D V, Calderwood DA. Regulation of integrin-mediated adhesions. Curr Opin Cell Biol 2015;36:41 -- 7. doi:10.1016/j.ceb.2015.06.009. [4] Sun Z, Guo SS, Fässler R. Integrin-mediated mechanotransduction. J Cell Biol 2016;215:445 -- 56. doi:10.1083/JCB.201609037. [5] Leckband DE, de Rooij J. Cadherin Adhesion and Mechanotransduction. Annu Rev Cell Dev Biol 2014;30:291 -- 315. doi:10.1146/annurev-cellbio- 100913-013212. [6] Changede R, Sheetz M. Integrin and cadherin clusters: A robust way to organize adhesions for cell mechanics. BioEssays 2017;39:1 -- 12. doi:10.1002/bies.201600123. [7] Alberts B, Johnson A, Lewis J, Raff M, Roberts K, Walter P. Molecular biology of the cell, 5th Edition, Garland Science. vol. 19. New York: 2007. [8] Keselowsky BG, García AJ. Quantitative methods for analysis of integrin binding and focal adhesion formation on biomaterial surfaces. Biomaterials 2005;26:413 -- 8. doi:10.1016/j.biomaterials.2004.02.050. [9] Geiger B, Bershadsky A, Pankov R, Yamada KM. Transmembrane crosstalk between the extracellular matrix-cytoskeleton crosstalk. Nat Rev Mol Cell Biol 2002;2:793 -- 805. doi:10.1038/35099066. [10] Lehenkari PP, Horton MA. Single integrin molecule adhesion forces in intact cells measured by atomic force microscopy. Biochem Biophys Res Commun 1999;259:645 -- 50. doi:10.1006/bbrc.1999.0827. [11] Lee IS, Marchant RE. Force measurements on platelet surfaces with high spatial resolution under physiological conditions. Colloids Surfaces B Biointerfaces 2000;19:357 -- 65. doi:10.1016/S0927-7765(00)00144-2. [12] Deshpande VS, McMeeking RM, Evans AG. A model for the contractility 61 of the cytoskeleton including the effects of stress-fibre formation and dissociation. Proc R Soc A Math Phys Eng Sci 2007;463:787 -- 815. doi:10.1098/rspa.2006.1793. [13] Besser A, Schwarz US. Coupling biochemistry and mechanics in cell adhesion: A model for inhomogeneous stress fiber contraction. New J Phys 2007;9:425. doi:10.1088/1367-2630/9/11/425. [14] Ingber DE. Cellular tensegrity: defining new rules of biological design that govern the cytoskeleton. J Cell Sci 1993;104:613 -- 27. [15] Cañadas P, Laurent VM, Oddou C, Isabey D, Wendling S. A cellular tensegrity model to analyse the structural viscoelasticity of the cytoskeleton. J Theor Biol 2002;218:155 -- 73. doi:10.1006/yjtbi.3064. [16] Ghaffari H, Saidi MS, Firoozabadi B. Biomechanical analysis of actin cytoskeleton function based on a spring network cell model. Proc Inst Mech Eng Part C J Mech Eng Sci 2017;0:1 -- 16. doi:10.1177/0954406216668546. [17] Schwarz US, Erdmann T, Bischofs IB. Focal adhesions as mechanosensors: The two-spring model. BioSystems 2006;83:225 -- 32. doi:10.1016/j.biosystems.2005.05.019. [18] Ley K, Laudanna C, Cybulsky MI, Nourshargh S. Getting to the site of inflammation: the leukocyte adhesion cascade updated. Nat Rev Immunol 2007;7:678 -- 89. doi:10.1038/nri2156. [19] Albelda SM. Role of integrins and other cell adhesion molecules in tumor progression and metastasis . J Tech Methods Pathol 1993;68:4 -- 17. [20] Rao CCG, Chianese D, Doyle GVG, Miller MC, Russell T, Sanders R, et al. Expression of epithelial cell adhesion molecule in carcinoma cells present in blood and primary and metastatic tumors. Int J Oncol 2005;27:49 -- 57. doi:10.3892/ijo.27.1.49. [21] Kobayashi H, Boelte KC, Lin PC. Endothelial cell adhesion molecules and cancer progression. Curr Med Chem 2007;14:377 -- 86. doi:10.2174/092986707779941032. [22] Orgovan N, Peter B, Bősze S, Ramsden JJ, Szabó B, Horvath R. Dependence of cancer cell adhesion kinetics on integrin ligand surface density measured by a high-throughput label-free resonant waveguide grating biosensor. Sci Rep 2014;4:4034. doi:10.1038/srep04034. [23] Salánki R, Hős C, Orgovan, Norbert, Péter B, Sándor N, Bajtay Z, Erdei 62 A, et al. Single Cell Adhesion Assay Using Computer Controlled Micropipette. PLoS One 2014;9:e111450. doi:10.1371/journal.pone.0111450. [24] Trepat X, Wasserman MR, Angelini TE, Millet E, Weitz DA, Butler JP, et al. Physical forces during collective cell migration. Nat Phys 2009;5:426 -- 30. doi:10.1038/nphys1269. [25] Legant WR, Choi CK, Miller JS, Shao L, Gao L, Betzig E, et al. Multidimensional traction force microscopy reveals out-of-plane rotational moments about focal adhesions. Proc Natl Acad Sci U S A 2013;110:881 -- 6. doi:10.1073/pnas.1207997110. [26] Huang N, Michel R, Voros J, Textor M, Hofer R, Rossi A, et al. Poly(l- lysine)-g-poly(ethylene glycol) Layers on Metal Oxide Surfaces: Surface- Analytical Characterization and Resistance to Serum and Fibrinogen Adsorption. Langmuir 2001;17:489 -- 98. doi:10.1021/la000736. [27] Roger M, Pasche S, Textor M, Castner DG. The Influence of PEG Architecture on Protein Adsorption and Conformation. Langmuir 2005;21:12327 -- 12332. doi:10.1016/j.biotechadv.2011.08.021.Secreted. [28] Faraasen S, Faraasen S, Vörös J, Vörös J, Csúcs G, Csúcs G, et al. Ligand- specific trageting of microspheres to phagocytes by surface modification with poly(l-lysine)-grafted PEG conjugate. Pharm Res 2003;20:237 -- 46. [29] Ogaki R, Zoffmann Andersen O, Jensen GV, Kolind K, Kraft DCE, Pedersen JS, et al. Temperature-induced ultradense PEG polyelectrolyte surface grafting provides effective long-term bioresistance against mammalian cells, serum, and whole blood. Biomacromolecules 2012;13:3668 -- 77. doi:10.1021/bm301125g. [30] VandeVondele S, Vörös J, Hubbell J. RGD-grafted poly-L-lysine-graft- (polyethylene glycol) copolymers block non-specific protein adsorption while promoting cell adhesion. Biotechnol Bioeng 2003;82:784 -- 90. doi:10.1002/bit.10625. [31] Koo LY, Irvine DJ, Mayes AM, Lauffenburger D, Griffith LG. Co- regulation of cell adhesion by nanoscale RGD organization and mechanical stimulus. J Cell Sci 2002;115:1423 -- 33. [32] Zhen G, Zurcher S, Falconnet D, Xu F, Kuennemann E, Textor M. NTA- Functionalized Poly(L-lysine)-g-Poly(Ethylene Glycol): A Polymeric Interface for Binding and Studying 6 His-tagged Proteins. 2005 IEEE Eng. Med. Biol. 27th Annu. Conf., vol. 1, 2005, p. 1036 -- 8. 63 doi:10.1109/IEMBS.2005.1616595. [33] Hua B, Han KY, Zhou R, Kim H, Shi X, Abeysirigunawardena SC, et al. An improved surface passivation method for single-molecule studies. Nat Methods 2014;11:1233 -- 6. doi:10.1038/nmeth.3143. [34] Kovacs B, Patko D, Szekacs I, Orgovan N, Kurunczi S, Sulyok A, et al. Flagellin based biomimetic coatings: From cell-repellent surfaces to highly adhesive coatings. Acta Biomater 2016;42:66 -- 76. doi:10.1016/j.actbio.2016.07.002. [35] Kovacs B, Patko D, Klein A, Kakasi B, Saftics A, Kurunczi S, et al. Bacteria repellent layer made of flagellin. Sensors Actuators B Chem 2018;257:839 -- 45. doi:10.1016/j.snb.2017.11.027. [36] Kovacs B, Saftics A, Biro A, Kurunczi S, Szalontai B, Kakasi B, et al. Kinetics and Structure of Self-Assembled Flagellin Monolayers on Hydrophobic Surfaces in the Presence of Hofmeister Salts: Experimental Measurement of the Protein Interfacial Tension at the Nanometer Scale. J Phys Chem C 2018;122:21375 -- 86. doi:10.1021/acs.jpcc.8b05026. [37] Kovacs N, Patko D, Orgovan N, Kurunczi S, Ramsden JJ, Vonderviszt F, et al. Optical anisotropy of flagellin layers: In situ and label-free measurement of adsorbed protein orientation using OWLS. Anal Chem 2013;85:5382 -- 9. doi:10.1021/ac3034322. [38] Khalili AA, Ahmad MR. A Review of cell adhesion studies for biomedical and biological applications. Int J Mol Sci 2015;16:18149 -- 84. doi:10.3390/ijms160818149. [39] Klebe RJ. Isolation of a collagen-dependent cell attachment factor. Nature 1974;250:248 -- 51. doi:10.1038/250248a0. [40] García AJ, Ducheyne P, Boettiger D. Quantification of cell adhesion using a spinning disc device and application to surface-reactive materials. Biomaterials 1997;18:1091 -- 8. doi:10.1016/S0142-9612(97)00042-2. [41] Kaplanski G, Farnarier C, Tissot O, Pierres A, Benoliel AM, Alessi MC, et al. Granulocyte-endothelium initial adhesion. Analysis of transient binding events mediated by E-selectin in a laminar shear flow. Biophys J 1993;64:1922 -- 33. doi:10.1016/S0006-3495(93)81563-7. [42] Christ K V, Turner KT. Methods to Measure the Strength of Cell Adhesion to Substrates. J Adhes Sci & Technol 2010;24:2027 -- 58. doi:10.1163/016942410X507911. 64 [43] Helenius J, Heisenberg C-P, Gaub HE, Muller DJ. Single-cell force spectroscopy. J Cell Sci 2008;121:1785 -- 91. doi:10.1242/jcs.030999. [44] Puech PH, Poole K, Knebel D, Muller DJ. A new technical approach to quantify cell-cell adhesion forces by AFM. Ultramicroscopy 2006;106:637 -- 44. doi:10.1016/j.ultramic.2005.08.003. [45] Hochmuth RM. Micropipette aspiration of living cells. J Biomech 2000;33:15 -- 22. doi:10.1016/S0021-9290(99)00175-X. [46] Shao J-Y, Xu G, Guo P. Quantifying cell-adhesion strength with micropipette manipulation: principle and application. Front Biosci 2004;9:2183 -- 91. doi:10.2741/1386. [47] Sung KL, Sung LA, Crimmins M, Burakoff SJ, Chien S. Determination of junction avidity of cytolytic T cell and target cell. Science 1986;234:1405 -- 8. doi:10.1126/science.3491426. [48] Potthoff E, Guillaume-Gentil O, Ossola D, Polesel-Maris J, LeibundGut- Landmann S, Zambelli T, et al. Rapid and Serial Quantification of Adhesion Forces of Yeast and Mammalian Cells. PLoS One 2012;7:e52712. doi:10.1371/journal.pone.0052712. [49] Roca-Cusachs P, Conte V, Trepat X. Quantifying forces in cell biology. Nat Cell Biol 2017;19:742 -- 51. doi:10.1038/ncb3564. [50] Grashoff C, Hoffman BD, Brenner MD, Zhou R, Parsons M, Yang MT, et al. Measuring mechanical tension across vinculin reveals regulation of focal adhesion dynamics. Nature 2010;466:263 -- 6. doi:10.1038/nature09198. [51] Ramsden JJ, Horvath R. Optical biosensors for cell adhesion. J Recept Signal Transduct 2009;29:211 -- 23. [52] Orgovan N, Salánki R, Sándor N, Bajtay Z, Erdei A, Szabó B, et al. In-situ and label-free optical monitoring of the adhesion and spreading of primary monocytes isolated from human blood: Dependence on serum concentration levels. Biosens Bioelectron 2014;54:339 -- 44. doi:10.1016/j.bios.2013.10.076. [53] Cottier K, Horvath R. Imageless microscopy of surface patterns using optical waveguides. Appl Phys B Lasers Opt 2008;91:319 -- 27. doi:10.1007/s00340-008-2994-6. [54] Peter B, Ungai-Salanki R, Szabó B, Nagy AG, Szekacs I, Bösze S, et al. High-Resolution Adhesion Kinetics of EGCG-Exposed Tumor Cells on 65 Biomimetic Interfaces: Comparative Monitoring of Cell Viability Using Label-Free Biosensor and Classic End-Point Assays. ACS Omega 2018;3:3882 -- 91. doi:10.1021/acsomega.7b01902. [55] Peter B, Farkas E, Forgacs E, Saftics A, Kovacs B, Kurunczi S, et al. Green tea polyphenol tailors cell adhesivity of RGD displaying surfaces: Multicomponent models monitored optically. Sci Rep 2017;7:1 -- 16. doi:10.1038/srep42220. [56] Giebel KF, Bechinger C, Herminghaus S, Riedel M, Leiderer P, Weiland U, et al. Imaging of cell/substrate contacts of living cells with surface plasmon resonance microscopy. Biophys J 1999;76:509 -- 16. doi:10.1016/S0006-3495(99)77219-X. [57] Lord MS, Modin C, Foss M, Duch M, Simmons A, Pedersen FS, et al. Monitoring cell adhesion on tantalum and oxidised polystyrene using a quartz crystal microbalance with dissipation. Biomaterials 2006;27:4529 -- 37. doi:10.1016/j.biomaterials.2006.04.006. [58] Dixon MC. Quartz Crystal Microbalance with Dissipation Monitoring: Enabling Real-Time Characterization of Biological Materials and Their Interactions. J Biomol Tech 2008;19:151 -- 8. [59] Modin C, Stranne A-L, Foss M, Duch M, Justesen J, Chevallier J, et al. QCM-D studies of attachment and differential spreading of pre- osteoblastic cells on Ta and Cr surfaces. Biomaterials 2006;27:1346 -- 54. doi:10.1016/j.biomaterials.2005.09.022. [60] Marx K, Zhou T, Schulze H, Braunhut SJ. A Quartz Crystal Microbalance cell biosensor: Detection of microtubule alterations in living cells at nM nocodazole concentrations. Biosens Bioelectron 16 2001;16:773 -- 82. [61] Christ K V., Williamson KB, Masters KS, Turner KT. Measurement of single-cell adhesion strength using a microfluidic assay. Biomed Microdevices 2010;12:443 -- 55. doi:10.1007/s10544-010-9401-x. [62] Walther BT, Ohman R, Roseman S. A quantitative assay for Intercellular Adhesion 1973;70:1569 -- 73. [63] Rubin K, Höök M, Öbrink B, Timpl R. Substrate adhesion of rat hepatocytes: Mechanism of attachment to collagen substrates. Cell 1981;24:463 -- 70. doi:https://doi.org/10.1016/0092-8674(81)90337-8. [64] Chu L, Tempelman LA, Miller C, Hammer DA. Centrifugation assay of Ig-E mediated cell adhesion to antigen coated gels. AIChE J 1994;40:692 -- 66 703. doi:https://doi.org/10.1002/aic.690400412. [65] McClay DR, Wessel GM, Marchase RB. Intercellular recognition: quantitation of initial binding events. Proc Natl Acad Sci U S A 1981;78:4975 -- 9. doi:10.1073/pnas.78.8.4975. [66] Lotz MM, Burdsal C, Erickson HP, McClay DR. Cell adhesion to fibronectin and tenascin: Quantitative measurements of initial binding and subsequent strengthening response. J Cell Biol 1989;109:1795 -- 805. doi:10.1083/jcb.109.4.1795. [67] Maheshwari G, Brown G, Lauffenburger D, Wells A, Griffith LG. Cell adhesion and motility depend on nanoscale RGD clustering. J Cell Sci 2000;113 ( Pt 1:1677 -- 86. [68] Angres B, Barth A, Nelson WJ. Mechanism for Transition from Initial to StableCell-Cell Adhesion: Kinetic Analysis of E-Cadherin- MediatedAdhesion Using a Quantitative Adhesion Assay. J Cell Biol 1996;134:549 -- 57. [69] Reyes CD, García AJ. A centrifugation cell adhesion assay for high- throughput screening of biomaterial surfaces. Wiley Period Inc J Biomed Mater Res 2003;67A:328 -- 333. [70] Thoumine O, Ott A, Louvard D. Critical centrifugal forces induce adhesion rupture or structural reorganization in cultured cells. Cell Motil Cytoskeleton 1996;33:276 -- 87. doi:10.1002/(SICI)1097- 0169(1996)33:4<276::AID-CM4>3.0.CO;2-7. [71] Thoumine O, Ott A. Comparison of the mechanical properties of normal and transformed fibroblasts. Biorheology 1997;34:309 -- 26. doi:10.3233/BIR-1997-344-505. [72] Channavajjala LS, Eidsath A, Saxinger WC. A simple method for measurement of cell-substrate attachment forces: application to HIV-1 Tat. J Cell Sci 1997;110:249 -- 56. [73] Prakobphol A, Burdsal CA, Fisher SJ. Quantifying the Strength of Bacterial Adhesive Interactions with Salivary Glycoproteins. J Dent Res 1995;74:1212 -- 8. doi:10.1177/00220345950740051101. [74] Wertheim JA, Forsythe K, Druker BJ, Hammer D, Boettiger D, Pear WS. BCR-ABL -- induced adhesion defects are tyrosine kinase -- independent. Blood 2002;99:4122 -- 31. [75] García AJ, Huber F, Boettiger D. Force required to break α5β1integrin- 67 fibronectin bonds in intact adherent cells is sensitive to integrin activation state. J Biol Chem 1998;273:10988 -- 93. doi:10.1074/jbc.273.18.10988. [76] Shi Q, Boettiger D. A Novel Mode for Integrin-mediated Signaling: Tethering Is Required for Phosphorylation of FAK Y397. Mol Biol Cell 2003;14:4306 -- 15. doi:10.1091/mbc.E03. [77] Gallant ND, Capadona JR, Frazier AB, Collard DM, García AJ. Micropatterned surfaces to engineer focal adhesions for analysis of cell adhesion strengthening. Langmuir 2002;18:5579 -- 84. doi:10.1021/la025554p. [78] Bowen WR, Fenton AS, Lovitt RW, Wright CJ. The measurement of Bacillus mycoides spore adhesion using atomic force microscopy, simple counting methods, and a spinning disk technique. Biotechnol Bioeng 2002;79:170 -- 9. doi:10.1002/bit.10321. [79] Goldstein AS, DiMilla PA. Application of fluid mechanic and kinetic models to characterize mammalian cell detachment in a radial-flow chamber. Biotechnol Bioeng 1997;55:616 -- 29. doi:10.1002/(SICI)1097- 0290(19970820)55:4<616::AID-BIT4>3.0.CO;2-K. [80] Dimilla PA, Stone JA, Quinn JA, Albelda SM, Lauffenburger AD. Maximal migration of human smooth muscle cells on fibronectin and type IV collagen occurs at an intermediate attachment strength. J Cell Biol 1993;122:729 -- 37. [81] Goldstein AS, DiMilla PA. Comparison of converging and diverging radial flow for measuring cell adhesion. AIChE J 1998;44:465 -- 73. doi:10.1002/aic.690440222. [82] Bearinger JP, Castner DG, Golledge SL, Rezania A, Hubchak S, Healy KE. P(AAm- co -EG) Interpenetrating Polymer Networks Grafted to Oxide Surfaces: Surface Characterization, Protein Adsorption, and Cell Detachment Studies. Langmuir 1997;13:5175 -- 83. doi:10.1021/la970101j. [83] Goldstein AS, DiMilla PA. Examination of membrane rupture as a mechanism for mammalian cell detachment from fibronectin-coated biomaterials. J Biomed Mater Res A 2003;67A:658 -- 66. doi:10.1002/jbm.a.10125. [84] Van Kooten TG, Schakenraad JM, Van der Mei HC, Busscher HJ. Development and use of a parallel-plate flow chamber for studying cellular adhesion to solid surfaces. J Biomed Mater Res 1992;26:725 -- 38. doi:10.1002/jbm.820260604. 68 [85] Weiss L. The measurement of cell adhesion. Exp Cell Res 1961;Suppl 8:141 -- 53. [86] Goldstein AS, Dimilla PA. Effect of adsorbed fibronectin concentration on cell adhesion and deformation under shear on hydrophobic surfaces. J Biomed Mater Res 2002;59:665 -- 75. doi:10.1002/jbm.1276. [87] Baumgartner JN, Yang CZ, Cooper SL. Physical property analysis and bacterial adhesion on a series of phosphonated polyurethanes. Biomaterials 1997;18:831 -- 7. doi:10.1016/S0142-9612(96)00197-4. [88] Van Kooten TG, Schakenraad JM, Van der Mei HC, Dekker A, Kirkpatrick CJ, Busscher HJ. Fluid shear induced endothelial cell detachment from glass - influence of adhesion time and shear stress. Med Eng Phys 1994;16:506 -- 12. doi:10.1016/1350-4533(94)90077-9. [89] Van Kooten TG, Schakenraad JM, Van der Mei HC, Busscher HJ. Influence of substratum wettability on the strength of adhesion of human fibroblasts. Biomaterials 1992;13:897 -- 904. doi:10.1016/0142- 9612(92)90112-2. [90] Lee J, Lee S, Khang G, Lee H. The Effect of Fluid Shear Stress on Endothelial Cell Adhesiveness to Polymer Surfaces with Wettability Gradient. J Colloid Interface Sci 2000;230:84 -- 90. doi:10.1006/jcis.2000.7080. [91] Martines E, McGhee K, Wilkinson C, Curtis A. A Parallel-plate flow chamber to study initial cell adhesion on a nanofeatured surface. IEEE Trans Nanobioscience 2004;3:90 -- 5. doi:10.1109/TNB.2004.828268. [92] Truskey GA, Proulx TL. Relationship between 3T3 cell spreading and the strength of adhesion on glass and silane surfaces. Biomaterials 1993;14:243 -- 54. doi:10.1016/0142-9612(93)90114-H. [93] Kapur R, Rudolph S. Cellular and cytoskeleton morphology and strength of adhesion of cells on self-assembled monolayers of organosilanes. Exp Cell Res 1998;244:275 -- 85. doi:10.1006/excr.1998.4156. [94] Messer RLW, Davis CM, Lewis JB, Adams Y, Wataha JC. Attachment of human epithelial cells and periodontal ligament fibroblasts to tooth dentin. J Biomed Mater Res - Part A 2006;79A:16 -- 22. doi:10.1002/jbm.a.30703. [95] Bakker DP, Van der Plaats A, Verkerke GJ, Busscher HJ, Van der Mei HC. Comparison of Velocity Profiles for Different Flow Chamber Designs Used in Studies of Microbial Adhesion to Surfaces. Appl Environ 69 Microbiol 2003;69:6280 -- 7. doi:10.1128/AEM.69.10.6280-6287.2003. [96] Roosjen A, Kaper HJ, van der Mei HC, Norde W, Busscher HJ. Inhibition of adhesion of yeasts and bacteria by poly(ethylene oxide)-brushes on glass in a parallel plate flow chamber. Microbiology 2003;149:3239 -- 46. doi:10.1099/mic.0.26519-0. [97] McClaine JW, Ford RM. Characterizing the adhesion of motile and nonmotile Escherichia coli to a glass surface using a parallel-plate flow chamber. Biotechnol Bioeng 2002;78:179 -- 89. doi:10.1002/bit.10192. [98] Schmolke H, Demming S, Edlich A, Magdanz V, Büttgenbach S, Franco- Lara E, et al. Polyelectrolyte multilayer surface functionalization of poly(dimethylsiloxane) (PDMS) for reduction of yeast cell adhesion in microfluidic devices. Biomicrofluidics 2010;4:044113. doi:10.1063/1.3523059. [99] Gervais T, El-Ali J, Günther A, Jensen KF. Flow-induced deformation of shallow microfluidic channels. Lab Chip 2006;6:500 -- 7. doi:10.1039/b513524a. [100] Lu H, Koo LY, Wang WM, Lauffenburger DA, Griffith LG, Jensen KF. Microfluidic shear devices for quantitative analysis of cell adhesion. Anal Chem 2004;76:5257 -- 64. doi:10.1021/ac049837t. [101] Young EWK, Wheeler AR, Simmons CA. Matrix-dependent adhesion of vascular and valvular endothelial cells in microfluidic channels. Lab Chip 2007;7:1759 -- 66. doi:10.1039/b712486d. [102] Kwon KW, Choi SS, Lee SH, Kim B, Lee SN, Park MC, et al. Label-free, microfluidic separation and enrichment of human breast cancer cells by adhesion difference. Lab Chip 2007;7:1461 -- 8. doi:10.1039/b710054j. [103] Choi JW, Rosset S, Niklaus M, Adleman JR, Shea H, Psaltis D. 3- Dimensional Electrode Patterning Within a Microfluidic Channel Using Metal Ion Implantation. Lab Chip 2010;10:783 -- 8. doi:10.1039/b917719a. [104] Mitchison JM, Swann MM. The mechanical properties of the cell surface. J Exp Biol 1954;31:443 -- 60. [105] Rand RP, Burton AC. Mechanical Properties of the Red Cell Membrane. I. Membrane Stiffness and Intracellular Pressure. Biophys J 1964;4:115 -- 35. doi:10.1016/S0006-3495(64)86773-4. [106] Chiou PY, Ohta AT, Wu MC. Massively parallel manipulation of single cells and microparticles using optical images. Nat Lett 2005;436:370 -- 2. 70 doi:10.1038/nature03831. [107] Shao JY, Hochmuth RM. Micropipette suction for measuring piconewton forces of adhesion and tether formation from neutrophil membranes. Biophys J 1996;71:2892 -- 901. doi:10.1016/S0006-3495(96)79486-9. [108] Shao J-Y, Xu J. A modified micropipette aspiration technique and its application to tether formation from human neutrophils. J Biomech Eng 2002;124:388 -- 96. doi:10.1115/1.1486469. [109] Sung KL, Frojmovic MM, O'Toole TE, Zhu C, Ginsberg MH, Chien S. Determination of adhesion force between single cell pairs generated by activated GpIIb-IIIa receptors. Blood 1993;81:419 -- 23. [110] Zhao H, Dong X, Wang X, Li X, Zhuang F, Stoltz JF, et al. Studies on single-cell adhesion probability between lymphocytes and endothelial cells with micropipette technique. Microvasc Res 2002;63:218 -- 26. doi:10.1006/mvre.2001.2390. [111] Evans E, Ritchie K, Merkel R. Sensitive force technique to probe molecular adhesion and structural linkages at biological interfaces. Biophys J 1995:2580 -- 7. doi:10.1016/S0006-3495(95)80441-8. [112] Merkel R, Nassoy P, Leung A, Ritchie K, Evans E. Energy landscapes of receptor-ligand bonds explored with dynamic force spectroscopy. Nature 1999;397:50 -- 3. doi:10.1038/16219. [113] Evans E, Ritchie K. Dynamic strength of molecular adhesion bonds. Biophys J 1997;72:1541 -- 55. doi:10.1016/S0006-3495(97)78802-7. [114] Evans E. Looking inside molecular bonds at biological interfaces with dynamic force spectroscopy. Biophys Chem 1999;82:83 -- 97. doi:10.1016/S0301-4622(99)00108-8. [115] Evans E, Ritchie K. Strength of a weak bond connecting flexible polymer chains. Biophys J 1999;76:2439 -- 47. doi:10.1016/S0006-3495(99)77399-6. [116] Evans E, Ludwig F. Dynamic strengths of molecular anchoring and material cohesion in fluid biomembranes. J Phys Condens Matter 2000;12:A315 -- 20. doi:10.1088/0953-8984/12/8A/341. [117] Evans E, Leung A, Hammer D, Simon S. Chemically distinct transition states govern rapid dissociation of single L-selectin bonds under force. Proc Natl Acad Sci U S A 2001;98:3784 -- 9. doi:10.1073/pnas.061324998. [118] Evans E. Probing the relation between force-lifetime-and chemistry in 71 single molecular bonds. Annu Rev Biophys Biomol Struct 2001;30:105 -- 28. doi:10.1146/annurev.biophys.30.1.105. [119] Spillmann C, Osorio D, Waugh R. Integrin activation by divalent ions affects neutrophil homotypic adhesion. Ann Biomed Eng 2002;30:1002 -- 11. doi:10.1114/1.1511241. [120] Lomakina EB, Waugh RE. Micromechanical tests of adhesion dynamics between neutrophils and immobilized ICAM-1. Biophys J 2004;86:1223 -- 33. doi:10.1016/S0006-3495(04)74196-X. [121] Környei Z, Beke S, Mihálffy T, Jelitai M, Kovács KJ, Szabó Z, et al. Cell sorting in a Petri dish controlled by computer vision. Sci Rep 2013;3:1088. doi:10.1038/srep01088. [122] Ungai-Salánki R, Gerecsei T, Fürjes P, Orgovan N, Sándor N, Holczer E, et al. Automated single cell isolation from suspension with computer vision. Sci Rep 2016;6:20375. doi:10.1038/srep20375. [123] Salánki R, Gerecsei T, Orgovan N, Sándor N, Péter B, Bajtay Z, et al. Automated single cell sorting and deposition in submicroliter drops. Appl Phyisics Lett 2014;105:083703. doi:10.1063/1.4893922. [124] Jani PK, Schwaner E, Kajdácsi E, Debreczeni ML, Ungai-Salánki R, Dobó J, et al. Complement MASP-1 enhances adhesion between endothelial cells and neutrophils by up-regulating E-selectin expression. Mol Immunol 2016;75:38 -- 47. doi:10.1016/j.molimm.2016.05.007. [125] Orgovan N, Ungai-salánki R, Lukácsi S, Sándor N, Bajtay Z, Erdei A, et al. Adhesion kinetics of human primary monocytes , dendritic cells , and macrophages : Dynamic cell adhesion measurements with a label-free optical biosensor and their comparison with end-point assays Adhesion kinetics of human primary monocytes , dendritic c. Biointerphases 2016;031001. doi:1116/1.4954789. [126] Sándor N, Lukácsi S, Ungai-Salánki R, Orgovan N, Erdei A, Bajtay Z. CD11c / CD18 Dominates Adhesion of Human Monocytes , Macrophages and Dendritic Cells over CD11b / CD18. PLoS One 2016;11:e0163120. doi:10.1371/journal.pone.0163120. [127] Zhang H, Liu KK. Optical tweezers for single cells. J R Soc Interface 2008;5:671 -- 90. doi:10.1098/rsif.2008.0052. [128] Neuman KC, Chadd EH, Liou GF, Bergman K, Block SM. Characterization of photodamage to Escherichia coli in optical traps. 72 Biophys J 1999;77:2856 -- 63. doi:10.1016/S0006-3495(99)77117-1. [129] Dai J, Sheetz MP. Mechanical properties of neuronal growth cone membranes studied by tether formation with laser optical tweezers. Biophys J 1995;68:988 -- 96. doi:10.1016/S0006-3495(95)80274-2. [130] Neuman KC, Nagy A. Single-molecule force spectroscopy: Optical tweezers, magnetic tweezers and atomic force microscopy. Nat Methods 2008;5:491 -- 505. doi:10.1038/nmeth.1218. [131] Dholakia K, Reece P. Optical micromanipulation takes hold. Nano Today 2006;1:18 -- 27. doi:10.1016/S1748-0132(06)70019-6. [132] Jähnke T, Rauch P. Optical tweezers for single-cell, multicellular investigations in the life sciences. Am Lab 2015. [133] Hénon S, Lenormand G, Richert A, Gallet F. A new determination of the shear modulus of the human erythrocyte membrane using optical tweezers. Biophys J 1999;76:1145 -- 51. doi:10.1016/S0006-3495(99)77279-6. [134] König K, Liang H, Berns MW, Tromberg BJ. Cell damage in near-infrared multimode optical traps as a result of multiphoton absorption. Opt Lett 1996;21:1090 -- 2. doi:10.1364/ol.21.001090. [135] Korda PT, Spalding GC, Grier DG. Evolution of a colloidal critical state in an optical pinning potential landscape. Phys Rev B 2002;66:024504. doi:10.1103/PhysRevB.66.024504. [136] Gross SP. Application of optical traps in vivo. Methods Enzymol 2003;361:162 -- 74. doi:10.1016/S0076-6879(03)61010-4. [137] Pavone FS, Antolini R, Sacconi L, Tolić-Nørrelykke IM, Stringari C. Optical micromanipulations inside yeast cells. Appl Opt 2005;44:2001 -- 7. doi:10.1364/ao.44.002001. [138] Castelain M, Pignon F, Piau JM, Magnin A, Mercier-Bonin M, Schmitz P. Removal forces and adhesion properties of Saccharomyces cerevisiae on glass substrates probed by optical tweezer. J Chem Phys 2007;127:135104. doi:10.1063/1.2772270. [139] Castelain M, Rouxhet PG, Pignon F, Magnin A, Piau JM. Single-cell adhesion probed in-situ using optical tweezers: A case study with Saccharomyces cerevisiae. J Appl Phys 2012;111:114701. doi:10.1063/1.4723566. [140] Castelain M, Pignon F, Piau JM, Magnin A. The initial single yeast cell 73 adhesion on glass via optical trapping and Derjaguin-Landau-Verwey- Overbeek predictions. J Chem Phys 2008;128:135101. doi:10.1063/1.2842078. [141] Kim ST, Shin Y, Brazin K, Mallis RJ, Sun ZYJ, Wagner G, et al. TCR mechanobiology: Torques and tunable structures linked to early T cell signaling. Front Immunol 2012;3:1 -- 8. doi:10.3389/fimmu.2012.00076. [142] Block SM, Goldstein LSB, Schnapp BJ. Bead movement by single kinesin molecules studied with optical tweezers. Nature 1990;348:348 -- 52. [143] Litvinov RI, Shuman H, Bennett JS, Weisel JW, Litvinov R, Shumant H, et al. Binding strength and activati on state of single fibrinogen-integrin pairs on I iving cells. PNAS 2002;99:7426 -- 31. [144] Thoumine O, Kocian P, Kottelat A, Meister JJ. Short-term binding of fibroblasts to fibronectin: Optical tweezers experiments and probabilistic analysis. Eur Biophys J 2000;29:398 -- 408. doi:10.1007/s002490000087. [145] W. H, B. A, J.H. T, R.G. L, K.A. A. Temporal effects of cell adhesion on mechanical characteristics of the single chondrocyte. J Orthop Res 2003;21:88 -- 95. [146] Li Z, Anvari B, Takashima M, Brecht P, Torres JH, Brownell WE. Membrane tether formation from outer hair cells with optical tweezers. Biophys J 2002;82:1386 -- 95. doi:10.1016/S0006-3495(02)75493-3. [147] Binnig G, Quate CF. Atomic Force Microscope. Phys Rev Lett 1986;56:930 -- 3. doi:10.1103/PhysRevLett.56.930. [148] Hinterdorfer P, Dufrêne YF. Detection and localization of single molecular recognition events using atomic force microscopy. Nat Methods 2006;3:347 -- 55. doi:10.1038/nmeth871. [149] Friedrichs J, Helenius J, Muller DJ. Quantifying cellular adhesion to extracellular matrix components by single-cell force spectroscopy. Nat Protoc 2010;5:1353 -- 61. doi:10.1038/nprot.2010.89. [150] Alsteens D, Gaub HE, Newton R, Pfreundschuh M, Gerber C, Müller DJ. Atomic force microscopy-based characterization and design of biointerfaces. Nat Rev Mater 2017;2:17008 1-16. doi:10.1038/natrevmats.2017.8. [151] Hugel T, Seitz M. The Study of Molecular Interactions by AFM Force Spectroscopy. Macromol Rapid Commun 2001;22:989 -- 1016. 74 [152] Dufrêne YF, Evans E, Engel A, Helenius J, Gaub HE, Müller DJ. Five challenges to bringing single-molecule force spectroscopy into living cells. Nat Methods 2011;8:123 -- 7. doi:10.1038/nmeth0211-123. [153] Moy VT, Florin E, Gaub HE. Intermolecular Forces and. Science (80- ) 1994;266:257 -- 60. [154] Friedrichs J, Legate KR, Schubert R, Bharadwaj M, Werner C, Müller DJ, et al. A practical guide to quantify cell adhesion using single-cell force spectroscopy. Methods 2013;60:169 -- 78. doi:10.1016/j.ymeth.2013.01.006. [155] Benoit M, Gabriel D, Gerisch G, Gaub HE. Discrete interactions in cell adhesion measured by single-molecule force spectroscopy. Nat Cell Biol 2000;2:313 -- 7. doi:10.1038/35014000. [156] Li F, Redick SD, Erickson HP, Moy VT. Force measurements of the alpha5beta1 integrin-fibronectin interaction. Biophys J 2003;84:1252 -- 62. doi:10.1016/S0006-3495(03)74940-6. [157] Lower SK, Hochella MF, Beveridge TJ. Bacterial recognition of mineral surfaces: nanoscale interactions between Shewanella and alpha-FeOOH. Science 2001;292:1360 -- 3. doi:10.1126/science.1059567. [158] Bowen W, Lovitt RW, Wright C. Atomic Force Microscopy Study of the Adhesion of Saccharomyces cerevisiae. J Colloid Interface Sci 2001;237:54 -- 61. doi:10.1006/jcis.2001.7437. [159] Razatos A, Ong Y, Sharma M GG. Molecular determinants of bacterial adhesion monitored by atomic force microscopy. Proc Natl Acad Sci U S A 1998;95:11059 -- 64. doi:10.1073/pnas.95.19.11059. [160] Guillaume-Gentil O, Potthoff E, Ossola D, Franz CM, Zambelli T, Vorholt JA. Force-controlled manipulation of single cells: From AFM to FluidFM. Trends Biotechnol 2014;32:381 -- 8. doi:10.1016/j.tibtech.2014.04.008. [161] Benoit M, Gaub HE. Measuring cell adhesion forces with the atomic force microscope at the molecular level. Cells Tissues Organs 2002;172:174 -- 89. doi:10.1159/000066964. [162] Fotiadis D, Scheuring S, Müller S, Engel A, Müller DJ. Imaging and manipulation of biological structures with the AFM. Micron 2002;33:385 -- 97. doi:10.1016/S0968-4328(01)00026-9. [163] Weisenhorn L, Hansma PK, Albrecht TR, Quate CF. Forces in atomic force microscopy in air and water. Appl Phys Lett 1989;54:2651 -- 3. 75 [164] Müller D, Büldt G, Engel A. Force-induced conformational change of bacteriorhodopsin. J Mol Biol 1995;249:239 -- 43. doi:10.1006/jmbi.1995.0292. [165] Müller DJ, Schabert F, Büldt G, Engel A. Imaging purple membranes in aqueous solutions at sub-nanometer resolution by atomic force microscopy. Biophys J 1995;68:1681 -- 6. doi:10.1016/S0006- 3495(95)80345-0. [166] Hansma HG, Vesenka J, Siegerist C, Kelderman G, Morrett H, Sinsheimer RL, et al. Reproducible imaging and dissection of plasmid DNA under liquid with the atomic force microscope. Science (80- ) 1992;256:1180 -- 4. doi:10.1126/science.256.5060.1180. [167] Müller DJ, Schoenenberger CA, Büldt G, Engel A. Immuno-atomic force microscopy of purple membrane. Biophys J 1996;70:1796 -- 802. doi:10.1016/S0006-3495(96)79743-6. [168] Taubenberger A, Cisneros DA, Friedrichs J, Puech P-H, Muller DJ, Franz CM. Revealing Early Steps of α2β1 Integrin-mediated Adhesion to Collagen Type I by Using Single-Cell Force Spectroscopy. Mol Biol Cell 2007;18:1634 -- 44. doi:10.1091/mbc.E06. [169] Dettmann W, Grandbois M, André S, Benoit M, Wehle AK, Kaltner H, et al. Differences in zero-force and force-driven kinetics of ligand dissociation from β-galactoside-specific proteins (plant and animal lectins, immunoglobulin G) monitored by plasmon resonance and dynamic single molecule force microscopy. Arch Biochem Biophys 2000;383:157 -- 70. doi:10.1006/abbi.2000.1993. [170] Müller DJ, Engel A. Voltage and pH - induced channel closure of porin OmpF visualized by atomic force microscopy. J Mol Biol Vol 1999;285:1347 -- 51. doi:https://doi.org/10.1006/jmbi.1998.2359. [171] Drake B, Prater CB, Weisenhorn a L, Gould S a, Albrecht TR, Quate CF, et al. Imaging crystals, polymers, and processes in water with the atomic force microscope. Science 1989;243:1586 -- 9. doi:10.1126/science.2928794. [172] Mou J, Yang J, Shao Z. Atomic force microscopy of cholera toxin B- oligomers bound to bilayers of biologically relevant lipids. J Mol Biol 1995;248:507 -- 12. doi:10.1006/jmbi.1995.0238. [173] Müller DJ, Baumeister W, Engel A. Conformational change of the hexagonally packed intermediate layer of Deinococcus radiodurans 76 monitored by atomic force microscopy. J Bacteriol 1996;178:3025 -- 30. [174] Friedrichs J, Helenius J, Müller DJ. Stimulated single-cell force spectroscopy to quantify cell adhesion receptor crosstalk. Proteomics 2010;10:1455 -- 62. doi:10.1002/pmic.200900724. [175] Beaussart A, Herman P, El-Kirat-Chatel S, Lipke PN, Kucharíková S, Van Dijck P, et al. Single-cell force spectroscopy of the medically important Staphylococcus epidermidis-Candida albicans interaction. Nanoscale 2013;5:10.1039/c3nr03272h. doi:10.1039/c3nr03272h. [176] Sénéchal A, Carrigan SD, Tabrizian M. Probing surface adhesion forces of enterococcus faecalis to medical-grade polymers using atomic force microscopy. Langmuir 2004;20:4172 -- 7. doi:10.1021/la035847y. [177] Chang K-C, Hammer DA. Influence of Direction and Type of Applied Force on the Detachment of Macromolecularly-Bound Particles from Surfaces. Langmuir 1996;12:2271 -- 82. doi:10.1021/la950690y. [178] Busscher HJ, Poortinga AT, Bos R. Lateral and perpendicular interaction forces involved in mobile and immobile adhesion of microorganisms on model solid surfaces. Curr Microbiol 1998;37:319 -- 23. doi:10.1007/s002849900385. [179] Parpura V, Haydon PG, Sakaguchi DS, Henderson E. Atomic force microscopy and manipulation of living glial cells. J Vac Technol A 1993;11:773 -- 5. doi:10.1116/1.578346. [180] Sagvolden G, Giaever I, Feder J. Characteristic Protein Adhesion Forces on Glass and Polystyrene Substrates by Atomic Force Microscopy. Langmuir 1998;14:5984 -- 7. doi:10.1021/la980271b. [181] Meister A, Gabi M, Behr P, Studer P, Vörös J, Niedermann P, et al. FluidFM: Combining atomic force microscopy and nanofluidics in a universal liquid delivery system for single cell applications and beyond. Nano Lett 2009;9:2501 -- 7. doi:10.1021/nl901384x. [182] Saftics A, Türk B, Sulyok A, Nagy N, Szekacs I, Horvath R, et al. Biomimetic dextran-based hydrogel layers for cell micropatterning over large areas using the FluidFM BOT technology. Langmuir 2019. doi:10.1021/acs.langmuir.8b03249. [183] Dörig P, Stiefel P, Behr P, Sarajlic E, Bijl D, Gabi M, et al. Force- controlled spatial manipulation of viable mammalian cells and micro- organisms by means of FluidFM technology. Appl Phys Lett 77 2010;97:023701. doi:https://doi.org/10.1063/1.3462979. [184] Potthoff E, Ossola D, Zambelli T, Vorholt JA. Bacterial adhesion force quantification by fluidic force microscopy. Nanoscale 2014;00:1 -- 3. doi:10.1039/c4nr06495j. [185] Amarouch MY, El Hilaly J, Mazouzi D. AFM and FluidFM Technologies: Recent Applications in Molecular and Cellular Biology. Scanning 2018;2018:1 -- 10. doi:10.1155/2018/7801274. [186] Grüter RR, Vörös J, Zambelli T. FluidFM as a lithography tool in liquid: Spatially controlled deposition of fluorescent nanoparticles. Nanoscale 2013;5:1097 -- 104. doi:10.1039/c2nr33214k. [187] Hirt L, Ihle S, Pan Z, Dorwling-Carter L, Reiser A, Wheeler JM, et al. Template-Free 3D Microprinting of Metals Using a Force-Controlled Nanopipette for Layer-by-Layer Electrodeposition. Adv Mater 2016;28:2311 -- 5. doi:10.1002/adma.201504967. [188] Ventrici De Souza J, Liu Y, Wang S, Dörig P, Kuhl TL, Frommer J, et al. Three-Dimensional Nanoprinting via Direct Delivery. J Phys Chem B 2018;122:956 -- 62. doi:10.1021/acs.jpcb.7b06978. [189] Wang JH, Li B. The principles and biological applications of cell traction force microscopy. Microsc Sci Technol Appl Educ 2010;29:449 -- 58. [190] Sabass B, Gardel ML, Waterman CM, Schwarz US. High resolution traction force microscopy based on experimental and computational advances. Biophys J 2008;94:207 -- 20. doi:10.1529/biophysj.107.113670. [191] Schwarz US, Balaban NQ, Riveline D, Bershadsky A, Geiger B, Safran SA. Calculation of forces at focal adhesions from elastic substrate data: The effect of localized force and the need for regularization. Biophys J 2002;83:1380 -- 94. doi:10.1016/S0006-3495(02)73909-X. [192] Hur SS, Zhao Y, Li YS, Botvinick E, Chien S. Live Cells Exert 3- Dimensional Traction Forces on Their Substrata. Cell Mol Bioeng 2009;2:425 -- 36. doi:10.1007/s12195-009-0082-6. [193] Wang JHC, Lin JS. Cell traction force and measurement methods. Biomechan Model Mechanobiol 2007;6:361 -- 71. doi:10.1007/s10237-006- 0068-4. [194] Ladoux B, Nicolas A. Physically based principles of cell adhesion mechanosensitivity in tissues. Reports Prog Phys 2012;75:116601. doi:10.1088/0034-4885/75/11/116601. 78 [195] Colin-York H, Eggeling C, Fritzsche M. Dissection of mechanical force in living cells by super-resolved traction force microscopy. Nat Protoc 2017;12:783 -- 96. doi:10.1038/nprot.2017.009. [196] Hall MS, Long R, Feng X, Huang Y, Hui CY, Wu M. Toward single cell traction microscopy within 3D collagen matrices. Exp Cell Res 2013;319:2396 -- 408. doi:10.1016/j.yexcr.2013.06.009. [197] Steinwachs J, Metzner C, Skodzek K, Lang N, Thievessen I, Mark C, et al. Three-dimensional force microscopy of cells in biopolymer networks. Nat Methods 2016;13:171 -- 6. doi:10.1038/nmeth.3685. [198] Yang Z. Validation of the Boussinesq equation for use in traction field determination. Comput Methods Biomech Biomed Engin 2011;14:1065 -- 70. doi:10.1080/10255842.2010.506642. [199] Yang Z, Lin JS, Chen J, Wang JHC. Determining substrate displacement and cell traction fields-a new approach. J Theor Biol 2006;242:607 -- 16. doi:10.1016/j.jtbi.2006.05.005. [200] Krishnan R, Park CY, Lin YC, Mead J, Jaspers RT, Trepat X, et al. Reinforcement versus fluidization in cytoskeletal mechanoresponsiveness. PLoS One 2009;4:e5486. doi:10.1371/journal.pone.0005486. [201] Li B, Li F, Puskar KM, Wang JHC. Spatial patterning of cell proliferation and differentiation depends on mechanical stress magnitude. J Biomech 2009;42:1622 -- 7. doi:10.1016/j.jbiomech.2009.04.033. [202] Chen J, Li H, SundarRaj N, Wang JHC. Alpha-smooth muscle actin expression enhances cell traction force. Cell Motil Cytoskeleton 2007;64:248 -- 57. doi:10.1002/cm.20178. [203] Gov NS. Traction forces during collective cell motion. HFSP J 2009;3:223 -- 7. doi:10.2976/1.3185785. [204] Ladoux B. Cells guided on their journey. Nat Phys 2009;5:377 -- 8. doi:10.1038/nphys1281. [205] Lakowicz JT. Principles of fluorescence spectroscopy. 3rd ed. 2006. [206] Clapp AR, Medintz IL, Mauro JM, Fisher BR, Bawendi MG, Mattoussi H. Fluorescence Resonance Energy Transfer between Quantum Dot Donors and Dye-Labeled Protein Acceptors. J Am Chem Soc 2004;126:301 -- 10. doi:10.1021/ja037088b. [207] Weiss S. Fluorescence spectroscopy of single biomolecules. Science (80- ) 79 1999;283:1676 -- 83. doi:DOI: 10.1126/science.283.5408.1676. [208] Stryer L, Hauglnd RP. Energy transfer: A spectroscopic ruler. Proc NAS 1967;58:719 -- 26. doi:10.1146/annurev.bi.47.070178.004131. [209] Kong HJ, Polte TR, Alsberg E, Mooney DJ. FRET measurements of cell- traction forces and nano-scale clustering of adhesion ligands varied by substrate stiffness. Proc Natl Acad Sci 2005;102:4300 -- 5. doi:10.1073/pnas.0405873102. [210] Freikamp A, Mehlich A, Klingner C, Grashoff C. Investigating piconewton forces in cells by FRET-based molecular force microscopy. J Struct Biol 2017;197:37 -- 42. doi:10.1016/j.jsb.2016.03.011. [211] Cost AL, Ringer P, Chrostek-Grashoff A, Grashoff C. How to Measure Molecular Forces in Cells: A Guide to Evaluating Genetically-Encoded FRET-Based Tension Sensors. Cell Mol Bioeng 2015;8:96 -- 105. doi:10.1007/s12195-014-0368-1. [212] Jurchenko C, Salaita KS. Lighting Up the Force: Investigating Mechanisms of Mechanotransduction Using Fluorescent Tension Probes. Mol Cell Biol 2015;35:2570 -- 82. doi:10.1128/MCB.00195-15. [213] Chang C-W, Kumar S. Vinculin tension distributions of individual stress fibers within cell-matrix adhesions. J Cell Sci 2013;126:3021 -- 30. doi:10.1242/jcs.119032. [214] Borghi N, Sorokina M, Shcherbakova OG, Weis WI, Pruitt BL, Nelson WJ, et al. E-cadherin is under constitutive actomyosin-generated tension that is increased at cell-cell contacts upon externally applied stretch. Proc Natl Acad Sci 2012;109:12568 -- 73. doi:10.1073/pnas.1204390109. [215] Cai D, Chen SC, Prasad M, He L, Wang X, Choesmel-Cadamuro V, et al. Mechanical feedback through E-cadherin promotes direction sensing during collective cell migration. Cell 2014;157:1146 -- 59. doi:10.1016/j.cell.2014.03.045. [216] Conway DE, Breckenridge MT, Hinde E, Gratton E, Chen CS, Schwartz MA. Fluid shear stress on endothelial cells modulates mechanical tension across VE-cadherin and PECAM-1. Curr Biol 2013;23:1024 -- 30. doi:10.1016/j.cub.2013.04.049. [217] Kelley M, Yochem J, Krieg M, Calixto A, Heiman MG, Kuzmanov A, et al. FBN-1, a fibrillin-related protein, is required for resistance ofthe epidermis to mechanical deformation during c. Elegans embryogenesis. 80 Elife 2015:1 -- 30. doi:10.7554/eLife.06565. [218] Morimatsu M, Mekhdjian AH, Adhikari AS, Dunn AR. Molecular tension sensors report forces generated by single integrin molecules in living cells. Nano Lett 2013;13:3985 -- 9. doi:10.1021/nl4005145. [219] Huebsch ND, Mooney DJ. Fluorescent resonance energy transfer: A tool for probing molecular cell-biomaterial interactions in three dimensions. Biomaterials 2007;28:2424 -- 37. doi:10.1016/j.biomaterials.2007.01.023. [220] Rahul R, Sungchul H, Taekjip H. A Practical Guide to Single Molecule FRET. Nat Methods 2013;5:507 -- 16. doi:10.1038/nmeth.1208.A. [221] Johnson AE. Fluorescence approaches for determining protein conformations, interactions and mechanisms at membranes. Traffic 2005;6:1078 -- 92. doi:10.1111/j.1600-0854.2005.00340.x. [222] Dobrucki JW, Kubitscheck U. Fluorescence Microscopy. Fluoresc Microsc 2005;2:910 -- 9. doi:10.1002/9783527687732.ch3. [223] Kirchner J, Kam Z, Tzur G, Bershadsky AD, Geiger B. Live-cell monitoring of tyrosine phosphorylation in focal adhesions following microtubule disruption. J Cell Sci 2002;116:975 -- 86. doi:10.1242/jcs.00284. [224] Kong HJ, Boontheekul T, Mooney DJ. Quantifying the relation between adhesion ligand-receptor bond formation and cell phenotype. Proc Natl Acad Sci 2006;103:18534 -- 9. doi:10.1073/pnas.0605960103. [225] Yamada S, Nelson WJ. Localized zones of Rho and Rac activities drive initiation and expansion of epithelial cell-cell adhesion. J Cell Biol 2007;178:517 -- 27. doi:10.1083/jcb.200701058. [226] Svoboda K, Block SM. Force and velocity measured for single kinesin molecules. Cell 1994;77:773 -- 84. doi:10.1016/0092-8674(94)90060-4. 81
1102.0615	1	1102	2011-02-03T08:07:12	Electromediated formation of DNA complexes with cell membranes and its consequences for gene delivery	[ "physics.bio-ph", "cond-mat.soft", "q-bio.SC" ]	Electroporation is a physical method to induce the uptake of therapeutic drugs and DNA, by eukaryotic cells and tissues. The phenomena behind electro-mediated membrane permeabilization to plasmid DNA have been shown to be significantly more complex than those for small molecules. Small molecules cross the permeabilized membrane by diffusion whereas plasmid DNA first interacts with the electropermeabilized part of the cell surface, forming localized aggregates. The dynamics of this process is still poorly understood because direct observations have been limited to scales of the order of seconds. Here, cells are electropermeabilized in the presence of plasmid DNA and monitored with a temporal resolution of 2 ms. This allows us to show that during the first pulse application, plasmid complexes, or aggregates, start to form at distinct sites on the cell membrane. FRAP measurements show that the positions of these sites are remarkably immobile during the application of further pluses. A theoretical model is proposed to explain the appearance of distinct interaction sites, the quantitative increase in DNA and also their immobility leading to a tentative explanation for the success of electro-mediated gene delivery.	physics.bio-ph	physics	Electromediated formation of DNA complexes with cell membranes and its consequences for gene delivery Jean-Michel Escoffre1 Institut de Pharmacologie et de Biologie Structurale, CNRS, Universit´e de Toulouse, UPS, Toulouse, France Institut de Pharmacologie et de Biologie Structurale and Laboratoire de Physique Th´eorique, CNRS, Thomas Portet1 Universit´e de Toulouse, UPS, Toulouse, France Cyril Favard1 Institut Fresnel, CNRS, Universit´es Aix-Marseille, Marseille, France Justin Teissi´e Institut de Pharmacologie et de Biologie Structurale, CNRS, Universit´e de Toulouse, UPS, Toulouse, France David S. Dean Laboratoire de Physique Th´eorique, CNRS, Universit´e de Toulouse, UPS, Toulouse, France Marie-Pierre Rols∗ Institut de Pharmacologie et de Biologie Structurale, CNRS, Universit´e de Toulouse, UPS, Toulouse, France Abstract Electroporation is a physical method to induce the uptake of therapeutic drugs and DNA, by eukaryotic cells and tissues. The phenomena behind electro-mediated mem- brane permeabilization to plasmid DNA have been shown to be significantly more com- plex than those for small molecules. Small molecules cross the permeabilized mem- brane by diffusion whereas plasmid DNA first interacts with the electropermeabilized part of the cell surface, forming localized aggregates. The dynamics of this process is still poorly understood because direct observations have been limited to scales of 1 1 0 2 b e F 3 ] h p - o i b . s c i s y h p [ 1 v 5 1 6 0 . 2 0 1 1 : v i X r a ∗Corresponding author Email address: [email protected], Fax : +33 (0) 561 175 994, Tel :+33 (0) 561 175 811 ( Marie-Pierre Rols ) 1The first three authors made an equal contribution to this paper Preprint submitted to Elsevier October 26, 2018 the order of seconds. Here, cells are electropermeabilized in the presence of plasmid DNA and monitored with a temporal resolution of 2 ms. This allows us to show that during the first pulse application, plasmid complexes, or aggregates, start to form at dis- tinct sites on the cell membrane. FRAP measurements show that the positions of these sites are remarkably immobile during the application of further pluses. A theoretical model is proposed to explain the appearance of distinct interaction sites, the quantita- tive increase in DNA and also their immobility leading to a tentative explanation for the success of electro-mediated gene delivery. Keywords: Electroporation; DNA transfer; modeling; fluorescence microscopy 1. Introduction Electropermeabilization is the phenomena by which the application of an electric field across a biological membrane renders it permeable to the passage of small and even macro molecules such as DNA [1, 2]. It is exploited in the clinical context where the permeabilization of cells to a therapeutic agent is required, for example in gene therapy and chemotherapy [3, 4, 5]. Despite the use of electropermeabilization in med- ical science, many questions remain open as to the underlying biophysicochemical phenomena which underpin its success. This is especially the case for the permeabi- lization of membranes to DNA where a number of interesting biological, chemical and physical factors remain to be understood. Given the size of the DNA, if the permeabi- lization is due to pores -- or conducting defects -- , as suggested by the standard theory of electroporation [6, 7], the pores must be relatively large due to (i) the relatively large size of the DNA and (ii) the large charge of DNA as dielectric exclusion must be over- come [8]. The cell membrane has a much more complex organization than a model lipid bilayer. One expects that the location of regions permeabilized to DNA will be determined not only by the local electric field but also by the local membrane com- position. In cell membranes, the regions most susceptible to permeabilization may be those containing certain types of transmembrane proteins or at the boundary between differing types of lipid domains [9]. Another key, physicochemical factor is how the field not only modifies the permeability of the membrane, but also how it influences 2 the movement of the DNA by electrophoresis. Large polyelectrolytes are strongly ad- vected by the electric field [10]. Experimental studies [11] have demonstrated that the physicochemical phenomena involved in electro-mediated membrane permeabilization to plasmid DNA are indeed significantly more complex than those for small molecules. Small molecules cross the permeabilized membrane directly mainly by post-pulse dif- fusion, whereas plasmid DNA first interacts with the electropermeabilized part of the cell surface resulting in the formation of localized aggregates. However the dynamics of this initial interaction is only partially understood because the direct observations of [11] were limited to time scales exceeding several seconds. In this paper, we have ana- lyzed the interaction between the cell membrane and strongly charged macromolecules (4.7 kbp plasmid DNA) upon the application of a permeabilizing electric field, at a tem- poral resolution of 2 ms. This enables us to address the following unresolved, and key, questions (i) What are the phenomena leading to formation of the sites where the DNA aggregates? (ii) What is the underlying dynamics behind their formation? (iii) To what extent are the sites localized in space, and over what time scales? and (iv) What is the biochemical mechanism behind this localization? We put forward a model for the transport of charged molecules in the presence of the electric field which provides an interpretation for most of our experimental findings on DNA and whose validity is also tested via additional experiments using the, much smaller, charged molecule Propidium Iodide. 2. Materials and methods 2.1. Cells Chinese hamster ovary (CHO) cells were used. The WTT clone was selected for its ability to grow in suspension or plated on Petri dishes or on a microscope glass coverslip. Cells were grown as previously described [12]. For microscopy experiments, 105 cells were put on a Lab-tek chamber 12 hours before electric pulse treatment with 1 ml of culture medium. 3 2.2. DNA staining A 4.7 kbp plasmid (pEGFP-C1, Clonetech, Palo Alto, CA) carrying the green flu- orescent protein gene controlled by the CMV promoter was stained stoechiometrically with the DNA intercalating dye TOTO-1 (or alternatively POPO-3) (Molecular Probes, Eugene, OR ) [13]. The plasmid was stained with 2.3x10−4 M dye at a DNA concen- tration of 1µg/µl for 60 minutes on ice. This concentration yields an average base pair to dye ratio of 5. Even if the labeling is not covalent, the equilibrium is dramatically in favor of the linked form. Plasmids were prepared from E. Coli transfected bacteria by using Maxiprep DNA purification system (Qiagen, Chatsworth, CA). 2.3. Electropermeabilization apparatus. Electropulsation was carried out with a CNRS cell electropulsator (Jouan, St Herblain, France) which delivers square-wave electric pulses. An oscilloscope (Enertec, St. Eti- enne, France) was used to monitor the pulse shape. The electropulsation chamber was built using two stainless-steel parallel rods (diameter 0.5 mm, length 10 mm, inter- electrode distance 5 mm) placed on a Lab-tek chamber. The electrodes were connected to the voltage generator. A uniform electric field was generated. The chamber was placed on the stage of an inverted digitized videomicroscope (Leica DMIRB, Wetzlar, Germany) or a confocal microscope (Zeiss, LSM 5 life, Germany). 2.4. Electropermeabilization Permeabilization of cells was performed by application of millisecond electric pulses required to transfer genes and to load macromolecules into cells. Cell viability was pre- served when using millisecond pulse duration by decreasing the electric field intensity [14, 15]. Penetration of PI (100 µM in a low ionic strength pulsing buffer: 10 mM phosphate, 1 mM MgCl2, 250 mM sucrose, pH 7.4) was used to monitor permeabiliza- tion. 10 pulses of 20 ms duration and 0.5 kV/cm amplitude were applied at a frequency of 1 Hz at room temperature. For plated cells, the culture medium was removed and replaced by the same buffer described above. 4 2.5. Microscopy Cells were observed with a Leica 100×, 1.3 numerical aperture oil immersion ob- jective. Images (and optical sections) were recorded with the CELLscan System from Scanalytics (Billerica, MA) fitted with a cooled CCD camera (Princeton Instrument, Trenton, NJ). This digitizing set up allowed a quantitative localized analysis of the flu- orescence emission as described previously [11]. This was done along the cell mem- brane. Images were taken at a 1 Hz frequency. For fast kinetics studies, a Zeiss LSM 5 Live confocal microscope was used. All measurements were performed at room temperature. Image sequences were acquired at a frequency of 500 fps. 2.6. FRAP experiments FRAP experiments were performed on a Zeiss LSM 510 confocal microscope. The 488 nm line of the Ar+ laser was used for excitation of TOTO-1. We used a sequential mode of acquisition with a 63×, 1.4 numerical aperture water immersion lens. After 50 pre-bleach scans, a region of interest (ROI) with a radius w = 1µm was bleached, and fluorescence recovery was sampled on 150 scans, i.e. 40 s [16]. 2.7. Actin cytoskeleton destabilization To examine the role of actin in the phenomenon of DNA membrane interaction, cells were incubated at 37o C for 1 hour with 1 µM Latrunculin A (Sigma) in culture medium. This protocol is known to efficiently disrupt the actin cytoskeleton [17]. 3. Experimental results 3.1. Electropulsation experiments To study membrane permeabilization by electric fields, and the associated pro- cesses of molecular uptake, we used PI and TOTO-1 labelled plasmid DNA. Membrane permeabilization was induced by applying 20 ms electric pulses of intensity 0.5 kV/cm. This protocol is known to induce both membrane permeabilization and DNA transfer, accompanied by an associated gene expression [11]. Membrane permeabilization was observed at the single cell level at a rate of 500 frames per second. This image acquisi- tion frequency made it possible to monitor the entire process of molecular uptake, both 5 during and after pulse application, with a good spatial resolution using confocal mi- croscopy. Particular attention was paid to the molecular uptake occurring after a single permeabilizing pulse and a series of 10 pulses. 3.2. Electropulsation experiments with PI As shown in Figs. 1A and 1C, and in agreement with previous studies [11], mem- brane permeabilization, as detected by the uptake of PI, was observed at sites of the cell membrane facing the two electrodes. The influx into the cells of PI, deduced from the associated fluorescence intensity, suggests free transport across the permeabilized re- gions of the membrane into the cytoplasm (no trapping in the region near the membrane is observed). The uptake of PI into cells started at the moment the pulse was applied and the concentration of PI in the cell increased for up to a minute (Fig. 1E), showing that the permeabilized membrane state, due to defect regions such as metastable pores, persists for some time after the field is cut. Indeed, the majority of molecules which were taken up entered the cell after the pulses. Analyzing the molecular uptake image by image led to the observation that PI uptake was both asymmetric and delayed: it was higher at the anode facing side of the cell and its entrance at the cathode facing side occurred after a 4-5 second delay. This asymmetry can be explained by the fact that PI has a charge +2e in solution and thus the electrophoretic force drives it toward the cathode. The post-pulse diffusion of PI into the cytoplasm of the permeabilized cell can be seen in Movie 1 (published in supporting information). 3.3. Electropulsation experiments with DNA Electro-induced uptake by cells in presence of plasmid DNA showed results that differed considerably from those for cells pulsed in presence of PI. DNA was only observed to interact with the membrane for electric fields greater than a critical value (0.25 kV/cm) which induces their permeabilization, i.e. leading to PI uptake. In the low field regime DNA simply flowed around the cells towards the anode. For permeabi- lizing fields, plasmid DNA was not observed to enter the cell during pulse application or during the minute that followed. However, DNA was seen to accumulate at distinct sites on the cathode facing side of the permeabilized cell membrane (Fig. 1D). At these 6 sites the DNA appeared to form aggregates, which became visible as soon as the elec- tric pulse was triggered (Fig. 2A, 2B). Intriguingly, the number and apparent size of these sites did not increase with the number of pulses applied. During the application of a single pulse, the fluorescence of the sites increased linearly in time showing that the quantity of DNA at each site increased linearly (Fig. 1F), but no such fluorescence increase was observed in the absence of the electric field. Therefore the accumulation of DNA at the interaction sites must be principally due to electrophoresis. The presence of stable sites, where membrane plasmid DNA interaction occurred, was observed across the entire cell population (Fig. 1D) (see Movie 2 in supporting information). The average distance between a site and its nearest neighbor was found to be about 1 µm. The diameter of the individual sites was found to be in the range of 300-600 nm (lower range limit due to optical diffraction). As shown in Fig. 2C, the total amount of DNA in the membrane region also increased linearly with the number of electric pulses. The final average amount of DNA localized in the membrane region correlated with the amount of DNA reaching the nucleus, as estimated by the rate of GFP expression, between 2 and 24 hours after pulse application (see supplementary information). As shown in Figs. 1E and 1F, part of the DNA interacting with the membrane was observed to be desorbed after the pulses but a large proportion stayed fixed, this could be due to an interaction between the membrane and DNA but also because the DNA forced into the pore electrophoretically is virtually immobile in the cell in the absence of the electric field. This last observation shows that the observed DNA membrane interaction could be, at least partially, due to (i) electrophoretic accumulation along with (ii) reduced mobility of the DNA in the pore region within the cell. We also studied the mobility of the DNA complexes at the interaction sites. In order to measure the in-membrane or lateral diffusion constant of the complexes, we carried out Fluorescence Recovery After Photobleaching (FRAP) experiments. The FRAP results showed that no exchange between DNA aggregates or with the bulk DNA took place and that the DNA was, on experimental time scales, totally immobile, its lateral diffusion coefficient being less than 10−16m2s−1. This could be explained by the reduced mobility of individual DNA molecules within the cell or because DNA 7 forms micro sized aggregates which are immobile due to their large size. It is unlikely that this immobility is due to an intrinsic immobility of pores, as the measured diffusion constant is orders of magnitude smaller than the typical values reported in the literature for transmembrane objects, such as proteins. One could argue that actin polymerization around inserted DNA may be responsible for its immobilization at the membrane level as well as in its subsequent traffic inside the cytoplasm. However, actin cytoskeleton perturbation with Latrunculin A did not seem to modify the features of DNA spot creation, their stability or immobility (see supplementary information). This implies that, despite the fact that actin causes a drastic decrease in the mobility of individual large DNA molecules inside cells, the initial formation of DNA spots and their immobility is not caused by the actin network. 4. Theoretical analysis 4.1. Theoretical model In order to better understand the features of the electro-mediated DNA uptake, we developed a theoretical model that we used to carry out numerical simulations. We assumed that the cell membrane was an infinitesimally thin non-conducting surface. This approximation is valid where the membrane thickness is small compared to all other length scales and the conductivities of the cell interior σi and exterior electropul- sation solution σe are much greater than the conductivity σm of the cell membrane. An electric field of magnitude E0 = E0 is applied in the z direction, perpendicular to the electrodes, and the local electric potential φ is given, in the steady state regime, by the solution to Laplace's equation [18] ∇ · σ∇φ = 0, (1) with the boundary conditions φ = −E0z as z → ∞. In what follows we will assume that the DNA is advected by the field given by the solution of Laplace's equation (1) and will neglect additional fields that would be caused by local accumulation of charge due to the DNA. This is because the applied fields are relatively large and because we expect the DNA to be strongly screened. Once the electric field has been computed the 8 DNA is advected by the field but also diffuses. The local concentration of DNA c(x, t) obeys the electrodiffusion equation = −∇ · j ∂c ∂t (2) where j is the thermodynamic current j = −D∇c + µcE, with D the local diffusion con- stant of the DNA, which depends on its environment, and is denoted by De outside the cell, Dm in the membrane and Di in the cell interior. The term µ is the electrophoretic mobility and is defined via the velocity v of the molecule in a (locally) uniform field E = −∇φ as v = µE. (3) As for the diffusion constant, the value of µ will also depend on the local environment as it depends on electrolyte concentration and viscosity. As the electrode system is in contact with a bath of the transported molecule we impose the boundary conditions ∇c = 0 as c → ∞ on equation (2) (note that this is compatible with the Ohmic behavior j = µcE far from the cell). We assume that Dm and µm = 0, which means that DNA cannot move into intact regions of membrane. 4.2. Modeling pores We will assume that when the field is applied, the membrane becomes permeabi- lized and micro sized pores are formed. For the ease of computation we will investigate what happens when a circular pore is formed at each face of the cell facing the elec- trodes, that facing the anode in the angular region θ ∈ [0, θp] and that facing the cathode in the region θ ∈ [π−θp, π]. In this region we assume that the conductivity is σi and that the diffusion constant is Di, effectively we have taken away the membrane from these regions and replace it by the cellular interior. We numerically compute the evolution of the concentration of marker molecules as a function of time using the experimental ap- plied field protocols (we assume that the membrane charging process is much quicker that any of the diffusion processes and thus change the electric field instantaneously). Estimation of theoretical parameters. We were able to follow the fluorescence of the DNA in the buffer solution in a uniform electric field and thus measure its velocity. Using equation (3). we estimated the electrophoretic mobility in solution to be µe = 9 10−8m2V−1s−1. This agrees with more precise measurements reported in the literature [19]. The diffusion constant of DNA in aqueous solution is estimated from the literature to be De = 10−12m2s−1[20]. The charge of the PI is taken to be q = +2e and its diffusion constant in aqueous solution is estimated to be De = 10−10m2s−1 (by estimating its effective radius and us- ing the Stokes formula for the diffusion constant of a solid sphere). The electrophoretic mobility for PI is estimated via the Stokes Einstein relation µ = qD kBT where kB is Boltz- mann's constant and T is the temperature. In the cell interior we estimate that for PI the effective diffusion constants and electrophoretic mobilities are smaller than their external values by a factor of 10 but for the larger DNA molecules they are smaller by a factor of 1 000 (in fact this factor is a lower bound) due largely to the interaction between the DNA and the actin network of the cell interior [20, 21]. Finally we took the radius of the spherical vesicle to be R = 8 µm as estimated from the average cell size. 4.3. Theoretical results and comparison with experiments We assume that the marker fluorescence is proportional to the concentration c(x, t). For the comparison with the experiments we took a pore angle θp = 3.6o as estimated by the size of the DNA/membrane interaction sites we observed. We computed the average fluorescence intensity inside the cell using Icell ∝ ccell. In Figs. 1G and 1H we show the predicted behavior for PI fluorescence for one pulse with applied field 0.5 kV cm−1 applied for 20 ms and see that it compares very well with that measured experimentally (Fig. 1 E). Similarly we see in the same figures the comparison between theory and experiment is also good for DNA (with the same pulse protocol), with the exception that the slight decrease in DNA fluorescence after the pulse has been applied is less pronounced in the theoretical curve of Fig. 1F. However some of the experi- mentally observed decrease may be due to photo bleaching of the fluorescent marker. Note that although we have only simulated one pore, if we assume that the pores be- have independently (as conduction channels in parallel) the overall form of the increase in fluorescence for several pores should be approximately of the same form, up to an overall change in the fluorescence levels, as that for single pore. 10 In the case of PI, our model also reproduced the other qualitative behavior seen in the experiments. Numerical calculations showed that the PI enters through the anode facing side of the vesicle during the pulse application and that its entry into the cathode facing side is delayed, as expected from the physical arguments given in the Results section. For DNA, computations showed that just after the field application there is virtually no uptake of DNA at the anode facing side of the cell, but that at the cathode facing side there is an accumulation of DNA near the surface of the cell in the region where the pore is. In our model the DNA accumulation, apparently, at the surface of the cell, is due to the strong (as compared to the case of PI) electrophoretic force which pushes the DNA into the hole opposite the cathode and then the much reduced mobility of the DNA in this region. For DNA most of the contribution to the total cell fluorescence comes from the region close to the membrane surface, thus justifying our comparison of Figs. 1E, 1F with Figs. 1G, 1H. 4.4. Constant number of interaction sites The experimental results show that after the first interaction sites become visible, no further sites appear during subsequent pulsation. An explanation for this is that the pores formed are conducting and thus the electric field across the membrane is lowered in the non-conducting regions of the membrane. A simple criterion for determining whether a membrane can be locally permeabilized is if the stress caused by the electric field causes the local surface tension to exceed the lysis tension of the membrane. In a simple electric model for the membrane this turns out to be equivalent to the local transmembrane voltage ∆φ exceeding a critical value ∆φc, which has values typically between 250 − 1000 mV [22, 23]. To see how the presence of a pore reduces the trans- membrane potential elsewhere we consider the simplified case of an infinitesimally thin, non-conducting, infinite flat membrane with a conducting pore of radius a. The potential drop across the membrane at a distance r from the center of the pore is [24] ∆φ(r) = 2∆φo ), where ∆φo is the potential drop far from the pore (or in the absence of the pore) and we assume that ∆φo > ∆φc . If pores can only be formed in regions where ∆φ > ∆φc one sees that this critical potential drop cannot be exceeded π arctan( √ r2−a2 a 11 (cid:16) π∆φc a 2∆φo cos (cid:17) from the center of the pore. Although a very approxi- within in a radius rc = mate estimate, this shows that we should expect pores to be separated from each other by a distance of the order of the pore size. This is indeed the case as can be seen in Fig. 2B, where the fluorescence peaks widths are of the same order as the inter-peak distance. 5. Discussion Despite the complexity of the system studied here, a number of our experimental observations can be explained qualitatively and to an extent quantitatively on a largely physical basis. According to the standard theory of electropermeabilization, the effect of the electric field is to render the cell membrane permeable to external molecules via the formation of micro sized pores. Another effect of the electric field on the DNA is an electrophoretic one, DNA is pushed toward the cathode facing side of the cell and, as our numerical calculations have shown, if a sufficiently large pore is present the DNA can be forced through it. However once the DNA is inside the cell it stays, on experimental time scales, very close to the surface either due to the reduced elec- trophoretic mobility and diffusion constant of individual DNA molecules caused by the actin cytoskeleton or because it forms aggregates which are immobile due to their large size. The number of DNA interactions sites remains constant even on increasing the number of pulses, we have argued that this is because these sites are conducting, the electric field elsewhere in the cell membrane is reduced by the presence of these con- ducting sites and thus does not exceed the threshold value necessary to form addition permeabilized regions and thus interaction sites. Results on cells where the actin cytoskeleton is disrupted also show spot formation and so we conclude that the principal mechanism for spot formation is the formation of aggregates where the DNA molecules are bound together and thus diffuse as a macro- scopic object with a very small diffusion constant. A possible mechanism for this aggregation is the presence of multivalent cations induced by the high concentration of DNA in the pore region [25]. To conclude we have been able to provide a detailed explanation of why gene ther- 12 apy using electropulsation is successful. The process of plasmid transfer through the cellular cytoplasm to the nuclear envelope is a complex process [26, 27]. In principle micro sized aggregates of DNA or vesicles filled with DNA could be too large to pass through the pores formed by electroporation. However individual DNA molecules, while they can pass through electropores, have a limited mobility within the cell and may well be totally degraded before reaching the nucleus. It is possible and worth in- vestigating the possibility that the actin cytoskeleton reacts to the presence of DNA ag- gregates and plays an important role in the subsequent intracellular transport. It seems reasonable that only aggregates beyond a certain size (a few hundred nano-meters) can induce a biological cellular response and can be transported by the cell. In addition, the fact that the DNA is in aggregate form means that the DNA in the center of the ag- gregate is relatively protected from degradation. Therefore, for gene therapy purposes, it is optimal for DNA to enter the cell as single molecules, but the subsequent trans- port toward the nucleus is, for biological (possibly by inducing a response of the actin cytoskeleton) and physical (diminishing enzymatic degradation) reasons, optimized if the DNA is in a micro-sized aggregate form. We thus see that a rather beautiful and subtle, and to an extent fortuitous, combi- nation of biological, chemical and physical factors may underpin the success of gene therapy via electropulsation. As our understanding of these underlying phenomena ad- vances we should be able to refine and optimize the protocols used in electro-mediated gene therapies. 6. Acknowledgements Our group belongs to the CNRS consortium Celltiss. This work has benefitted from the financial support of the Association Franc¸aise contre les Myopathies, the Re- gion Midi-Pyrenees and the Institut Universitaire de France. We wish to thank Gabor Forgacs for useful discussions on this work. 13 References [1] Neumann, E., A.E. Sowers, and C.A. Jordan, 1989. Electroporation and electro- fusion in cell biology (Plenum, New York). [2] Weaver J.C., 1995. Electroporation theory. Concepts and mechanisms. Methods Mol. Biol. 55:3 -- 28. [3] Belehradek, M., C. Domenge, B. Luboinski, S. Orlowski, J. Belehradek and L. Mir, 1993. Electrochemotherapy, a new antitumor treatment. First clinical phase I-II trial. Cancer 72:3694 -- 3700. [4] Daud, A.I., R.C. DeConti, S. Andrews, P. Urbas, A.I. Riker, V.K. Sondack, P.N. Munster, D.M. Sullivan, K.E. Ugen, I.J. Messina and R. Heller, 2008. Phase I trial of interleukin-12 plasmid electroporation in patients with metastatic melanoma. J. Clin. Oncol. 26:5896 -- 5903. [5] Rols M.P., 2006. Electropermeabilization, a physical method for the delivery of therapeutic molecules into cells. Biochim. Biophys. Acta 1758:423 -- 428. [6] Pastushenko, V.F., Y.A. Chizmadzhev and V.B. Arekelyan, 1979. Electric break- down of bilayer membranes II. Calculation of the membrane lifetime in the steady-state diffusion approximation. Bioelectrochem. Bioenerg. 6:53-62. [7] Powell, K.T. and J.C. Weaver, 1986. Transient aqueous pores in bilayer mem- branes: a statistical theory. Bioelectrochem. Bioenerg. 15 : 211-227. [8] V.A. Parsegian, 1969. Energy of an ion crossing of a low dielectric membrane: Solutions to four relevant electrostatic problems. Nature 221:844 -- 846. [9] Antonov, V.F., W. Petrov, A.A. Molnar, D.A. Predvoditelev and A.S. Ivanov, 1980. Appearance of single-ion channels in unmodifed lipid bilayer-membranes at the phase transition temperature. Nature 283:585 -- 586. [10] Gabriel, B. and J. Teissi´e, 1999. Time courses of mammalian cell electroper- meabilization observed by millisecond imaging of membrane property changes during the pulse. Biophys. J. 76:2158 -- 2165. 14 [11] Golzio, M., J. Teissi´e and M.P. Rols, 2002. Direct visualization at the single-cell level of electrically mediated gene delivery. Proc. Natl. Acad. Sci. U S A 99:1292 -- 1297. [12] Rols, M.P., D. Coulet and J. Teissi´e, 1992. Highly efficient transfection of mam- malian cells by electric field pulses. Application to large volumes of cell culture by using a flow system. Eur. J. Biochem. 206:115 -- 121. [13] Rye, H.S., S. Yue, D.E. Wemmer, M.A. Quesada, R.P. Haugland, R.A. Mathies and A.N. Glazer, 1992. Stable fluorescent complexes of double-stranded DNA with bis-intercalating asymmetric cyanine dyes: properties and applications. Nu- cleic Acids Res. 20:2803 -- 2812. [14] Wolf, H., M.P. Rols, E. Boldt, E. Neumann and J. Teissi´e, 1994. Control by pulse parameters of electric field-mediated gene transfer in mammalian cells. Biophys. J. 66:524 -- 531. [15] Golzio, M., M.P. Mora, C. Raynaud, C. Delteil, J. Teissi´e and M.P. Rols, 1998. Control by osmotic pressure of voltage-induced permeabilization and gene trans- fer in mammalian cells. Biophys. J. 74:3015 -- 3022. [16] Causeret, M., N. Taulet, F. Comunale, C. Favard and C. Gauthier-Rouviere, 2005. N-cadherin association with lipid rafts regulates its dynamic assembly at cell-cell junctions in C2C12 myoblasts. Mol. Biol. Cell. 16:2168 -- 2180. [17] Sun, M., N. Northup, F. Marga, T. Huber, F.J. Byfield, I. Levitan and G. Forgacs, 2007. The effect of cellular cholesterol on membrane-cytoskeleton adhesion. J. Cell Sci. 120:2223 -- 2231. [18] Landau, L.D. and E.M. Lifshitz, 1975. Electrodynamics of Continuous Media (Pergamon Press, Oxford). [19] Stellwagen, N.C., C. Gelfi and P.G. Righetti, 1997. The free solution mobility of DNA. Biopolymers 42:687 -- 703. 15 [20] Lukacs, G.L., P. Haggie, O. Seksek, D. Lechardeur, N. Freedman and A.S. Verk- man, 2000. Size-dependent DNA mobility in cytoplasm and nucleus. J. Biol. Chem. 275:1625 -- 1629. [21] Dauty, E. and A.S. Verkman, 2005. Actin cytoskeleton as the principal determi- nant of size-dependent DNA mobility in cytoplasm. J. Biol. Chem. 280:7823 -- 7828. [22] Teissi´e, J. and M.P. Rols, 1993. An experimental evaluation of the critical po- tential difference inducing cell membrane electropermeabilization. Biophys. J. 65:409 -- 413. [23] Portet, T., F. Camps Febrer, J.M. Escoffre, C. Favard, M.P. Rols and D.S. Dean, 2009. Visualization of membrane loss during the shrinkage of giant vesicles under electropulsation. Biophys. J. 96:4109-4121. [24] Winterhalter, M. and W. Helfrich, 1987. Effect of voltage on pores in membranes. Phys. Rev. A 36:5874-5876. [25] Bloomfield, V.A., D.M. Crothers and I. Tinoco, 1999. Nucleic acids: structures, properties and functions (University Science Books, Sausalito). [26] Vaughn, E.E., J.V. DeGiulio and D.A. Dean, 2006. Intracellular trafficking of plasmids for gene therapy: mechanisms of cytoplasmic movement and nuclear import. Curr Gene Ther. 6:671-681. [27] Lam, A.P. and D.A. Dean, 2010. Progress and prospects: nuclear import of non- viral vectors. Gene Ther. 12:439-447. 7. Figures 16 Figure 1: Results and theoretical predictions of uptake of PI and DNA by electropulsed cells. 1A and 1B: cells in the presence of PI and DNA before electropulsation, there is clearly no uptake of either marker. 1C: uptake of PI by cells after being electropulsed. 1D: interaction of DNA and cells after being electropulsed. 1E: long time behavior of PI (red) and DNA (green) uptake measured from the corresponding fluorescence. 1F: same data shown over a shorter time scale. 1G: long term uptake of PI and DNA as predicted by the electrodiffusion model. 1H: corresponding behavior over shorter time scale. 17 0.00.10.20.30.40.50.00.20.40.60.81.0-10010203040500.00.20.40.60.81.01.2-10010203040500.00.20.40.60.81.00.00.10.20.30.40.50.00.20.40.60.81.0FEG Marker Amount (A.U.)t (s)HA+- Marker Amount (A.U.)t (s)ABCD Fluorescence Intensity (A.U.)t (s)EPEPEP DNA PI DNA PI DNA PI DNA PI Fluorescence Intensity (A.U.)t (s)EPFigure 2: Evolution of DNA fluorescence at interaction sites as a function of the number of elec- tropulses. 2A raw imaging data. 2B: quantified increase in DNA fluorescence based on a digital analysis of 2A. 2C: experimentally measured increase of total fluorescence due to DNA uptake as a function of the num- ber of applied pulses. 2D: increase in DNA fluorescence as a function of the number of pulses as predicted by the electrodiffusion model. 18 0246810120.000.050.100.150.200246810120.000.050.100.150.205 µm0123456789105 µm012345678910601234587910601234587910 DNA amount (A.U.)Number of electric pulsesx 10-22ABD Relative Fluorescence Intensity (A.U.)Number of electric pulsesC
1911.12407	1	1911	2019-11-27T20:26:40	A Dynamical Model of Oncotripsy by Mechanical Cell Fatigue: Selective Cancer Cell Ablation by Low-Intensity Pulsed Ultrasound (LIPUS)	[ "physics.bio-ph", "q-bio.CB" ]	The method of oncotripsy, first proposed in [S. Heyden and M. Ortiz (2016). Oncotripsy: Targeting cancer cells selectively via resonant harmonic excitation. Journal of the Mechanics and Physics of Solids, 92:164-175], exploits aberrations in the material properties and morphology of cancerous cells in order to ablate them selectively by means of tuned low-intensity pulsed ultrasound (LIPUS). We propose a dynamical model of oncotripsy that follows as an application of cell dynamics, statistical mechanical theory of network elasticity and 'birth-death' kinetics to describe processes of damage and repair of the cytoskeleton. We also develop a reduced dynamical model that approximates the three-dimensional dynamics of the cell and facilitates parametric studies, including sensitivity analysis and process optimization. We show that the dynamical model predicts---and provides a conceptual basis for understanding---the oncotripsy effect and other trends in the data of [D. R. Mittelstein, J. Ye, E. F. Schibber, A. Roychoudhury, L. T. Martinez, M. H. Fekrazad, M. Ortiz, P. P. Lee, M. G. Shapiro, M. Gharib (2019). Selective Ablation of Cancer Cells with Low Intensity Pulsed Ultrasound. BioRxiv] for cells in suspension, including the dependence of cell-death curves on cell and process parameters.	physics.bio-ph	physics	A DYNAMICAL MODEL OF ONCOTRIPSY BY MECHANICAL CELL FATIGUE: SELECTIVE CANCER CELL ABLATION BY LOW-INTENSITY PULSED ULTRASOUND (LIPUS) E. F. SCHIBBER1, D. MITTELSTEIN1, M. GHARIB1, M. SHAPIRO1, P. LEE2 AND M. ORTIZ1 Abstract. The method of oncotripsy, first proposed in [1], exploits aberrations in the material properties and morphology of cancerous cells in order to ablate them selectively by means of tuned low-intensity pulsed ultrasound (LIPUS). We propose a dynamical model of oncotripsy that follows as an application of cell dynamics, statistical mechanical theory of network elasticity and 'birth-death' kinetics to describe pro- cesses of damage and repair of the cytoskeleton. We also develop a reduced dynamical model that approximates the three-dimensional dy- namics of the cell and facilitates parametric studies, including sensitivity analysis and process optimization. We show that the dynamical model predicts -- and provides a conceptual basis for understanding -- the on- cotripsy effect and other trends in the data of Mittelstein et al. [2] for cells in suspension, including the dependence of cell-death curves on cell and process parameters. 1. Introduction The method of oncotripsy, first proposed in [1], exploits aberrations in the material properties and morphology of cancerous cells, cf., e. g., Figs. 1 and 2, in order to ablate them selectively by means of tuned low-intensity ultrasound. A wealth of observational evidence reveals that a substantial size difference between normal nuclei, with an average diameter of 7 to 9 microns, and malignant nuclei, which can reach a diameter of over 50 microns, often characterizes malignancy [3]. Using atomic force microscopy, [4] reported the stiffness of live metastatic cancer cells taken from pleural fluids of patients with suspected lung, breast and pancreas cancer. They found that the cell stiffness of metastatic cancer cells is more than 70% softer than the benign cells that line the body cavity. Swaminathan et al. [5] applied a magnetic tweezer system to measure that stiffness of human ovarian cancer cell lines and found that cells with the highest invasion and migratory potential are up to five times softer than healthy cells [5]. Experimental investigations of hepatocellular carcinoma cells (HCC) have also found that an increase in Key words and phrases. oncotripsy, ultrasound, LIPUS, biomechanics, fatigue. 1 2 E. F. SCHIBBER ET AL. Figure 1. Optical images showing deformability on three breast cells due to a constant stretching laser power of 600mW. Deformability increases in the cancerous MCF-7 and ModMCF-7 cells in comparison to the healthy cell MCF-10. Reproduced from [8] and [9]. Figure 2. (a-d) Healthy lymphocyte cells from non-Acute Lymphoblastic Leukemia patients. (e-h) Probable lym- phoblast cells showing marked differences in size and mor- phology with respect to the healthy cells. Reproduced from [10]. stiffness of the extracellular matrix (ECM) promotes HCC proliferation [6] and advances malignant growth [7]. Owing to these and other similar observed aberrations in material proper- ties and morphology attendant to malignancy, the eigenfrequencies at which cell resonance occurs are expected to differ markedly between healthy and cancerous cells. In a recent numerical study [1], Heyden and Ortiz have shown that HCC natural frequencies lie above those of healthy cells, with a typical gap in the lowest natural frequency of about 37 kHz. For instance, they computed the fundamental frequency of HCC to be of the order 80 kHz and of the order of 43 kHz for the healthy cells. Heyden and Ortiz [1] posited that, by exploiting this spectral gap, cancerous cells can be selectively ab- lated by means of carefully tuned ultrasound while simultaneously leaving normal cells intact, an effect that they referred to as oncotripsy. Specifically, ONCOTRIPSY 3 by studying numerically the vibrational response of HCC and healthy cells, Heyden and Ortiz [1] found that, by carefully tuning the frequency of the harmonic excitation, lysis of the HCC nucleolus membrane could be induced selectively at no risk to healthy cells. They also estimated the acoustic den- sity required for oncotripsy to operate to be in the Low-Intensity Ultrasound (LIPUS) range. This low-intensity requirement sets oncotripsy apart from High-Intensity Focused Ultrasound (HIFU), which acts via thermal ablation and is non-specific, with no selectivity for cancer cells. The first numerical calculations of Heyden and Ortiz [1] neglected vis- coelasticity and damping in the cell and ECM. Under these conditions, the resonant response of the cells exhibits rapid linear growth in time and the cells are predicted to attain lysis relatively quickly. However, experimental studies suggest that the material behavior of the different cell constituents is viscoelastic [11, 12, 13, 14]. In a subsequent study [15], Heyden and Ortiz investigated the influence of viscoelasticity on the oncotripsy effect. They assumed Rayleigh damping and estimated the damping coefficients from dy- namic atomic force microscopy (AFM) experiments on live fibroblast cells in buffer solutions [16]. They concluded that, for these cells, the main effect of viscoelasticity is a modest reduction in the resonant natural frequencies of the cells and an equally modest increase of the time to lysis of the cancerous cells. On the basis of these results, they speculated that oncotripsy remains viable when viscoelasticity is taken into account. Following these leads, Mittelstein et al. [2] have endeavored to assess the oncotripsy effect in carefully designed laboratory tests involving a number of cancerous cell lines in aqueous suspension. They have developed a system for testing oncoptripsy that includes a tunable source of ultrasonic transduc- tion in signal communication with a control system that allows control of several parameters, including frequency and pulse duration, of the ultrasonic transduction. Transducers were selected to produce ultrasound pulses in the frequency range of approximately 100 kHz to 1 MHz, a pulse duration range of 1 ms to 1 s, acoustic intensity up to 5 W/cm2, and output pressure up to 2 MPa. The instrumentation of the system allows the measurement of esti- mated cell death rates as a function of frequency, pressure, pulse duration, duty cycle and number of cycles. In agreement with the original oncotripsy concept, the experiments con- firm that the application of low-intensity pulsed ultrasound (LIPUS) can indeed result in high death rates in the cancerous cell population selec- tively, i. e., simultaneously with small or zero death rates among healthy cells. The death and survival rates depend critically on the frequency of the ultrasound, indicative of a dynamical response of the cells. The oncotripsy effect is maximum at a certain frequency, and diminishes at both larger and smaller frequencies, also indicative of a resonant response of the cells. However, under the conditions of the experiments, cell death is observed to require the application of a much larger number of ultrasound cycles than anticipated by either [1] or [15], suggesting that the dynamics of cells in 4 E. F. SCHIBBER ET AL. aqueous suspension is much more heavily damped than estimated in [15] based on the AFM measurements of [16]. The observations reported by Mittelstein et al. [2] suggest that, under the conditions of the experiment, cell death occurs though a process of slow accumulation of damage over many cycles, instead of the rapid rupture of one of the cell membranes, as hypothesized in [1]. Figure 3. Live yellow fluorescent protein (YFP) tagged actin network staining of cells before and 5 min after exposure to 290 kPa acoustic pressure showing massive fiber disruption (repro- duced from [17]). Scale bar 10 µm. A number of experimental investigations suggest a mechanistic basis for the oncotripsy effect. The susceptibility of the cytoskeleton dynamics to therapeutic ultrasound, at strains of the order of 10−5 and frequencies in the MHz range, has been noted by [17]. At low acoustic intensities, no struc- tural network changes are observed over the duration of the experiments. By contrast, at sufficiently high acoustic intensities the actin network is progres- sively disrupted and disassembles within three minutes following exposure, Fig. 3. This disruption is accompanied by a 50% reduction in cell stiffness. Remarkably, after exposure to moderate acoustic intensities the stiffness of the cell gradually recovers and returns to its initial value. The mechanisms of actin stress-fiber repair have been extensively studied and are reasonably well-understood at present, cf., e. g., [18, 19] and references therein. By contrast, at high acoustic intensities no recovery takes place after cessation over the span of observation. To gain insight into the biomolecular mechanisms of LIPUS cytodisrup- tion, Mittelstein et al. [2] examined CT-26 cells after 2-minute LIPUS treat- ment at 500 kHz and focal pressure of 1.4 MPa. To evaluate the effect of LIPUS on the cytoskeleton, they plated CT-26 cells after LIPUS and per- formed confocal microscopy immediately after insonation. Confocal images show the actin cytoskeleton, stained with phalloidin-conjugated green dye, as a ring on the cell periphery, Fig. 4. This ring is disrupted and shows dimin- ished fluorescence for a 30 ms pulse duration, suggesting that cytodisruption ONCOTRIPSY 5 Figure 4. Confocal microscopy of CT-26 cells immediately after LIPUS treatment at 500kHz, focal pressure of 1.4MPa and pulse durations 0 ms (control), 1 ms and 30 ms (reproduced from [2]). Dead cells stained red with fixable LIVE/DEAD, actin cytoskele- ton stained green using phalloiding, and nucleus stained blue with DAPI. Confocal images shows disrupted actin cytoskeleton ring and significantly decreased actin stain intensity. Microscopy sug- gests LIPUS cytodisruption is coupled with persistent cytoskeletal disruption. is coupled with persistent cytoskeleton disruption. These observations are consistent with reports for other systems that LIPUS disrupts the cellu- lar cytoskeleton [20, 21]. In contrast, with 1 ms pulse duration, the actin cytoskeleton appears unchanged from the negative control. Mittelstein et al. [2] conclude that these observations suggest that LIPUS induces actin cytoskeletal disruption and activates apoptotic cell-death pathways. In the present work, we argue that these competing mechanisms of cy- toskeletal disruption and self-repair, when coupled to the -- possibly resonant -- dynamics of the cells over many insonation cycles, underlie the oncotripsy observations of Mittelstein et al. [2]. Based on this hypothesis, we develop a plausible theoretical model of oncotripsy that accounts for several of the key experimental observations of Mittelstein et al. [2], including the dependence of the cell death rates on frequency, pulsing characteristics and number of cycles. We posit that, under the conditions of the experiments, cells in sus- pension subjected to LIPUS act as frequency-dependent resonators and that the evolution of the cells is the result of competing mechanisms of high-cycle cumulative damage and healing of the cytoskeleton. We recall that struc- tural materials can fail at load levels well below their static strength through processes of slow incremental accumulation of damage when subjected to a Ki67+Ø1102030100Bcl2+Ø1102030100Calreticulin+Ø11020301000.00.51.0ApoptoticCellsØ11020301000.00.51.0CellDeathØ11020301000.00.51.0(a)NoUS1msPD30msPDDeadCellActinDAPI(b)ActinIntensityØ130012PD(ms)PD(ms)*******PD(ms)6 E. F. SCHIBBER ET AL. large number (millions) of loading cycles, a phenomenon known as mechan- ical fatigue [22]. Likewise, whereas one single LIPUS pulse is unlikely to cause significant cytoskeletal damage, we posit that over millions of cycles damage can accumulate to levels that render the cell unviable and cause it to die. By analogy to structural materials, we refer to the hypothesized necrosis mechanism as mechanical cell fatigue. We note that, whereas the elasticity, rheology and remodeling of the cy- toskeleton have been extensively studied in the past (cf., e. g., [23, 24, 25, 26] and references therein), no model of cumulative damage and mechanical cell fatigue appears to have been as yet proposed. The model proposed in this work follows as an application of cell dynamics, statistical mechanical the- ory of network elasticity and 'birth-death' kinetics to describe processes of damage and repair of the cytoskeleton. We also develop a reduced dynam- ical model that approximates the three-dimensional dynamics of the cell and facilitates parametric studies, including sensitivity analysis and process optimization. The reduced dynamical system encompasses the relative mo- tion of the nucleus with respect to the cell membrane and a state variable measuring the extent of damage to the cytoskeleton. The cell membrane is assumed to move rigidly according to the particle velocity induced in the water by the insonation. The dynamical system evolves in time as a re- sult of structural dynamics and kinetics of cytoskeletal damage and repair. The resulting dynamics is complex and exhibits behavior on multiple time scales, including the period of vibration and attenuation, the characteristic time of cytoskeletal healing, the pulsing period and the time of exposure to the ultrasound. We show that this multi-time scale response can effectively be accounted for by recourse to WKB asymptotics and methods of weak convergence [27]. We also account for cell variability and estimate the at- tendant variance of the time-to-death of a cell population using simple linear sensitivity analysis. The reduced dynamical model predicts, analytically up to quadratures, the response of a cell population to LIPUS as a function of fundamental cell properties and process parameters. We show, by way of partial validation, that the reduced dynamical model indeed predicts -- and provides a conceptual basis for understanding -- the oncotripsy effect and other trends in the data of Mittelstein et al. [2], including the dependence of cell-death curves on pulse duration and duty cycle. 2. Experimental basis We begin with a brief summary of the experimental system developed by Mittelstein et al. [2], as well as data and observations resulting from the study that are directly relevant to the present work. Their original publication maybe consulted for a complete account. 2.1. Experimental system. The experimental setup, Fig. 5a, was devel- oped to investigate the response of cells in aqueous suspension to ultrasound insonation [2]. Suspension cells are placed with a mylar film pocket that is ONCOTRIPSY 7 (a) (b) Figure 5. Experimental setup of Mittelstein et al. a) Schematic drawing of the LIPUS system and high frame-rate cam- era setup enabling cellular imaging at a frame rate of 5 MHz. b) Schematic of pulse duty cycle. [2]. submerged within a water bath. The cells within the pocket are thus in acoustic contact with the ultrasound transducer. The investigation indi- cated that the cell-disruption effect through low intensity pulsed ultrasound (LIPUS) requires the presence of spatial standing waves, which are generated by the reflection of the ultrasound wave off of an acrylic or metal acoustic reflector. Several hypotheses for the requirement of a standing wave are explored in [2]. The transducer in the water tank is positioned directly in- cident with the mylar pocket such that the acoustic axis is perpendicular with the optical axis which is illuminated by laser light. The mylar pocket is supported by a three-sided acrylic frame. One side of this frame serves as an acoustic reflector to form the standing waves. A water immersion pan-fluor objective is lowered into water bath and a series of prism mirrors and converging lenses deliver the image into a high-speed camera. Images are acquired 100 ms after the arrival time of the pulse to observe the effect of prolonged ultrasound exposure. The experiments aimed to isolate the mechanical effects of ultrasound by preventing local heating from taking place. In order to maintain low intensity ultrasound conditions (Ispta < 5 W/cm2), pulsed ultrasound was performed as shown in Fig. 5b. LIPUS was applied at a 10% duty cycle. However the pulsing parameters were varied in order to investigate their role on ultrasound cytodisruption. The pulse duration corresponds to the length of each pulse during which the ultrasound is on. By varying the pulse duration, while maintaining a constant duty cycle, the pulsing pattern of the ultrasound applied to the cells can be modified while maintaining constant acoustic energy deposited on the cells. To further investigate the effects of modifying ultrasound parameters on cytodisruption, three different transducers operating at 300 kHz, 500 kHz, and 670 kHz were used during Pulse durationListening timePulse repetition period8 E. F. SCHIBBER ET AL. this investigation. To provide consistent comparisons, they were configured to produce a peak negative pressure of 1.4 MPa at their focus in free water. (a) (b) Figure 6. a) Frames from video captured by Mittelstein et al. [2] and processed with Ncorr[28] (scale bar 10 microns). b) Measured velocity of K-562 cell under an incident plane wave of focal pressure amplitude P0 = 1.4 MPa and excita- tion frequency f0 = 670 kHz. 2.2. Cell motion. The recordings show that the entire field-of-view oscil- lates in the direction of ultrasound propagation with minimal observable cell membrane deformation, Fig. 6a. The damping out of cell membrane oscilla- tions is expected given the exceedingly low Reynolds number characteristic of the cell dynamics in aqueous suspension. Fig. 6b shows the measured tra- jectory of a K-562 cell upon insonation of focal pressure of P0 = 1.4 MPa, frequency f0=670 kHz and wavelength λ = 2.2 mm. As may be seen from the figure, the cell undergoes an ostensibly harmonic motion. The period of the motion is T = 1.4 µs, which corresponds to a frequency of f = 714 kHz. In addition, the amplitudes of the motion in the x- and y-directions are ux = 0.23 µm and uy = 0.022 µm, respectively, for a total displace- y = 0.231 µm and a velocity amplitude of ment amplitude of u = v = 2πf u = 1.037 m/s. By way of reference, the particle velocity amplitude of the medium is v0 = P0/ρ0c0 = 0.97 m/s, where ρ0 = 1000 Kg/m3 is the mass density of water and c0 = 1450 m/s is its sound speed. We thus conclude that, as expected for the long wavelength of the insonation relative to the cell size, the cells move ostensibly at the particle velocity of the fluid. u2 x + u2 (cid:113) 05101520-1.5-1.0-0.50.00.51.01.5Time(μs)Velocity(μm/μs)ONCOTRIPSY 9 Morphology Tissue Disease Source Cell Line K-562 Lymphoblast Lymphocyte U-937 Monocyte Lymphocyte T-Cells Lymphocyte Peripheral blood cells, isolated CD3+ mye- Chronic ologeneous leukemia Pleura/pleural effusion, lymphocyte, myeloid Human cell line Human cell line Human primary cells Table 1. Haematopoietic and lymphoid malignancies tumor cells used in the experiments of Mittelstein et al. [2] classified by morphology, type and disease. (a) (b) Figure 7. Tests of cancerous K562 and U937 cells and healthy CD4 T-cells at a PNP of 0.7MPa and a time of expo- sure of 60 seconds, showing the effect of frequency and pulse duration on cell death rates. In call cases, the pulse duration is 10% of the total pulse repetition period. (a) Cell death fraction vs pulse duration, and (b) cell death fraction at 20 ms pulse duration vs type. Reproduced from [2]. 2.3. Cell-death data. The experimental study of Mittelstein et al. [2] reveals that LIPUS conditions at specific frequencies and pulsing parameters can indeed achieve cell-selective cytodisruption. This capability to tune ultrasound parameters to cause selective disruption in cancer cells while sparing healthy cells appears to be a novel finding and fits with many of the predictions of the oncotripsy theory. The morphology, type and related disease for each cell line are listed in Table 1. 300 kHzPD (ms)Cell Death3300.00.51.0670 kHzPD (ms)4400.00.51.0K562U937T Cell10 E. F. SCHIBBER ET AL. (a) (b) Figure 8. Tests of cancerous cell K562 at a free field pres- sure of 0.7 MPa, pulse duration of 100 ms and duty cycle of 10%, showing the effect of frequency and number of cy- cles. (A) Cell death vs number of cycles (B) Cell death at 1.8 million cycles. Unpublished data from Mittelstein et al. [2]. Figure 7 demonstrates that cells can have varying responses to ultrasound depending on the ultrasound waveform. All data points in this figure rep- resent cell death assessed using LIVE/DEAD assays after exposure to an equal dosage of acoustic energy, though administered with different signal frequencies and pulse durations. These tests were all performed on cells in suspension for an exposure time of 60 seconds, a duty cycle of 10%, and in a spatial standing wave setup with a free field pressure of 0.7 MPa. Re- markably, high cell-death rates are observed for both the cancerous K-562 and U937 lines at 500 kHz signal frequency and 20 ms pulse duration while, under identical conditions, the control T-cells remain nearly unaffected (see Fig. 7b). These observations bear out the oncotripsy effect, as a frequency- dependent resonant response -- and eventual death -- of cells under harmonic excitation, and its selectivity. The data in Fig. 7 also shows a strong dependence of the cell response on pulse duration, with cell death enhanced at higher pulse durations. We take this dependence to suggest that the cell response is the result of two competing effects with vastly different characteristic times: damage accu- mulation during the on-part of the cycle and cell repair and healing during the entire time of exposure. The efficiency of the duty cycle may then be expected to depend sensitively on the relative values of the pulsing period and the characteristic times for damage accumulation and healing. Figure 8 shows data from tests of cancerous K562, showing the effect of frequency and the number of cycles. In all cases, the pulse duration is 10% of the total pulse repetition period, or a duty factor of 0.1. As may be seen from these figures, cell death does not occur instantly but requires a certain exposure time to occur. We take this observation to suggest that death occurs by a process of damage accumulation over many insonation 0.05.0×1061.0×1071.5×1072.0×1070.00.51.01.5Number of CyclesCell Death670 kHz500 kHz3005006700.00.20.40.60.8Frequency (kHz)Cell DeathK562ONCOTRIPSY 11 Figure 9. Schematic of the computational model, consisting of a random three-dimensional network of filaments spanning a rigid and heavy nucleus and a cell membrane oscillating rigidly with the surrounding fluid. Left: Electron micrograph of F-actin (adapted from [29]). cycles. It is also evident from the figures that some cells die relatively early whereas others require considerably large number of cycles to die. These observations are suggestive of a broad variability in the susceptibility of the cell population to LIPUS. 3. Oncotripsy model We proceed to develop a theoretical framework in which to understand and rationalize the preceding observations. The framework explored in this work is based on the following assumptions: i) For cells in suspension subjected to ultrasound, the aqueous medium damps out and suppresses the outer membrane vibrations, which translates rigidly at the particle velocity of the water. ii) The internal structures of the cell, including its nucleus, respond as a resonator and vibrate in sync with the applied ultrasound. iii) For sufficiently large pulse amplitudes, the cytoskeleton sustains cu- mulative mechanical damage that increases with successive cycles. iv) At all times during exposure to ultrasound, the cytoskeleton can repair itself at a rate proportional to the level of damage sustained. v) The cell ceases to be viable and dies when the amount of cumulative damage to the cytoskeleton exceeds a critical threshold. The various elements of the theory are next developed in turn. 3.1. Three-dimensional structure. In mammalian cells, the nucleus, as the largest cellular organelle, occupies about 10 % of the total cell volume [30, 31]. It is surrounded by the cytosol, a viscoelastic solid containing sev- eral subcellular structures such as the Golgi apparatus, the mitochondrion, and the endoplasmic reticulum. The cytosol and other organelles contained within the plasma membrane, for instance mitochondria and plastids, form 𝑎𝑎𝑏𝑏1𝜇𝜇𝜇𝜇12 E. F. SCHIBBER ET AL. the so-called cytoplasm. The nucleus is bounded by the nuclear envelope and contains the nucleoplasm, a viscoelastic solid similar in composition to the cytosol. It furthermore comprises the nucleolus, which constitutes the largest structure within the nucleus and consists of proteins and RNA. In the present work, we neglect the organelles within the cytosol, which is ide- alized as a uniform viscous matrix containing the cytoskeleton. The nucleus is likewise idealized as rigid and we omit explicit consideration of the nucle- oplasm. Given the focus on cytoskeletal dynamics, we additionally neglect the effect of the nuclear and cellular membranes. 3.2. Cytoskeleton elasticity. The cytoskeleton is a system of filaments in the cell that radiates from the nucleus and is anchored at the plasma membrane. In eukaryotic cells, the filament network has three major com- ponents: microtubules, intermediate filaments and microfilaments, Fig. 9a. Microfilaments are polymers of the protein actin, microtubules are composed of the protein tubulin and intermediate filaments are composed of various proteins, depending on the type of cell. The cytoskeleton confers elasticity to the cell, mediates the movement of the cells, helps to support the cy- toplasm and responds against external mechanical stimuli. In particular, microfilaments and intermediate filaments act as cables to support tension loads while microtubules act as beams in compression [32], in analogy to tensegrity structures [33, 34, 35, 36]. According to the network theory of elasticity in statistical mechanics [37, 38], the cytoskeleton may be modeled as an amorphous network of cross- linked fibers. The fibers consist of many freely-jointed segments and are far from full extension. It is further assumed that the cross-linking points move according to the local macroscopic deformation. In addition, the cytoskele- ton is assumed to be embedded in a viscous matrix. A standard analysis (cf., e. g., [37]) then gives the free-energy density per unit volume of the network as (1) A(F, T ) = µ(T ) 2 KIJ (CIJ + C−1 IJ ), up to inconsequential additive constants. In (1), µ(T ) is a temperature- dependent shear modulus, F is the local deformation gradient, C = F T F is the right Cauchy-Green deformation tensor and T is the absolute tem- perature (cf., e. g., [37, 39] for background on continuum mechanics). An analysis of the configurational entropy of the fibers [37, 38] gives the shear modulus as (2) µ(T ) = 2nl2 b2 kBT, where n is the number of fibers per unit volume, b is the segment length, l is the end-to-end distance of the fibers and kB is Bolzmann's constant. In ONCOTRIPSY 13 addition, the structure tensor K in (1) follows as (3) KIJ = S2 p(ξ)ξI ξJ dΩ, where ξ is the unit vector pointing from one end of the fiber to the other, or fiber direction, p(ξ) is the fraction of chains in the ensemble of direction ξ, S2 is the unit sphere and dΩ is the element of solid angle. The density p(ξ) is subject to the normalization condition (4) p(ξ) dΩ = 1. S2 The distribution function p(ξ) describes the structure of the cytoskeletal network and is assumed fixed and known. For instance, Smolyakov et al. [40] used single-cell force spectroscopy to test mechanical properties of four breast cancer 11 cell lines and found that the most invasive cells, MDA- MB231, contain actin fibers that are distributed randomly throughout the cell without any particular structure of preferred direction. For an isotropic fiber distribution of this type, p = 1/4π, and the structure tensor (3) re- duces to the identity. Under these conditions, the free-energy density (1) specializes to (5) 2 where tr denotes the matrix trace. A(F, T ) = µ(T ) (cid:0)tr(C) + tr(C−1)(cid:1), 3.3. Cytoskeletal damage and healing. The experimental observations of Mittelstein et al. [2] for cells in suspension, Section 2, reveal that cell death requires the application of a large number (millions) of insonation pulses, which in turn suggests that, under the conditions of the experi- ment, cell death is the result of a process of slow damage accumulation. Indeed, Mizrahi et al. [17] observed that, whereas the cytoskeletal actin fibers are catastrophically disrupted under the action of ultrasound stim- ulation of sufficiently high intensity, Fig. 3, under low-intensity ultrasound cellular responses exhibit gradual damage accumulation and sometimes com- plete recovery following insonation cessation. Confocal microscopy of CT-26 cells assessed after LIPUS treatment reported in [2] also reveals that LIPUS cytodisruption is coupled with persistent cytoskeletal disruption, cf. Fig. 4. Whereas cytoskeletal elasticity has been extensively studied in the past, processes of damage accumulation in the cytoskeleton under LIPUS actua- tion, or high-cycle cell fatigue, appear to be as yet poorly understood. Build- ing on past work on failure of polymer networks [41, 42, 43], we develop a model of cumulative cell damage that accounts for the gradual disrup- tion and repair of cytoskeletal fibers. This competition between disruption ('death') and repair ('birth') is a classical example of a 'birth-death' process in evolutionary dynamics, cf., e.g., [44]. We assume that the mechanism of damage accumulation to the cytoskele- ton is the progressive disruption of the actin fibers. In order to account for 14 E. F. SCHIBBER ET AL. the attendant loss of stiffness, we introduce a damage variable q(ξ) ranging from 0 to 1 such that q(ξ) = 0 when all the fibers with direction ξ are intact and q(ξ) = 1 when all the fibers with direction ξ are broken. We addition- ally assume that the breaking of the fibers requires a certain energy to be supplied. We represent these effects by means of a free-energy density of the form (6) A(F, T, q) = p(ξ) S2 where (7) (cid:16) µ(T ) (1−q(ξ))2(cid:0)λ2(ξ)+λ−2(ξ)−2(cid:1)+ λ(ξ) =(cid:112)CIJ ξI ξJ 2 β 2 (cid:17) q2(ξ) dΩ, is the stretch ratio of the fibers of direction ξ and β is a constant. We note from (6) that the effect of a damage field q(ξ) is to decrease the free-energy density of the fibers of direction ξ by a factor (1 − q(ξ))2 at an energy cost of (β/2)q2(ξ). Additionally, damage relaxes the stresses in the network by reducing the stiffness of the fibers. Evidently, in the absence of damage, q(ξ) = 0, (6) reduces to (1), as required. Following the method of Coleman and Noll [45], the thermodynamic driv- ing forces for damage follow as (8) f (ξ) = − ∂A ∂q(ξ) = p(ξ) (cid:16) µ(T )(1 − q(ξ))(cid:0)λ2(ξ) + λ−2(ξ) − 2(cid:1) − βq(ξ) (cid:17) . We see from this expression that, by the choice (6) of free-energy density, the driving force (8) comprises two terms. The first term represents the energy-release rate due to the disruption of the fibers and, therefore, pro- motes damage. The second term represents the energetic cost of disrupting the fibers, which hinders damage and promotes healing. Assuming linear kinetics, we obtain the damage evolution law (9) α q(ξ) = f (ξ), where α is a kinetic coefficient. The kinetic relation (9), in combination with the driving forces (8), define an evolution of the cytoskeletal state as a balance between 'birth' and 'death' processes. Thus, the energy-release term µ(T )(1−q(ξ))(cid:0)λ2(ξ)+λ−2(ξ)−2(cid:1) in proportionally to the energy µ(T )(cid:0)λ2(ξ) + λ−2(ξ) − 2(cid:1) of the fibers. The the driving force induces progressive damage ('death') of the fiber population additional factor (1 − q(ξ)) brings the driving force to zero at full damage q(ξ) = 1 and ensures that q(ξ) ≤ 1 at all times. By contrast, the energetic cost term −βq(ξ) in the driving force tends to restore ('birth') the fiber population and thus accounts for healing. Built into the form of (8) is the assumption that the rate of healing is proportional to the extent of damage. In particular, the healing rate vanishes for q(ξ) = 0, which it ensures that q(ξ) ≥ 0 at all times. ONCOTRIPSY 15 3.4. Cell viscosity. Another source of resistance to cell deformation arises from the viscosity of the cytoplasm. This viscosity damps resonant vibra- tions within the cell and limits their amplitude. On average, the cytoplasm viscosity does not differ significantly from that of water [46, 47], but the distribution of intracellular viscosity is highly heterogeneous. Full maps of subcellular viscosity have been successfully constructed via fluorescent ratio- metric detection and fluorescence lifetime imaging [48]. However, this degree of detail is beyond the scope of this study. Instead, we assume an average viscosity uniformly distributed over the cytoplasm. Further assuming linear viscosity, the viscous Cauchy stress in the cytoplasm follows as (cid:19) (cid:18) κ − 2 3 (10) σij = η(vi,j + vj,i) + η div v δij, where η is the shear viscosity, κ is the bulk viscosity, v is the velocity field, a comma denotes partial differentiation and div v is the divergence of the velocity field. 3.5. Reduced model. The preceding model of cytoplasm elasticity, dam- age, healing and viscosity can be taken as a basis for a fully three-dimensional analysis of cell motion, e. g., by means of the finite-element method, cf. [49]. However, parametric and sensitivity studies are greatly facilitated by re- duced models. We develop a reduced dynamical model of cell deformation and damage based on the following assumptions: i) Spherical geometry of cell and nucleus. ii) Rigid translational motion of the cell membrane. iii) Heavy and rigid nucleus. iv) Ansatze for the cytoplasm deformation and damage fields. We note that, under the conditions of interest here, a Rayleigh treatment of the acoustic scattering problem is justified in view of the large wavelength of the ultrasound waves compared to the cell size. We specifically consider a spherical cell of radius b containing a concentric spherical nucleus of radius a. We assume that the cell moves under the action of planar waves and executes a translational motion according to the particle velocity of the aqueous medium. We attach a moving cartesian reference frame to the center of the cell such that the x3 axis is aligned with the direction of motion. We additionally introduce a spherical coordinate system (r, ϕ, θ), such that (11) x1 = r sin θ cos ϕ, x2 = r sin θ sin ϕ, x3 = r cos θ, where r is the radius, ϕ is the azimuthal angle and θ the inclination. In these spherical coordinates, the domain of the cytoplasm in its undeformed configuration is ϕ ∈ [0, 2π), θ ∈ [0, π) and r ∈ [a, b]. The nucleus is assumed to translate rigidly through a time-dependent displacement u(t) relative to the cell membrane. In addition, a material point in the cytoplasm initially 16 E. F. SCHIBBER ET AL. (a) (b) Figure 10. Deformation ansatz used in model reduction. a) Cross section of the reference configuration of the cell, showing nucleus (inner circle) and two concentric material spheres to aid in the visualization of the deformation. b) Deformed configuration of the cell after a displacement of the nucleus. at location (x1, x2, x3) in the undeformed configuration is assumed to be at location (12) y1 = x1, y2 = x2, y3 = x3 + u(t), b − r b − a following the displacement of the nucleus. In this ansatz, a spherical material shell of radius r in the undeformed configuration translates rigidly to another spherical shell of the same radius centered at u(t) (b − r)/(b − a) following the displacement of the nucleus, cf. Fig. 10. 3.5.1. Dynamics without damage. Inserting this ansatz into the free-energy density (1) and assuming small relative displacements u(t), we obtain, after a trite calculation, (13) A = µ 2 u2(t) (b − a)2 (3 + cos(2θ)), and the total free energy of the cytoskeleton evaluates to (14) A(u(t)) = Ar2 sin θ dr dθ dϕ = 16π (b3 − a3)µ 2π π b 0 0 a which, in the absence of damage, supplies a potential for the relative dis- placement of the nucleus. Likewise, the velocity field of the cytoplasm follows by time differentiation of the ansatz (12), with the result u2(t) (b − a)2 , 9 b − r b − a (15) v1 = 0, v2 = 0, v3 = u(t). ONCOTRIPSY 17 Inserting this velocity field into the viscosity law (10) and assuming small relative displacements of the nucleus gives, after a straightforward calcula- tion, the dissipation per unit undeformed volume (cid:0)5η + 6κ − (η − 6κ) cos(2θ)(cid:1) u2(t) (b − a)2 , (16) D = 1 2 σijvi,j = 1 24 and the total dissipation follows as (17) D( u(t)) = 2π π Dr2 sin θ dr dθ dϕ = 0 0 (b3 − a3)(4η + 3κ) 2π 27 u2(t) (b − a)2 . (cid:17) Finally, the total kinetic energy of the cell follows as (18) K(t, u(t)) = 1 2 m0 + 2π 15 ρ(b − a)(6a2 + 3ab + b2) (v(t) + u(t))2, where m0 is the mass of the nucleus, ρ is the density of the cytoplasm and v(t) is the prescribed velocity of the cell membrane. An appeal to the Lagrange-D'Alembert principle gives the equation of motion (19) ∂A ∂u Inserting (14), (17) and (18) into (19), we obtain (t, u(t)) + ∂D ∂ u ∂K ∂ u ( u(t)) + d dt (u(t)) = 0. mu(t) + c u(t) + ku(t) = −m v(t), b a (cid:16) (20) where (21) m = m0+ 2π 15 ρ(b−a)(6a2+3ab+b2), c = b3 − a3 (b − a)2 (4η+3κ), 4π 27 b3 − a3 (b − a)2 µ, 32π 9 k = are the total mass, damping coefficient and stiffness of the cell. Eq. (20) de- scribes a damped and forced harmonic oscillator, with the material velocity v(t) of the aqueous medium supplying the forcing. 3.5.2. Dynamics with damage. Suppose now that the cell undergoes dam- age. In general, damage patterns may be expected to arise at two levels: inhomogeneously over the cytoplasm; and damage along preferential fiber directions at every material point. Such degree of complexity requires a full three-dimensional analysis for its elucidation, cf. [49]. In order to simplify the dynamics, we simply assume that damage is isotropic at all material points, i. e., the damage parameter q is independent of direction ξ; and independent of position over the cytoskeleton. By this simple ansatz, the state of damage of the cell is characterized by a single state variable q(t). An immediate extension of (14) then gives the total free energy of the cell as (22) A(u(t), q(t)) = (b3 − a3)(1− q(t))2µ (b3 − a3) q2(t). 16π u2(t) (b − a)2 + 4π 3 β 2 9 18 E. F. SCHIBBER ET AL. (a) (b) Figure 11. Axisymmetric finite-element results at 0.1 ms expo- sure using the full three-dimensional damage model [49]. The cell is insonated at 1.4MPa focal pressure and 500 kHz frequency. a) Magnitude of axial displacement in microns. b) Damage averaged over all fiber directions. Likewise, the total dissipation (17) extends to (23) D( u(t), q(t)) = (b3 − a3)(4η + 3κ) 2π 27 u2(t) (b − a)2 + 4π 3 (b3 − a3) α 2 q2(t). The Lagrange-D'Alembert principle then gives the coupled equations (t, u(t)) + ( u(t), q(t)) + (u(t), q(t)) = 0, ∂A ∂u ∂K ∂ u d dt ∂D ∂ q ∂D ∂ u ∂A ∂q ( u(t), q(t)) + (u(t), q(t)) = 0. (24a) (24b) (25a) (25b) Inserting (22), (23) and (18) into (24), we now obtain mu(t) + c u(t) + (1 − q(t))2ku(t) = −m v(t), n q(t) + dq(t) = (1 − q(t))ku2(t), with m, c and k as before and (26) n = 4π 3 (b3 − a3) α, d = 4π 3 (b3 − a3) β. The first of these equations represents a forced and damped harmonic oscil- lator in which the stiffness depends on the instantaneous state of damage. The second governs the kinetic evolution of the damage state, including damage accumulation and healing. (cid:882)(cid:484)(cid:890)(cid:882)(cid:484)(cid:886)(cid:882)u(µm)(cid:882)(cid:484)(cid:882)(cid:890)(cid:888)(cid:887)(cid:882)(cid:484)(cid:882)(cid:886)(cid:885)(cid:885)(cid:882)qONCOTRIPSY 19 The accuracy of the reduced model just derived can be assessed by means of comparisons with finite-element implementations of the full model. Fig. 11 shows a typical axisymmetric calculation in which a cell is insonated at 1.4MPa focal pressure and 500 kHz frequency over 0.1 ms [49]. As may be seen from the figure, the damage to the cytoskeleton is localized at poles of the cell. Despite this patterning, the nuclear displacements and average cy- toskeletal damage predicted by the reduced model are found to be within 7% of the full-field finite-element calculations. Given the level of observational error, this accuracy may reasonably be deemed adequate for all practical purposes. Further details of the error analysis may be found in [49]. 3.6. WKB dynamics. Under the conditions of interest here, the dynam- ics described by system (24) is characterized by two disparate time scales: the period of oscillation and the characteristic time for damage evolution, the former much smaller than the latter. This two-time structure suggests analyzing the problem by means of WKB asymptotics [27]. We consider a generic duty cycle such as shown inset in Fig. 5a, starting at time t0 and consisting of an on-period ending at time t1 and an off-period ending at time t2. The duration of the on-period, or pulse duration, is T1 = t1−t0, the duration of the off cycle, or listening time, is T2 = t2−t1 and the total duration of the duty cycle, or pulse repetition period, is T = t2−t0. We specifically assume harmonic excitation of the form (27) v(t) = V eiωt, during the on-period and v(t) = 0 during the off-period. In (27), V is a complex amplitude and ω is the insonation frequency. We begin by analyzing the equation of motion (25a), which we rewrite in the form (28) where ω0 =(cid:112)k/m is the natural frequency of the undamaged cell and ζ is u(t) + 2ζω0 u(t) + (1 − q(t))2ω2 0u(t) = − v(t), the damping ratio. During the on-period of the duty cycle, we have (29) u(t) + 2ζω0 u(t) + (1 − q(t))2ω2 0u(t) = −iωV eiωt, where, for convenience, we extend the equation to the complex domain. Assume now that the period of oscillation T0 = 2π/ω0 is much smaller than the pulse duration T1. Assume additionally that the frequency ω of insonation is comparable to ω0. Finally, suppose that the variation of the damage state variable q(t) is slow and on the scale of the pulse duration T1. Under these conditions, the solution u(t) can be obtained by performing a WKB asymptotic analysis in the small parameter T0/T1. We note that, for fixed q(t), eq. (29) is a linear second-order ordinary differential equation and, therefore, its solution is the sum of the general homogeneous solution and a particular solution. Owing to the presence of damping, with damping coefficient ζ of O(1), the homogeneous solution decays on the scale of T0 20 E. F. SCHIBBER ET AL. and can be safely neglected. We seek a particular equation of the form (30a) (30b) (30c) u(t) = A(t)eiωt, u(t) = ( A(t) + iωA(t))eiωt, u(t) =(cid:0) A(t) + 2iω A(t) − ω2A(t)(cid:1)eiωt. Inserting these expressions into (29) and retaining leading-order terms only, we obtain (31) − ω2A(t) + 2iζω0ωA(t) + (1 − q(t))2ω2 0A(t) = −iωV. Solving for the amplitude A(t), we find (32) A(t) = iωV ω2 − (1 − q(t))2ω2 0 − 2iζω0ω . Finally, inserting into (30a) we obtain iωV eiωt ω2 − (1 − q(t))2ω2 u(t) = (33) asymptotically as T0/T1 → 0. We observe from (33) that the nucleus exe- cutes rapid oscillations relative to the cell membrane over the pulse duration in sync with the ultrasound excitation, with amplitude modulated by the damage variable q(t). 0 − 2iζω0ω , Next, we turn to the damage evolution equation (25b). Inserting solution (33) into (25b) gives (34) n q(t) + dq(t) = (cid:0)ω2 − (1 − q(t))2ω2 k(1 − q(t))ω2V 2 (cid:1)2 + 4ζ 2ω2 0 , 0ω2 which is now fully expressed in terms of the damage variable q(t). Conve- niently, eq. (34) is separable and admits the explicit solution q (35) t = t0 + q0 n dξ (cid:0)ω2 − (1 − ξ)2ω2 k(1 − ξ)ω2V 2 (cid:1)2 + 4ζ 2ω2 0 0ω2 , − dξ where we write q0 = q(t0). Alternatively, the equation of evolution (34) can be recast in terms of dimensionless variables as dq dτ (τ ) + q(τ ) = (cid:0)w2 − (1 − q(τ ))2(cid:1)2 + 4ζ 2w2 (1 − q(t))w4ε , t − t0 tr , tr = n d , w = ω ω0 , ε = kV 2 dω2 0 mV 2 d , = (36) where (37) τ = whereupon (35) becomes (38) τ = q0 q dξ (cid:0)w2 − (1 − ξ)2(cid:1)2 + 4ζ 2w2 (1 − ξ)w4ε . − ξ ONCOTRIPSY 21 From this reparametrization, we observe that the evolution of damage de- pends on the following dimensionless parameters: i) The ratio of the elapsed time to the relaxation time tr for healing; ii) the ratio w between the fre- quency of insonation and the undamaged natural frequency; iii) the energy deposited by insonation relative to the energy cost of repair; and iv) the cell damping ratio. It is also interesting to note that the damage state variable attains a steady-state maximum value qmax when (1 − qmax)w4ε (39) qmax = (cid:0)w2 − (1 − qmax)2(cid:1)2 + 4ζ 2w2 (cid:0)w2 − (1 − qmax)2(cid:1)2 + 4ζ 2w2 , which expresses a balance between damage accumulation and healing. From this relation, the energy intensity required to attain a maximum level of damage qmax follows as w4 ε(qmax) = (40) As expected, ε(qmax) reduces to zero as qmax → 0 and diverges to infinity as qmax → 1. We also note that, by virtue of the existence of a steady state at qmax, the integral in (38) is well-defined and finite in the range q0 ≤ q < qmax and diverges to infinity at q = qmax, indicating that the steady state is attained only asymptotically at infinite time. qmax 1 − qmax . (a) (b) Figure 12. Example of cell response to harmonic excita- tion. a) Damage state variable vs. time. b) Relative nucleus displacement and amplitude vs. time. Parameters: tr = 1, ω = ω0 = 100, ζ = 1, qmax = 1/2. Fig. 12 shows an example of the WKB dynamics just elucidated for pa- rameters: tr = 1, ω = ω0 = 100, ζ = 1, qmax = 1/2. As may be seen from Fig. 12, the state of damage of the cell evolves on the scale of the relaxation time tr for healing and tends asympotically to qmax. The rela- tive displacement of the nucleus is damped out on the shorter time scale 0.00.10.20.30.40.50.60.00.10.20.30.40.5TimeDamage0.00.10.20.30.40.50.60.7-0.04-0.020.000.020.04TimeDisplacement22 E. F. SCHIBBER ET AL. 1/ζω0 and simultaneously amplified by the loss of stiffness due to damage on the time scale tr. The competition between these two opposing effects results in a well-defined steady-state amplitude, which follows from (31) by taking the limit of q(t) → qmax. Correspondingly, the phase-space tra- jectory (u(t), u(t)) converges to a stable limit cycle. The ability of WKB asymptotics to characterize the fast oscillations of the system and their slow modulation in time is remarkable. During the off period, the governing equations (25) reduce to (41a) (41b) mu(t) + c u(t) + (1 − q(t))2ku(t) = 0, n q(t) + dq(t) = 0. Again, we assume that the duration T2 of the off-period is much larger than the natural period of vibration T0. Under these assumptions, in off-period we have (42) u(t) = 0, q(t) = q1e−(t−t1)/tr , outside a short transient decaying on the scale of T0 immediately following t1. Thus, modulo short transients during the off-period the cell is quiescent and repairs itself exponentially on the time scale of tr. 3.7. Fractional-step approximation of high-cycle limit. Of special in- terest is the case in which the amount damage accumulated over each duty cycle is small. Thus, in the experiments of Mittelstein et al. [2] the death of a significant fraction of the population requires the application of a large number of duty cycles of insonation. Correspondingly, the number of in- sonation pulses required to cause cell death is large, i. e., T /tr (cid:28) 1. We proceed to obtain an effective equation describing the evolution of the sys- tem over larger numbers of duty cycles, or high-cycle limit. The effective equation follows by an appeal to the method of fractional steps [50]. We recall that the duty cycle under consideration consists of an on-period of scaled duration τ1 = T1/tr and an off-period of scaled duration τ2 = T2/tr. The entire scaled duration of the duty cycle is τ1 + τ2. Assuming τ1 (cid:28) 1, over a single on-period (36) gives (43) q1 ≈ q0 + τ1 (cid:32) (cid:0)w2 − (1 − q0)2(cid:1)2 + 4ζ 2w2 (1 − q0)w4ε (cid:33) − q0 . (cid:33) Likewise, with τ2 (cid:28) 1 over the subsequent off-period (41b) gives (44) q2 ≈ (1 − τ2)q1. Compounding the preceding relations and keeping terms of first order in τ1 and τ2 gives (45) q2 ≈ q0 + τ1 (cid:0)w2 − (1 − q0)2(cid:1)2 + 4ζ 2w2 (1 − q0)w4ε − q0 − τ2q0. (cid:32) (cid:32) Rearranging terms gives the relation q2 − q0 τ1 + τ2 ≈ λ (cid:0)w2 − (1 − q0)2(cid:1)2 + 4ζ 2w2 (1 − q0)w4ε (46) where (47) ONCOTRIPSY 23 (cid:33) − q0 − (1 − λ)q0, λ = τ1 τ1 + τ2 , 1 − λ = τ2 τ1 + τ2 , are the on-time fraction of the duty cycle, or duty factor, and the off-time fraction, respectively. Formally passing to the limit in (46) gives the differ- ential equation (48) dq dτ (τ ) + q(τ ) = (cid:0)w2 − (1 − q(τ ))2(cid:1)2 + 4ζ 2w2 λ(1 − q(τ ))w4ε , which approximates slow damage evolution over larger numbers of duty cycles, or high-cycle limit. Again, the differential equation (48) is separable with solution (49) τ = 0 , − ξ q dξ λ(1 − ξ)w4ε (cid:0)w2 − (1 − ξ)2(cid:1)2 + 4ζ 2w2 (cid:0)w2 − (1 − qmax)2(cid:1)2 + 4ζ 2w2 (cid:0)w2 − (1 − qmax)2(cid:1)2 + 4ζ 2w2 λ(1 − qmax)w4ε , which is explicit up to a quadrature. As in the case of steady insonation, we note that the system attains a steady state at a maximum level of damage (50) qmax = at which point damage accumulation and healing balance each other. The energy intensity required to attain a maximum level of damage qmax follows as qmax 1 − qmax ε(qmax, λ) = (51) As expected, ε(qmax, λ) reduces to zero as λ → 0 and reduces to (40) for λ = 1. We also note that the integral in (49) is well-defined and finite in the range q0 ≤ q < qmax and diverges to infinity at q = qmax, indicating that the steady state is attained only asymptotically. λw4 . The convergence of the damage evolution to the high-cycle limit as the pulse repetition period T becomes much smaller than the characteristic time tr for healing is illustrated in Fig. 13, which corresponds to the choice of parameters: tr = 10, λ = 1/10, ω = ω0 = 100, ζ = 1/10, qmax = 1/2. Figs. 13a-c show the evolution of the damage state variable obtained by solving directly the WKB eq. (36) and eq. (41b) for T = 1, 1/10 and 1/100, respectively. As expected, damage accumulates during the off-period and otherwise relaxes at all times, resulting in a characteristic saw-tooth profile. Fig. 13d shows the corresponding evolution of the damage state variable predicted by the effective fractional-step eq. (48). Evidently, the high-cycle 24 E. F. SCHIBBER ET AL. (a) (b) (c) (d) Figure 13. Convergence of the damage evolution to the high-cycle limit as pulse repetition period T becomes much smaller than the characteristic time tr for healing, cf. Fig. 5a (inset). Parameters: tr = 10, λ = 1/10, ω = ω0 = 100, ζ = 1/10, qmax = 1/2. a) T = 1. b) T = 1/10. c) T = 1/100. d) Damage evolution predicted by the high-cycle limit equa- tion (48). limiting curve is smooth and represents a weak limit of the damage evolution curves as the number of duty cycles tends to infinity, respectively, the pulse duration cycle tends to zero. 3.8. Cell death. We recall that the state variable q(t) measures the amount of damage sustained by a cell at time t. A plausible assumption is that a cell becomes unviable and dies when q(t) attains a critical value qc. In light of our previous discussion, this condition cannot be met if qmax ≤ qc, i. e., if the maximum accumulated damage induced by insonation is less that the critical value. Conversely, it follows from (51), that cell death requires a 0.00.20.40.60.81.00.000.050.100.150.200.25TimeDamage0.00.20.40.60.81.00.000.050.100.15TimeDamage0.00.20.40.60.81.00.000.020.040.060.080.100.12TimeDamage0.00.20.40.60.81.00.000.020.040.060.080.100.12TimeDamageONCOTRIPSY 25 minimum level of energy deposition (52) ε ≥ ε(qc, λ). If this condition is met, then in the high-cycle limit the time-to-death of a cell follows from (49) as (53) τc = 0 qc dξ (cid:0)w2 − (1 − ξ)2(cid:1)2 + 4ζ 2w2 λ(1 − ξ)w4ε , − ξ otherwise τc = +∞ and the cell survives for all time. The corresponding number of insonation pulses is (54) Nc = n d τc T , were T is the total pulse duration. As noted in the introduction, this type of system failure by slow damage accumulation over many cycles is observed in other systems, notably inert structural materials, in which context it is known as high-cycle mechanical fatigue [22]. The number of loading cycles to failure is correspondingly known as the fatigue life of the material. In this analogy, cell death by slow damage accumulation over many cycles may be thought of as a form of mechanical cell fatigue, and the number of cycles Nc to death as the fatigue life of the cell. 3.9. Variability within a cell population. A typical population of can- cerous cells exhibits broad variation in geometry and mechanical properties. This variability is strongly suggested by the cell-death curves observed by Mittelstein et al. [2], which show that some cells die much earlier than oth- ers. In order to capture this gradual cell necrosis, we regard the parameters governing the evolution of the cells as random and a cell population as a sam- ple drawn from the probability distribution of the parameters. By virtue of the variability of the sample, parts of the population have a relatively short time-to-death and die early, whereas other parts have a comparatively longer time-to-death and die later, resulting in the gradual estimated cell death curves observed experimentally, Fig. 8. The statistics of the time-to-death can be estimated simply by means of a linear sensitivity analysis, cf., e. g., [51]. We see from (53) that the time- to-death tc = τctr depends on the cell parameters (tr, ω0, ζ, qc), respectively, the relaxation time for healing, the natural frequency of vibration and the damping ratio; and on the process parameters (ε, ω, λ), respectively, the energy intensity, frequency and on-period fraction of the insonation. For simplicity, we assume that the process parameters can be controlled exactly and are uncertainty-free. Contrariwise, the cell parameters define a mul- tivariate random variable X ≡ (tr, ω0, ζ, qc), with probability distribution reflecting the variability of the cell population. 26 E. F. SCHIBBER ET AL. Owing to the randomness of the cell population, the time-to-death tc itself defines a random variable Y . In terms of these random variables, (53) defines a relation of the form (55) Y = f (X). In order to estimate the variability in the time-to-death random variable Y , we make a small-deviation approximation Y ≈ f ( ¯X) + Df ( ¯X)(X − ¯X) + h.o.t., (57) is the mean value of the cell parameters and Df ( ¯X) are sensitivity param- eters. The average time-to-death then follows as ¯X = E(X) ≡ (¯tr, ¯ω0, ¯ζ, ¯qc) ¯Y = E(Y ) ≈ f ( ¯X) + h.o.t. (56) where (58) (60) In addition, a measure of the variability of Y is given by the variance (59) Y = E((Y − ¯Y )2) = Df ( ¯X)T E((X− ¯X)⊗(X− ¯X))Df ( ¯X) = Df ( ¯X)T ΣDf ( ¯X), σ2 where Σ = E((X − ¯X) ⊗ (X − ¯X)) is the covariance matrix of the cell parameters. We note that, for small deviations, the mean time-to-death of the cell pop- ulation is obtained by evaluating (53) at the mean value ¯X = (¯tr, ¯ω0, ¯ζ, ¯qc) of the cell parameters, cf. eq. (58), with the result (61) ¯tc = ¯tr 0 ¯qc dξ (cid:0) ¯w2 − (1 − ξ)2(cid:1)2 + 4¯ζ 2 ¯w2 λ(1 − ξ) ¯w4ε , − ξ where we write ¯w = ω/¯ω0 and we assume that (52) is satisfied with qc = ¯qc. Likewise, the requisite sensitivity parameters Df ( ¯X) follow by differenti- ating (58) with respect to the cell parameters and evaluating the resulting integrals at their mean value. Simple forms of the probability distribution of tc are fully determined by Y . For instance, if we hypothesize a gamma distribu- the statistics ¯Y and σ2 tion (62) p(Y ) = 1 Γ(k)θk Y k−1e−Y /θ, then the parameters of the distribution follow as σ2 Y = kθ2. ¯Y = kθ, (63) The fraction of the cell population with a time-to-death less or equal to t is given by the cumulative distribution function F (t) = P (Y ≤ t). (64) ONCOTRIPSY 27 (a) (b) Figure 14. Dead-cell fraction vs. time curves obtained from Y = 1, ¯tc = 1/2, 1, 2, 4 and 8. the Gamma-distribution. a) σ2 b) ¯tc = 1, σ2 Y = 1, 1/2, 1/4, 1/8 and 1/16. For the gamma distribution (62), we have (65) F (t) = 1 − Γ (k, t/θ) Γ(k) , where Γ is the gamma function. The resulting dead-cell fraction vs. time curves are illustrated in Fig. 14. 500 kHz ¯Y (sec) σY (sec) 670 kHz ¯Y (sec) σY (sec) 30.5 46.6 49.4 71.36 Table 2. Mean and standard deviation obtained by fitting to cell-death time data [2] for cell line K-562 at focal pres- sure 1.4 MPa, pulse duration 100ms, 10% duty cycle at two insonation frequencies 500 kHz and 670 kHz. By way of example, Fig. 15 shows a least-squares fit of the cell-death time data of [2] using the function F (t) obtained from the Γ distribution, eq. (65). The data corresponds to the cell line K-562 at focal pressure 1.4 MPa, pulse duration 100 ms and 10% duty cycle. The mean and standard deviation derived from the fit are listed in Table 2. As may be seen from the figure, the Γ distribution provides an adequate fit of the data. 4. Comparison with experiment We proceed to assess the ability of the proposed dynamical model to account for the experimentally observed trends summarized in Section 2. 02468100.00.20.40.60.81.0TimeDead-cellfraction012340.00.20.40.60.81.0TimeDead-cellfraction28 E. F. SCHIBBER ET AL. Figure 15. Γ-distribution fit of cell-death time data [2] for cell line K-562 at focal pressure 1.4 MPa, pulse duration 100ms, 10% duty cycle at two insonation frequencies 500 kHz and 670 kHz. . (a) (b) Figure 16. Damage accumulation rate as a function of in- sonation frequency. Parameters: ω0 = 1, ε = 1, λ = 1, tr = 1, ζ = 1/10, 2/10, 3/10, 4/10, 5/10. a) Pristine cell, q = 0. b) Damaged cell, q = 1/10. 4.1. Qualitative comparison. We note that the experimentally observed dead-cell fraction vs. time curves exhibit the sigmoidal form predicted by the proposed dynamical model, cf. Figs. 8 and 14, which can be used to fit the experimental curves. More importantly, the model explains the observed dead-cell fraction curves as a result of cell-to-cell variability, specifically, the random distribution of times-to-death in the cell population. Furthermore, ▲▲▲500kHz▲670kHz01020304050600.00.20.40.60.81.0Time(sec)Dead-cellfraction0.60.81.01.21.41.61.82.00510152025FrequencyDamagerate0.60.81.01.21.41.61.82.0051015FrequencyDamagerateONCOTRIPSY 29 the time-to-death of an individual cell is predicted by the model explicitly as a function of cell parameters (tr, ω0, ζ, qc) and process parameters (ε, ω, λ), e. g., through eq. (53) in the high-cycle limit. Owing to the variability of the cell population, the cell parameters may be assumed to be random and, by an appeal to linear sensitivity analysis, the mean and variance of the cell time- to-death can be related to the mean values and covariance matrix of the cell parameters, eqs. (53) and (59). Thus, if the statistics of the cell parameters is known, the time-to-death statistics and, correspondingly, the dead-cell fraction curves, are given explicitly by the model. In this manner, the model relates the observed dead-cell fraction curves to fundamental mechanical properties of the cell such as mass, stiffness, viscosity and damage tolerance. The dynamical model also predicts the dependence of the dead-cell frac- tion curves on pulse duration observed experimentally, Fig. 7. Indeed, this trend is exhibited by the damage evolution curves shown in Fig. 13. A careful inspection of these curves shows that the maximum level of damage attained within the insonation cycles decreases as the pulse duration de- creases relative to the characteristic time for healing. Thus, for long pulses the cells have time to accumulate large amounts of damage during the on- period of the pulse. For shorter pulses, the extent of damage accumulation is comparatively less. If the pulse duration is comparable to -- or smaller than -- the relaxation time for healing, the cell does not have sufficient time to recover during the off-period of the cycle, and the trend persists over repeated cycles. Therefore, according to the model the dependence of the dead-cell fraction curves on pulse duration is the result of a delicate interplay between the pulse repetition period and pulse duration, the cell dynamics, which determines the rate at which damage accumulates and the kinetics of cell healing, which determines the rate at which damage is restored. The dynamical model also exhibits the oncotripsy effect, i. e., the insonation- frequency dependence of the cell response and the window of opportunity for selective cell ablation. Fig. 16 shows the damage accumulation rate q computed from (48) as a function of insonation frequency, damping ratio and state of damage. The parameters used in the figure are: ω0 = 1, ε = 1, λ = 1, tr = 1, ζ = 1/10, 2/10, 3/10, 4/10, 5/10, q = 0, 1/10. As may be seen from the figure, the damage rate peaks sharply in the vicinity of the undamped resonant frequency ω = ω0. The damage accumulation rate is largest for a pristine cell, q = 0, and persists, albeit somewhat reduced, after the cell sustains damage, q = 1/10. This frequency dependence is clearly apparent in the experimental data, Fig. 8b. 4.2. Quantitative comparison. A quantitative comparison between the predicted cell death times and experimental data provides a measure of val- idation of the model. We recall that the death time tr of a cell characterized by parameters X ≡ (tr, ω0, ζ, qc) is given analytically by (53). We regard X as a multivariate random variable with a certain probability distribution reflecting the variability of the cell population. Owing to this variability, 30 E. F. SCHIBBER ET AL. the time-to-death tc itself defines a random variable Y , in terms of which (53) is to be regarded as a response function of the form (55). In order to exercise the linearized sensitivity framework formulated in Section 3.9, we need to know the average values ¯X of the parameters X for a given cell population and their covariance matrix Σ. In lieu of direct characterization, we estimate these statistics as follows. We begin by as- suming that the parameters X are independent and log-normal distributed with unknown mean ¯X and diagonal covariance matrix Σ. From this distri- bution, we generate a random sample {Xi, i = 1, . . . , N} of size N = 1000, compute the corresponding cell-death times {ti, i = 1, . . . , N} using (53) and evaluate the fraction of the cell population with a time-to-death less or equal to t as, cf. eq. (64), (66) F (t) = 1 N #{ti ≤ t, i = 1, . . . , N}, where # is the counting measure. The statistics ¯X and Σ are then obtained by means of a least-square fit to the data. The results are listed in Table 3. tr (sec) ω0 (rads/sec) ζ 0.7 0.136 Mean 0.175 0.0136 Standard deviation Table 3. Estimated mean, standard deviation and sensitiv- ities of cell parameters. qc 100 10 3142 393 Figure 17. Comparison of predicted cell-death fraction with experimental data from [2] for a focal pressure of 1.4MPa, pulse duration 100ms, duty cycle 10% and frequen- cies 500kHz and 670kHz. The experimental data is repre- sented through the Γ-distribution fit shown in Fig. 15. ExperimentsMonteCarloSensitivity01020304050600.00.20.40.60.81.0Time(sec)Dead-cellfractionExperimentsMonteCarloSensitivity01020304050600.00.20.40.60.81.0Time(sec)Dead-cellfractionONCOTRIPSY 31 Finally, we are in a position to compare predicted cell-death curves with the experimental data. Fig. 17 shows computed cell-death curves for log- normal independent cell population parameters, with mean values and stan- dard deviations as in Table 3, together with experimental data from [2]. The predicted curves are computed directly via Monte Carlo based on a sample of size N = 1000 and by means of the linearized-sensitivity approx- imation. As may be seen from the figure, the linearized-sensitivity curve closely approximates the Monte Carlo curve, which establishes the validity of the linearized-sensitivity approximation under the conditions of the ex- periments. In addition, both the linearized-sensitivity and the Monte Carlo curves match closely the experimental data, which provides a measure of validation of the model. 5. Discussion and concluding remarks The proposed dynamical model provides a rational basis for understand- ing the oncotripsy effect posited by Heyden and Ortiz [1] under the condi- tions of the experiments of Mittelstein et al. [2]. An important difference between those experiments and the scenario initially contemplated in [1] is that in the experiments of Mittelstein et al. [2] the cells are in aqueous suspension, whereas the analysis of Heyden and Ortiz [1] is concerned with cells embedded in a solid extracellular matrix (ECM). In aqueous suspen- sion, the cells experience an exceedingly viscous environment, which is likely to suppress any vibrations of the cell membrane. The response of the cells to ultrasound stimulation is thus reduced to that of an internal resonator. Heyden and Ortiz [1] pointed out that the spectral gap between cancerous and healthy cells depends sensitively on the mechanical properties of the ECM and that the changes in those properties experienced by the cancer- ous tissue are a key contributing factor to the opening of a spectral gap. In addition, for cells embedded in an ECM, membrane rupture provides an additional lysis mechanism which is absent in cells in suspension. These con- siderations suggest the need for an independent experimental assessment of the oncotripsy effect in cancerous tissues, preferrably in vivo. The proposed dynamical model also reveals the dependence of oncotripsy on fundamental cell parameters and on process parameters. The cell pa- rameters of the model can be calibrated from cell-death data for specific cell lines. Alternatively, fundamental cell properties such as stiffness and viscosity can be measured independently. The calibrated model can then be used as a tool for optimizing process parameters for maximum therapeutic effect. Most importantly, theoretical understanding such as provided by the proposed dynamical model is key for interpreting experimental observations and formulating new and improved clinical therapies. In this regard, a number of possible therapies suggest themselves as possi- ble clinical applications of oncotripsy. Thus, due to genomic instability and 32 E. F. SCHIBBER ET AL. being in different states within the cell cycle, cancer cells are highly hetero- geneous at any given moment. As such, it is unlikely that an entire cancer cell population can be killed by a single set of acoustic parameters. This suggests exploiting oncotripsy in connection with other synergistic cancer therapies such as immunogenic cell death (ICD). In this combination, on- cotripsy does not need to kill every last cancer cell to be effective, as long as it can induce ICD of sufficient cancer cells to trigger the host immune sys- tem to kill remaining cancer cells (abscopal effect). Again, these and other fundamental questions suggest worthwhile directions for further research. Acknowledgements The support of the California Institute of Technology through the Rothen- berg Innovation Initiative and through the Caltech -- City of Hope Biomedical Research Initiative is gratefully acknowledged. References [1] S Heyden and M Ortiz. Oncotripsy: Targeting cancer cells selectively via resonant harmonic excitation. Journal of the Mechanics and Physics of Solids, 92:164 -- 175, 2016. [2] D. R. Mittelstein, J. Ye, E. F. Schibber, A. Roychoudhury, L. T. Martinez, M. H. Fekrazad, M. Ortiz, P. P. Lee, M. G. Shapiro, and M. Gharib. Selective ab- lation of cancer cells with low intensity pulsed ultrasound. bioRxiv, 2019. doi: https://doi.org/10.1101/779124. [3] J. J. Berman. Precancer: The Beginning and the End of Cancer. Jones & Bartlett Publishers, London, United Kingdom, 1st edition, 2011. [4] S. E. Cross, Y.-S. Jin, J. Rao, and J. K. Gimzewski. Nanomechanical analysis of cells from cancer patients. Nature Nanotechnology, 2:780 -- 783, 2007. [5] V. Swaminathan, K. Mythreye, E. T. O'Brien, A. Berchuck, G. C. Blobe, and R. Su- perfine. Mechanical stiffness grades metastatic potential in patient tumor cells and in cancer cell lines. Cancer Research, 71(15):5075 -- 5080, 2011. [6] J. Schrader, T. T. Gordon-Walker, R. L. Aucott, M. van Deemter, A. Quaas, S. Walsh, D. Benten, S. J. Forbes, R. G. Wells, and J. P. Iredale. Matrix stiffness modulates proliferation, chemotherapeutic response, and dormancy in hepatocellular carcinoma cells. Hepatology, 53(4):1192 -- 1205, 2011. [7] K. R. Levental, H. Yu, L. Kass, J. N. Lakins, M. Egeblad, and J. T. Erler. Matrix crosslinking forces tumor progression by enhancing integrin signaling. Cell, 139:891 -- 906, 2009. [8] S. Suresh. Biomechanics and biophysics of cancer cells. Acta Biomaterialia, 3(4):413 -- 438, 2007. [9] J. Guck, S. Schinkinger, B. Lincoln, F. Wottawah, S. Ebert, M. Romeyke, D. Lenz, H. M. Erickson, R. Ananthakrishnan, D. Mitchell, J. Ka, S. Ulvick, and C. Bilby. Op- tical Deformability as an Inherent Cell Marker for Testing Malignant Transformation and Metastatic Competence. Biophysics Journal, 88(5):3689 -- 3698, May 2005. [10] R. Donida Labati, V. Piuri, and F. Scotti. ALL-IDB : The Acute Lymphoblastic Leukemia Image Database for Image Processing. IEEE International Conference on Image Processing (ICIP), pages 2045 -- 2048, 2011. [11] P. Panorchan, J. S. H. Lee, T. P. Kole, Y. Tseng, and D. Wirtz. Microrheology and rock signaling of human endothelial cells embedded in a 3d matrix. Biophysical Journal, 91:3499 -- 3507, 2006. ONCOTRIPSY 33 [12] F. Guilak, J. R. Tedrow, and R. Burgkart. Viscoelastic properties of the cell nucleus. Biochemical and Biophysical Research Communications, 269:781 -- 786, 2000. [13] G. Zhang, M. Long, Z.-Z. Wu, and W.-Q. Yu. Mechanical properties of hepatocellular carcinoma cells. World Journal of Gastroenterology, 8(2):243 -- 246, 2002. [14] P. A. Janmey. The cytoskeleton and cell signaling: component localization and me- chanical coupling. Physiol Rev, 78:763 -- 781, 1998. [15] S Heyden and M Ortiz. Investigation of the influence of viscoelasticity on oncotripsy. Computer Methods in Applied Mechanics and Engineering, 314, 09 2016. [16] A. Cartagena and A. Raman. Local viscoelastic properties of live cells investigated us- ing dynamic and quasi-static atomic force microscopy methods. Biophysical Journal, 106:1033 -- 1043, 2014. [17] N. Mizrahi, E. H. Zhou, G. Lenorman, R. Krishnan, D. Weihs, J. P. Butler, D. A. Weitz, and E. Fredberg, J. J.and Kimmel. Low intensity ultrasound perturbs cy- toskeleton dynamics. Soft Matter, 8(8):2438 -- 2443, February 2012. [18] M.A. Smith1, L. M. Hoffman, and M. C. Beckerle. Lim proteins in actin cytoskeleton mechanoresponse. Trends in Cell Biologyl, 24(10):575 -- 583, October 2014. [19] M. Nakamura, Dominguez A. N. M., J. R. Decker, A. J. Hull, J. M. Verboon, and S. M. Parkhurst. Into the breach: how cells cope with wounds. Open Biolology, page 180135, 2018. http://dx.doi.org/10.1098/rsob.180135. [20] S. Noriega, G. Hasanova, and A. Subramanian. The effect of ultrasound stimulation on the cytoskeletal organization of chondrocytes seeded in three-dimensional matrices. Cells Tissues Organs, 197(1):14 -- 26, 2013. [21] M. Samandari, K. Abrinia, M. Mokhtari-Dizaji, and A. Tamayol. Ultrasound induced strain cytoskeleton rearrangement: An experimental and simulation study. Journal of Biomechanics, 60:39 -- 47, 2017. [22] S. Suresh. Fatigue of Materials. Cambridge University Press, Cambridge, Mass., 1998. [23] C T Lim, E H Zhou, and S T Quek. Mechanical models for living cells -- a review. J Biomech, 39(2):195 -- 216, 2006. [24] M. R. K. Mofrad. Rheology of the cytoskeleton. Annual Review of Fluid Mechanics, 41:433 -- 453, 2009. [25] Daniel A Fletcher and R Dyche Mullins. Cell mechanics and the cytoskeleton. Nature, 463(7280):485 -- 492, 2010. [26] M. H. Jensen, E. J. Morris, and D. A. Weitz. Mechanics and dynamics of reconstituted cytoskeletal systems. Biochim Biophys Acta, 1853(11 0 0):3038 -- 3042, November 2015. [27] C. M. Bender and S. A. Orszag. Advanced Mathematical Methods for Scientists and Engineers. International series in pure and applied mathematics. McGraw-Hill, 1978. [28] J. Blaber, B. Adair, and A. Antoniou. Ncorr: Open-Source 2D Digital Image Corre- lation Matlab Software. Experimental Mechanics, 55(6):1105 -- 1122, 2015. [29] M.L. Gardel, K. Kasza, C. Brangwynne, J. Liu, and D. A. Weitz. Mechanical Response of Cytoskeletal Networks, volume 89 of Methods in Cell Biology, chapter 19, pages 487 -- 519. Elsevier Inc., 2008. [30] B. Alberts, A. Johnson, J. Lewis, M. Raff, K. Roberts, and P. Walter. Molecular Biology of the Cell. Garland Science, New York, 4th edition, 2002. [31] H. Lodish, A. Berk, P. Matsudaira, C. A. Kaiser, M. Krieger, M. P. Scott, S. L. Zipursky, and J. Darnell. Molecular Cell Biology. WH Freeman, New York, 5th edi- tion, 2004. [32] Shaofan Li and Bohua Sun. Advances in Cell Mechanics. Springer, 2011. [33] D E Ingber. Tensegrity II. How structural networks influence cellular information processing networks. Journal of Cell Science, 116(8):1397 -- 1408, 2003. [34] Dieter Kardas, Udo Nackenhorst, and Daniel Balzani. Computational model for the cell-mechanical response of the osteocyte cytoskeleton based on self-stabilizing tenseg- rity structures. Biomechanics and Modeling in Mechanobiology, 12(1):167 -- 183, 2013. 34 E. F. SCHIBBER ET AL. [35] Sara Barreto, Casper H Clausen, Cecile M Perrault, Daniel A Fletcher, and Damien Lacroix. A multi-structural single cell model of force-induced interactions of cytoskele- tal components. Biomaterials, 34(26):6119 -- 6126, 2013. [36] J Li, M Dao, C T Lim, and Subra Suresh. Spectrin-level modeling of the cytoskeleton and optical tweezers stretching of the erythrocyte. Biophysical Journal, 88(5):3707 -- 3719, 2005. [37] J. H. Weiner. Statistical Mechanics of Elasticity. Dover Publications, Mineola, N. Y., 2nd edition, 2002. [38] P. J. Flory. Statistical Mechanics of Chain Molecules. Hanser Publishers, Munich, 1989. [39] J. E. Marsden and T. J. R. Hughes. Mathematical Foundations of Elasticity. Dover Civil and Mechanical Engineering Series. Dover, 1994. [40] G. Smolyakov, B. Thiebot, Campillo., S. Labdi, C. Severac, J. Pelta, and E. Dague. Elasticity, Adhesion, and Tether Extrusion on Breast Cancer Cells Provide a Signa- ture of Their Invasive Potential. ACS Applied Materials and Interfaces, 8(41):27426 -- 27431, 2016. [41] D. Balzani and M. Ortiz. Relaxed incremental variational formulation for damage at large strains with application to fiber-reinforced materials and materials with truss- like microstructures. Journal for Numerical Methods in Engineering, 92(6):551 -- 570, 2012. [42] S. Heyden, S. Conti, and M. Ortiz. A nonlocal model of fracture by crazing in poly- mers. Mechanics of Materials, 90:131 -- 139, 2015. [43] S. Heyden, B. Li, K. Weinberg, S. Conti, and M. Ortiz. A micromechanical damage and fracture model for polymers based on fractional strain-gradient elasticity. Journal of the Mechanics and Physis of Solids, 74:175 -- 195, 2015. [44] M. A. Nowak. Evolutionary dynamics: exploring the equations of life. Belknap Press of Harvard University Press, Cambridge, Mass., 2006. [45] B. Coleman and W. Noll. The thermodynamics of elastic materials with heat conduc- tion and viscosity. Archive for Rational Mechanics and Analysis, 13:167 -- 179, 1963. [46] K. Fushimi and A. S. Verkman. Domain of cell cytoplasm measured by picosecond polarization microfluorimetry. Journal of Cell Biology, 112:719 -- 725, 1991. [47] K. Luby-Phelps, S. Mujumdar, R. B. Mujumdar, L. A. Ernst, W. Galbraith, and A. S. Waggoner. A novel fluorescence ratiometric method confirms the low solvent viscosity of the cytoplasm. Byophisical Journal, 65:236 -- 242, 1993. [48] T. Liu, X. Liu, D. R. Spring, X. Xuhong Qian, J. Cui, and Z. Xu. Quantitatively mapping cellular viscosity with detailed organelle information via a designed pet fluorescent probe. Scientific Reports, 4:5418, 2014. [49] E. F. Schibber. High-Cycle Dynamic Cell Fatigue with Applications on Oncotripsy. PhD thesis, California Institute of Technology, Pasadena, California, USA, 2019. [50] N. N. Yanenko. The method of fractional steps: Solution of problems of mathematical physics in several variables. Springer, 1971. [51] T. J. Sullivan. Introduction to Uncertainty Quantification, volume 63 of Texts in Applied Mathematics. Springer, 2015. 1Division of Engineering and Applied Science, California Institute of Tech- nology, 1200 East California Boulevard, Pasadena, CA 91125, USA., 2 Depart- ment of Immuno-Oncology, City of Hope National Medical Center, 1500 E Duarte Rd, Duarte, CA 91010, USA E-mail address: [email protected]
1212.3112	1	1212	2012-12-13T10:16:32	Geometric interpretation of phyllotaxis transition	[ "physics.bio-ph", "q-bio.TO" ]	The original problem of phyllotaxis was focused on the regular arrangements of leaves on mature stems represented by common fractions such as 1/2, 1/3, 2/5, 3/8, 5/13, etc. The phyllotaxis fraction is not fixed for each plant but it may undergo stepwise transitions during ontogeny, despite contrasting observation that the arrangement of leaf primordia at shoot apical meristems changes continuously. No explanation has been given so far for the mechanism of the phyllotaxis transition, excepting suggestion resorting to genetic programs operating at some specific stages. Here it is pointed out that varying length of the leaf trace acts as an important factor to control the transition by analyzing Larson's diagram of the procambial system of young cottonwood plants. The transition is interpreted as a necessary consequence of geometric constraints that the leaf traces cannot be fitted into a fractional pattern unless their length is shorter than the denominator times the internode.	physics.bio-ph	physics	Geometric interpretation of phyllotaxis transition Faculty of Engineering, Shizuoka University, 3-5-1 Johoku, Hamamatsu 432-8561, Japan Takuya Okabe 2 1 0 2 c e D 3 1 ] h p - o i b . s c i s y h p [ 1 v 2 1 1 3 . 2 1 2 1 : v i X r a Abstract The original problem of phyllotaxis was focused on the regular arrangements of leaves on mature stems represented by common fractions such as 1/2, 1/3, 2/5, 3/8, 5/13, etc. The phyllotaxis fraction is not fixed for each plant but it may undergo stepwise transitions during ontogeny, despite contrasting observation that the arrangement of leaf primordia at shoot apical meristems changes continuously. No explanation has been given so far for the mechanism of the phyllotaxis transition, excepting suggestion resorting to genetic programs operating at some specific stages. Here it is pointed out that varying length of the leaf trace acts as an important factor to control the transition by analyzing Larson's diagram of the procambial system of young cottonwood plants. The transition is interpreted as a necessary consequence of geometric constraints that the leaf traces cannot be fitted into a fractional pattern unless their length is shorter than the denominator times the internode. Keywords: Schimper-Braun's law, Populus deltoides, Fibonacci numbers, golden ratio, phyllotaxy 1. INTRODUCTION The spiral arrangement of leaves on a stem, phyllotaxis, is represented by the fraction of the circumference of the stem traversed by the spiral in passing from one leaf to the next. Braun and Schimper discovered that leaves are lined up in ranks parallel to the stem so that the fraction is lit- erally represented by a common fraction (Braun (1835)). In a 3/8 phyllotaxis, for instance, every eighth leaf comes over one below it after three turns of the spiral, so that eight straight ranks are visible along the stem (Fig. 1). To this day, the following list of the most common frac- tions has been circulated in books and websites for non- specialists; 1/2 for elm, lime and linden, 1/3 for beech and hazel, 2/5 for oak, cherry, apple, holly and plum, 3/8 for poplar, rose and pear, 5/13 for almond, etc. Such corre- spondence tables seem to have existed already in the mid- dle of the nineteenth century (Henfrey (1870)). Sometimes willow is listed in 5/13 (Coxeter (1961)) and sometimes in 3/8 (Adam (2006)). As a matter of fact, it had been remarked since early times that even an individual plant sometimes makes transitions between different fractions (Braun (1835)). Therefore, the phyllotaxis fraction is not a determined characteristic of each species. Most notably, Larson (1980) has revealed the manner in which the vascu- lar system is rearranged through the phyllotaxis transition by mapping arrangement of the leaf traces, the portion of vascular bundles of leaves that resides in the stem (Fig. 2). By contrast, it has been commonly accepted that the ar- rangement of leaf primordia in the bud or at the shoot api- cal meristem does not conform to any fractional number; Email address: [email protected] (Takuya Okabe) primordia do not appear as radial rows. The angular diver- gence between successive primordia stays close to a unique "ideal" angle of 137.5◦, which is the golden mean 0.3820 of 360◦ (Church (1920); Richards (1951)). The golden mean is a mathematical limit of the sequence of the above frac- tions. Much attention has been paid to mathematics of these numbers and to mechanisms of leaf primordia for- mation by which the ideal angle is achieved and regulated (Adler et al. (1997)). In contrast, the phyllotaxis transi- tion has been left unexplained without attracting interest from researchers. The fractional phyllotaxis and the phyl- lotaxis transition are two sides of the same problem. A key point noted in the present study is that phyllotaxis in the bud changes continuously whereas phyllotaxis on the ma- ture stem changes stepwise. Evidently, the arrangement of primordia at the apical growing point must be a determin- ing factor of leaf arrangement on the mature stem. But what causes the transition from 2/5 to 3/8, for instance? There must be another factor. Larson (1980) has sug- gested that the transition is programmed in the plant to occur at specific stages of ontogeny. In another view, phyl- lotaxis of primordium formation and vascularization are controlled by some higher-level system (Romberger et al. (1993)). Kuhlemeier (2007) remarks that virtually nothing is known about the molecular mechanisms that underlie the transitions between different spiral systems (e.g. 3/8 to 5/13 patterns), except that larger meristems seem to have higher Fibonacci numbers. The present paper aims at pointing out that this phenomenon is consistently ex- plained by considering geometry of growing leaf traces. By means of a full quantitative analysis, which surely is not standard in this field of research, it is shown (1) that the positions of the phyllotaxis transition are located based Figure 2: Diagram of the procambial system of a typical cottonwood plant compiled by Larson (1980). The vascular cylinder is displayed as if unrolled and laid flat. The ordinate is the Leaf Plastochron Index (LPI) for leaves, whereas the abscissa corresponds to the an- gular coordinate in a full turn about the stem axis. Each leaf is entered by three traces; a central (×), right (N), and left (△) traces. Phyllotaxis orders, 1/2, 1/3, 2/5, 3/8 and 5/13, are indicated by the right vertical axis. By virtue of the convention to take LPI as the ordi- nate, the vertical scale of the diagram is normalized such that differences in height between two successive leaves, or internodes, are a unit of length in the vertical direc- tion. Therefore, the vertical component of a line segment in Fig. 2 represents not its actual length but an effec- tive length relative to the internode length, namely the length measured in internode units. Accordingly, vertical lengths of the leaf traces in internode units are directly evaluated by applying a digitizing ruler to the diagram. The lower end points of right and left traces are located without ambiguity, for they are connected to other types of traces. Lengths of central traces are fixed by decom- posing the whole pattern into clusters consisting of three adjacent traces, the right trace of leaf n, the central trace of leaf n + 2 and the left trace of leaf n − 1, where n for LPI is an integer. For illustration purposes, the clusters from n = 0 to 7 are shown in Fig. 5. The length of the leaf traces thus obtained is plotted against LPI in Fig. 3, where reading errors are of the order of an internode at most. On the other hand, divergence angles are evaluated from the horizontal, angular coordinates of the exit points of the leaf traces denoted by the symbols. Evaluated angles show rhythmic, systematic variations around the "ideal" angle, which are typically observed in a quantitative anal- ysis (Okabe (2012a)). The systematic variations, which are correlated with the angular positions, are suppressed apparently by deleting from the diagram a blank rectan- gular strip of a narrow width along the left ordinate. The width of the strip is determined so as to minimize the standard deviation of divergence angles for ten youngest central traces (LPI less than −6). Results for the diver- gence angle thus corrected are shown in Fig. 4. The width Figure 1: A young poplar in a 3/8 phyllotaxis with eight vertical ranks of leaves. on the angular positions at which leaf traces exit from the vascular cylinder and (2) that the phyllotaxis transition is caused as a necessary consequence of change in size of leaf traces relative to internode length. A caveat: this paper deals with the mechanism of the phyllotaxis transition in spiral systems. Although a com- mon keyword of "phyllotaxis" may suggest, it should not be confused with mechanisms of phyllotactic primordium formation, for which considerable advances have been made over the last decade (Kuhlemeier (2007)). Models of the latter category deal with continuous changes of apical meris- tems, while they are not concerned with the fractional ex- pression of a phyllotactic pattern. 2. MATERIAL AND METHODS The analysis is based on a diagram of the procambial system of a cottonwood plant (Populus deltoides) recon- structed by Larson (1980), which is reproduced in Fig. 2. The vascular cylinder is displayed as if unrolled and laid flat. The ordinate is the leaf plastochron index (LPI) for numbering leaves. Each leaf has three traces: central, right and left traces exit the vascular cylinder at positions de- noted by symbols ×, △ and N, respectively. For more details, see Larson (1980) and references cited therein. 2 ) s e d o n r e t n i ( e c a r t f a e l f o h t g n e L 13 12 11 10 9 8 7 6 5 4 3 2 1 0 2/5 Order of phyllotaxis 3/8 5/13 12 8 4 0 -4 -8 -12 -16 Leaf Plastochron Index Figure 3: Vertical length of leaf traces in Fig. 2 is plotted against the Leaf Plastochron Index. Orders of phyllotaxis, 2/5, 3/8 and 5/13, are denoted at the same position as indicated in Fig. 2. See Fig. 2 for trace designations. The phyllotaxis transitions from 2/5 to 3/8 and from 3/8 to 5/13 are caused by the traces indicated by a solid arrow and a dashed arrow, respectively. Horizontal dashed lines at 5 and 8, separating the different phyllotaxis regimes, are drawn for reference sake (see Results and Discussion). Figure 4: Divergence angle between the traces at LPI n and n + 1 is plotted against LPI n. See Fig. 2 for trace designations. The inset shows the angle between the central traces at LPI n and n + 5: for ideal patterns of 2/5, 3/8 and 5/13 orders, it should be zero, 45 and 28 degrees, respectively. The positions at which the phyllotaxis transitions from 2/5 to 3/8 and from 3/8 to 5/13 occur are indicated by solid and dashed arrows, respectively. ratio of the deleted strip is 0.023, whereby the standard deviation of the divergence angle is suppressed from 4.2◦ to 0.89◦. The correction is made for ease of understand- ing implications of Fig. 4. It does not affect the results discussed below qualitatively. 3. RESULTS Results in Fig. 3 indicate that the three traces grow in length steadily all alike. This behavior is consistent with other quantities reported by Larson (1980). According to Fig. 3, three traces at LPI 2 have almost the same length of about five internodes. This means that the traces at LPI 2 extend down to the height of LPI 7 (see Fig. 2). As indi- cated by the right ordinate of Fig. 2, the phyllotaxis order fraction changes from 2/5 through 3/8 to 5/13 as we climb up the stem, or as LPI decreases. The order fractions are indicated at the top of Fig. 3 at the same LPI coordinates as in Fig. 2 by Larson (1980). Horizontal dashed lines at five and eight internodes in Fig. 3 are drawn to separate regimes of different phyllotaxis orders (see below). Divergence angle between the leaf traces at LPI n and n+1 is plotted against LPI n in Fig. 4. The inset shows the angle between the central traces at LPI n and n + 5, that is, the net angle of inclination of 5-parastichies. The angle should become zero, 45 and 28 degrees for 2/5, 3/8 and 5/13 ideal patterns, respectively; for instance, the ideal an- gle for 3/8 is 360×3/8×5 = 360×2−45, which is congruent to −45. The inset of Fig. 4 clearly indicates stepwise tran- sitions between the three distinct fraction regimes. Thus, the positions of the phyllotaxis transition are located based on the exit points, or the bases, of the leaf traces, i.e., with- out inspecting internal changes in the vascular structure. This crucial property for us is brought to light probably because the subject plants are grown under controlled uni- form conditions (Larson (1980). The cause of the transition is traced back by close in- spection based on the quantitative results. The transition from 3/8 to 5/13, indicated by a dashed arrow in the in- set of Fig. 4, is caused by an irregular decrease of diver- gence angle at LPI 2, which therefore is also indicated by a dashed arrow in the main figure. Similarly, a solid arrow in the inset indicates the transition from 2/5 to 3/8, which is ascribed to an increase of divergence angle at LPI 7, a solid arrow in the main figure. In connection with the latter transition, it is worth a remark that divergence angle in the 2/5 phyllotaxis regime is not held at an ideal constant value of 144◦ (360 × 2/5 = 144). This means that five ranks (orthostichies) of a real 2/5 pattern is not equally spaced. According to Fig. 4, two full turns (360◦ × 2) of the 2/5 pattern is divided roughly into unequal parts of 140◦ × 4 + 160◦ × 1, instead of a regular spacing with 144◦ × 5. The figure shows that the irregular shift at LPI 7 is shared with LPI 12. This is just as expected for the 2/5 arrangement (7 + 5 = 12). Similarly, a cycle of 360◦×3 of a 3/8 pattern is divided into 137◦×7+120◦×1, as indicated by the dashed arrow. Thus, the quantitative analysis reveals that divergence angle on a stem is a secondary property, as it is very unlikely that the exceptional angles of 160◦ and 120◦ are intrinsic to the plant. Accordingly, the pattern of Fig. 2 is to be viewed as a result of secondary processes. The final step is to identify the cause of the irregular shift in divergence angle. For brevity, let the central (×), 3 0R 3R 1R 6R 5R 7C Figure 5: A schematic excerpt from Fig. 2 near the transition from 2/5 to 3/8. The right trace at LPI 5 is denoted as 5R, and 7C signifies the central trace at LPI 7. According to Fig. 4, the traces 5R and 7C are shifted slightly to the right as compared to the preceding (lower) traces. This figure shows that the shift (solid arrow) is caused by a longer trace 0R intervening between 3R and 5R, thereby the transition is initiated. If the length of 0R were shorter than five internodes, 0R should have been aligned with 5R to keep the 2/5 order, as the preceding 1R is with 6R. right (N) and left (△) trace of LPI n be denoted as nC, nR and nL, respectively. At the transition from 2/5 to 3/8, the irregular shift of 7C (solid arrow) is accompanied by 5R on the right side (Fig. 4). Inspection of Fig. 2 reveals that this collective shift is caused as a result of 0R intervening between 3R and 5R. This is illustrated in Fig. 5, a schematic excerpt from Fig. 2. For this reason, the trace 0R is indicated by a solid arrow in Fig. 3 as the very cause of the phyllotaxis transition from 2/5 to 3/8. A horizontal line at five is drawn in Fig. 3 to indicate that 0R interferes with 5R if only the former length exceeds 5 − 0 = 5 internodes (cf. Fig. 5). Indeed, the filled triangle at the solid arrow in Fig. 3 lies well above the horizontal line at five internodes. Similarly, the trace causing the transition from 3/8 to 5/13 is indicated by a dashed arrow in Fig. 3, where the upper horizontal line at eight is drawn as a threshold length for the transition. 4. DISCUSSION The trace length represented in internode units is an important geometric factor as it imposes constraints on possible fractional patterns to be realized (Okabe (2011, 2012b)). The geometric effect invoked in interpreting the above results is schematically illustrated in Fig. 6. During growth of the shoot apex, the precursor of the vascular sys- tem associated with a leaf primordium develops to become the leaf trace (Romberger et al. (1993)). Both lengths of the leaf trace and internode change continuously during growth. In this view, the phyllotaxis order to be realized on the mature stem is determined depending on the trace 4 Figure 6: Distinct patterns of 2/5 and 3/8 orders on the mature stem cylinders (bottom) result from similar arrangements of leaf primordia at the shoot apical meristems (top). Each dot represents a leaf node, through which a dotted line is drawn to demarcate internodes. Leaf traces (line segments) are aligned in a 2/5 phyllotaxis pattern if their length is greater than three and less than five internodes (left), whereas a 3/8 phyllotaxis results if it is greater than five and less than eight internodes (right). length represented in internode units. Top two patterns in Fig. 6 represent the initial arrangements of the most common case of 137.5◦ angular divergence, whereas their growth rates in the radial direction are different, i.e., the two patterns are characterized with different plastochron ratios (Richards (1951)). At this point, the difference is not a qualitative but a quantitative one. Leaf traces (line segments) in the left pattern traverse about four intern- odes, while those in the right pattern traverse about six internodes. As the figure shows, a qualitative difference is brought about through the quantitative difference in the trace length: geometric constraints tend to achieve either a 2/5 or 3/8 phyllotaxis on the mature stem depending on whether the trace length is shorter or longer than five in- ternodes. Experimentally, the difference would be judged on whether leaf trace 0 is identified or not in the cross section at the level of node 5. In a similar manner, the threshold value of the trace length for the transition from 3/8 to 5/13 order is eight internodes, the denominator of the lower order fraction. These are indicated by horizon- tal dashed lines in Fig. 3, as already remarked. In terms of the threshold length, the phyllotaxis transition is inter- preted consistently without resort to operations of elab- orate mechanisms like genetic programs; the phyllotaxis transitions are caused because changing length of the leaf trace happens to cross the threshold values of five and eight internodes. As illustrated in Fig. 6, this view provides a consistent explanation of empirical observation that large meristems result in arrangements of higher order fractions. Unequal distribution of divergence angle noted in Fig. 4 is Henfrey, A., 1870. An elementary course of botany: structural, phys- iological and systematic. J. Van Voorst. Kuhlemeier, C., 2007. Phyllotaxis. TRENDS in Plant Science 12, 143 -- 150. Larson, P. R., 1980. Interrelations between phyllotaxis, leaf devel- opment and the primary-secondary vascular transition in Populus deltoides. Annals of Botany 46, 757 -- 769. Okabe, T., 2011. Physical phenomenology of phyllotaxis. Journal of Theoretical Biology 280, 63 -- 75. Okabe, T., 2012a. Systematic variations in divergence angle. Journal of Theoretical Biology 313, 20 -- 41. Okabe, T., 2012b. Vascular phyllotaxis transition and an evolution- ary mechanism of phyllotaxis. http://arxiv.org/abs/1207.2838 . its quantitative expression and relation to growth in the apex. Philosophical Transactions of the Royal Society of London. Series B 225, 509 -- 564. Richards, F. J., 1951. Phyllotaxis: Romberger, J., Hejnowicz, Z., Hill, J., 1993. Plant structure: func- tion and development : a treatise on anatomy and vegetative development, with special reference to woody plants. Springer- Verlag. circumstantial evidence of a unique intrinsic angle close to 137.5◦ and secondary distortions therefrom. Fibonacci numbers 5 and 8 enter as the threshold val- ues because leaf 0 appears close to leaves 5 and 8 (Fig. 6). This in turn is a mathematical consequence of the golden- mean angle at the apex. Mathematically, any number can be approximated by a common fraction with any as- signed degree of accuracy. The greater the denominator, the better the approximation. A fast-converging sequence of approximate fractions is uniquely determined for a given number (Hardy and Wright (1979)). For the golden mean 0.3820, it is 1/2, 1/3, 2/5, 3/8, 5/13, etc., the main se- quence of phyllotaxis. The denominators of these approx- imate fractions comprise the index differences of nearby leaves, namely the threshold values for the trace length. Thus, the geometric interpretation predicts a correlation between the fraction index of the phyllotaxis order and the trace length represented in internode units. Indeed, it has been remarked as a general rule; the higher phyllotaxis order is associated with the longer leaf trace (Girolami (1953); Esau (1965)). To conclude, the plant in its ma- turity, as it were, achieves rational approximations to a divergence angle at the apex in conformity with the leaf- trace length in internode units. Mathematically, the golden mean is the worst "approx- imable" real number in the sense that the sequence of the approximate fractions converges most badly. In this con- nection, it has been commonly mentioned, despite objec- tions, that the golden-mean divergence is advantageous be- cause leaves are distributed most evenly to sunlight. When viewed in the context of this study, the golden-mean diver- gence distributes the leaf traces most efficiently to coordi- nate the vascular system. Although "phyllotaxis" is the arrangement of leaves on a stem according to dictionaries, the geometric view on the leaf-trace organization, the ar- rangement of leaves in a stem, may shed a new light on the long-standing problem of phyllotaxis in vascular plants. References Adam, J., 2006. Mathematics in Nature: Modeling Patterns in the Natural World. Princeton University Press. Adler, I., Barab´e, D., Jean, R. V., 1997. A history of the study of phyllotaxis. Annals of Botany 80, 231 -- 244. Braun, A., 1835. Dr. Carl Schimper's Vortrage uber die Moglichkeit eines wissenschaftlichen Verstandnisses der Blattstellung, nebst Andeutung der hauptsachlichen Blattstellungsgesetze und Ins- besondere der Neuentdeckten Gesetze der Aneinanderreihung von Cyclen Verschiedene. Flora 18, 145 -- 191. Church, A. H., 1920. On the interpretation of phenomena of phyl- lotaxis. Botanical memoirs. Hafner Pub. Co. Coxeter, H. S. M., 1961. Introduction to Geometry. Wiley, New York and London. Esau, K., 1965. Vascular differentiation in plants. New York: Holt, Rinehart and Winston. Girolami, G., 1953. Relation between phyllotaxis and primary vas- cular organization in linum. American Journal of Botany 40, 618 -- 625. Hardy, G., Wright, E., 1979. An Introduction to the Theory of Num- bers. Oxford Science Publications. Clarendon Press. 5
1606.00683	1	1606	2016-06-02T14:10:17	Long-range interactions and phase defects in chains of fluid-coupled oscillators	[ "physics.bio-ph", "cond-mat.soft", "math.DS", "physics.flu-dyn" ]	Eukaryotic cilia and flagella are chemo-mechanical oscillators capable of generating long-range coordinated motions known as metachronal waves. Pair synchronization is a fundamental requirement for these collective dynamics, but it is generally not sufficient for collective phase-locking, chiefly due to the effect of long-range interactions. Here we explore experimentally and numerically a minimal model for a ciliated surface; hydrodynamically coupled oscillators rotating above a no-slip plane. Increasing their distance from the wall profoundly effects the global dynamics, due to variations in hydrodynamic interaction range. The array undergoes a transition from a traveling wave to either a steady chevron pattern or one punctuated by periodic phase defects. Within the transition between these regimes the system displays behavior reminiscent of chimera states.	physics.bio-ph	physics	Long-range interactions and phase defects in chains of fluid-coupled oscillators Douglas R. Brumley1,2, Nicolas Bruot3,4, Jurij Kotar4, Raymond E. Goldstein5, Pietro Cicuta4 and Marco Polin6∗ 1Ralph M. Parsons Laboratory, Department of Civil and Environmental Engineering, 2Department of Civil, Environmental and Geomatic Engineering, ETH Zurich, 8093 Zurich, Switzerland Massachusetts Institute of Technology, Cambridge, MA 02139, USA 3Institute of Industrial Science, University of Tokyo, 4-6-1 Komaba, Meguro-ku, Tokyo 153-8505, Japan 4Cavendish Laboratory, University of Cambridge, Cambridge CB3 0HE, UK 5Department of Applied Mathematics and Theoretical Physics, University of Cambridge, Centre for Mathematical Sciences, Wilberforce Road, Cambridge CB3 0WA, UK 6Physics Department, University of Warwick, Gibbet Hill Road, Coventry CV4 7AL, UK. Eukaryotic cilia and flagella are chemo-mechanical oscillators capable of generating long-range co- ordinated motions known as metachronal waves. Pair synchronization is a fundamental requirement for these collective dynamics, but it is generally not sufficient for collective phase-locking, chiefly due to the effect of long-range interactions. Here we explore experimentally and numerically a minimal model for a ciliated surface; hydrodynamically coupled oscillators rotating above a no-slip plane. Increasing their distance from the wall profoundly effects the global dynamics, due to variations in hydrodynamic interaction range. The array undergoes a transition from a traveling wave to either a steady chevron pattern or one punctuated by periodic phase defects. Within the transition between these regimes the system displays behavior reminiscent of chimera states. The ability of ensembles of oscillators to achieve collec- tive motions is fundamental in biological processes rang- ing from the initiation of heartbeats to the motility of microorganisms. The emergent properties of coupled os- cillators can vary dramatically depending on the intrinsic properties of the oscillators and the nature of the cou- pling between them [1]. Flashing fireflies equally and in- stantaneously coupled to one another [2] can support very different behaviors to chemical micro-oscillators which are coupled only locally, and subject to time delays [3]. Eukaryotic cilia and flagella are chemo-mechanical os- cillators which generate a variety of collective motions that can be quantified with high-speed microscopy in mi- crofluidic environments [4–6]. The molecular biology of these internally driven filaments is virtually identical in green algae [5], protists [7] and humans [8], and the flows they generate fulfill crucial roles in development, motil- ity, sensing and transport. When close together, the mu- tual interaction between their oscillatory flow fields can cause them to beat in synchrony [9], and larger ensembles of flagella demonstrate striking collective motions in the form of metachronal waves (MWs) [10–12], akin to the 'Mexican wave' propagating around a packed stadium. Many surrogate models for flagellar dynamics have been proposed [12–22], typically with a set geometry which fixes the range and coupling between oscillators. Here we relax this condition and study a linear ar- ray of colloidal oscillators [23] driven in circular trajec- tories at a controllable height above a no-slip wall. The oscillator couplings can be modified continuously from being primarily through nearest neighbors to a regime involving significant long-range interactions. As a func- tion of rotor properties, a traveling wave found at small heights becomes either a chevron pattern or is punctu- ated by phase defects at large ones. The transition is not a gradual morphing between the two profiles, but rather a process involving generation and propagation of defects along the strip, where phase-locked and non-phase-locked subgroups of oscillators can coexist. A behavior arising from long-range interactions whose amplitude is modu- lated by the distance from the wall [16], these dynam- ics are reminiscent of chimera states, in which oscilla- tors split into phase-locked and desynchronized clusters [24, 25]. In our experiments, silica colloids of radius a = 1.74 µm (BangsLab, USA) suspended in a water-glycerol solution of viscosity µ = 6 mPas within a 150 µm-thick sample, are captured and driven by feedback-controlled time-shared (20 kHz) optical tweezers (OTs) based on acousto-optical deflection of a 1064 nm-wavelength diode-pumped solid-state laser (CrystaLaser IRCL-2W- 1064) as previously described [26, 27]. The OTs de- scribe a planar array of circular trajectories (Fig. 1a) of radius R = 1.59 µm and center-to-center separation (cid:96) = 9.19 µm, a distance h above the sample bottom, with 4.2(4) µm ≤ h ≤ 51.7(4) µm. This configuration, which reflects the capabilities and limitations of our OT setup, is similar to arrays of nodal cilia, but differs from the more common situation where the ciliary beating plane is perpendicular to an organism's surface. The OTs are arranged so that a colloidal particle on the ith trajec- tory (i ∈ {0, . . . , N − 1}) experiences a radial harmonic potential with spring constant λ = 2.06 ± 0.06 pN/µm resisting excursions from the prescribed radius, and a constant tangential force of magnitude Fi = FdrDi−0.5 leading to rotation. Before each experiment we cali- brate Fdr (cid:39) 2.23 pN (see [28]; typical variation ±2%). D (cid:54)= 1 is used to break left-right symmetry along the chain and induce a stable traveling wave for small h [11]. For the detuning adopted here, D = 1.01, the period 6 1 0 2 n u J 2 ] h p - o i b . s c i s y h p [ 1 v 3 8 6 0 0 . 6 0 6 1 : v i X r a 2 slower and faster respectively [12, 13]. The timescale for synchronization depends on the spring constant λ and the strength of hydrodynamic interactions between ro- tors, which is itself a function of height h and spacing (cid:96). The dynamics become richer if a discrepancy between the rotor's intrinsic frequencies is introduced (D (cid:54)= 1), for then the coupling must be sufficiently strong to over- come the natural tendency for the rotor's phase difference χ = φ1 − φ0 to drift. Bifurcation plots in Fig. 1b show, for different h, the average phase drift between two oscillators as a func- tion of D. The behavior is typical of a saddle-node bi- furcation: the oscillators phase-lock until D reaches a critical value D∗(h) and then drift with a monotonically increasing speed. D∗(h) increases with h, reflecting the strengthening of inter-rotor hydrodynamic coupling with increasing distance from the wall. The phase-locking be- havior is summarized in Fig. 1c, where the experimental synchronization boundary is based on a threshold of 5 slips in the whole experiment ( χav = 0.131 rad/s). As h is increased, the rotor pair moves deeper into the synchro- nized region: the coupling between the two strengthens due to lower hydrodynamic screening from the wall, lead- ing to an enhanced stability of the synchronized state. This is reproduced by simulations (Fig. 1c) up to small shift in D, likely the result of a finite value of a/d and In the limit a, R (cid:28) (cid:96), the evolu- experimental noise. tion of the phase difference χ = φ1 − φ0 can be derived by a generalization of previous arguments [13, 28]. As phase-locking is slow compared to the rotation period, we average over this fast timescale and find (cid:2)2A(β) + B(β)(cid:3) sin χ, χ = F1 − F0 R0ζ0ζw − 3a 4(cid:96) F0F1 λζ0R2 0 where, A(β) = 1 − X − β2 X = 1/(cid:112)1 + β2, and β = 2h/(cid:96). From Eq. (1), the aver- (cid:19)2 age phase drift χav for non-phase-locked states reads 2 X 3, B(β) = 1 − X 3 + 3β2 (cid:115)(cid:18) F1 − F0 (cid:18) 3a 2 X 5, (1) (cid:19)2 − χav = R0ζ0ζw F0F1 λζ0R2 0 4(cid:96) (2A(β) + B(β)) . (2) Given the functional form of the frequency detuning, Fi = FdrDi−0.5, Eq. (2) can be solved explicitly to yield the critical detuning D∗(h) (solid line in Fig. 1c). The theoretical and numerical solutions for the boundary in Fig. 1c slightly under- and over-estimate the data, re- spectively, owing to neglect of temporal variations in the inter-particle spacing and the finite size of the beads, re- spectively. Both also neglect thermal fluctuations. We now turn to the dynamics of a linear array of 6 ro- tors, with the ith rotor centered at x = (il, 0, h). This is the longest controllable chain with our active-feedback- based OTs. The dynamics are studied experimentally as a function of h, but numerical simulations allow wider ex- FIG. 1. (color online). Experimental setup and results. (a) Microspheres a distance h above a no-slip boundary are driven in circular trajectories by time-sharing optical tweezers. (b) Average phase drift χ = φ1 − φ0 for a rotor pair vs. detuning D for h = 4.2(•), 6.7(•), 11.7(•), 16.7(•), 31.7(•), 51.7(•) µm. (c) Phase diagram showing experimental regions of synchrony (blue) and drift (red), the boundary from hydrodynamic sim- ulations (dashed) and theory from Eq. (2) (solid). a of individual oscillators varies between τ ∼ 0.5 s and ∼ 1 s across the explored range of h. The oscillators are imaged under a Nikon inverted Eclipse Ti-E with a 60× Nikon Plan Apo VC water immersion objective (NA = 1.20), and recorded for up to 1200 s using an AVT Marlin F131B CMOS camera set at 229 fps. The parti- cle's positions are measured using a profile image cor- relation with subpixel resolution, and used to track the rotor phases {φi(t)} (Fig. 1a). Isolated oscillators rotate with a height-dependent angular velocity ωi = Fi/Rζ0ζw, where ζ0 = 6πµa is the sphere's bulk drag coefficient, h + O(a3/h3) accounts for the pres- and ζw(h) = 1 + 9 16 ence of the wall [29]. Experimental results are compared with deterministic hydrodynamic simulations based on the Oseen approximation to inter-particle coupling [11]. Unavoidable delays in the OT's feedback response intro- duce a mismatch between experiments and simulations which for a pair of oscillators is corrected by increasing the simulation value of λ by a factor κ relative to the experimental one. We estimate κ = 2.21 in agreement with previous reports [27]. Consider first two rotors separated by a distance (cid:96). For rotors with instantaneous positions {xi} and veloc- ities {vi}, the hydrodynamic drag on the ith rotor is ], where F ext is the net external force acting on the jth sphere and G(xi, xj) is the Green's function in the presence of the no-slip wall. For identical rotors (detuning D = 1), hy- drodynamic coupling eventually leads to synchrony pro- vided λ < ∞, by perturbing the angular velocities of the two rotors so that the leading and lagging rotors become given by −ζ(xi)· [vi−(cid:80) j(cid:54)=i G(xj, xi)· F ext j j 3 FIG. 2. (color online). Results for the linear array of driven colloidal oscillators. (a) Kymographs showing sin φi at three heights above the wall. With increasing h, the traveling wave becomes frustrated, with the introduction of wobbles (arrows) and phase defects (circles). (b) Numerical results from model. (c) Fraction of total coupling corresponding to interacting with different neighbors, as a function of h. Shaded red region represents experimental parameter regime. ploration of parameters, including changes in the radial stiffness λ, which governs the coupling strength [9, 11– 13, 27] as in Eq. (1). In both experiments and simulations we introduce a mild frequency bias D = 1.01, typical also of Volvox colonies [12], which breaks the transla- tional symmetry and induces a MW for h (cid:46) 10 µm. At all heights studied, this value of D is deep within the synchronized region of parameter space for two rotors. FIG. 3. (color online). Experimental phase dynamics. (a,b) Phase difference relative to the first rotor, φi − φ0, at h = 11.7 µm. (c) Wobbles are characterized by their magnitude W (radians) and timescale τ /(cid:104)T(cid:105) (normalized by rotor period), shown as a function of h in panels (d) and (e). (f) Probability that a propagating wobble ends at rotor i, resulting in a slip. Figure 2a shows that at h = 6.7(4) µm the rotors phase-lock in a stable MW whose direction is set by the frequency bias. With increasing h, defects (phase slips) emerge, giving rise to a net drift in the cumulative phase difference between rotors at opposite ends of the chain. Phase defects always propagate in the direction of the fastest oscillator. At these intermediate heights, the phase profile also displays "wobbles" – perturbations to the MW that are not accompanied by a phase defect. Numerical results shown in Fig. 2b capture the traveling wave at h = 5 µm, the presence of defects and their prop- agation direction, and wobbles at larger heights. At the largest height, h = 50 µm, defects no longer propagate through the chain, and rotors 3-5 remain phase-locked. The phase dynamics of wobbles and defects are shown in Fig. 3a for h = 11.7(4) µm. The first 25 seconds of the time-series show fluctuations in φi − φ0 (wobbles), even while the system is frequency-locked. Fluctuations start at the first oscillator pair and travel unidirectionally along the chain (Fig. 3b) with a preserved, soliton-like signature [30]. Occasionally, they terminate within the chain with a slip (Fig. 3a); these are the phase defects observed in kymographs. Both wobbles and defects are characterized by initial excursions of amplitude W and recurrence time τ (Fig. 3c) which depend on h (Fig. 3d,e). The typical time τ ∼ 10(cid:104)T(cid:105) (where (cid:104)T(cid:105) (cid:39) 1 s is the average period) depends less strongly on h than does W , which shows a pronounced growth (Fig. 3d) mirroring the increased probability that a wobble will terminate in a slip within the chain, causing a defect (Fig. 3f). Although their position can vary, defects tend to cluster, in this case at the middle of the chain (position i = 2), as seen also in simulations of longer chains [28]. The hydrodynamic coupling between two rotors in- creases monotonically with h; for an isolated pair, this manifests in more robust synchronization at larger heights. For a chain of rotors, increasing h has the re- verse effect, disrupting the stable MW with wobbles and punctuating it with periodic phase defects (Fig. 2). The hydrodynamic coupling between every pair of rotors in the chain grows as h is increased. For just two rotors, 4 FIG. 4. (color online). (a) Average phase drift per beat between end oscillators (measured in beats) as a function of height above the wall and radial spring stiffness. Shown also are four representative kymographs. (b) Time-averaged amplitude ¯A and (c) angle ¯Ψ of the complex order parameter Z = AeiΨ. The axes are the same as in a. (d) Same as a but with hydrodynamic interactions truncated to nearest neighbor. Parameters used include a = 1.74 µm, (cid:96) = 9.19 µm, R = 1.59 µm, viscosity µ = 6 mPas. Simulations correspond to 0 ≤ t ≤ 2000 s. The dashed white line shows the value of λ corresponding to Fig. 2. Eq. (1) shows that equivalent changes to the hydrody- namic coupling can be achieved through modification of the mean interparticle separation l. For the chain of 6 rotors, in which longer range hydrodynamic interactions also occur, changes to h and l are no longer equivalent. The peculiar dynamics observed arise from a change in the relative contributions of interactions with differ- ent neighbors. The no-slip wall has the effect of screening the hydrodynamic interactions in a way that qualitatively changes as a function of β = 2h/(cid:96). This is an important determinant of MW stability, as observed also in simula- tions of colloidal "rowers" [16]. Figure 2c shows the mag- nitude of the coupling of a given oscillator with its jth nearest neighbor, estimated with Eq. (1), normalized by the total interaction strength with the first 5 neighbors. Although all pairwise couplings grow monotonically with h, the relative magnitude of the nearest neighbor inter- actions actually diminishes. Conversely, the relative im- portance of all others increases with h. Hydrodynamic disturbances parallel to the wall decay as u ∼ r−j where j = 1 and 3 represent the far (β (cid:29) 1) and near (β (cid:28) 1) asymptotic limits [16]. For the end rotor the magnitude of the coupling with the nth nearest neighbor, normalized i=1 i−j. For β (cid:28) 1 the interactions are dominated by nearest neighbor, with S(1) = 0.84, while for β (cid:29) 1, S(1) = 0.44 (see black curve in Fig. 2c). We test the hypothesis that the breakdown of the traveling wave is due to long-range hydrodynamic interactions through simulations in which interactions are truncated at nearest neighbors, and find the abundance of defects is significantly reduced. Impor- tantly, the dynamics are nearly insensitive to h, with a maximum relative variation in end-to-end drift speed of just 3% between h = 5 µm and 1000 µm (Fig. 4) [28]. by the total coupling strength is S(n) = n−j/(cid:80)5 Additional numerical simulations permit the wider ex- N−1 ploration of parameter space. Figure 4a shows the av- (cid:80)N−1 erage end-to-end phase drift per beat as a function of λ and h. We also compute the complex order parameter n=0 eiχn where χn = φn+1 − φn Z = AeiΨ = 1 [16, 31]. Using the average values ¯A and ¯Ψ for t > 200 s [Fig. 4b,c], and following the experimental path (dotted line), we see the stable traveling wave at small h 3 de- veloping and, as h increases, defects and wobbles along the whole chain first 4 and then localized to the ini- tial half of the chain 2 , with the remaining three os- cillators constantly phase-locked. At values of λ smaller than the experimental one, however, we observe richer dynamics. For λ (cid:46) 2.5 pN/µm, the pattern morphs con- tinuously between different types of complete synchro- nization, going from a MW 3 to a chevron-like pat- tern 1 . These transitions happen without the emer- gence of defects [11]. For 2.5 pN/µm < λ < 3 pN/µm the system shows reentrant behavior with defects only at intermediate heights, separating a MW region from a chevron-like region. The order parameter angle ¯Ψ (Fig. 4c) identifies clearly the stable MW (yellow/orange) and chevron (dark blue) regions of parameter space. For a fixed h (cid:38) 50 µm, increasing λ results in a monotonic decrease in ¯A owing to the reduced rotor compliance. Conversely, the end-to-end phase drift exhibits a strong peak around λ = 4.5 pN/µm, where the rotors slip ap- proximately one beat in every five, despite an intrinsic frequency difference of just 5%. These nontrivial dynam- ics emerge due to the combination of phase slips induced by long-range interactions, and rapid healing of phase de- fects through orbit compliance. The complete absence of these features from the simulations with nearest neigh- bor coupling alone (Fig. 4d) highlights the role played by competition between interactions at different ranges. Changing h is then a simple and accessible way to mod- ulate their relative strength (see Fig. 2). 20130571 (2013). 5 Large arrays of cilia are synonymous with no-slip boundaries, and in many cases, the spacing between these organelles is comparable to their length [12], so that effec- tively h/(cid:96) ∼ 1 (see Fig. 4a). Our results suggest that flag- ella of Volvox may then be balancing the need to extend out into the fluid enough to generate a vigorous thrust, with the screening of long-rage hydrodynamic interac- tions necessary to stabilize MWs on the colony surface. As a result, ensembles of flagella in Volvox [11] (but see also numerical simulations [20]) may operate in a regime naturally prone to the emergence of metachronal phase defects, which are indeed observed experimentally [12]. We are grateful to T.J. Pedley and D. Bartolo for useful discussions. This work was supported by a Hu- man Frontier Science Program Cross-Disciplinary Fellow- ship (DRB), a Wellcome Trust Senior Investigator Award (REG), and the EU ERC CoG Hydrosync (PC). ∗ [email protected] [1] F. Dorfler and F. Bullo, Automatica 50, 1539 (2014). [2] R. E. Mirollo and S. H. Strogatz, SIAM J. Appl. Math. 50, 1645 (1990). [3] M. Toiya, H. O. Gonz´alez-Ochoa, V. K. Vanag, S. Fraden and I. R. Epstein, J. Phys. Chem. Lett. 1, 1241 (2010). [4] K. Son, D. R. Brumley, and R. Stocker, Nat. Rev. Micro. 13, 761 (2015). [5] R. E. Goldstein, Annu. Rev. Fluid Mech. 47, 343 (2015). [6] G. Quaranta, M.-E. Aubin-Tam, and D. Tam, Phys. Rev. Lett.115, 238101 (2016). [7] M. A. Sleigh, The Biology of Cilia and Flagella, (Perga- mon Press, Oxford, 1962). [8] B. Button, L. Cai, C. Ehre, M. Kesimer, D. B. Hill, J. K. Sheehan, R. C. Boucher, and M. Rubinstein, Science 337, 937 (2012). [9] D. R. Brumley, K. Y. Wan, M. Polin, and R. E. Gold- stein, eLife 3, e02750 (2014). [10] E. W. Knight-Jones, Quart. J. Micro. Sci. 95, 503 (1954). [11] D. R. Brumley, M. Polin, T. J. Pedley, and R. E. Gold- stein, Phys. Rev. Lett. 109, 268102 (2012). [12] D. R. Brumley, M. Polin, T. J. Pedley, and R. E. Gold- stein, J. R. Soc. Interface 12, 20141358 (2015). [13] T. Niedermayer, B. Eckhardt, and P. Lenz, Chaos 18, 37128 (2008). [14] A. Vilfan and F. Julicher, Phys. Rev. Lett. 96, 058102 (2006). [15] N. Uchida and R. Golestanian Phys. Rev. Lett. 106, 058104 (2011). [16] C. Wollin and H. Stark, Eur. Phys. J. E 34, 42 (2011). [17] S. Gueron and K. Levit-Gurevich, Proc. Natl. Acad. Sci. USA 96, 12240 (1999). [18] M. Cosentino Lagomarsino, P. Jona, and B. Bassetti, Phys. Rev. E 68, 021908 (2003). [19] N. Osterman and A. Vilfan, Proc. Natl. Acad. Sci. USA 108, 15727 (2011). [20] J. Elgeti and G. Gompper, Proc. Natl. Acad. Sci. USA 110, 4470 (2013). [21] N. Bruot, and P. Cicuta, J. R. Soc. Interface 10, [22] I. Kavre, A. Vilfan, and D. Babic, Phys. Rev. E 91, 031002 (2015). [23] N. Bruot and P. Cicuta, Annu. Rev. Condens. Matt. Phys. 7, 323 (2016). [24] D. M. Abrams and S. H. Strogatz, Phys. Rev. Lett. 93, 174102 (2004). [25] E. A. Martens, S. Thutupalli, A. Fourri`ere, and O. Hal- latschek, Proc. Natl. Acad. Sci. USA 110, 10563 (2013). [26] M. Leoni, J. Kotar, B. Bassetti, P. Cicuta, and M. C. Lagomarsino, Soft Matter 5, 472 (2009). [27] J. Kotar, L. Debono, N. Bruot, S. Box, D. Phillips, D. Simpson, S. Hanna, and P. Cicuta, Phys. Rev. Lett. 111, 228103 (2013). [28] Supplementary Material. [29] J. Happel and H. Brenner, Low Reynolds Number Hydro- dynamics (Kluwer, Dordrecht, 1991). [30] K. Ahnert and A. Pikovsky, Chaos 18, 37118 (2008). [31] A. Pikovsky, M. Rosenblum, and J. Kurths, Synchroniza- tion: A Universal Concept in Nonlinear Sciences (Cam- bridge University Press, Cambridge, 2003). [32] Blake, J. R. (1971) Mathematical Proceedings of the Cam- bridge Philosophical Society 70(2), 303–310. SUPPLEMENTARY MATERIAL Single rotor force calibration For a single bead of radius a in a viscous fluid, situ- ated at a distance h from an infinite no-slip boundary, the external applied force F is related to its velocity v according to F = ζ · v, where ζ is the anisotropic drag matrix, given by [14] 16h (I + ezez) + O((a/h)3)(cid:3). (cid:2)I + 9a ζ = ζ(h) = ζ0 (S1) The coefficient ζ0 = 6πµa is the drag on the sphere in an unbounded fluid of viscosity µ (equivalent to setting h → ∞). We are interested only in trajectories which are parallel to the no-slip wall (v · ez = 0). For a constant applied driving force Fdr, the sphere's speed v = v is given by v (cid:39) Fdr ζ0(1 + 9 16 , a h ) (S2) implying a monotonic increase of the sphere's speed with h for a given Fdr. Each set of experiments involves studying the colloidal oscillators at a number of differ- ent heights h. For each set, the center of the trajec- tory, its radius and the driving and radial forces are cal- ibrated, for each individually loaded rotor, at the height of h = 22 µm. These are then checked for independence on h. Figure S1 shows, after a full calibration, the speed of an individually loaded colloidal oscillator at 6 differ- ent heights together with the prediction from Eq. (S2) using a constant driving force. The two agree well for Fdr = 2.23 pN. 6 ζ0ζwR0ωi, where ζ0 = 6πµa is the bulk drag coefficient, and ζw is the correction due to the presence of the wall. For sufficiently small a/h this correction can be written as ζw = 1+(9a/16h)+O((a/h)3). The first thing required is the generic expression of the 'Blakelet', i.e. the Stokeslet on top of a bounding wall. This is given by [32]: (cid:20)(cid:18) δij v21,i = Fj 8πµ rirj r3 + 2h(δjαδαk − δj3δ3k) + r ∂ (cid:19) (cid:19) − (cid:18) δij (cid:26) hRi R3 − R ∂Rk + RiRj R3 (cid:18) δi3 + RiR3 R3 R (cid:19)(cid:27)(cid:21) . (S3) Here r = ((cid:96), 0, 0), r = r; R = ((cid:96), 0, 2h), R = R; α ∈ {1, 2}. In our case, the point force will be in the (x, y) plane. Furthermore, we are only interested in the component of the velocity in the same plane, since the trajectories are constrained to lie at z = h. The back- ground flow is due to the motion of the other sphere, which is: (cid:20)(cid:18) F (cid:18) F R3 − 3 + r (cid:19) (cid:19) − (cid:18) F (cid:21) R . r(F · r) r3 R(F · R) R5 v21, = 1 8πµ − 2h2 (cid:19) R(F · R) + R3 (S4) In the limit as h → 0, this reduces to Eq. (16) in [13]. Being interested only in the component, and because F = F, we can substitute R(F · R) = (r + 2he3)(F · r) → r(F · r). (S5) Calling, as in [13], s = F/8πµ we get v21, = A(β)s + B(β)r(s · r) r (S6) where r = r/r = ex, called n21 in [13]; β = 2h/(cid:96), and 2 − β2 2 1 + β2 (cid:18) 1 (cid:18) 1 (cid:19) 1 (cid:19) 3 2 1 + β2 + 3β2 2 (cid:18) 1 (cid:18) 1 1 + β2 2 (cid:19) 3 (cid:19) 5 2 1 + β2 , (S7) . (S8) A(β) = 1 − B(β) = 1 − Notice that, due to the nearby wall, the strength of the Stokeslet s is written in terms of the sphere's velocity as: s = 3 4 aζ0ζwRi φieφi. (S9) The derivation of the equations of motion follows the same procedure outlined in Appendix A of [13]. The FIG. S1. Calibration of an individual colloidal oscillator, mov- ing under the influence of a harmonic potential in an opti- cal tweezer. Experimental results (dots) are shown alongside the prediction of Eq. (S2), with which the driving force of Fdr = 2.23 pN can be extracted. Hydrodynamic interaction of two rotors at an arbitrary distance from a no-slip plane The fluid disturbance produced by the motion of a sphere parallel to a no-slip wall depends on its height above the planar boundary. For two such spheres situ- ated in the fluid, it is important to calculate the strength of the hydrodynamic interactions between them, and the subsequent effects on their dynamics. We consider two spheres of radius a driven around circular orbits of ra- dius R0 which are parallel to a no-slip wall. The orbit's centers are located at positions (x, y, z) = (0, 0, h) and ((cid:96), 0, h) respectively. The plane z = 0 represents the no- slip boundary, with the semi-infinite domain z > 0 filled with fluid of viscosity µ. (eφ1, eR1) and (eφ2, eR2) are the unit vectors of the local cylindrical frame of reference of each single rotor. This reference frame is centered at the center of the orbit. The displacement of each sphere from the center of its trajectory will be expressed in its cylindrical frame of reference as (R1, φ1) and (R2, φ2) re- spectively. Each rotor is subject to a constant tangential driving force Fi = FieRi, and also to a radial spring force with stiffness λ, which suppresses excursions from the equilibrium radius R0. The spring stiffness is assumed to be large enough that the radial degree of freedom is slaved to the angular degree of freedom. That is, know- ing (φ1, φ2), we know the instantaneous value of (R1, R2). It will be our goal to derive the equations of motion of the spheres, without making any assumptions about the relative magnitudes of (cid:96) and h. We assume that sphere radius and trajectory radius are both small compared to other length scales (a, R0 (cid:28) h, (cid:96)). Correspondingly, the two orbits are sufficiently far from each other that we can neglect the variation in the relative separation between the spheres as they move. The separation vector will always be taken to be (cid:96)ex. We will write the relation between the sphere's angular velocity ωi and the tangential driving force Fi, as Fi = occur. We use the following relation 7 (cid:90) 2π 0 (cid:2)a− b sin χ(cid:3)−1 (cid:115) (∆ω)2 − dχ = 2π√ a2 − b2 , for a > b (S23) (cid:18) 3a 4l ζ0ζ 2 w λ ω1ω2 (cid:2)2A(β) + B(β)(cid:3)(cid:19)2 . to find the time-averaged phase drift (S24) For each rotor (now indexed by i ∈ {0, 1}), the intrinsic angular frequency is given by ωi = Fi/(ζ0ζwR0) and so the above equations reads: (cid:115)(cid:18) F1 − F0 R0ζ0ζw (cid:19)2 − (cid:18) 3a 4(cid:96) F0F1 λζ0R2 0 (cid:2)2A(β) + B(β)(cid:3)(cid:19)2 . χav = (S25) Since a detuning factor D is included so that the driv- ing force is Fi = FdrDi−1/2, the above equation can be written as (cid:18) 3a 4(cid:96) Fdr λR0 (cid:2)2A(β) + B(β)(cid:3)(cid:19)2 . χav = Fdr R0ζ0 (D − 1)2 D ζ 2 w − (cid:115) (S26) The threshold value of D beyond which the rotors' phase difference will drift can be calculated explicitly. (S16) Varying chain length In order to assess the generality of the results presented in the main text, we used numerical simulations to ex- plore the effect of changing the number of rotors, N , present in the linear array (see Fig. 1a). Figure S2 shows the average phase drift (measured in beats per beat) with respect to the first rotor, along chains of different length, N ∈ {2, 15}. Each chain has a fixed detuning of 5% be- tween the end rotors. For each height h = 10 µm and h = 100 µm, simulations were conducted with full hydro- dynamic coupling and nearest neighbor coupling only. The results for N = 6 are representative of the dynam- ics across a range of chain lengths. For chains in which rotors are coupled through nearest neighbor interactions, the rotors tend to phase-lock in clusters of 2-5 rotors. As discussed in the main text, the nearest neighbor results are fairly insensitive to changes in h, shown here by the similarly between the results of Fig. S2b and d. In stark contrast, the chains in which rotors are fully coupled to one another through hydrodynamic interactions exhibit qualitatively different behavior at different heights. + B(β) sin(φ1 + φ2)]. (S11) χav = only things we need to calculate are: eφ1 · v12 = eR1 · v12 = 3a 8(cid:96) 3a 8(cid:96) ζwR2 φ2[(2A(β) + B(β)) cos(φ1 − φ2) + B(β) cos(φ1 + φ2)], (S10) ζwR2 φ2[(2A(β) + B(β)) sin(φ1 − φ2) The rest of the calculation can be carried out in exactly the same way as in [13] and the final result is φ1 = ω1 − ρω2J(φ1, φ2; β) − ραω1ω2K(φ1, φ2; β), φ2 = ω2 − ρω1J(φ2, φ1; β) − ραω1ω2K(φ2, φ1; β), (S12) (S13) where now ρ = 3aζw/8(cid:96), α = ¯ωζ0ζw/λ, and J(φi, φj : β) = −[(2A(β) + B(β)) cos(φi − φj) + B(β) cos(φi + φj)] K(φi, φj : β) = (2A(β) + B(β)) sin(φi − φj) + B(β) sin(φi + φj). (S14) (S15) For example, this means that the phase difference χ = φ2 − φ1 and phase sum Φ = φ1 + φ2 evolve according to χ = (ω2 − ω1)[1 + ρJ(φ1, φ2; β)] − 2ραω1ω2(2A(β) + B(β)) sin(χ), Φ = (ω1 + ω2)[1 + ρ(2A(β) + B(β)) cos χ + ρB(β) cos Φ] − 2ραω1ω2B(β) sin Φ. (S17) To the first order in the small quantities ∆ω = ω2 − ω1 and ρ, and averaging over a "natural" timescale of the fast variable Φ we get χ = ∆ω − 2αω1ω2 ρ(2A(β) + B(β)) sin χ, (cid:104) Φ(cid:105) = (ω1 + ω2) [1 + ρ(2A(β) + B(β)) cos χ] . (S18) (S19) These functions can be rewritten as χ = ∆ω + D(χ), (cid:104) Φ(cid:105) = (ω1 + ω2) + S(χ), (S20) (S21) where D(χ) = D0 sin χ and S(χ) = S0 cos χ. As in [13], time has been rescaled according to the mean angular speed ¯ω, and both ωi are measured in units of ¯ω. Rewrit- ing Eq. (S18) in dimensional units yields (cid:2)2A(β) + B(β)(cid:3) sin(χ). χ = ∆ω − 3a 4(cid:96) ζ0ζ 2 w λ ω1ω2 (S22) This is of the form χ = ∆ω − C sin(χ), with C > 0. If ∆ω < C then a stable fixed point χ = 0 exists. Con- versely, for ∆ω > C, a cycle-averaged phase drift will 8 Truncation of hydrodynamic interactions Figure S3 shows the results of deterministic numerical simulations, with hydrodynamic interactions truncated to be nearest neighbor in nature (see also Fig. 4d). The dynamics are almost completely insensitive to changes in h, across several orders of magnitude. FIG. S2. (color online). Average phase drift with respect to the first rotor, for chains of different lengths N ∈ {2, 15}, at two different heights h ∈ {10 µm, 100 µm}, and subject to either full hydrodynamic coupling or nearest neighbor inter- actions only. The end-to-end detuning is fixed at 5% in each case, the radial spring stiffness is λ = 4.5 pN/µm, and all other parameters are as in Fig. 4. FIG. S3. (color online). Kymographs showing the phase sin φi along the linear chain of model rotors, coupled hydrodynam- ically through the Blake tensor, but with interactions arti- ficially restricted to be nearest neighbor. The radial spring stiffness is λ = 4.5 pN/µm and all other parameters are as in Fig. 4.
1004.0883	1	1004	2010-04-06T16:08:26	Influence of spatially modified tissue on atrial fibrillation patterns: Insight from solutions of the FitzHugh-Nagumo equations	[ "physics.bio-ph", "physics.med-ph", "q-bio.TO" ]	We study the interplay between traveling action potentials and spatial inhomogeneities in the FitzHugh-Nagumo model to investigate possible mechanisms for the occurrence of fibrillatory states in the atria of the heart. Different dynamical patterns such as ectopic foci, localized and meandering spiral waves are found depending on the characteristics of the inhomogeneities. Their appearance in dependence of the size and strength of the inhomogeneities is quantified by phase diagrams. Furthermore it is shown that regularly paced waves in a region R, that is connected by a small bridge connection to another region L with perturbing waves emanating from an additional pacemaker, can be strongly disturbed, so that a fibrillatory state emerges in region R after a transient time interval. This finding supports conjectures that fibrillatory states in the right atrium can be induced by self-excitatory pacemakers in the left atrium.	physics.bio-ph	physics	Influence of spatially modified tissue on atrial fibrillation patterns: Insight from solutions of the FitzHugh-Nagumo equations Claudia Lenk1,∗ Mario Einax1, and Philipp Maass2† 1Institut fur Physik, Technische Universitat Ilmenau, 98684 Ilmenau, Germany 2Fachbereich Physik, Universitat Osnabruck, Barbarastrasse 7, 49069 Osnabruck, Germany (Dated: 6 April 2010) Abstract We study the interplay between traveling action potentials and spatial inhomogeneities in the FitzHugh-Nagumo model to investigate possible mechanisms for the occurrence of fibrillatory states in the atria of the heart. Different dynamical patterns such as ectopic foci, localized and meandering spiral waves are found depending on the characteristics of the inhomogeneities. Their appearance in dependence of the size and strength of the inhomogeneities is quantified by phase diagrams. Furthermore it is shown that regularly paced waves in a region R, that is connected by a small bridge connection to another region L with perturbing waves emanating from an additional pacemaker, can be strongly disturbed, so that a fibrillatory state emerges in region R after a transient time interval. This finding supports conjectures that fibrillatory states in the right atrium can be induced by self-excitatory pacemakers in the left atrium. PACS numbers: 87.10.Ed, 87.18.Hf 0 1 0 2 r p A 6 ] h p - o i b . s c i s y h p [ 1 v 3 8 8 0 . 4 0 0 1 : v i X r a ∗Electronic address: [email protected] †Electronic address: [email protected]; URL: http://www.statphys.uni-osnabruck.de 1 I. INTRODUCTION Atrial fibrillation (AF) is the most frequently appearing heart arrhythmia since it accounts for one third of all hospitalizations caused by heart arrhythmia in the industrialized countries [1]. During AF the electric conduction system of the heart is disturbed and an increased rate of activation by a factor of 3-12 compared to normal sinus rhythm occurs. Special spatio-temporal patterns of the electric potential like spiral waves, mother waves or ectopic foci are thought to be underlying generating mechanisms of AF [2 -- 5]. These patterns are often located near physiologically modified regions of the heart tissue in the left atrium [5 -- 8]. The question hence arises, how these physiologically modified regions can be responsible for the generation of spiral waves or ectopic foci and how they influence the properties of these patterns. To tackle these questions, we study generating mechanism for AF on the basis of the FitzHugh-Nagumo model [9], which is a simple model for action potential generation and propagation. By modeling physiologically modified regions using a spatial variation of the parameters characterizing cell properties like excitability or resting state stability, we calcu- late phase diagrams, which specify the type of spatio-temporal excitation pattern in depen- dence of the extent of the modified region and the strength of the modification. Thereupon we investigate how self-excitatory sources as spiral waves or ectopic foci with rather regular dynamics in one region can induce irregular, fibrillatory excitation patterns in some other region. Irregular, fibrillatory states are often observed in the right atrium [7, 8, 10] and it was conjectured that these are caused by the perturbation of regular waves generated by the sinus node by waves emanating from an additional pacemaker like a spiral wave or ectopic foci in the left atrium. II. MODEL The FitzHugh-Nagumo (FHN) equations [9] are a set of two coupled nonlinear ordinary differential equations, which describe excitable media via an inhibitor-activator mechanism. They were originally developed by searching for a simplified version of the Hodgkin-Huxley equations for electric pulse propagation along nerves [11]. When combined with a spatial 2 diffusion term, the equations are ∂u ∂t ∂v ∂t = D(cid:18)∂2u ∂x2 + ∂2u ∂y2(cid:19) + c(v + u − u3 3 + z) = − 1 c (u − a + bv) . (1) This set of partial differential equations serves as a prototype for a large variety of reaction- diffusion systems, which occur, for example, in chemical reactions as the Bhelousov- Zhabotinsky reaction [12, 13] or the catalysis of carbon monoxide [14, 15], in population dynamics [16], in biology in connection with aggregation processes [17] or plancton dynam- ics [19], as well as in the spreading of forest fires [20]. Here we will use Eqs. (1) in their original context as a model to investigate the spatio- temporal evolution of electric excitations in the heart. In this approach the variable u is roughly associated with the membrane potential and the variable v with the ion currents through the cell membrane. The resting state is given by the pair of values u = u0 = 1.2 and v = v0 = −0.6. The diffusion coefficient D describes the coupling between the cells, and z is an applied external current (stimulus). The influence of the parameters a, b and c can be inferred by numerical solutions of Eqs. (1) without the diffusive term. The parameter values have to be limited to some range in order to generate excitability, and their detailed effect on the pulses is complicated due to mutual interdependencies originating from the nonlinearity in Eq. (1). Roughly speaking, a affects the length of the refractory period, b influences the stability of the resting state, and c controls the excitability and strength of the cells' response to a stimulus. To capture the propagation and form of a typical action potential, the following set of parameters can be used: D = D0 = 0.1, a = a0 = 0.7, b = b0 = 0.6, and c = c0 = 5.5. These values will be associated with a "healthy tissue" in the following. Figure 1 shows the time development of u and v during an excitation after a stimulus with these parameters. Note that the variable −u mirrors the form of a pulse in an usual representation of an ECG recording. Ectopic foci and spiral waves are thought to be caused and influenced by physiologically modified regions of the tissue, which in the modeling correspond to spatial variations of the parameters. To simplify the analysis, we fix a = a0 and D = D0, and consider variations of 3 the parameters b and c according to b(x, y) = b0 − ∆b exp(−p(x − x0)2 + (y − y0)2/ξb) , c(x, y) = c0 − ∆c exp(−p(x − x0)2 + (y − y0)2/ξc) , (2) (3) where the amplitudes ∆b, ∆c characterize the strength, and the correlation lengths ξb and ξc characterize the spatial range of modification. The calculations are carried out on a two-dimensional simulation area of size 20 × 20, which represents an isolated section of atrial heart tissue, as it is used often in experiments [21 -- 23]. The boundary conditions of the simulation area are of von Neumann type, i. e. ∂u/∂n = 0, where ∂/∂n denotes the normal derivative. To solve the two nonlinear coupled partial differential equations (1) we use the finite element method (FEM) with a triangulation consisting of 4225 nodes and 8192 triangles, and a constant integration time step ∆t = 0.01. A simulation time of 1 corresponds to a time of roughly 5 to 5.5 ms. The nonlinearity u3(~x, t) in Eq. (1) is treated as an inhomogeneity, which means that for u(~x, ti) the value u(~x, ti−1) of the preceding time step is used. III. GENERATING MECHANISMS A. Ectopic activity Ectopic foci are regions in the atria, which generate activation waves emanating from self-excitatory hyperactive cells. In these cells the transmembrane potential raises without external stimulation until the threshold value is reached and an action potential results. In optical mapping studies and spatially resolved ECG recordings, ectopic foci are often localized in the regions of the pulmonary veins [7, 8]. To model a tissue with physiologically modified properties that result in ectopic activity, we fix c = c0 (∆c = 0) and vary the resting state stability b around the center of the simulation area (x0 = y0 = 10) according to Eq. (2). Initially the system is in the excitable resting state (u = u0 and v = v0). Figure 2a shows the resulting activation pattern for ∆b = 0.4 and ξb = 0.8. The modified tissue is self-excitatory and acts as a pacemaker for activation waves, which propagate radially. In the time evolution shown in Fig. 3, u decreases until the threshold value uth ≃ 0.6 for activation is reached and an action potential with a 4 steep fall in u occurs. In response to this activation, the inhibitor variable v increases and pulls u back to a value even larger than its initial value (overshoot) before u returns to it, and the self-excitatory process starts anew. In order to systematically characterize the occurrence of ectopic activity, we calculate a phase diagram, where in dependence of ξb and ∆b regions of ectopic activity can be distin- guished from that without self-excitatory behavior. The results in Fig. 2b show that there exists a minimal ∆bmin ≃ 0.25, below which no ectopic activity occurs. The corresponding value (b0 −∆bmin) ≃ 0.5 is the critical value of resting state stability in the FHN equations (1) in the absence of the diffusion term. For fixed ∆b > ∆bmin the ectopic activity vanishes, when ξb falls below the dashed transition line in Fig. 2b. In this region the diffusive current from the modified tissue to the surrounding causes the initial decrease of u to become so slow, that the counter-regulation by v eventually hinders u to reach its activation threshold, cf. Fig. 3. Only small oscillation of u around a reduced resting state value can be seen in Fig. 3, which become weaker with growing time. The temporal-spatial pattern of the activation in the phase of ectopic activity is char- acterized by the frequency of the ectopic focus. We calculate this frequency as the inverse mean time interval between consecutive action potentials. As shown in Fig. 4, the frequency becomes larger with increasing ∆b (at fixed ξb) and ξb (at fixed ∆b), and it tends to saturate for large ξb. With increasing ξb, the diffusive current of the inner cells of the modified tissue decreases and thus a larger frequency is obtained. In the saturation limit the frequency is nearly the same as in the absence of diffusion and thus is mainly determined by the refractory period. B. Spiral waves In this section we study the influence of physiologically modified regions, called "obsta- cles" henceforth, on spiral wave behavior. It was observed that spiral waves in the atria can be generated by a perturbation of the propagation of planar excitation waves by anatomic obstacles as, for example, the pulmonary veins, the venae cavae, the pectinate muscle bun- dles or some localized region of modified tissue [6, 21, 24, 25]. These regions are considered as not fully excitable and are thus modeled as regions with a reduced parameter c according to Eq. (3). 5 In an experiment by Ikeda and coworkers [21] a nearly rectangular area of atrial tissue was placed on an electrode plaque in a tissue bath. Holes with different diameters were created and a reentrant wave was initiated by cross-field stimulation. The resulting behavior of the wavefront was, amongst others, classified according to whether the spiral is anchored by the obstacle, and by the relationship between hole size and cycle length of the reentry. It was observed that for large obstacle sizes (6, 8 and 10 mm), the reentrant wave attaches to the obstacle, leading to a linear increase of the cycle length with the hole diameter. For small obstacle diameters below about 4 mm, by contrast, meandering spirals with a tip getting variably closer to or further away from the hole were found. In this case the cycle length becomes independent of the hole diameter. Similar results were observed by Lim and coworkers [24]. They analyzed the behavior of spiral waves near holes with diameters ranging from 0.6 to 2.6 mm and obtained a higher attachment rate for larger obstacle diameters as well as a positive linear correlation of the reentry conduction velocity and wave length with the obstacle diameter in the case of attached spirals For smaller obstacle diameters the spiral waves were found to attach to and detach from the obstacle. The missing anchoring for small hole sizes was explained in [21] by invoking a "source-sink relationship". The "source" is the activation wavefront and provides a diffusive current to the surrounding tissue in the resting state, which constitutes the "sink". The sink becomes larger for smaller obstacles, where more cells become depolarized by the activation wavefront. If the source-to-sink ratio is decreased below a certain critical value, the wavefront detaches from the obstacle. To elucidate these experimental findings, we perform numerical calculations for a ge- ometry corresponding to the experiments with the following initial state and parameters settings: the modified region, is, as in the previous Sec. III A, placed in the center of the simulation area at x0 = y0 = 10. Initially a "planar" (linear) wave is generated by inducing a current z in the stripe 9.5 ≤ x ≤ 10, 0 ≤ y ≤ 10, and by setting the area 0 ≤ x ≤ 9.5, 0 ≤ y ≤ 10 into a refractory state with u = 1.6 and v = 0, while the rest of the simulation area is in the resting state (u = u0, v = v0). This initial state resembles the activation pat- tern directly after application of a cross-field or paired-pulse stimulation (two rectangular pulses). At the "upper part" of the initial planar wavefront (9.5 ≤ x ≤ 10, y = 10), diffusive currents flow "radially" in all forward directions (y > 10), while at the "right boundary" (x = 10, 0 ≤ y ≤ 9.5) the diffusive currents can flow only in positive x direction (due to 6 the refractory state in the area 0 ≤ x ≤ 9.5, 0 ≤ y ≤ 10). This higher loss by diffusion leads to a smaller propagation speed of the initial wavefront at its upper boundary compared to its right boundary. As a consequence, the wavefront becomes curved, and a reentrant spiral wave develops for all reductions ∆c in excitability and obstacle sizes ξc, in accordance with the experimental observations. Figure 5 shows activation patterns for ξc = 2 and a) ∆c = 4.5, and b) ∆c = 1.5. The stronger reduction of excitability in Fig. 5a leads to an anchoring of the spiral wave, while in Fig. 5b the spiral is meandering. To analyze the parameter regimes of the occurrence of anchored or meandering spiral waves, we perform a frequency analysis for different values of ∆c and ξc. Therefore, we determine the peak positions in the time series of u at 8 positions far away from the center of the spiral and calculate the peak-to-peak intervals. The frequency of one point is one over the mean of the peak-to-peak intervals and the mean cycle length is one over the average of all these local frequencies. The results in Fig. 6 show that, as in the experiments, attached spiral waves occur for large ∆c > ∼ 3 and for sufficiently large ξc > ξ⋆ c , where ξ⋆ c decreases with increasing ∆c. For these anchored spirals, the frequency is proportional to f = η/2πξc, where η ≃ 0.82 is the conduction velocity in the FHN model. Accordingly, 1/f increases linearly with ξc for ξc > ξ⋆ c in Fig. 6. For small ∆c < ∼ 3, only meandering spirals are observed. The transition from large to small ∆c reflects the transition from anatomical to functional reentry [26], as it has been reported in medical studies [24]. The fact that for small ξc always meandering spirals occur, can be interpreted by the small source-sink ratio [21]. The same mechanism can also lead to meandering spirals for large ξc if ∆c becomes small. Note, that nevertheless the spiral wave is not anchored to the obstacle, its movement is still influenced by the obstacle. IV. INDUCED FIBRILLATORY STATES IN THE RIGHT ATRIUM In previous studies on interactions of paced waves with self-excitatory waves, the influence of the pacing on a spiral wave was studied [27, 28] with the aim to suggest a possible therapy to suppress fibrillation or tachycardia. The pacing was applied to the region, where the spiral wave was located. It was found that the pacing leads to an annihilation of the reentrant activity or to a shift of the spiral core [28 -- 31]. Here we investigate the perturbation of regular paced waves from a source representing 7 the sinus node by waves emanating from an additional pacemaker located in a distant region. Electrocardiogram recordings and their frequency analyses show that regular excitation pat- terns are often observed in the left atrium, where additional pacemakers like spiral waves or ectopic foci are located [5, 8], and that at the same time irregular, fibrillatory-like states in the right atrium occur [7]. This led to the conjecture that fibrillatory states can be induced in the right atrium by self-excitatory pacemakers in the left atrium. In this connection it is important to better understand how a fibrillatory state can occur, if regular paced waves, as generated by the sinus node, are disturbed by additional pacemaker waves. To this end, we consider the waves to be located in spatially separated regions that are connected by a small region. To be specific, we choose a simulation area of size 21 × 10, which is divided into three regions (see Fig. 7). The rectangular area L with 0 ≤ x ≤ 10, 0 ≤ y ≤ 10 representing the left atrium, the rectangular area R with 11 ≤ x ≤ 21, 0 ≤ y ≤ 10 representing the right atrium, and the small bridge B with 10 < x < 11, 4 < y < 6 representing the connection between the atria. The grid used in the finite element calculations consists of in total 8871 nodes and 17350 triangles. We focus on situations where the pacemaker in the region L is located far outside the left part of the simulation area, so that the resulting wavefronts become "planar" (linear). In the simulation they are generated by application of a stimulating current z = −1 with duration tz = 1 = 100∆t and a period 1/fpert in the region x ≤ 0.5 and 0 ≤ y ≤ 10. The activation waves representing the pacemaker in region R are generated by the application of a current z = −1 with duration tz = 1 and period 1/fpace in the region 11 ≤ x ≤ 21 and y ≤ 0.5. The irregularity of the resulting patterns in region R is quantified by calculat- ing the Shannon entropy of the distribution of local activation frequencies for every grid point in R. To this end we divide the frequency range into Nb bins of size ∆ =min(cid:8)exp [0.626 + 0.4 ln(Ng − 1)]−1 , 0.01(cid:9) [32, 33] and calculate the probabilities n(fl ≤ f ≤ fl + ∆) pl = Ng , (4) of finding frequency f in bin l, where Ng is the total number of grid points. The normalized entropy then is given by s = S Smax = −PNb l=1 pl ln pl ln Nb , (5) 8 For a single frequency (pl = δl,l0), s = 0, while for a chaotic activation pattern with a uniform distribution (pl = 1/Nb), s = 1. For small perturbation frequencies (fpert ≤ 0.1) the influence of the activation wavefronts from the additional pacemaker onto the sinus node waves is almost negligible. Small defor- mations of the linear wavefronts are observed, but the measured frequencies are close to the pacing frequency, and the overall spatiotemporal pattern in R is regular. With increasing perturbation frequency the spatiotemporal pattern in region R becomes more irregular and a breakup of the regularly paced waves can occur. Figure 8 shows the time evolution of the excitation patterns for a perturbation frequency fpert = 0.105 and a pacing frequency fpace = 0.091. For small times, the waves are only slightly perturbed and the pattern remains regular, as can be seen from the four consecutive snapshots in Fig. 8a. At a later time, however, the perturbation by the waves from region L results in a breakup of the waves close to the bridge B in region R, as can be seen from the four consecutive snapshots in Fig. 8b. The onset of this breakup was found to occur at a time t ≃ 160. In order to investigate how the spatial irregularity is reflected in the time evolution, we consider three different points P1 = (11.49, 5.34), P2 = (11.33, 6.34) and P3 = (11.94, 6.15) in region R. The time evolution of u for these three points is shown for two different perturba- tion frequencies f (1) curve, P3: blue dash-dotted curve). For the lower frequency f (1) points is regular, see Fig. 9a. For the higher frequency f (2) pert = 0.105 in Fig. 9 (P1: black solid curve, P2: red dashed pert the evolution at all three pert = 0.1 and f (2) pert, by contrast, the break ups of the waves seen in Fig. 8b yield unsuccessful activations during refractory periods, as can be seen, for example, at time t ≃ 175 in point P2 (red dashed curve) and at time t = 179 in point P1 (black solid curve). These unsuccessful activations are caused by a rapid pacing of the region by the curled wave. Another feature is that the shape of the action potential varies. This can be seen, for example, at point P1 (black solid line) when comparing the pulses at t ≃ 160 and t ≃ 194. It is important to note that these irregularities are hardly observed at point P3 (blue dash-dotted curve), showing that they exhibit a spatial heterogeneity. The total irregularity in region R quantified by the normalized Shannon entropy s of the local frequency distribution is shown in Fig. 10 as a function of the frequency fpert of the perturbing waves from region L. For small perturbation frequencies fpert s equals the unperturbed case, while for fpert > ∼ 0.1, s sharply increases until it reaches a < ∼ 0.1 the entropy maximum at fpert ≃ 0.1075. For higher fpert a return to more regular activation pattern is 9 found, indicating that the disturbance is most pronounced if fpert is close to fpace. To conclude, the disturbance of the wavefronts in region R by waves emanating from an additional source in region L and propagating through the bridge region B can lead to irregular, fibrillatory-like activation patterns in region R. On the other hand, the waves in region L are almost unaffected by the waves in R with primary wavefront orthogonal to the cross section of the bridge. The irregularities in region R are most pronounced at a certain perturbation frequency fpert. How this value is influenced by the geometry of the bridge and the parameters characterizing the cell properties remains further investigation. V. SUMMARY The influence of physiologically modified regions on the generation and properties of spatio-temporal activation patterns was investigated on the basis of the FitzHugh-Nagumo equations with von Neumann boundary conditions, in particular the occurrence of ectopic foci and spiral waves under spatial inhomogeneities of the parameters characterizing the cell properties. It was shown that the reduction ∆b of the resting state stability in circular regions of the tissue can lead to ectopic activity. A minimal size of hyperactive tissue is necessary for ectopic activity to occur, as well as a minimal strength of the reduction of resting state stability with respect to the "healthy" reference value. With increasing size ξb of the hyperactive tissue, the frequency of the ectopic focus first increases and eventually saturates. The saturation frequency depends on the strength of the modification ∆b. For spiral wave patterns it was found that an anchoring of the wave to the obstacle can occur. To uncover this mechanism, an obstacle was modeled as a patch of modified tissue with reduced excitability by a reduction of the parameter c in the FHN equations. The obstacle was placed in the middle of a two-dimensional square simulation area and a planar excitation wave was generated aside of the obstacle in front of a refractory region, which represents an activation pattern observed after cross-field stimulation in experiments. As in the experiments, reentrant waves are observed. These exhibit either functional or anatomical reentry in dependence of the obstacle size and reduction strength ∆c of excitability. An analysis of the spiral wave frequency in dependence of the obstacle size yields results in accordance with the experimental observations. Finally we studied the question, if and how fibrillatory-like states can arise in the right 10 atrium due to the presence of self-excitatory spiral waves or ectopic foci in the left atrium. To this end the simulation area was separated into two rectangular regions L and R connected by a small bridge B. Planar excitation waves were generated with different frequencies in the left region L to model a pacemaker far outside the left part of the simulation area. Planar excitation waves resembling stimulation by the sinus node were generated by periodic application of a stimulating current at one boundary in the right part R of the simulation area. For small perturbation frequencies in L, the disturbance of the waves in R turned out to be small. For higher perturbation frequencies , the waves in R become significantly disturbed and the spatio-temporal activation pattern eventually becomes irregular. The time evolution of the activation variable u, representing the electric potential in the FHN equations, shows features in close resemblance to the ones found in intra-atrial electrocardiogram recordings during fibrillation in the right atrium. The spatial variation of the excitation frequency was quantified in terms of an entropy, which showed, for a given pacing frequency, a maximum as a function of the perturbation frequency. Further investigations will focus on the influence of the geometry of the bridge and the wavefronts as well as analyse the behavior for different pacing frequencies. The reliability of the measure of irregularity s should be analysed and if necessary other methods to characterise the system behavior should be searched. Acknowledgments C. L. thanks the Thuringian government for financial support. 11 [1] V. Fuster, L. E. Ryden, D. S. Cannom, H. J. Crijns, A. B. Curtis, K. A. Ellenbogen, J. L. Halperin, J.-Y. Le Heuzey, G. N. Kay, J. E. Lowe, S. B. Olsson, E. N. Prystowsky, J. L. Tamargo, S. Wann, S. C. Smith, A. K. Jacobs, C. D. Adams, J. L. Anderson, E. M. Antman, S. A. Hunt, R. Nishimura, J. P. Ornato, R. L. Page, B. Riegel, S. G. Priori, J.-J. Blanc, A. Budaj, A. J. Camm, V. Dean, J. W. Deckers, C. Despres, K. Dickstein, J. Lekakis, K. McGregor, M. Metra, J. Morais, A. Osterspey, J. L. Zamorano, Europace 8, 651 (2006). [2] S. Nattel, Nature 415, 219 (2002). [3] S. Nattel, D. Li, L. Yue, Annu. Rev. Physiol. 62, 51 (2000). [4] S. Nattel, L. H. Opie, Lancet 367, 262 (2006). [5] R. Mandapati, A. Skanes, Y. Chen, O. Berenfeld, J. Jalife, Circulation 101, 194 (2000). [6] T.-J. Wu, M. Yashima, F. Xie, C. A. Athill, Y.-H. Kim, M. C. Fishbein, Z. Qu, A. Garfinkel, J. N. Weiss, H. S. Karagueuzian, P.-S. Chen, Circ. Res. 83, 448 (1998). [7] J. Sahadevan, K. Ryu, L. Peltz, C. M. Khrestian, R. W. Stewart, A. H. Markowitz, A. L. Waldo, Circulation 110, 3293 (2004). [8] P. Sanders, O. Berenfeld, M. Hocini, P. Jais, R. Vaidyanathan, L.-F. Hsu, S. Garrigue, Y. Takahashi, M. Rotter, F. Sacher, C. Scavee, R. Ploutz-Snyder, J. Jalife, M. Haisaguerre, Circulation 112, 789 (2005). [9] R. FitzHugh, Biophys. J. 1, 445 (1959). [10] S. Lazar, S. Dixit, F. E. Marchlinski, D. J. Callans, E. P. Gerstenfeld, Circulation 110, 3181 (2004). [11] A. L. Hodgkin, A. F. Huxley, J. Physiol. 117, 500 (1952). [12] S. M. Tobias, E. Knobloch, Phys. Rev. Lett. 80, 4811 (1998). [13] M. Bar, M. Eiswirth, Phys. Rev. E 48, R1635 (1993). [14] S. Jakubith, H. H. Rotermund, W. Engel, A. von Oertzen, G. Ertl, Phys. Rev. Lett. 65, 3013 (1990). [15] G. Ertl, Science 254, 1750 (1991). [16] M. G. Clerc, D. Escaff, V. M. Kenkre, Phys. Rev. E 72, 056217 (2005). [17] K. J. Lee, E. C. Cox, R. E. Goldstein, Phys. Rev. Lett. 76, 1174 (1996). [18] J. D. Murray, Mathematical Biology, Springer (1988). 12 [19] L. S. Schulman, P. E. Seiden, Science 233, 425 (1986). [20] V. Mendez, J. E. Llebot, Phys. Rev. E 56, 6557 (1997). [21] T. Ikeda, M. Yashima, T. Uchida, D. Hough, M. C. Fishbein, Circ. Res. 81, 753 (1997). [22] S. Iravanian, Y. Nabutovsky, C.-R. Kong, S. Saha, N. Bursac, L. Tung, Am. J. Physiol. Heart Circ. Physiol. 285, H449 (2003). [23] T. Ikeda, L. Czer, A. Trento, C. Hwang, J. J. C. Ong, D. Hough, M. C. Fishbein, W. J. Mandel, H. S. Karagueuzian, P.-S. Chen, Circulation 96, 3013 (1997). [24] Z. Y. Lim, B. Maskara, F. Aguel, R. Emokpae, Jr., L. Tung, Circulation 114, 2113 (2006). [25] A. M. Pertsov, J. M. Davidenko, R. Salomonsz, W. T. Baxter, J. Jalife, Circ. Res. 72, 631 (1993). [26] L. Boersma, J. Brugada, C. Kirchhof, M. Allessie, Circulation 89, 852 (1994). [27] G. V. Obisov, A. T. Stamp, J. J. Collins, Control of Oscillations and Chaos (Proc.) 3, 453 (2000). [28] J. M. Davidenko, R. Salomonsz, A. M. Pertsov, W. T. Baxter, J. Jalife, Circ. Res. 77, 1166 (1995). [29] K. Agladze, M. W. Kay, V. Krinsky, N. Savazyan, Am. J. Physiol. Heart Circ. Physiol. 293, H503 (2007). [30] Y.-Q. Fu, H. Zhang, Z. Cao, B. Zheng, G. Hu, Phys. Rev. E 72, 046206 (2005). [31] G. Gottwald, A. Pumir, V. Krinsky, Chaos 11, 487 (2001). [32] R. Otnes, L. Enochson, Digital Time Series Analysis, John Wiley & Sons (1972) [33] M. Rosenblum, A. Pikovsky, J. Kurths, C. Schafer, P. A. Tass, Handbook of Biological Physics 4, 279 (2001). [34] D. Cysarz, H. Bettermann, P. Van Leeuwen, Am. J. Physiol. Heart Circ. Physiol. 278, H2163 (2000). 13 a) b) 2 1 0 -1 u v 0.6 0.4 0.2 0 -0.2 -0.4 -0.6 0 2 4 6 simulation time 8 10 12 14 FIG. 1: Time evolution of a) u and b) v calculated with the FHN equations and the parameters representing a healthy tissue. 14 a) y 16 12 8 4 b) 0,5 0,4 b ∆ 0,3 0,2 4 8 x 12 16 0,5 1 ξ b 1,5 2 2,5 FIG. 2: (color online) a) Activation pattern of the ectopic focus at ∆b = 0.4 and ξb = 0.8. The lines represent isolines for u = −0.8 and four times t = 5 (black solid line), 7 (red dotted line), 9 (blue dashed line), and 13 (green dash-dotted line). Initially (t = 0) the system is in the resting state (u = u0, v = v0). b) Phase diagram of ectopic activity for modifications according to Eq. (2). Red diamonds and black circles refer to the occurrence and absence of ectopic activity, respectively. The dotted line is drawn as a guide to the eye and marks the transition between the regions of ectopic activity and absence of self-excitatory behavior. 15 u 2 1 0 -1 -2 0,5 v 0 -0,5 0 10 20 t 30 40 50 FIG. 3: (color online) Time evolution of u and v for ∆b = 0.4, and two values ξb = 0.8 (black solid line) and ξb = 0.5 (red dashed line). For ξb = 0.8, u decreases below the threshold value uth ≃ 0.6 and an action potential is generated. For ξb = 0.5 the decrease of u is slower and less steep due to the stronger diffusive current compared to ξb = 0.8. As a consequence the response of u is more susceptible to the initial decrease of v, which is not distinguishable for ξb = 0.5 and ξb = 0.8. Thus, u does not reach the threshold value and relaxes with small oscillations to a value u ≃ 0.9. 16 2 f 0 1 6,25 5,75 5,25 4,75 ∆b = 0.5 ∆b = 0.45 ∆b = 0.4 ∆b = 0.35 ∆b = 0.3 ∆b = 0.25 0.5 1 1.5 ξ b 2 2.5 FIG. 4: (color online) Frequency of ectopic activity in dependence of the size ξb of the modified tissue for various modification strength ∆b of the resting state stability. A frequency of 0.053 in the simulation corresponds to a frequency of roughly 10 Hz. 17 a) 14 12 y 10 8 6 6 b) 14 12 y 10 8 6 6 ξ 8 10 x 12 14 ξ 8 10 x 12 14 FIG. 5: a) Isolines for u = −0.8 at four different times t = 82 (black solid line), 87 (red dotted line), 92 (blue dashed line) and t = 97 (dash-dotted line) for an obstacle with ξc = 2 and ∆c = 4.5 (marked by the black circle). The spiral wave is pinned and rotates around the obstacle (anatomical reentry). b) Isolines for u = −0.8 at four different times t = 82 (black solid line), 84.5 (red dotted line), 87 (blue dashed line) and t = 89.5 (dash-dotted line) for an obstacle with ξc = 2 and ∆c = 1.5. The spiral wave rotates around a moving center not corresponding to but influenced by the obstacle and is not attached to the obstacle (functional reentry). 18 ∆c = 4.5 ∆c = 3.5 ∆c = 2.5 ∆c = 1.5 30 f / 1 20 10 1 2 ξ c 3 4 FIG. 6: (color online) Mean cycle length 1/f as a function of obstacle size ξc for four different strength of the modification of excitability ∆c. A cycle length of 15 corresponds to a frequency of roughly 12 Hz. For the two smaller values ∆c = 1.5 and ∆c = 2.5 the cycle length is independent of ξc (functional reentry). For the two larger values ∆c = 3.5 and ∆c = 4.5 a transition from functional to anatomical reentry occurs, when ξc exceeds a threshold value ξ⋆ c that increases with decreasing ∆c. The dashed lines are fits to the data for mean cycle lengths independent of ξc (functional reentry) and mean cycle lengths proportional to ξc (anatomical reentry). 19 y 10 8 6 4 2 0 0 P2 P3 P1 5 10 x 15 20 FIG. 7: Illustration of the simulation area. The red dotted line describes a regular paced wavefront of the sinus node. The blue dashed line is a wavefront of the perturbing pacemaker. P1, P2 and P3 mark the "observation points", for which the time evolution of u is shown in Figure 9. 20 a) t = 120 t = 124 b) t = 184 t = 188 t = 128 t = 132 t = 192 t = 196 FIG. 8: Time evolution of excitation pattern observed by perturbation of a regular pacemaker with frequency fpace = 0.091 in the right part R of the simulation area by an additional pacemaker with frequency fpert = 0.105 in the left part L through a bridge region B. The blue solid lines are isolines for u = −0.8. The shaded regions represent the boundaries of the bridge between the two parts L and R. a) For small times the excitation in R is regular. b) At later times breakups of waves in region R occur close to the bridge region B, resulting in irregular excitation patterns. a) u b) u 2 1 0 -1 -2 2 1 0 -1 -2 150 160 170 t 180 190 200 FIG. 9: (color online) Time evolution of the activation variable u at three different points of the simulation area: P1 = (11.49, 5.34) (black solid line), P2 = (11.33, 6.34) (red dashed line) and P3 = (11.94, 6.15) (blue dash-dotted line) for a pacing frequency fpace = 0.091 and two different perturbation frequencies a) fpert = 0.1, and b) fpert = 0.105. We have checked that this behaviour remains qualitatively the same even at a three times longer simulation time which suggests that this time evolution corresponds to a stationary state. 21 s 0.32 0.3 0.28 0.26 0.09 0.095 0.1 0.105 fpert 0.11 0.115 FIG. 10: (color online) Normalized entropy of the local frequency distribution in region R as a function of the frequency fpert of the perturbing waves in region L. The pacing frequency in region R is fpace = 0.091. The dotted (black) line through the data points is a guide to the eye and the dashed line marks the value of s for the regularly paced system without perturbation by an additional pacemaker. 22
1705.05274	1	1705	2017-05-15T14:48:32	Anomalous behavior of membrane fluidity caused by copper-copper bond coupled phospholipids	[ "physics.bio-ph" ]	Membrane fluidity, well-known to be essential for cell functions, is obviously affected by copper. However, the underlying mechanism is still far from being understood, especially on the atomic level. Here, we unexpectedly observed that a decrease in phospholipid (PL) bilayer fluidity caused by Cu2+ was much more significant than those induced by Zn2+ and Ca2+, while a comparable reduction occurred in the last two ions. This finding clearly disagrees with the placement in the periodic table of Cu just next to Zn and far from Ca. The physical nature was revealed to be a special attraction between Cu+ cations, which can induce a motif forming of two phospholipids coupled by Cu-Cu bond (PL-diCu-PL). Namely, upon Cu2+ ion binding to a negatively charged phosphate group of lipid, Cu2+ was reduced to Cu+. The special attraction of the cations then caused one Cu+ ion simultaneously binding to two lipids and another Cu+, resulting in the formation of PL-diCu-PL structure. In contrast, this attraction cannot occur in the cases of Zn and Ca ions due to their electron structure. Remarkably, besides lipids, the phosphate group widely exists in other biological molecules, including DNA, RNA, ADP and ATP, which would also induce the similar structure of Cu ions with the molecules. Our findings thus provide a new view for understanding the biological functions of copper and the mechanism underlying copper-related diseases.	physics.bio-ph	physics	Anomalous behavior of membrane fluidity caused by copper-copper bond coupled phospholipids Xiankai Jiang1†, Jinjin Zhang2†, Bo Zhou3, Xiaojuan Hu2, Zhi Zhu1, Chao Chang4, Junhong Lü2, Bo Song1 1Terahertz Technology Innovation Research Institute, Shanghai Key Lab of Modern Optical System, Terahertz Science Cooperative Innovation Center, School of Optical-Electrical Computer Engineering, University of Shanghai for Science and Technology, Shanghai 200093, China 2Division of Physical Biology and CAS Key Laboratory of Interfacial Physics and Technology, Shanghai Institute of Applied Physics, Chinese Academy of Science, Shanghai 201800, China 3School of Electronic Engineering, Chengdu Technological University, Chengdu 611730, China 4Key Laboratory for Physical Electronics and Devices of the Ministry of Education, Xi'an Jiaotong University, Xi'an 710049, China †These authors contributed equally to this work. Correspondence to: [email protected], [email protected] Keywords: Attraction between Cu ion, membrane fluidity, phospholipid, Cu-lipid motif Abstract Membrane fluidity, well-known to be essential for cell functions, is obviously affected by copper. However, the underlying mechanism is still far from being understood, especially on the atomic level. Here, we unexpectedly observed that a decrease in phospholipid (PL) bilayer fluidity caused by Cu2+ was much more significant than those induced by Zn2+ and Ca2+, while a comparable reduction occurred in the last two ions. This finding clearly disagrees with the placement in the periodic table of Cu just next to Zn and far from Ca. The physical nature was revealed to be a special attraction between Cu+ cations, which can induce a motif forming of two phospholipids coupled by Cu-Cu bond (PL-diCu-PL). Namely, upon Cu2+ ion binding to a negatively charged phosphate group of lipid, Cu2+ was reduced to Cu+. The special attraction of the cations then caused one Cu+ ion simultaneously binding to two lipids and another Cu+, resulting in the formation of PL-diCu-PL structure. In contrast, this attraction cannot occur in the cases of Zn and Ca ions due to their electron structure. Remarkably, besides lipids, the phosphate group widely exists in other biological molecules, including DNA, RNA, ADP and ATP, which would also induce the similar structure of Cu ions with the molecules. Our findings thus provide a new view for understanding the biological functions of copper and the mechanism underlying copper-related diseases. Significance Statement Membrane fluidity, critically essential for cell functions, is obviously affected by copper, but the underlying molecular mechanism is poorly understood. We propose that a special attraction between Cu+ ions can cause a motif forming of two phospholipids (PLs) coupled by a Cu-Cu bond (PL-diCu-PL), following Cu2+ reduced to Cu+ upon binding to PLs. The resulted motif of PL-diCu-PL plays an unusual role in the fluidity of lipid bilayer. Our findings provide a new view for understanding biological functions of copper. Introduction Proper fluidity of the biological membrane is critically essential for numerous cell functions, such as adapting to the thermal stress of the environment of the microorganism (1), the binding of peripheral proteins associated at the lipid surface (2), reaction rates of enzymes (1), and even cell signaling and phagocytosis (3). Both in-vivo and in-vitro evidences have indicated that copper, as a biologically trace element, plays an important role in the membrane, especially in regard to its fluidity (4-8). However, the underlying mechanism is still far from being understood, partially because researches have been majorly devoted to the interactions of alkali and alkaline earth metal ions with phospholipids as well as the influences on the lipid bilayer (9-18). Traditionally, the effect of metal ions on the membrane was majorly attributed to the electrostatic attraction with lipid headgroups (19). This can explain the impact of divalent metal ions on membrane fluidity more than that of monovalent ones but cannot be applied to the differences of the influences of divalent ions, such as Ca2+, Mg2+, Zn2+ and Cu2+. Recently, Cremer and his coworkers studied effects of the Cu2+ ion on a bilayer comprised of both phosphatidylcholine (PC) and phosphatidylserine (PS), and proposed that the ion was specifically bound to PS (20). Meanwhile, this binding was only stable under basic conditions, but not at acidic pH values. Further investigations suggested that a complex of Cu(PS)2 formed upon Cu2+ binding to PS molecules, which did not alter the net negative charge on the membrane (21). This differed from the manner and impact of Ca2+ or Mg2+ binding. Very recently, it was determined that the cis isomer of the Cu(PS)2 complex was preferred to the trans one (22). Additionally, a synergetic effect of Cu2+ with Ca2+ were proposed, which potentially triggered the transition of PS membrane from fluid phase to soft solid phase (23). Besides those, influences of the Cu2+ ion on a bilayer consisting of PC and phosphatidylethanolamine (PE) have also been explored (24). It was suggested that Cu2+ could stably bind to the amine moieties of PE lipids, while other transition metal ions to PE bound in a similar manner. Noticeably, all these Cu2+-lipid interactions specifically relate to the existence of amine moiety in the headgroup of the lipid. Here, we propose an amine-independent copper-phospholipid motif in the membrane and apply it to illuminate our measurements of the anomalous effect of copper on the fluidity of a bilayer composed of PC and phosphatidylglycerol (PG). We observed that the decrease of the PC/PG bilayer fluidity caused by Cu2+ ions was much more significant than those induced by Zn2+ and Ca2+ ions, while a comparable reduction occurred in the last two cases. A model of two phospholipids coupled by a Cu-Cu bond (diCu) was built to explain the unexpected behavior of the bilayer fluidity induced by the Cu2+ ion. Namely upon the interaction of two Cu2+ ions with two phospholipids, one ion preferred simultaneously binding with the phosphate groups of the lipids and another ion after the Cu2+ ions were reduced to Cu+. The underlying physics was then revealed to be an anomalous 3d10-3d10 attraction between Cu+ cations, which was resulted from a special 3d10 closed shell of the outermost electron structure in Cu+. In contrast, this attraction cannot occur in the cases of Zn and Ca ions due to their electron structures. Moreover, Ångström-resolution atomic force microscope (AFM) imaging also supported the formation of diCu coupled to two lipids. Results and Discussions Cu2+ ion caused anomalous fluidity of PC/PG bilayer. The fluorescence recovery after photobleaching (FRAP) method involves the production of a concentration gradient of fluorescent molecules by irreversibly bleaching a portion of fluorophores in the observed region. The disappearance of this gradient over time is an indicator of the mobility of the fluorophores in the membrane as the fluorophore diffuses from the adjacent unbleached regions of the membrane into the bleached zone. We chose to use PC/PG/NBD (73:25:2) supported lipid bilayers for the FRAP measurements due to their large size, which allows their visualization in the microscopic field. The supported lipid bilayers appear large and uniformly fluorescent when observed through the confocal microscope. The inset in Fig. 1A shows fluorescence images of a representative FRAP experiment performed on such lipid bilayers: the dark circular region represents the bleached spot immediately after bleaching (0 s, bleach) and after the recovery of fluorescence at 80 s and 160 s (postbleach). The scanning parameters for all FRAP experiments were optimized to ensure no significant fluorescence photobleaching due to repeated imaging. Nonlinear curve fitting analysis of NBD fluorescence recovery kinetics after bleaching from the experimental data using the equations described in previous studies (25, 26) in the absence and presence of a 20 mM CuCl2/ZnCl2/CaCl2 treatment on PC/PG bilayer mixtures are shown in Fig. 1A. The corresponding lateral diffusion rates of the NBD probe evaluated from Fig. 1A are displayed in Fig. 1B. The rate was 0.64 ± 0.03 for the sample incubated with CuCl2, 1.12 ± 0.03 for ZnCl2, 1.24 ± 0.03 for CaCl2, and 1.80 ± 0.11 for the control (incubated with NaCl2). These data indicate that the interaction of Cu2+/Ca2+/Zn2+ ions with a PC/PG bilayer suppress the motion of the membrane. This can be attributed to the fact that divalent metal ions bind with the negatively charged lipid PG more stably than Na+ (the control) (19), hence decreasing the electrostatic repulsion of the lipids, resulting in a close packing of lipid molecules. Subsequently, the constraints imposed on the displacement of lipids due to their enhanced order in the presence of metal ions in membranes lead to a reduced rate of lateral diffusion. Remarkably, an unexpected order (Fig. 1B) of metal-ion impacts on the diffusion rate of lipids in the membranes was observed: Cu2+ > Zn2+ ~ Ca2+ > Na+ (control). This indicates that Cu2+ plays a distinguished role in suppressing the mobility of lipids, more than the other ions, while a comparable effect occurs with the incubations of Zn2+ and Ca2+. This result is obviously inconsistent with the placement in the periodic table of Cu just next to Zn and far from Ca. Analyses of mechanism underlying the anomalous fluidity of the bilayer. To illustrate the mechanism under the anomalous influence of the Cu2+ ion on the phospholipid membrane, we have studied the interactions of two Cu ions with two PG molecules. H3C-[PO4]--CH3, with a phosphate group as a large portion of its composition, was employed as a simplified model of the phospholipid (PL). [Cu(H2O)5]2+ was applied to simulate the Cu ion because it was found to coordinate five water molecules in solution (27). First, two hydrated Cu2+ ions interact with two phospholipids, respectively. We called the resultant state "State I" (Fig. 2C), denoted by [PL-Cu(aq)]+, in which the hydrated Cu cation bound the negatively charged oxygen atom in the phosphate group of the lipid. The label aq stands for the water molecules in the hydrated group. Using Eq. 1, the binding energy of a Cu2+ ion in this state was calculated by an ab initio method based on density functional theory (DFT) with the solvation effect of the outer water environment, E I binding (Cu) = = 1   2 ([PL-Cu(aq)] ) E + − E − (PL ) − E (Cu(aq) + 2 ). 2 ([PL-Cu(aq)] ) 2 (PL ) 2 (Cu(aq) E E E + − − − + 2 )   (1) E(PL-), E(Cu(aq)2+) and E([PL-Cu(aq)]+) indicate energies of the phospholipid, hydrated Cu2+ ion and their binding state I, respectively. As presented in Fig. 3A, the binding strength reached -37.14 kcal/mol, meaning that the hydrated Cu ion can bind to the oxygen of the phosphate group. Natural-bond-orbital (NBO) analysis (28) showed that the Wiberg bond order (29) was 0.374 for the Cu-O bond (Table S1 in SI), suggesting a chemical bond is occurring with a few covalent characteristics. This chemical bond can consequently provide the stable binding of the Cu ion with the O of phospholipid in solution. Second, two hydrated Cu2+ ions simultaneously bind with two phospholipids, resulting in a diCu coupled lipid pair (Fig. 2D). The resulting conformation, referred to "State II", had two positive charges (denoted by [PL-diCu(aq)-PL]2+). The optimized Cu-O and Cu-Cu bond lengths in the resulting PL pair were 1.90 Å and 2.58 Å, respectively. The frequencies of Cu-related vibration modes were 0.4 × 1010 Hz, 1.7 × 1010 Hz and 3.1 × 1010 Hz. The binding energy of a Cu2+ ion in State II was calculated via Eq. 2 as follows, E II binding (Cu) = = E di ([PL- Cu(aq)-PL] 2+ 1   2 ([PL- Cu(aq)-PL] E 2+ di ) / 2 + 2 )   (2) − E − ) 2 (PL ) 2 (Cu(aq) − (Cu(aq) − (PL ) E E E + 2 ), − − where E([PL-diCu(aq) -PL]2+) denotes the energy of the copper-phospholipid complex in State II. Surprisingly, the binding strength of a Cu ion in PL-diCu-PL reached -38.88 kcal/mol, which is surprisingly greater than the strength of -37.14 kcal/mol in State I. Moreover, the Wiberg bond order was 0.421 for the Cu-Cu binding in State II (Fig. 2B), suggesting that a chemical bond with a definite covalent characteristic is formed (more information shown in the following part with Fig. 4). All of these results suggest that the Cu ions can also bind to the phospholipids in State II, similar to in State I, and even more stably than in State I with the presence of water. It should be noted that State II [PL-diCu-PL]2+ can be taken as the coupling of two groups [PL-Cu]+ (the structure in State I) (see Fig. 2C,D). The positive charges of the two groups would cause a Coulomb repulsion and subsequently hinder the formation of State II. Hence, we performed an electron-structure analysis to reveal the physics underlying the formation of State II. The Cu ion in the structure [PL-diCu-PL]2+ had an NBO charge of only +0.97 e, indicating that the Cu2+ ion is reduced to Cu+ with an electron configuration of [Ar]3d104s0 upon binding to the negatively charged phospholipid. Additionally, electrons were observed at the midpoint of two Cu ions with a density of 0.043 e/Å3 (Fig. 4A left, and Table S2 in SI). The pattern of electron localized function (ELF) was approximately square in the region close to the Cu ion (the yellow squares under the yellow balls in the lower-left region of Fig. 4B), suggesting that the outermost electrons majorly occupy the 3d orbital. The ELF value was 0.254 at the midpoint of two Cu ions (Fig. 4B upper-left), indicating that there exists an electron pair localized in this area with a probability of 25.4%. Further NBO analysis showed a chemical bond clearly occurring between the two Cu ions (Fig. 4C left). The bond was composed of 50% of the valance orbitals of each Cu ion, in which the ratio of Cu 3d orbital reached 84.3%. Thus, the 3d electrons of the Cu ions substantially contributed to the Cu-Cu bond. All of these results suggest that a strong 3d10-3d10 attraction (30, 31) occurs between the ions after the Cu2+ is reduced, upon binding to the negatively charged phosphate group of lipid. Therefore, it is this attractive force that suppresses the Coulomb repulsion between two [PL-Cu]+, resulting in the formation of [PL-diCu-PL]2+. Notably, the 3d10-3d10 attraction of the Cu+ ions has already been reported and applied in inorganic materials (32-34). For comparison, we have also studied the binding characteristics of other metal ions M (M = Na+, K+, Mg2+, Ca2+, Zn2+ and Ag2+) with H3C-[PO4]--CH3 fragments. It was found that only Ag2+ presented the same behavior as Cu2+ upon binding to the phospholipids, while the others did not show this manner. We first calculated the binding energies of hydrated M ions with H3C-[PO4]--CH3 in States I and II, respectively. The results are presented in Fig. 2A. For State I, the binding energies were -23.14 kcal/mol for Na+(aq), -22.37 kcal/mole for K+(aq), -33.86 kcal/mol for Mg2+(aq), -26.14 kcal/mol for Ca2+(aq), -33.29 kcal/mol for Zn2+(aq), and -38.94 kcal/mol for Ag2+(aq). These results indicate that all of these ions can bind to the phosphate group of a lipid in State I, and the binding of divalent ions is more stable than that of monovalent ions. For State II, negative values were observed only in the binding energies of the hydrated Ag2+ ion (-45.10 kcal/mol) and Zn2+ ion (-22.13 kcal/mol), and not in the other ions. Moreover, a larger binding strength in State II than in State I occurred with the Ag2+ ion but not with the Zn2+ ion. Additionally, the Wiberg bond order of M-M binding was 0.337 for the Ag ions in State II and was less than 0.1 for the others, suggesting that a chemical bond only occurs in a PL-diAg-PL structure with a covalent characteristic, but not in the cases of other metal ions. Therefore, with the competition of State I, State II is stable only for Ag2+ ions in the presence of water, but not for the other ions. The NBO charge of the Ag2+ ion was +0.93 in State II, denoting that Ag2+ is reduced to Ag+ upon binding to the phosphate group of lipid, and then, a closed shell [Kr]4d105s0 of electron configuration occurs in the Ag ion. Therefore, the formation of the Ag+-Ag+ chemical bond can be attributed to the 4d10-4d10 attraction, similar to the case of [PL-diCu-PL]2+. We thus conclude that the similar behavior of Ag ion and Cu ion in State II results from the structure of valence electrons, which is the same for Group 11 metals. This special property consequently leads to the distinguished difference of these ions with Zn2+, Ca2+, Mg2+, K+ and Na+ ions upon binding to phospholipids. Based on the previous DFT calculations, we have improved the CHARMM36 force field specifically for Cu-Cu and Cu-O bindings in PL-diCu-PL structure (35) (detailed information presented in Table S3 and the corresponding discussion of SI) and performed classical molecular dynamics simulations (MD) to explore the influences of the Cu-containing structure on lipid bilayer. A molecular model, consisting of two PG molecules coupled by two Cu+ ions (Fig. 5A), was employed in the MD. The positive charge of the PL-diCu-PL structure was neutralized by Cl- ions in the presence of water molecules. Through a 1.0-μs simulation of each sample, the lipids assembled to form a pattern of stripes (Fig. 5B-C). A segment of assembled PL-diCu-PL structures are shown in Fig. 5D. Two Cl- ions were observed over and under the Cu-O plane, respectively, indicating that Cl- ions are involved in the assembly of the membrane due to an electrostatic attraction. To study the order of the degree of assembled lipids in a membrane, we applied three parameters: the angles of O-O direction in phospholipid along the x (θ) and z (φ) axes (upper-left inset of Fig. 5E) for the orders parallel to the membrane surface and the position (dz) of the phosphorus atom in the membrane along the z-axis for the order perpendicular to the surface. The coordinate origin of the z-axis was set at the midpoint of two layers. First, in the distribution of lipids according to the parameters θ and φ with presence of Cu ions (the orange curves in Fig. 5E), two peaks were observed at θ = ±60.5°, and φ = ±42.5°. In the distribution according to dz, two peaks were located at approximately ±2.0 nm. The plus/minus symbols above resulted from two opposite O-O directions in the PL-diCu-PL structure (Fig. 5A) for the cases of the angles θ and φ and from the two layers of the membrane for dz. In contrast, without the presence of Cu ions, there was no peak for θ, and a broad peak for φ (at -90° or +90°, due to a periodic condition), clearly suggesting that the lipids are out of order in the direction parallel to the membrane surface. The full-width half-maximum (FWHM) of the peak was then employed to quantitatively study the order. The FWHM of the θ-related peak was 180° (no peak) without Cu ions and 42° with Cu ions. The reduction from 180° to 42° obviously indicates that Cu ions induce a change in lipids from a disordered state to an ordered state. Similar observations occurred for the other parameters. The FWHM of φ-related peak decreased from 63° (without Cu ions) to 25° (with Cu ions), and the FWHM of dz-related peak was reduced from 0.7 nm (without Cu ions) to 0.4 nm (with Cu ions). Moreover, the narrow peaks (FWHM = 25°) at φ = ±42.5° with Cu ions further suggested that for the assembled lipids, most of the Cu-O planes were parallel to each other, with an angle of approximately 42.5° (Fig. 5D). We thus conclude that Cu ions can clearly enhance the order of lipids in a membrane, especially in the directions parallel to the membrane surface. Roughness of the membrane surface has also been calculated by the mean square deviation (MSD) of phosphorus-atom displacements along the z-axis of the membrane. The values in the presence and absence of Cu2+ ions were 0.20 ± 0.02 nm and 0.26 ± 0.04 nm, respectively (Fig. 5F). The significant difference between them obviously suggests that Cu ions suppress the roughness of membrane. Moreover, AFM measurements further supported the theoretical results above. As shown in Fig. 5G-H, a pattern of clear stripes was observed in the CuCl2-incubated PC/PG bilayer and not in the control (incubated with NaCl2). The roughness of the membrane was 0.12 ± 0.01 nm for the presence of CuCl2 and 0.11 ± 0.08 nm in its absence, and the P value between them was 0.015. The AFM data above indicate that Cu ions can significantly suppress the roughness of a lipid bilayer. It is noted that the difference of roughness values between AFM and MD results can be attributed to that a substrate for lipid bilayer was applied in AFM measurements but not in MD simulations. From both the theoretical and experimental analyses, we conclude that a pattern of stripes can be assembled in the membrane with help of the special Cu-lipid interaction and the Cl ions involved. This significantly enhanced order then imposes a strong constraint on displacement of lipid molecules in the bilayer, clearly reducing the fluidity of membrane, which results in the anomalous effect of CuCl2 previously observed by FRAP measurements. Conclusion In summary, we proposed a motif of diCu coupled phospholipids, which can illuminate the anomalous decrease of membrane fluidity caused by Cu2+, compared to those by Zn2+ and Ca2+. The mechanism under the motif formation was further revealed. Upon the Cu2+ ion interacting with the lipid, Cu2+ was reduced to Cu+. After that, one Cu+ ion preferred simultaneous binding to two phospholipids and another Cu+, due to the anomalous 3d10-3d10 attraction between the metal ions. In contrast, this attraction cannot occur in the cases of Zn and Ca ions due to their electronic structures. It is worth noting that besides lipids, the phosphate group also widely exists in other biological molecules, such as DNA, RNA, ADP, ATP, and enzymes. Therefore, as a kernel of the motif, the structure of diCu coupled phosphate groups and the anomalous Cu-Cu attraction (see Fig. 2D) will provide a new direction for understanding the biological function of copper as a trace element essential to our life, as well as to the mechanisms of copper-related clinical diseases. Materials and Methods Preparation of lipid bilayers for confocal microscope imaging. The supported lipid bilayers were prepared from the negatively charged lipid DOPG (1,2-dioleoyl-sn-glycero-3-(phospho-rac-(1-glycerol))), neutral lipid DOPC (1,2-dioleoyl-sn-glycero-3-phosphocholine) and headgroup labeled NBD-PE (1,2-diphytanoyl-sn-glycero-3-phosphoethanolamine-N-(7-nitro-2-1,3-benzoxadiazol- 4-yl)). DOPC, DOPG and NBD-PE solutions in chloroform were mixed to achieve a DOPC/DOPG/NBD-PE molar ratio of 73:25:2; the solvent was evaporated under nitrogen and the dried lipid film was suspended in TBS buffer (50 mM Tris, 150 mM NaCl) to a concentration of 1 mM. The lipid suspension was then sonicated to clarity, yielding a suspension of small unilamellar vesicle liposomes. The small unilamellar vesicle suspension was then exposed to a clean glass surface (a microscope petri dish was first etched by plasma (Harrick, PDC-32G, 4 minutes in air/vacuum) then was cleaned by 1% hydrofluoric acid and thoroughly rinsed with deionized water and dried under nitrogen) and incubated for 1 hour at room temperature to form lipid bilayers. The excessive unfused liposomes were removed from rinsing with excess of the buffer. The 20 mM CuCl2/CaCl2/ZnCl2 solution was added to the lipid bilayers before imaging. Confocal microscope experimental setup and data acquisition. A commercial confocal microscope (Leica TCS SP5) was used for the fluorescence recovery after photobleaching (FPAR) measurements. A 488-nm Argon laser was used as the excitation source. The sample was illuminated and the fluorescence emission was collected by a 63× oil immersion objective. The dimensions of the acquired regions were 180 μm × 180 μm. The bleaching pulse was applied by rapidly scanning a focused laser beam over an area with a dimension of 8 μm × 8 μm for 30 s with an interval of 1s at full laser power. Immediately after bleaching, the region of 180 μm × 180 μm was recorded for 300 s with an interval of 5 s at low excitation energy. Ab initio calculations. Our ab initio calculations based on the density functional theory (DFT) as well as the electron structure analyses were implemented in the Gaussian09 package (36). The geometry optimizations and vibrational frequencies of all compounds were carried out at DFT level, employing the M06L functional (37). A mixed basis set GEN (SDD basis sets for Cu, Zn and Ag atoms, and 6-31+G(d,p) set for other atoms) was applied for all the calculations in this study. The optimized stationary points were identified as minima or first-order saddle points. Solvation effects of outer water environment were taken into account by calculating the single-point energies of the optimized configurations under the integral-equation- formalism polarizable continuum model (IEFPCM) of solvation (38) at the same level of theory as used in the gas-phase optimizations. To investigate the coordination effects on bond strength and charge distribution, the natural-bond-orbital (NBO) method was used for all complexes. Orbital populations and Wiberg bond orders were calculated with the NBO 3.0 program implemented in Gaussian 09. MD simulations. The initial configuration of lipid bilayer system was generated by MemBuilder server (39). A total of 128 DMPG lipid molecules were placed periodically in each lipid bilayer, and the number of water molecules per lipid was 45. According to the optimized structure obtained from the DFT calculations, the headgroup of two adjacent lipid molecules were linked by the Cu-O (0.1906 nm) and Cu-Cu (0. 2584 nm) bonds with the angle O-Cu-O (172.53°). The harmonic potential force constant for the bond Cu-O, Cu-Cu stretching and the bond-angle O-Cu-O vibration are 483660 kJ mol-1nm-2, 135450 kJ mol-1nm-2 and 800 kJ mol-1rad-2, respectively. A corresponding number of Cl- ions was added to neutralize the system. We performed the MD simulations for the system relaxation in an NPT ensemble at high temperature 320 K for 600 ns. After that, we selected one conformation per 20 ns in the time interval from 500 ns to 600 ns, and obtained five samples as initial structures. Finally, The MD simulations were performed in an NPT ensemble at the temperature of 303 K with 1.0-μs for each sample. All simulations were performed using GROMACS 5.1 (40) with a time step of 2 fs. The CHARMM36 force field for lipids (41, 42) and the CHARMM TIP3P water model (43) were used. The particle mesh Ewald (PME) method (44, 45) was used to treat long-range electrostatic interactions, whereas the van der Waals interactions were treated with a 1.0 nm - 1.2 nm force-based switching function (46). The temperature was maintained at 303 K using Nosé-Hoover thermostat (47, 48) with a coupling constant of 1 ps, and the pressure was kept constant at 1 bar using semi-isotropic Parrinello-Rahman barostat (49) with a coupling constant of 5 ps and a compressibility of 4.5 × 10−5 bar−1. Periodic boundary conditions were applied in the three directions. After a series of minimization and equilibration steps suggested by CHARMM-GUI (46), the data were collected every 2 ps during the next 1 μs production run. Preparation of Lipid Bilayers for AFM imaging. The negatively charged lipid, 1,2-Dimyristoyl-sn-glycero-3-phosphorylglycerol (DMPG), and neutral lipid, 1,2-Dimyristoyl-sn-glycero-3-phosphorylcholine (DMPC) were purchased from Avanti Polar Lipids (Alabaster, AL) and used without further purication. Lipid bilayers on freshly cleaved mica surface were prepared following the vesicle fusion method. Briefly, DMPC/DMPG (3:1) mixtures were first dissolved in chloroform, followed by evaporation of the solvent under nitrogen. After that lipid mixtures were dissolved in 50 mM Tris, 150 mM NaCl to a concentration of 1.5 mg/ml and sonicated in a bath sonicator until clear to form small vesicles. A 20 μL droplet of the vesicle solution was then applied to a freshly cleaved fragment of mica, incubated for about 2 h at room temperature, and then the sample was incubated at 35 ℃ for 40 min to fluidize the lipid, a necessary step to form the bilayer. AFM Imaging. The sample was placed within the AFM (Nano III, Veeco) and imaged in the contact mode using DNP tips (Bruker), with a spring constant of 0.06 N/m. The scan rate was 10 Hz and the applied force was minimized to about 0.1 nN. Acknowledgments. This work is supported by National Natural Science Foundation of China (Nos. 11474298, 11405250), Key Research Program of Frontier Sciences of the Chinese Academy Sciences (No. QYZDJ-SSW-SLH019) and the Tianjin Supercomputer Center of China. Author Contributions B.S. and J.L. conceived the idea and jointly supervised the project. J.L. designed the experiments. J.Z. and X.H. carried out the experiments, J.L. and B.S. performed the data analysis. The theoretical simulations and calculations were designed by B.S., carried out by X.J., B.Z, Z.Z. and B.S.. B.S. performed the theoretical analysis. C.C. performed some analysis. B.S. and J.L. co-wrote the paper. All authors discussed the results and commented on the manuscript. Competing interests: The authors declare no competing financial interests. References 1. Gennis RB (1989) Biomembranes: Molecular Structure and Function (Springer Verlag, New York). 2. Heimburg T, Marsh D (1996) Thermodynamics of the interaction of proteins with lipid membranes. Biological Membranes: A Molecular Perspective from Computation and Experiment, eds Merz KM, Roux B (Birkhäuser, Boston), pp 405-462. 3. Helmreich EJ (2002) Environmental influences on signal transduction through membranes: a retrospective mini-review. Biophys Chem 100(1):519-534. 4. Garcia JJ, Martínez-Ballarín E, Millán-Plano S, Allué JL, Albendea C, Fuentes L, Escanero JF (2005) Effects of trace elements on membrane fluidity. J Trace Elem Med Biol 19(1):19-22. 5. Suwalsky M, Ungerer B, Quevedo L, Aguilar F, Sotomayor CP (1998) Cu2+ ions interact with cell membranes. J Inorg Biochem 70(3):233-238. 6. Abu-Salah KM, Al-Othman AA, Lei KY (1992) Lipid composition and fluidity of the erythrocyte membrane in copper-deficient rats. Br J Nutr 68(2):435-443. 7. Rock E, Gueux E, Mazur A, Motta C, Rayssiguier Y (1995) Anemia in copper-deficient rats: role of alterations in erythrocyte membrane fluidity and oxidative damage. Am J Physiol 269(1):1245-1249. 8. Ohba S, Hiramatsu M, Edamatsu R, Mori I, Mori A (1994) Metal ions affect neuronal membrane fluidity of rat cerebral cortex. Neurochem Res 19(3):237-241. 9. Fukuma T, Higgins MJ, Jarvis SP (2007) Direct imaging of lipid-ion network formation under physiological conditions by frequency modulation atomic force microscopy. Phys Rev Lett 98(10):106101. 10. Sheikh KH, Giordani C, Kilpatrick JI, Jarvis SP (2011) Direct submolecular scale imaging of mesoscale molecular order in supported dipalmitoylphosphatidylcholine bilayers. Langmuir 27(7):3749-3753. 11. Ferber UM, Kaggwa G, Jarvis SP (2011) Direct imaging of salt effects on lipid bilayer ordering at sub-molecular resolution. Biophys Struct Mech 40(3):329-338. 12. Yi M, Nymeyer H, Zhou HX (2008) Test of the Gouy-Chapman theory for a charged lipid membrane against explicit-solvent molecular dynamics simulations. Phys Rev Lett 101(3):038103. 13. Khavrutskii IV, Gorfe AA, Lu B, Mccammon JA (2009) Free energy for the permeation of Na+ and Cl- ions and their ion-pair through a zwitterionic dimyristoyl phosphatidylcholine lipid bilayer by umbrella integration with harmonic fourier beads. J Am Chem Soc 131(5):1706-1716. 14. Gurtovenko AA, Vattulainen I (2008) Effect of NaCl and KCl on Phosphatidylcholine and Phosphatidylethanolamine Lipid Membranes: insight from Atomic-Scale Simulations for Understanding Salt-Induced Effects in the Plasma Membrane. J Phys Chem B 112(7):1953-1962. 15. Vácha R, et al. (2009) Effects of alkali cations and halide anions on the DOPC lipid membrane. J Phys Chem A 113(26):7235-7243. 16. Vácha R, et al. (2010) Mechanism of Interaction of Monovalent Ions with Phosphatidylcholine Lipid Membranes. J Phys Chem B 114(29):9504-9509. 17. Kagawa R, Hirano Y, Taiji M, Yasuoka K, Yasui M (2013) Dynamic interactions of cations, water and lipids and influence on membrane fluidity. J Membr Sci 435(10):130–136. 18. Bilkova E, Pleskot R, Rissanen S, Sun S, Czogalla A, Cwiklik L, Rog T, Vattulainen I, Cremer PS, Jungwirth P (2017) Calcium directly regulates phosphatidylinositol 4,5-bisphosphate headgroup conformation and recognition. J Am Chem Soc 139(11):4019-4024. 19. Sasaki DY (2003) Control of membrane structure and organization through chemical recognition. Cell Biochem Biophys 39(2):145-161. 20. Monson CF, Cong X, Robison A, Pace HP, Liu C, Poyton M F, Cremer PS (2012) Phosphatidylserine reversibly binds Cu2+ with extremely high affinity. J Am Chem Soc 134(18):7773-7779. 21. Xiao C, Poyton MF, Baxter AJ, Pullanchery S, Cremer PS (2015) Unquenchable surface potential dramatically enhances Cu2+ binding to phosphatidylserine lipids. J Am Chem Soc 137(24):7785-7792. 22. Kusler K, Odoh SO, Silakov A, Poyton MF, Pullanchery S, Cremer PS, Gagliardi L (2016) What is the preferred conformation of phosphatidylserine-copper(II) complexes? A combined theoretical and experimental investigation. J Phys Chem B 120(50):12883-12889. 23. Shirane K, Kuriyama S, Tokimoto T (1984) Synergetic effects of Ca2+ and Cu2+ on phase transition in phosphatidylserine membranes. Biochim Biophys Acta 769(3):596-600. 24. Poyton MF, Sendecki AM, Xiao C, Cremer PS (2016) Cu2+ binds to phosphatidylethanolamine and increases oxidation in lipid membranes. J Am Chem Soc 138(5):1584-1590 . 25. Pucadyil TJ, Mukherjee S, Chattopadhyay A (2007) Organization and dynamics of NBD-Labeled lipids in membranes analyzed by fluorescence recovery after photobleaching. J Phys Chem B 111(8):1975-1983. 26. Macháň R, Foo YH, Wohland T (2016) On the equivalence of FCS and FRAP: simultaneous lipid membrane measurements. Biophys J 111(1):152-161. 27. Pasquarello A, Petri I, Salmon PS, Parisel O, Car R, Toth E, Powell DH, Fischer HE, Helm L, Merbach A (2001) First solvation shell of the Cu(II) aqua ion: evidence for fivefold coordination. Science 291(5505):856-859. 28. Reed AE, Curtiss LA, Weinhold F (1988) Intermolecular interactions from a natural bond orbital, donor-acceptor viewpoint. Chem Rev 88(6): 899-926. 29. Wiberg KB (1968) Application of the pople-santry-segal CNDO method to the cyclopropylcarbinyl and cyclobutyl cation and to bicyclobutane. Tetrahedron 24(3): 1083-1096. 30. Pyykkö P, Runeberg N, Mendizabal F (1997) Theory of the d10–d10 closed-shell attraction: 1. dimers near equilibrium. Chem Eur J 3(9):1451-1457. 31. Pyykkö P (1997) Strong closed-shell interactions in inorganic chemistry. Chem Rev 97(3):597-636. 32. Katz MJ, Sakai K, Leznoff DB (2008) The use of aurophilic and other metal-metal interactions as crystal engineering design elements to increase structural dimensionality. Chem Soc Rev 37(9):1884-1895. 33. Liu JF, Min X, Lv JY, Pan FX, Pan QJ, Sun ZM (2014) Ligand-controlled syntheses of copper(I) complexes with metal-metal interactions: crystal structure and relativistic density functional theory investigation. Inorg Chem 53(20):11068-11074. 34. Willcocks AM, Robinson TP, Roche C, Pugh T, Richards SP, Kingsley AJ, Lowe JP, Johnson AL. (2012) Multinuclear copper(I) guanidinate complexes. Inorg Chem 51(1):246-257. 35. Jiang X, Gao J, Huynh T, Huai P, Fan C, Zhou R, Song B (2014) An improved DNA force field for ssDNA interactions with gold nanoparticles. J Chem Phys 140(23): 234102. 36. Frisch M, et al. (2009) Gaussian 09, Revision A, Gaussian. Inc., Wallingford CT. 37. Zhao Y, Truhlar DG (2006) A new local density functional for main-group thermochemistry, transition metal bonding, thermochemical kinetics, and noncovalent interactions. J Chem Phys 125(19):194101. 38. Scalmani G, Frisch MJ (2010) Continuous surface charge polarizable continuum models of solvation. I. General formalism. J Chem Phys 132(11):114110. 39. Ghahremanpour MM, Arab SS, Aghazadeh SB, Zhang J, van der Spoel D, (2014) MemBuilder: a web-based graphical interface to build heterogeneously mixed membrane bilayers for the GROMACS biomolecular simulation program. Bioinformatics 30(3):439-441. 40. Abraham MJ, et al. (2015) GROMACS: high performance molecular simulations through multi-level parallelism from laptops to supercomputers. SoftwareX 1:19-25. 41. Pastor RW,MacKerell AD (2011) Development of the CHARMM force field for lipids. J Phys Chem Lett 2(13):1526-1532. 42. Klauda JB, et al. (2010) Update of the CHARMM all-atom additive force field for lipids: validation on six lipid types. J Phys Chem B 114(23):7830-7843. 43. Jorgensen W L, Chandrasekhar J, Madura JD, Impey RW, Klein ML (1983) Comparison of simple potential functions for simulating liquid water. J Chem Phys 79(2):926-935. 44. Essmann U, et al. (1995) A smooth particle mesh Ewald method. J Chem Phys 103(19):8577-8593. 45. Darden T, York D, Pedersen L (1993) Particle mesh Ewald - an n.log(n) method for Ewald sums in large systems. J Chem Phys 98(12):10089-10092. 46. Lee J, et al. (2016) CHARMM-GUI input generator for NAMD, GROMACS, AMBER, OpenMM, and CHARMM/OpenMM simulations using the CHARMM36 additive force field. J Chem Theory Comput 12(1):405-413. 47. Hoover WG (1985) Canonical dynamics: equilibrium phase-space distributions. Phys Rev A 31(3):1695-1697. 48. Nose S (1984) A unified formulation of the constant temperature molecular-dynamics methods. J Chem Phys 81(1):511-519. 49. Parrinello M, Rahman A (1981) Polymorphic transitions in single-crystals - a new molecular-dynamics method. J Appl Phys 52(12):7182-7190. Figure Legends Fig. 1. Fluidity of a lipid bilayer in the presence of metal ions. (A) Normalized fluorescence intensity. The red circle, green triangle, blue inverted triangle and black rectangle indicate the fluorescence data with incubation of Cu2+, Zn2+, Ca2+ and control (Na+), respectively. The data are fitted with the curves with corresponding colors. The insets are the fluorescence-recovery images with incubation of Cu2+ at the times 0 s, 80 s and 240 s. (B) Rates of fluorescence recovery. The rate of incubation with CuCl2 is obviously less than those with other buffers, while the rates with CaCl2 and ZnCl2 are comparable (dark-red dashed line). Fig. 2. Ways of two hydrated Cu2+ ions binding with two phospholipids. The cyan, brown, red, white and yellow balls represent carbon, phosphorus, oxygen, hydrogen and copper, respectively. (A) A phosphate group in a phospholipid (PL). (B-D) A simplified model, H3C-[PO4]--CH3, of a phospholipid for studying the interaction with Cu2+ ions. (B) Initial state. Four molecular groups (two negatively charged fragments H3C-[PO4]--CH3 and two hydrated copper cations [Cu(H2O)5]2+) are separated by a large distance. (C) State I. Two hydrated Cu2+ ions bind to two phospholipids, respectively, which results in two [PL-Cu(aq)]+ structures. (D) State II. Two hydrated Cu2+ ions bind simultaneously with two phospholipids, forming a [PL-diCu(aq)-PL]2+ structure. Fig. 3. Binding energies (A) and bond orders (B) for hydrated metal ion Mn+ in States I and II. n = 1 for M = Na and K, while n = 2 for M = Mg, Ca, Zn, Cu and Ag. The bond orders (B) suggest that a chemical bond with a covalent characteristic occurs for Cu and Ag and an ionic bond occurs for the other metals. Moreover, the binding behavior of Cu2+ in State II significantly differs from that of Zn2+, although Zn is the element just next to Cu in the periodic table. Fig. 4. Electron analyses of Cu ions (left column) in State II by comparison with Zn ions (right column). The cyan, brown, red, white, yellow and dark gray balls represent carbon, phosphorus, oxygen, hydrogen, copper and zinc, respectively. (A) Electron densities. The green cloud denotes the electron density with an isosurface of 0.04 e/Å3. (B) Electron localized function (ELF). Upper: One-dimensional ELF along the metal-metal direction. The gray area indicates the region of core electrons, where ELF decays and quickly vanishes because a pseudo potential is employed in DFT calculations. Lower: Two-dimensional ELF. The pattern of ELF is approximately square in the area close to a Cu ion (lower-left) and is a circle in the area close to a Zn ion (lower-right). These suggest that the outermost electrons majorly occupy the 3d orbital for Cu and the 4s orbital for Zn. (C) Natural bond orbital between metal ions. The orange and light blue clouds indicate the orbital with an isosurface of 0.04 e/Å3. A metal-metal bond orbital occurs for Cu ions but not for Zn ions. For clarity, water molecules are not shown. Fig. 5. Assembly and order-analyses of phospholipids in membranes caused by Cu ions. (A-F) Lipid assembly induced by PL-diCu-PL structure through MD simulations. The light blue, brown, red, yellow and silver balls represent carbon, phosphorus, oxygen, copper and chlorine, respectively. For clarity, hydrogen atoms and water molecules are not shown. (A) Applied PL-diCu-PL structure in the PG-bilayer simulations. (B-C) Typical conformation of assembled lipids in membrane with side (B) and top (C) views. The blue box represents the periodic boundaries applied in the simulations. The tails of lipids and the atoms in the terminal of the lipid heads are represented by light blue curves and blue balls, respectively, for clarity. (D) A segment of assembled lipids. Two Cl- ions (silver balls) are observed over and under (dotted lines) the Cu-O plane, respectively. (E-F) Analyses of PL-diCu-PL induced assembly based on the degrees of order (E) and roughness (F). (E) The upper-left inset shows the applied angles (θ, φ) for the order degree in x-y plane. Upper and middle: Distributions of lipids according to the angles θ and φ, respectively. The black dotted lines denote the locations ±42.5° of two peaks. Lower: Distribution of phosphorus atoms in lipids along the z direction (dz), where the coordinate origin is set at the midpoint of two layers. The full-width half-maximum (FWHM) is presented for peaks in presence (orange value) and absence (blue value) of Cu ions. (G-I) AFM measurements. AFM images without (G) and with (F) incubation of CuCl2, and a comparison of the roughness of membrane surfaces (I). (H) The light green curve represents a stripe clearly in the membrane. (I) The label over double-arrow line indicates the P value < 0.05 of significant difference between the two groups of data. This suggests a significant difference in roughness of the membranes with and without CuCl2. Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5
1607.05742	1	1607	2016-07-19T20:07:20	Tension-dependent Free Energies of Nucleosome Unwrapping	[ "physics.bio-ph", "physics.chem-ph", "q-bio.BM" ]	Nucleosomes form the basic unit of compaction within eukaryotic genomes and their locations represent an important, yet poorly understood, mechanism of genetic regulation. Quantifying the strength of interactions within the nucleosome is a central problem in biophysics and is critical to understanding how nucleosome positions influence gene expression. By comparing to single-molecule experiments, we demonstrate that a coarse-grained molecular model of the nucleosome can reproduce key aspects of nucleosome unwrapping. Using detailed simulations of DNA and histone proteins, we calculate the tension-dependent free energy surface corresponding to the unwrapping process. The model reproduces quantitatively the forces required to unwrap the nucleosome, and reveals the role played by electrostatic interactions during this process. We then demonstrate that histone modifications and DNA sequence can have significant effects on the energies of nucleosome formation. Most notably, we show that histone tails are crucial for stabilizing the outer turn of nucleosomal DNA.	physics.bio-ph	physics	Tension-dependent Free Energies of Nucleosome Unwrapping Joshua Lequieu,† Andrés Córdoba,† David C. Schwartz,‡ and Juan J. de Pablo∗,†,¶ Institute for Molecular Engineering University of Chicago, Chicago, IL 60637 USA, Laboratory for Molecular and Computational Genomics, Department of Chemistry, Laboratory of Genetics, and UW-Biotechnology Center, University of Wisconsin-Madison, Madison, WI 53706, USA, and Materials Science Division Argonne National Laboratory, Argonne, IL 60439 USA E-mail: [email protected] Abstract Nucleosomes form the basic unit of compaction within eukaryotic genomes and their loca- tions represent an important, yet poorly understood, mechanism of genetic regulation. Quanti- fying the strength of interactions within the nucleosome is a central problem in biophysics and is critical to understanding how nucleosome positions influence gene expression. By compar- ing to single-molecule experiments, we demonstrate that a coarse-grained molecular model of the nucleosome can reproduce key aspects of nucleosome unwrapping. Using detailed simu- lations of DNA and histone proteins, we calculate the tension-dependent free energy surface ∗To whom correspondence should be addressed †Institute for Molecular Engineering University of Chicago, Chicago, IL 60637 USA ‡Laboratory for Molecular and Computational Genomics, Department of Chemistry, Laboratory of Genetics, and UW-Biotechnology Center, University of Wisconsin-Madison, Madison, WI 53706, USA ¶Materials Science Division Argonne National Laboratory, Argonne, IL 60439 USA 1 corresponding to the unwrapping process. The model reproduces quantitatively the forces required to unwrap the nucleosome, and reveals the role played by electrostatic interactions during this process. We then demonstrate that histone modifications and DNA sequence can have significant effects on the energies of nucleosome formation. Most notably, we show that histone tails are crucial for stabilizing the outer turn of nucleosomal DNA. Introduction Eukaryotic genomes are packaged into a compact, yet dynamic, structure known as chromatin. The basic building block of chromatin is the nucleosome, a disk-like structure consisting of 147 base pairs of DNA wrapped into 1.7 superhelical turns around proteins known as histones.1,2 These histone proteins form what is known as the histone octamer, a stable protein complex consisting of two copies of histone proteins H2A, H2B, H3 and H4. The surface of the histone octamer is highly positive, which interacts favorably with the negative backbone of DNA. As a result, at sufficiently low ionic conditions, nucleosomes are stable and spontaneously form. The locations of nucleosomes along the genome play a central role in eukaryotic regulation. DNA segments incorporated into nucleosomes are inaccessible to other DNA binding proteins, including transcription factors and polymerases, and thus nucleosomes must be disrupted in order for the cellular machinery to access nucleosomal DNA. As such, the positions occupied by nucle- osomes provide an additional, important mechanism by which eukaryotic genomes are regulated. Indeed, past work has demonstrated that deregulation of these processes is implicated in numer- ous diseases, including cancer.3 -- 5 Quantifying the strength of interactions within the nucleosome structure and the forces required to disrupt them is of fundamental importance to understanding the delicate dynamics of chromatin compaction. Optical-trapping single-molecule techniques have been particularly effective at probing multi- ple interactions that underlie the nucleosome. In these experiments, chromatin fibers6 -- 9 or single nucleosomes10 -- 15 are subjected to pico-newton scale forces, thereby providing the ability to pre- cisely perturb the native nucleosome structure. By analyzing the deformations that result from 2 these forces, one can infer the underlying strength of binding energies within the nucleosome. Following the initial work by Mihardja et al.,10 a consensus is emerging11,13,15 in which a single nucleosome is disrupted in two stages. In the first, at 3 pN, the outer wrap of DNA is removed from the histone surface. This first wrap is removed gradually and is considered to be an equilibrium process, where spontaneous unwrapping and rewrapping events can be observed under a constant force. The second transition occurs at forces 8-9 pN and occurs rapidly via so-called "rips", where the remaining wrap of DNA is suddenly released. More recently, these transitions have been shown to depend on torque (i.e. DNA supercoiling via twist),13 and to occur antisymmetrically due to variability in the bound DNA sequence.15 Several theoretical and computational studies have sought to help interpret these experimental results. Following the initial work of Kuli´c and Schiessel,16 most current treatments represent the nucleosome as an oriented spool, and the unbound DNA as a semiflexible worm-like chain.17,18 While earlier studies were only able to detect a single distinct unwrapping transition,16,19 consis- tent with the first experiments,6 more recent work17,18 has been able to reproduce the two transi- tions observed by Mihardja et al.10 By relying on simple, primarily analytic models, these studies have provided significant insights into the fundamental physics that govern interactions within the nucleosome. Such approaches, however, have necessarily had to invoke assumptions and intro- duce adjustable parameters in order to describe experiments17,18 (e.g. the DNA-histone binding energy). This limits their ability to predict nucleosomal behavior under different conditions, such as variations in DNA sequence or ionic environment, without resorting to additional experimental data. Additionally, these models cannot explicitly account for histone modifications, which are central to nucleosome positioning and higher-order chromatin structure.9,20 -- 23 A complementary approach, which should in principle enable prediction of nucleosomal inter- actions under a wide array of situations, could rely on molecular models where the nucleosome can be assembled or disassembled explicitly. Though these approaches are particularly promising, their success has been frustrated by the inability to access the experimentally relevant time scales of stretching, typically rates of 100nm/second. Clearly, these time scales are inaccessible to atomistic 3 simulations, yet even a highly coarse-grained spool-like model of the nucleosome only achieved stretching rates several orders of magnitude too fast.19 There is therefore a need to develop mod- els and methodologies to facilitate more direct comparisons between optical-trapping experiments and molecular-level calculations. If successful, such models could reveal the subtle, yet incredibly important effects of DNA-sequence and histone modifications on nucleosome stability. In this work, we build on a recently proposed coarse-grained model of the nucleosome24 -- 26 to examine its response to external perturbations. A computational framework is proposed in which the tension-dependent response of the nucleosome is examined at equilibrium, thereby providing access to the free energy of nucleosome unwrapping under tension. Our results are found to be in agreement with experimental measurements by Mihardja et al.,10 and serve to demonstrate that it is indeed possible to reproduce the absolute binding free energies of nucleosome formation in terms of purely molecular-level information, without resorting to adjustable parameters. Importantly, that model is then used to predict the impact of DNA sequence and histone modifications on the relative free energies of binding. Results A schematic of our simulation setup is shown in Figure 1a. As with optical-trapping experiments, the "state" of the nucleosome is represented by two parameters: the tension (or force) exerted on the DNA molecule, t , and the extension of the DNA ends, r. To facilitate comparison with experiments,10 the ends of the DNA are not torsionally constrained. Figure 1b shows instantaneous configurations of the nucleosome model for five different values of extension, r. Consistent with previous observations, the outer wrap of the nucleosome is first removed (A → T1 → B), followed by the inner wrap (B → T2 → C). In order to quantify these transitions, we examine the tension-dependent free energy of nucle- osome unwrapping. By calculating the tension-dependent free energy instead of a simple force- extension curve, as done previously,19,28 we can determine the true equilibrium behavior of the 4 Figure 1: Model of Nucleosome Unwrapping. a) Coarse-grained topology of Nucleosome. DNA is represented by 3SPN2.C,24 the histone proteins by AICG.26 Both the end-to-end extension, r, and tension, t , are constrained during a simulation. b) Unwrapping process. During extension, the wraps of DNA around histone proteins are removed one by one. T1 and T2 denote the transition states separating the first (A ↔ B) and second (B ↔ C) unwrapping events. Figures were generated using VMD.27 5 ) N p ( t , i n o s n e T 14 12 10 8 6 4 2 0 30 25 20 15 10 5 0 ) T k ( A 200 400 600 800 Extension, r (Å) Figure 2: Tension-dependent free energy surface of nucleosome unwrapping for 601 positioning sequence. The free energy surface demonstrates minima at extensions of r ≈ 120, ≈ 420, ≈ 700, depending on tension. As tension increases, the minimum-energy extension shifts to larger values. Consistent with Mihardja et al.,10 two transitions are observed. unwrapping process. Additionally, by performing simulations at equilibrium, we circumvent the issue of time scales that frustrate comparisons of traditional non-equilibrium molecular simulations to optical pulling experiments. A representative two-dimensional tension-extension free energy surface for the 601 positioning sequence29 is shown in Figure 2. Rather than increasing linearly with tension, the extension is quantized into three well defined vertical bands, located at ≈ 120, 420, and 700, corresponding to states "A", "B" and "C" in Figure 1. At low tension (t < 3pN), a low extension (r < 200) is preferred. As tension is increased (t ≈ 4 − 8pN), the minimum free energy shifts to intermediate values of extension (r ∼ 420). At higher tension (t > 8pN), the minimum free energy shifts to larger values of extension (r = 700). The free energy penalty of low tension and high extension t = 12, r = 200) results in large t = 3, r = 700) or high tension with low extension (e.g. (e.g. energy barriers > 40 kT . The tension-dependent transition can also be visualized by plotting one-dimensional "slices" 6 a) a) 100 Relative extension, r/L0 0.2 0.4 0.6 1 0.8 t =0 pN 2 pN 4 pN 6 pN 8 pN 10 pN 12 pN 80 60 40 20 0 1 0.8 0.6 0.4 0.2 0 ) T k ( A b) b) y t i l i b a b o r P Full Wrap Partial Wrap Unwrapped 200 300 400 500 600 700 Extension, r (Å) 1* 2* Full Wrap Partial Wrap Unwrapped 0 2 4 8 6 Tension, t (pN) 10 12 14 Figure 3: a) Free energy versus extension for different values of tension with the 601 positioning sequence. Based on the locations of the transition states, T1 and T2, three basins can be defined: "Fully Wrapped", "Partially Wrapped", and "Unwrapped". L0 represents the contour length of the DNA molecule. b) Probability of observing the nucleosome in each free energy basin for different tensions. The "Fully" and "Partially" Wrapped states are at equilibrium (i.e. equal probability) when t ∗ 2 = 8.9 pN. Error bars represent standard deviation across four independent simulations. 1 = 3.2 pN. The "Partially" and "Unwrapped" states are at equilibrium when t ∗ 7 t t of the free energy surface at different values of tension (Figure 3a). Visualizing the data in this way clearly demonstrates that there are three basins of nucleosome extension: "Fully Wrapped", "Partially Wrapped" and "Unwrapped". The basin that is favored depends on the tension applied to the DNA ends. As tension increases, the free energy minima shifts first from the "Fully Wrapped" to the "Partially Wrapped" basin, and then to the "Unwrapped" basin. The boundaries of these basins are defined by the locations of the transition states (i.e. local maximum in the free energy) that separate neighboring basins. The transition states separating the A → B transition, T1, and the B → C transition, T2, are shown in Figure 1b. Once these three basins are defined, we can determine the precise tension at which the outer and inner DNA turns unwrap from the nucleosome. This is obtained by converting the tension- dependent free energy into probabilities, and then integrating these probabilities to determine the total probability of finding the system in each basin (see Materials and Methods). The correspond- ing results are shown in Figure 3b; it can be appreciated that the probability of finding the system in the "Fully Wrapped" or "Partially Wrapped" basin is equivalent when t ≈ 3.2 pN. Thus, when t ≈ 3.2 pN the outer turn of nucleosomal DNA is in equilibrium (in a statistical mechanics sense) with its unbound state. We define this tension as t ∗ 1 . Similarly, the probability of the nucleosome in the "Partially Wrapped" and "Unwrapped" basins is the same (i.e. the inner wrap is in equilibrium when t ≈ 8.9 pN, defined as t ∗ 2 ). These values are in quantitative agreement with those measured by Mihardja et al.,10 who observed that the outer and inner DNA loops were removed at 3 pN and 8-9 pN, respectively. A complementary approach to estimate t ∗ 2 , is to determine the tension at which the free energy barriers of the forward and reverse reactions are equal.17 Figure 4 shows the corresponding 1 and t ∗ tension-dependent free energy barriers of the outer (A ↔ T1 ↔ B) and inner unwrapping (B ↔ T2 ↔ C) events. At low tension, the energy barriers for the forward reactions, A → B and B → C, dominate and the forward (i.e. unwrapping) reaction rate is low. As tension increases, the energy barriers for the forward reactions decrease, while those for the reverse increase, thereby causing the unwrapping reaction to proceed at a higher rate. When the energy barriers of the forward and 8 ) T k ( † A , s t h g e h i r e i r r a B 60 50 40 30 20 10 0 A fi B fi B fi C fi T1 T1 T2 T2 1* = 3.3pN 2* = 8.5pN 0 2 4 6 8 10 12 14 Tension, t (pN) Figure 4: Free energy barrier heights of nucleosome unwrapping for 601 positioning sequence. Solid lines represent the unwrapping (forward) reactions, dotted lines represent wrapping (reverse) reactions. When the unwrapping and wrapping barriers are equal, the two basins are at equilibrium 1 = 3.3 pN for the outer wrap, and t ∗ with one another. This is found when t ∗ 2 = 8.5 pN for the inner wrap. D A†(t ∗ 1 ) = 4 kT and D A†(t ∗ 2 ) = 16 kT . Error bars represent standard deviation across four independent simulations. 9 D t t reverse reactions are equal, the two basins are at equilibrium (in a transition state theory sense) and t ∗ 1 and t ∗ in excellent agreement with the probability-based analysis of Figure 3b. 2 can be determined. These unwrapping tensions are estimated to be 3.3 pN and 8.5 pN, The magnitude of the free energy barrier also helps explain the observation by Mihardja et al.10 that the outer turn of DNA can be removed reversibly, while the inner turn cannot. Since the energy barrier separating the "Fully" and "Partially" wrapped states is only ≈ 5 kT , the system can quickly transition between states when held at t = t ∗ 1 . In contrast, the "Partial Wrap" and "Unwrapped" states are separated by an energy barrier of ≈ 18 kT , indicating that even at equilibrium the P ↔ U transition occurs slowly. Thus, removal of the outer wrap may appear to be reversible on the time scales of a typical optical trapping experiment, while the inner wrap may not. Further, because force-extension curves are usually obtained via optical trapping by pulling a nucleosome at a fixed velocity, the experiments may not observe a P → U transition until t > t ∗ experiments to overestimate the value of t ∗ We also note that the barrier estimates in this work (D A† agreement to those predicted by Sudhanshu et al.17 (D A† 2 . This would cause the 2 , and lead to a sudden, irreversible "ripping" event. 2 = 16 kT ) are in excellent 1 = 4 kT,D A† 1 ≈ 6 kT,D A† 2 ≈ 15 kT ). Electrostatics, Sequence Dependence, Histone Modifications Having validated the proposed model against experimental data,10 we now examine the influence of ionic environment, DNA sequence, and histone modifications on the stability of the nucleosome. Such variations can have a significant impact on nucleosome formation, and the precise molecular origins of their impact is still poorly understood. We first investigate the origins of the tension-dependent mechanical response by exploring the role of DNA-DNA electrostatic repulsion on the stability of the nucleosome structure. Past the- oretical work16,18 has suggested that DNA-DNA repulsion within the nucleosome is central to its tension-dependent response. Other studies, however, have observed that DNA-DNA electro- static repulsion is unimportant and that the correct response can be achieved by accounting for the tension-dependent orientation of the free DNA ends.17 Since our proposed model explicitly 10 Figure 5: a) Schematic of proposed model with DNA-DNA repulsion removed. b) Resulting t ∗ 2 represent tension-dependent free energy barriers for 601 positioning sequence. change relative to complete model. Error bars represent standard deviation across three inde- pendent simulations. 1 and D t ∗ includes both contributions, we can directly evaluate the importance of DNA-DNA repulsion on nucleosome unwrapping. To examine this effect, we disable DNA-DNA electrostatic repulsion in our model between base-pairs separated by more than 20 base pairs. Only disabling electrostatics between distant regions of DNA was necessary to avoid implicitly lowering the persistence length of DNA by neglecting Coulombic interactions between neighboring base-pairs. All electrostatics responsible for DNA-histone affinity however, remain intact. Our results are summarized in Figure 5a,b. As anticipated,16 removal of DNA-DNA repulsive interactions has a greater impact on the outer DNA loop (D loop (D t ∗ 1 = +1.7pN) than on the inner DNA t ∗ 2 = +1.1pN). However in the absence of DNA-DNA repulsions, the qualitative features of the tension-dependent response remain unchanged. These results indicates that while DNA-DNA repulsions play a role in nucleosome disassembly, they are not primarily responsible for the two unwrapping steps observed in experiments. We next examine the impact of DNA sequence on the relative binding free energies of nucle- osome formation. Optical trapping experiments could in principle be used to probe the sequence- dependent energies within the nucleosome, but recent literature studies have been limited to the 601 positioning sequence10,11,13 and slight variations.15 Instead, competitive reconstitution as- 11 D ) l o m / J k ( . i m s A 25 20 15 10 5 0 -5 -10 -15 r=0.88 r=0.78 Freeman et al. 25 This Work -2 4 2 0 6 D Gexpt. (kJ/mol) 8 Figure 6: Sequence-dependent binding free energies. Squares denote model proposed by Freeman et al.25 (obtained at 300K and 150mM ionic strength). Circles denote model proposed in this work, obtained at 277K and vanishing ionic strength (for consistency with Ref.30). Despite differing solution conditions and DNA-Protein interactions, both models reproduce the relative binding free energies of nucleosome formation. The DNA sequences used here are given in Ref.25 says are the dominant experimental technique for characterization of sequence-dependent relative binding free energies.30,31 To compare model predictions to these experiments, we use the tech- nique employed by Freeman et al.,25 where the relative binding free energies of different DNA sequences are assessed computationally using alchemical transformations and thermodynamic in- tegration (see Materials and Methods). A comparison of predicted and experimental free energies, shown in Figure 6, indicates that, as with previous work,25 the model adopted here accurately reproduces the binding free energies of many different sequences. In general, the key predictor of binding free energy is the sequence-dependent shape of the DNA molecule (i.e. minor groove widths and intrinsic curvature). Sequences that bind strongly (low D motifs (e.g. TA base steps) that impart a shape that favorably "fits" underlying histone structure.31 In contrast, weakly binding sequences (large D A) do not posses these periodic motifs. A) posses periodic sequence In addition to DNA sequence, the modification of histone tails is widely considered to be the single most important determinant of chromatin structure.20 Methylated and acetylated histones 12 D D D D D Figure 7: a) Schematic of proposed model with histone tails removed. b) Resulting tension- dependent free energy barriers for 601 positioning sequence. D t ∗ 2 represent change relative to model with intact histone tails. Error bars represent standard deviation across three independent simulations. 1 and D t ∗ are enriched at promoters of highly expressed genes and are thought to play a role in the strong positioning of certain nucleosomes21,22. Histone tails are central to nucleosome-nucleosome inter- actions and their modification has important implications on chromatin's three-dimensional struc- ture23. Experiments9 have also established that removal of histone tails has a significant impact on the stability of the nucleosome. To examine the role of histone tails on nucleosome stability at a molecular level, we return to our earlier analysis and calculate the tension-dependent free energy of nucleosome unwrap- ping. Our results can be compared to the optical trapping experiments of Brower-Toland et al.,9 where arrays of 17 nucleosomes were disassembled for different histone tail modifications, includ- ing complete removal via trypsin digest or post-translationally via acetylation. In the model, we perform this trypsin digest in silico (see Figure 7a), and calculate the resulting tension-dependent response (Figure 7b). Our results indicate that removal of histone tails significantly destabilizes the outer loop, causing the "Partially Wrapped" (c.f. Figure 1b) state to be energetically favored at zero tension. Additionally, histone tail removal shifts the inner loop unwrapping force, t ∗ 2 , to 2.6pN, 5.9 pN lower than when histone tails were intact (c.f. Figure 4). Our results are consistent with measurements that reported a 3 pN reduction in "peak force" upon tail removal and a 60% reduction in the outer turn length.9 In this case, quantitative agreement is not expected since these experiments lacked sufficient spatial resolution to resolve the individual release of the outer DNA 13 turn. Conclusion In this work we have demonstrated that a molecular-model of the nucleosome, composed of two coarse-grained models of DNA and proteins,24 -- 26 can be combined parameter-free to accurately reproduce the tension-dependent response of nucleosome unwrapping. This model quantitatively reproduces the unwrapping forces observed in experiments10 and the barrier heights predicted by prior theoretical studies.17 We then demonstrated that this model can be used to examine, without adjustment, the role of subtle phenomena in nucleosome formation such as DNA-DNA Coulombic repulsion, DNA-sequence, and histone tail modifications. Our proposed approach opens up a new avenue for theoretical examinations of nucleosome stability. As a first step, this model can aid the interpretation of recent optical pulling experiments where nucleosome are subjected to torque,13 and is suitable to examine subtle features within the nucleosome such as asymmetric unwrapping.15 Yet the potential of our approach extends beyond nucleosome pulling experiments, and can begin to elucidate many unsolved questions within chro- matin biophysics. How does the methylation of specific histone tails (and not others) enhance the positioning of certain nucleosomes? What are the free energies of different folded chromatin structures, and how do histone modifications effect this energy landscape? What is the role of DNA sequence on these processes, and do certain DNA sequences dispose chromatin to differ- ent "folds"? The approach presented in this work represents an important step towards answering these questions. Methods The model adopted in this work relies on a coarse-grained model of DNA24,25 and proteins,26 which are combined to represent the nucleosome. Both models were developed independently, but they are implemented at the same level of description, thereby facilitating their concerted use. 14 Specifically, for DNA we use the 3SPN coarse-grained representation, where each nucleotide is described by three force sites located at the phosphate, sugar and base.24,32 -- 34 For the histone pro- teins, we use the "Atomistic-Interaction based Coarse-Grained model" (AICG), where the protein is represented by one site per amino acid located at the center of mass of the sidechain.26 Interactions between the 3SPN2.C and AICG models included electrostatic and excluded vol- ume effects. Phosphate sites with 3SPN were assigned a charge of −0.6 as described previously.34 Each protein site was given the net charge of that residue at physiological pH (i.e. +1 for Arg, Hys and Lys; -1 for Asp and Glu, 0 for others). As with prior work,25 the effective charge of interac- tions between DNA and protein sites was scaled by a factor of 1.67 to bring the local charge of the phosphates back to -1. We note that DNA-Protein interactions in this work differ slightly from those employed by Freeman et al.25 where, in addition to electrostatics, a small Lennard-Jones attraction was added between all DNA and protein sites. The strength of this attraction was very weak (e Pro−DNA = 0.25kJ/mol) and was originally included to reduce fluctuations within the nu- cleosome structure. Here we demonstrate that this weak interaction is unnecessary; by omitting it, both the relative and absolute formation free energies of the nucleosome can be reproduced. The combined model is effectively parameter-free: both the model of DNA and Protein are included as originally proposed without any additional terms. Electrostatic forces are introduced at the level of Debye-Hückel theory. All simulations were performed in the canonical ensemble using a Langevin thermostat and 150mM ionic strength. As an initial condition, we combine the 1KX5 crystal structure2 of the nucleosome core particle with a proposed configuration of exiting DNA35,36 to form a 223 base-pair structure, with 147 base-pairs bound to the histone proteins and 38 flanking bases on each side. When using the 601 positioning sequence,29 the flanking bases were chosen as polyA. This configuration was only used as the initial configuration, and no information from either structure was directly encoded into the nucleosome model. To extract the tension-dependent free energy surface, two constraints were applied to the nu- cleosomal model. First, a constant force (i.e. tension) was applied to each end of DNA in order 15 to mimic the experimental setup of optical-trapping experiments. Then, harmonic constrains were applied to the end-to-end extension of the DNA molecule, and umbrella sampling was performed to determine the free-energy as a function of DNA extension.37,38 This simulation framework re- sults in free energy "surfaces" that are not truly continuous for different values of tension. They are instead a compilation of two-dimensional "curves" that are plotted co-currently to construct the "surface" presented in Figure 2. The relative free energy of binding for different DNA sequences (D A) was calculated as de- scribed in detail previously.25 Briefly, a thermodynamic cycle was defined that represents the rel- ative sequence-dependent free energy of nucleosome formation, D A, as the difference between the free energy difference of two DNA sequences in the bulk, D Abulk, and bound to the histone proteins, D Abound (i.e. D A = D Abulk −D Abound); D Abulk and D Abound are determined by thermody- namic integration. The DNA sequences analyzed are given explicitly in the original paper. Acknowledgement This work was supported by NIST through the Center for Hierarchical Materials Assembly (CHi- MaD). Computational resources were provided by the Midway computing cluster at the University of Chicago and the University of Wisconsin -- Madison Center for High Throughput Computing. The authors gratefully acknowledge Professor Andrew Spakowitz for helpful discussions and Ralf Everaers and Sam Meyer for providing the configuration of the 223 base-pair nucleosome. Supporting Information Available This material is available free of charge via the Internet at http://pubs.acs.org/. 16 D D D References (1) Luger, K.; Mader, A. W.; Richmond, R. K.; Sargent, D. F.; Richmond, T. J. Nature 1997, 389, 251 -- 260. (2) Davey, C. a.; Sargent, D. F.; Luger, K.; Maeder, A. W.; Richmond, T. J. J. Mol. Biol. 2002, 319, 1097 -- 113. (3) Hendrich, B.; Bickmore, W. Hum. Mol. Genet. 2001, 10, 2233 -- 2242. (4) Bhaumik, S. R.; Smith, E.; Shilatifard, A. Nat. Struct. Mol. Biol. 2007, 14, 1008 -- 1016. (5) Sadri-Vakili, G.; Bouzou, B.; Benn, C. L.; Kim, M.-O.; Chawla, P.; Overland, R. P.; Glajch, K. E.; Xia, E.; Qiu, Z.; Hersch, S. M.; Clark, T. W.; Yohrling, G. J.; Cha, J.-H. J. Hum. Mol. Genet. 2007, 16, 1293 -- 1306. (6) Brower-Toland, B. D.; Smith, C. L.; Yeh, R. C.; Lis, J. T.; Peterson, C. L.; Wang, M. D. Proc Natl Acad Sci U S A 2002, 99, 1960 -- 1965. (7) Cui, Y.; Bustamante, C. Proc. Nat. Acad. Sci. USA. 2000, 97, 127 -- 132. (8) Gemmen, G. J.; Sim, R.; Haushalter, K. a.; Ke, P. C.; Kadonaga, J. T.; Smith, D. E. Journal of molecular biology 2005, 351, 89 -- 99. (9) Brower-Toland, B.; Wacker, D. A.; Fulbright, R. M.; Lis, J. T.; Kraus, W. L.; Wang, M. D. J. Mol. Biol. 2005, 346, 135 -- 146. (10) Mihardja, S.; Spakowitz, A. J.; Zhang, Y.; Bustamante, C. Proc. Nat. Acad. Sci. USA. 2006, 103, 15871 -- 6. (11) Kruithof, M.; van Noort, J. Biophys. J. 2009, 96, 3708 -- 3715. (12) Hall, M. A.; Shundrovsky, A.; Bai, L.; Fulbright, R. M.; Lis, J. T.; Wang, M. D. Nat. Struct. Mol. Biol. 2009, 16, 124 -- 9. 17 (13) Sheinin, M. Y.; Li, M.; Soltani, M.; Luger, K.; Wang, M. D. Nat. Commun. 2013, 4, 2579. (14) Mack, A. H.; Schlingman, D. J.; Ilagan, R. P.; Regan, L.; Mochrie, S. G. J. Mol. Biol. 2012, 423, 687 -- 701. (15) Ngo, T. T. M.; Zhang, Q.; Zhou, R.; Yodh, J. G.; Ha, T. Cell 2015, 160, 1135 -- 1144. (16) Kuli´c, I. M.; Schiessel, H. Phys. Rev. Lett. 2004, 92, 228101. (17) Sudhanshu, B.; Mihardja, S.; Koslover, E. F.; Mehraeen, S.; Bustamante, C.; Spakowitz, a. J. Proc. Nat. Acad. Sci. USA. 2011, 108, 1885 -- 90. (18) Mollazadeh-Beidokhti, L.; Mohammad-Rafiee, F.; Schiessel, H. Biophys. J. 2012, 102, 2235 -- 2240. (19) Wocjan, T.; Klenin, K.; Langowski, J. J. Phys. Chem. B 2009, 113, 2639 -- 2646. (20) Jenuwein, T.; Allis, C. D. Science (New York, N.Y.) 2001, 293, 1074 -- 80. (21) Kurdistani, S. K.; Tavazoie, S.; Grunstrin, M. Cell 2004, 117, 721 -- 733. (22) Barski, A.; Cuddapah, S.; Cui, K.; Roh, T. Y.; Schones, D. E.; Wang, Z.; Wei, G.; Chepelev, I.; Zhao, K. Cell 2007, 129, 823 -- 837. (23) Zhou, V. W.; Goren, A.; Bernstein, B. E. Nat. Rev. Genet. 2011, 12, 7 -- 18. (24) Freeman, G. S.; Hinckley, D. M.; Lequieu, J. P.; Whitmer, J. K.; de Pablo, J. J. The Journal of Chemical Physics 2014, 141, 165103. (25) Freeman, G. S.; Lequieu, J. P.; Hinckley, D. M.; Whitmer, J. K.; de Pablo, J. J. Phys. Rev. Lett. 2014, 113, 168101. (26) Li, W.; Wolynes, P. G.; Takada, S. Proc. Nat. Acad. Sci. USA. 2011, 108, 3504 -- 9. (27) Humphrey, W.; Dalke, A.; Schulten, K. VMD: Visual Molecular Dynamics. 1996. 18 (28) Kenzaki, H.; Takada, S. PLoS Comput. Biol. 2015, 11, e1004443. (29) Lowary, P. T.; Widom, J. Journal of molecular biology 1998, 276, 19 -- 42. (30) Thåström, A.; Lowary, P. T.; Widom, J. Methods (San Diego, Calif.) 2004, 33, 33 -- 44. (31) Segal, E.; Fondufe-Mittendorf, Y.; Chen, L.; Thåström, A.; Field, Y.; Moore, I. K.; Wang, J.- P. Z.; Widom, J. Nature 2006, 442, 772 -- 8. (32) Knotts, T. A.; Rathore, N.; Schwartz, D. C.; de Pablo, J. J. J. Chem. Phys. 2007, 126, 084901. (33) Sambriski, E. J.; Schwartz, D. C.; de Pablo, J. J. Biophys. J. 2009, 96, 1675 -- 90. (34) Hinckley, D. M.; Freeman, G. S.; Whitmer, J. K.; de Pablo, J. J. J. Chem. Phys. 2013, 139, 144903. (35) Hussain, S.; Goutte-Gattat, D.; Becker, N.; Meyer, S.; Shukla, M. S.; Hayes, J. J.; Ever- aers, R.; Angelov, D.; Bednar, J.; Dimitrov, S. Proc. Nat. Acad. Sci 2010, 107, 9620 -- 9625. (36) Meyer, S.; Becker, N. B.; Syed, S. H.; Goutte-Gattat, D.; Shukla, M. S.; Hayes, J. J.; An- gelov, D.; Bednar, J.; Dimitrov, S.; Everaers, R. Nucleic acids research 2011, 39, 9139 -- 54. (37) Kumar, S.; Rosenberg, J. M.; Bouzida, D.; Swendsen, R. H.; Kollman, P. A. Journal of Computational Chemistry 1995, 16, 1339 -- 1350. (38) Kästner, J. Wiley Interdiscip Rev Comput Mol Sci 2011, 1, 932 -- 942. 19
1009.4860	3	1009	2011-02-08T15:47:41	Phase separation and near-critical fluctuations in two-component lipid membranes: Monte Carlo simulations on experimentally relevant scales	[ "physics.bio-ph", "cond-mat.soft" ]	By means of lattice-based Monte Carlo simulations, we address properties of two-component lipid membranes on the experimentally relevant spatial scales of order of a micrometer and time intervals of order of a second, using DMPC/DSPC lipid mixtures as a model system. Our large-scale simulations allowed us to obtain important results previously not reported in simulation studies of lipid membranes. We find that, within a certain range of lipid compositions, the phase transition from the fluid phase to the fluid-gel phase coexistence proceeds via near-critical fluctuations, while for other lipid compositions this phase transition has a quasi-abrupt character. In the presence of near-critical fluctuations, transient subdiffusion of lipid molecules is observed. These features of the system are stable with respect to perturbations in lipid interaction parameters used in our simulations. The line tension characterizing lipid domains in the fluid-gel coexistence region is found to be in the pN range. When approaching the critical point, the line tension, the inverse correlation length of fluid-gel spatial fluctuations, and the corresponding inverse order parameter susceptibility of the membrane vanish. All these results are in agreement with recent experimental findings for model lipid membranes. Our analysis of the domain coarsening dynamics after an abrupt quench of the membrane to the fluid-gel coexistence region reveals that lateral diffusion of lipids plays an important role in the fluid-gel phase separation process.	physics.bio-ph	physics	Phase separation and near-critical fluctuations in two-component lipid membranes: Monte Carlo simulations on experimentally relevant scales Jens Ehrig, Eugene P Petrov and Petra Schwille Biophysics, BIOTEC, Technische Universitat Dresden, Tatzberg 47/49, 01307 Dresden, Germany E-mail: [email protected] Abstract. By means of lattice-based Monte Carlo simulations, we address properties of two-component lipid membranes on the experimentally relevant spatial scales of order of a micrometer and time intervals of order of a second, using DMPC/DSPC lipid mixtures as a model system. Our large-scale simulations allowed us to obtain important results previously not reported in simulation studies of lipid membranes. We find that, within a certain range of lipid compositions, the phase transition from the fluid phase to the fluid -- gel phase coexistence proceeds via near-critical fluctuations, while for other lipid compositions this phase transition has a quasi-abrupt character. In the presence of near-critical fluctuations, transient subdiffusion of lipid molecules is observed. These features of the system are stable with respect to perturbations in lipid interaction parameters used in our simulations. The line tension characterizing lipid domains in the fluid -- gel coexistence region is found to be in the pN range. When approaching the critical point, the line tension, the inverse correlation length of fluid -- gel spatial fluctuations, and the corresponding inverse order parameter susceptibility of the membrane vanish. All these results are in agreement with recent experimental findings for model lipid membranes. Our analysis of the domain coarsening dynamics after an abrupt quench of the membrane to the fluid -- gel coexistence region reveals that lateral diffusion of lipids plays an important role in the fluid -- gel phase separation process. PACS numbers: 87.14.Cc, 87.15.ak, 87.16.dj, 87.16.dt Keywords: Monte Carlo simulation; lipid membrane; DMPC; DSPC; phase diagram; critical fluctuations; dynamic scaling; line tension New Journal of Physics, 2011, accepted for publication J Ehrig, E P Petrov and P Schwille 2 1. Introduction The membrane plays a very important role in the cell, not only by defining its boundaries and boundaries of the cell organelles, but also by taking an active part in cell functioning, as many key processes take place in or across the cell membrane. The plasma membrane is a very complex system comprised of lipids, polysaccharides, and proteins, all strongly interacting with each other. These lipid -- lipid and lipid -- protein interactions induce the lateral microheterogeneity of the cell membrane, which is believed to be crucial in providing its high functionality [1]. The idea of the function- related lateral microheterogeneity of the membrane was formulated by Simons and Ikonen [2] in the form of the concept of lipid rafts. The original concept of lipid rafts involved the dynamic clustering of sphingolipids and cholesterol, supposedly forming moving platforms in the cell membrane to which certain proteins would attach [2]. Presently, the understanding is that the interaction of lipids with membrane proteins and the cytoskeleton, as well as effects of local curvature, play an important role in organization of the lateral microheterogeneity of the cell membrane [3 -- 6]. Consequently, the term membrane rafts appears to be more appropriate [7]. In spite of the importance of lipid -- protein interactions, the behaviour of the lipid component of the cell membrane is still believed to govern its properties to a large extent. In agreement with that, it was found that studies of model membrane systems -- both in vitro [8] and in silico, i.e., via numerical simulations [9 -- 12] -- give valuable information on the lateral organization and dynamics of lipid molecules, which provides a deeper insight into the structure and function of plasma membranes. Numerical simulations are an extremely fruitful approach to membrane studies, giving access to information which is either difficult or even impossible to access using present-day experimental techniques. Depending on particular aims and goals, a wide arsenal of simulation techniques has been developed to study model membrane systems. A particular choice of the simulation approach largely depends on the amount of the fine molecular details, as well as spatial scales and time ranges of relevance for a particular problem addressed in a study. Broadly, the approaches to numerical simulations of lipid membranes include molecular dynamics methods and their coarse-grained versions, mean-field based continuum model simulations, and lattice-based methods. Molecular dynamics simulations provide the most detailed approach to describe the structure and dynamics of membranes. They employ an atomistic description of the lipid membrane and thus, with the presently available computational resources, are able to cover spatial scales up to a few nm and time scales of ≤ 1 µs [13 -- 16]. By using coarse-grained molecular dynamics simulations which sacrifice the very fine atomistic details for the sake of a better computational efficiency [9, 12, 17] these scales can be extended to ≈ 100 nm and ≈ 10 µs. Further coarse-graining [18, 19], by representing a cluster of a few ten lipid molecules as one particle, allows for simulations of membranes consisting of ∼ 6000 coarse-grained particles, which is equivalent to the total number of ∼ 60000 lipid molecules in a simulated membrane. On the other hand, this type of J Ehrig, E P Petrov and P Schwille 3 coarse-graining does not allow one to address, e.g., diffusional motion of individual lipid molecules in the membrane. Mean field-based continuum simulations [20, 21], based on solving evolution equations for a compositional or phase field, are primarily concerned with the issue of the phase separation behaviour of lipid membranes [22 -- 24]. These simulations address much larger spatial scales of order of a few hundred nm. On the other hand, this approach does not seem to be able to reproduce the complete phase diagram of a real lipid mixture. Additionally, these simulations do not provide information on lateral diffusion in the membrane. Continuum simulations of the lipid membrane organization can be made more realistic by using lipid interaction parameters and lipid chain order parameter libraries extracted from molecular dynamics simulations [25]. For ternary lipid membrane systems, these simulations can currently address spatial scales of a few tens of nm [25]. In view of the continuum character of the model, they do not, however, provide information on the lateral diffusion of membrane components. Lattice-based Monte Carlo simulations constitute the computationally least demanding way to numerical modelling of lipid membranes. Not surprisingly, the first lattice-based simulations of the single-component lipid membrane date back to the early l980s [26]. Later, this approach based on either the ten-state Pink model or Ising-like models was successfully used to study phase transitions in one- and two-component lipid membranes, as well as the effects of lipid -- protein and lipid -- drug interactions on the phase separation in lipid membranes (see, e.g., [27 -- 32]). Although these lattice-based approaches do not even attempt at reproducing the membrane structure and dynamics in full detail on the atomic and molecular scale, they provide a surprisingly realistic description of the phase separation and dynamics in lipid membranes. Additionally, because of the simplified description of the membrane, lattice-based simulations are capable of describing the membrane behaviour on much larger spatial scales and substantially longer time intervals than the above-mentioned atomistic and coarse- grained simulations with the same computational expenses. While the previously published lattice-based simulations addressed one- or two-component lipid membranes on spatial scales from ten to hundred nm, the recent rapid progress in computer hardware presently allows one to substantially expand the spatial scales and time intervals addressed in lattice-based simulations. The aim of the present study is to address the properties of model lipid membranes using a reasonably simple, but still realistic enough, model on the experimentally relevant spatial scales of order of a micrometer and time scales of order of a second. Simulations on these spatial scales and time intervals should provide results which can be directly compared with experimental results obtained using optical microscopy-based methods routinely used in membrane studies. Our goal was, therefore, to develop a lattice-based Monte Carlo simulation approach that would not only provide an adequate description of the membrane phase behaviour and membrane dynamics, but in addition would be computationally efficient in order to reach the required spatial and temporal J Ehrig, E P Petrov and P Schwille 4 Figure 1. Confocal fluorescence microscopy image of the upper pole of a giant unilamellar vesicle (DMPC/DSPC 50/50) exhibiting fluid -- gel phase separation. Gel- phase domains, which coarsen and coalesce with time, appear on the image as dark areas. Scale bar: 20 µm. scales. We chose to study membranes consisting of a binary mixture of DMPC (1,2- dimyristoyl-sn-glycero-3-phosphocholine, Tm = 297 K) and DSPC (1,2-distearoyl-sn- glycero-3-phosphocholine, Tm = 328 K), which for a range of lipid compositions and temperatures shows coexistence of fluid and gel phases (figure 1). The phase diagram of this system has been extensively studied experimentally (see, e.g., [31 -- 39]). Some aspects of the phase behaviour of the same lipid mixture have been previously addressed in a series of lattice-based Monte Carlo simulations (see, e.g., [31, 32, 39]). Therefore, the choice of the well-studied DMPC/DSPC lipid systems as an object of our study allowed for a direct comparison of the results of our numerical simulations with experimental data, as well as with results obtained in simulation studies of the same lipid system carried out by other groups. In our recent work, using the same MC simulation approach as in the present paper, we addressed the aspects of lateral diffusion in the membrane with a focus on the effects of near-critical fluctuations in lipid membranes and the effects of cytoskeleton [40]. In the present paper, the main emphasis is on the phase diagram of the DMPC/DSPC lipid system, including the near-critical behaviour of the membrane close to its critical point, as well as the properties of membrane domains, including the line tension of the phase interface, and domain growth kinetics. 2. Materials and Methods 2.1. Experimental The lipids 1,2-dimyristoyl-sn-glycero-3-phosphocholine (DMPC), 1,2-distearoyl-sn- glycero-3-phosphocholine (DSPC) and the fluorescent lipid marker 1,2-dipalmitoyl-sn- J Ehrig, E P Petrov and P Schwille 5 glycero-3-phosphoethanolamine-N-(lissamine rhodamine B sulfonyl) (N-Rhod-DPPE) were purchased from Avanti Polar Lipids (Alabaster, AL), diluted in a chloroform- methanol (2:1) mixture, aliquoted, and stored at −20◦C in argon atmosphere. Giant unilamellar vesicles (GUVs) were prepared by electroformation on a platinum wire [41] and were used for visualization of phase separation in two-component DMPC/DSPC lipid membranes by means of laser scanning fluorescence microscopy. Multilamellar vesicle suspensions were obtained from rehydration of a dry lipid film. Differential scanning calorimetry (DSC) experiments were performed on suspensions of DMPC/DSPC multilamellar vesicles using a VP-DSC calorimeter (MicroCal, Northampton, MA) with a scan rate of 2 − 3 K/h and a medium ("mid") feedback. The temperatures of the onset and completion of the phase transition were determined from experimentally measured excess heat capacity curves using an empirical approach known as the tangent method (see, e.g., [11, 31, 32]). To do that, a tangential line is drawn to a heat capacity curve at its low-temperature (onset of the phase transition) or high-temperature (completion of the phase transition) slopes, and the respective temperature is determined as the zero-crossing point of this tangential line. Plotting the temperatures of the onset and completion of the phase transition as a function of the composition of the lipid mixture allows one to construct an empirical experimental phase diagram of the lipid mixture. 2.2. Monte Carlo Simulations The approach to lattice-based MC simulations of a two-component membrane employed in the present study is generally similar to the one described previously [31, 32, 39]. To achieve a higher computational efficiency in Monte Carlo simulations of membrane dynamics on the experimentally relevant spatial scales (of order of a micrometer) and time intervals (of order of a second), where a particular type of lipid packing and fine molecular details should be of little importance, we make one more step toward simplification of the description within the framework of the membrane lattice model. To represent a membrane, instead of a triangular lattice of lipid chains [31, 32], we use a square lattice of lipid molecules (figure 2). In this representation, neither the orientation of a lipid molecule within the membrane plane nor the states of individual lipid chains of the lipid molecule are accounted for. The model assumes that each of the lipid molecules placed on a square lattice with periodic boundary conditions can exist in either gel or fluid state, and can move laterally via next-neighbour exchange. Formally, this model is an Ising model with a conserved order parameter (representing lipids) decorated by an Ising model with a non-conserved order parameter (representing lipid states). The fundamental step of the MC simulation consists of two sub-steps. In the first sub-step, an attempt is made to change the state of a randomly chosen lipid (from gel to fluid or vice versa). The second sub-step is an attempt to exchange the positions of a randomly chosen pair of next neighbours on the lattice. J Ehrig, E P Petrov and P Schwille 6 Figure 2. Square lattice model used in this work to simulate lipid membranes consisting of a mixture of two lipids (DMPC and DSPC in the present study). Individual lipid molecules occupy the nods of a square lattice. Each of the two lipids (orange, cyan) can exist in either gel or fluid conformational state (black, white). The combined representation via four-colour coding makes it possible to see the complete configuration of the lattice in a single snapshot. ∆G = ∆N F For each sub-step the change in the Gibbs free energy 2 (∆E2 − T ∆S2) + ∆N GF 12 wGG 1 (∆E1 − T ∆S1) + ∆N F 22 + ∆N GG 11 wGF 11 12 + ∆N GF 12 + ∆N FF +∆N GF 22 wGF 12 wFF 12 wGF 12 + ∆N GF 21 wGF 21 , (1) is calculated. Here, ∆Ei and ∆Si are the changes in the internal energy and entropy of a molecule of lipid i when it switches its state from gel to fluid, wmn ij are the next-neighbour interaction parameters of lipids i and j being in states n and m, respectively, ∆N F is i the change in the number of molecules of lipid i in the fluid state and ∆N mn is the change in the number of next-neighbour contacts of lipids i and j being in states n and m, respectively. In the simulation, the attempts of the state change and next-neighbour exchange are accepted with probability p = 1 for ∆G < 0 and p = exp (−∆G/RT ) for ∆G ≥ 0. For an L × L lattice, one MC cycle consists of a chain of L2 elementary MC steps. After each cycle, the enthalpy of the lattice is calculated as follows: H = N F 21 wGF 21 . MC simulations naturally give access to a number of thermodynamic parameters of the system. For example, the excess heat capacity C(T ) can be calculated from the variance of equilibrium fluctuations of the total lattice enthalpy H as follows: 2 ∆E2 + N GF 1 ∆E1 + N F 22 + N GG 11 + N GF 12 + N GF 12 + N FF 12 wGG 12 + N GF 12 wGF 12 wFF 11 wGF 22 wGF ij C(T ) =(cid:10)(H − (cid:104)H(cid:105))2(cid:11)/RT 2. To adjust the lipid interaction parameters, we used the approach previously described by Sug´ar et al. [29, 31]: temperature-dependent excess heat capacity curves were obtained from MC simulations for a range of membrane compositions and compared with experimentally measured heat capacity curves, and the parameters wmn ij were varied until a reasonable agreement with experimental DSC data was achieved. As in the previous studies [31, 32], the procedure of tuning the lipid interaction parameters results in a single set of wmn ij which is used to describe the membrane at all temperatures and lipid compositions. Since a simpler lattice representation of the lipid system (lipid LipidsDMPCDSPCLipid statesgelfluidDMPCDSPCgelfluidCombined representationJ Ehrig, E P Petrov and P Schwille 7 Table 1. Thermodynamic parameters of lattice-based MC simulations of the DMPC/DSPC lipid membranes used in the present work and previous publications [31, 32].a Thermodynamic Parameters MC study ∆E1 ∆E2 (Jmol-1) (Jmol-1) ∆S1 ∆S2 (Jmol-1K-1) (Jmol-1K-1) wGF 11 (Jmol-1) wGF 22 (Jmol-1) wGF 12 (Jmol-1) wGF 21 (Jmol-1) wGG 12 (Jmol-1) wFF 12 (Jmol-1) Ref. [31], 12700 21970 42.65 67.02 1353 1474 1548 1715 565 335 triangular lattice of lipid chains Ref. [32], 13165 25370 44.31 77.36 1353 1474 1548 1716 607 251 triangular lattice of lipid chains Present work, 26330 50740 88.65 154.70 1827 1622 4025 4460 1412 502 square lattice of lipids [31] and Hac et al. a Simulations by Sug´ar et al. [32] were carried out on the triangular lattice of lipid chains. The parameters used in the present work differ from those of [31, 32] due to the fact that simulations were carried out on a square lattice of lipid molecules. Notice further, that the parameters in [31, 32] are given per mole of lipid chains while in the present work they are given per mole of lipid molecules. molecules arranged on a square lattice) was used in our simulations compared to the previous studies [31, 32] (lipids represented as dimers of acyl chains arranged on a triangular lattice), the lipid interaction parameters wmn in our study differ from those used in previous publications [31, 32] (table 1). In the present paper, unless explicitly stated otherwise, all simulations were carried out using this set of lipid interaction parameters. ij For each of the membrane compositions and temperatures addressed in the present study, simulations started with a random initial configuration of the membrane where lipids were randomly distributed on the lattice and all assigned to be in the fluid state (i.e., configurations corresponding to an infinitely high temperature). After that, the system was abruptly quenched to a particular target temperature, and simulations were carried out while keeping the temperature constant. The whole time-dependent evolution of the membrane starting with the abrupt temperature quench was recorded to study the domain coarsening and growth dynamics. To calculate the excess heat capacities and correctly address the other stationary properties of the membrane, including lipid diffusion dynamics, line tension of lipid domains, correlation lengths and susceptibilities, the membrane was first equilibrated. To achieve fast equilibration, a third substep, consisting in position exchange of two randomly chosen lipid molecules, was added to the Monte Carlo procedure. This approach, which was suggested in [42] and successfully applied in [31, 39, 43], is known to be very efficient in driving the system toward the equilibrium configuration. This procedure was propagated typically for 1.5×105 MC cycles, which for the lattice sizes used in this work is substantially longer J Ehrig, E P Petrov and P Schwille 8 than the typical time required by the total lattice enthalpy H to reach its equilibrium value at a given temperature in the presence of the random lipid exchange substep. After the completion of this procedure, the random lipid exchange substep is switched off, and the equilibrated system is ready for studies of its equilibrium properties. To ensure that the system has reached the full equilibrium, we verified whether the mean total lattice enthalpy relaxed to a constant value. In cases where equilibrium membrane configurations within the fluid -- gel phase coexistence region of the phase diagram were required, the complete equilibration was additionally checked by visual inspection of membrane snapshots. To represent the translational motion dynamics of lipids in the fluid and gel states in a realistic way, a rate function controlling the likelihood of next-neighbour exchange depending on the number of gel neighbours was introduced, as suggested in [32], to ensure a lower mobility of lipids in the gel environment. In particular, for two neighbouring lipids the next-neighbour exchange is accepted with a probability r = exp [− (Ngel neighbours/Nneighbours) (∆E/kBT )], where Nneighbours (Ngel neighbours) is the number of the next neighbours (in the gel state) for the pair of lipids attempting to exchange their positions; in our simulations on the square lattice of lipids Nneighbours = 6. The quantity ∆E ≥ 0 corresponds to the activation energy barrier for the next- neighbour exchange in the all-gel environment compared to the all-fluid environment. In our simulations, ∆E was adjusted to provide ca. 40 times slower translational diffusion of lipids in the gel phase compared to the fluid phase. Since the transition probability between the gel and fluid conformations of the lipid molecule in our simulations depends only on its immediate environment, and does not involve any time delay, it follows that at least the equilibrium properties of the membrane should not depend on the rate function. Simulations were typically carried out on 600 × 600 or 400 × 400 lattices, which corresponds to a membrane patch consisting of 3.6 × 105 and 1.6 × 105 lipid molecules, respectively. MC simulations were run for up to 2× 107 MC cycles, which allowed us to collect the necessary data for the analysis of thermodynamic properties of the membrane and translational diffusion of lipid molecules. The spatial scale and time intervals addressed in our simulations can be obtained by relating the simulation lattice size to the lipid head group dimensions, and subsequently finding the MC step duration by relating translational diffusion of lipids to experimental data. By assuming the lipid head group having a size of ca. 0.8 × 0.8 nm2 and using experimental data on DMPC diffusion coefficient in the fluid phase (3 − 6 µm2/s at T = 304.5 K [44 -- 46]), we find that one MC cycle of our simulations corresponds to ≈ 50 ns. Thus, our simulations describe the behaviour of a membrane fragment of a size of ≈ 0.48 × 0.48 µm2 over time intervals of ≈ 1 s. These are indeed the experimentally relevant scales we planned to achieve in our simulations. The program code is fully original and is written in Fortran completely from scratch except for the random number generation routine. The use of a good pseudo-random number generator is essential for the success of a Monte Carlo simulation study. In our J Ehrig, E P Petrov and P Schwille 9 simulations we therefore used the Mersenne Twister routine [47] providing sequences of pseudo-random numbers equidistributed in 623 dimensions characterized by an extremely long period of 219937− 1 ≈ 4.3× 106001. The Compaq Visual Fortran Compiler Ver. 6.6a (Compaq Computer Corporation, Houston, TX) or the Intel Visual Fortran Compiler Professional Edition Ver. 11.1 (Intel Corporation, Santa Clara, CA) was used for compilation. The use of the Intel compiler provides a ∼ 30% faster performance for the same code on the same workstation. Simulation results were analyzed using original dedicated routines written in MATLAB Ver. R2007b (The MathWorks, Natick, MA) and Fortran. Monte Carlo simulations were carried out on a workstation (Intel Core2 Quad Extreme CPU X9770 3.2 GHz, 4 GB RAM) running under Windows XP. Under these conditions, a simulation run on a 600 × 600 lattice for 2 × 107 MC cycles takes about 700 h (Compaq compiler) or 460 h (Intel compiler) of CPU time. The simulation method very naturally offers the opportunity of parallel implementation, which can further substantially speed-up the computation. Recent advances in the use of GPU- based computations [48, 49] also offer impressive speed-up of lattice based simulations and are planned to be implemented in our future work. 2.3. Data Analysis 2.3.1. Analysis of membrane configurations. To analyze the spatial distributions of lipids and lipid states, the radial autocorrelation function G(r) and the circularly averaged structure function S(k) were calculated from the radial distribution function g(r) as follows [50 -- 52] (cid:42)(cid:88) (cid:88) G(r) = g(r) − 1, g(r) = ¯ρ−2 S(k) = 1 + ¯ρF {G(r)} , j(cid:54)=i i (cid:43) δ(ri)δ(rj − r) , (2) (3) (4) where ¯ρ is the spatially averaged particle density, and F {·} denotes the Fourier transform. The fact that the particles cannot take arbitrary positions but rather occupy sites on the square lattice is taken into account to avoid artifacts in spatial and angular averaging. When studying equilibrium membrane properties, structure functions S(k) were averaged over several hundred equilibrium configurations of the membrane at a given temperature and composition. 2.3.2. Analysis of lipid diffusion and simulation of FCS experiments. To study the lateral diffusion of lipid molecules, positions (x, y) of a small fraction (< 0.05%) of lipid molecules were recorded, and the time- and ensemble-averaged mean-square displacement MSD(τ ) was calculated (cid:8)[x(t) − x(t + τ )]2 + [y(t) − y(t + τ )]2(cid:9), (5) MSD(τ ) = 1 tmax − τ tmax−τ(cid:88) t=1 J Ehrig, E P Petrov and P Schwille 10 where τ , t, and tmax are integers giving time in units of MC cycles; here, τ denotes the time lag, and tmax is the total length of the lipid molecule trajectory. Additionally, fluorescence correlation spectroscopy (FCS) [53] measurements were simulated: the tracked particles were assumed to be fluorescent, and the autocorrelation function G(τ ) = (cid:104)δF (t)δF (t + τ )(cid:105)/(cid:104)F(cid:105)2 of fluorescence intensity fluctuations δF (t) = F (t) − (cid:104)F(cid:105) around the mean intensity (cid:104)F(cid:105) in a 2D Gaussian detection volume 0) was calculated. The detection spot size r0 was 31 lattice units ≈ 25 exp(−2r2/r2 nm, the size experimentally achievable using the stimulated emission depletion (STED) FCS technique [54] (for more details see [40]). FCS curves were averaged over nine different positions on the lattice and analyzed using the model G(τ ) = G(0) 1 1 + (τ /τD)β , (6) where the characteristic decay time τD is the so-called diffusion time, which is related to the diffusion coefficient and the detection spot size (in case of normal diffusion, τD = r2 0/(4D)), and the exponent β is used to describe the deviation of the autocorrelation curve from the normal diffusion model. For β = 1 expression (6) corresponds to normal diffusion, while for 0 < β < 1 it provides a simple way to describe anomalous subdiffusion in FCS [53]. 3. Results 3.1. Phase Diagram of the DMPC/DSPC system based on heat capacity data The experimentally measured heat capacity curves in figure 3(a) capture the well-known feature of two-component lipid mixtures (see, e.g., [11]), namely, that in a two-lipid system with a fixed composition the all-fluid and all-gel states of the membrane are separated by a temperature range where coexistence of the fluid and gel phases takes place. This temperature range can span several tens of degrees. By contrast, phase coexistence is not observed in single-lipid systems, which, upon cooling down below the phase transition temperature, directly undergo a transition from the fluid to gel state. for the simulation of DMPC/DSPC on the square lattice, we find that our MC simulations reproduce well the experimentally measured excess heat capacity curves C(T ), as shown in figure 3(a, b). After successfully adjusting the interaction parameters wmn ij Historically, in the lipid studies, one usually speaks about broadening of the transition in a two-lipid system (here, by transition, one means the fluid -- gel transition, and the broadening is understood as compared to a single-lipid system), and the onset and completion temperatures determined from experimental data are usually discussed, with a fluid -- gel phase coexistence region located between these temperatures [31, 33, 37]. Going beyond the empirical experimental approach, one finds that the transition from a single-phase state of the membrane to a two-phase coexistence state is actually a phase transition by itself. A two-lipid system upon cooling from a high temperature at which J Ehrig, E P Petrov and P Schwille 11 Figure 3. (a, b) Excess heat capacity curves for DMPC/DSPC lipid membranes for a range of compositions (DMPC/DSPC = 0/100, 10/90, 20/80, ..., 90/10, 100/0) measured experimentally by differential scanning calorimetry (a) and obtained in our MC simulations (b). (c) Empirical phase diagram of DMPC/DSPC constructed from the excess heat capacity curves. Experimental data: present work (• ), [33] ((cid:3)), [34] (◦ ), [35] (♦), [36] ((cid:79)), [37] ((cid:47)), [38] ((cid:77)), [31, 39] ((cid:46)), our reanalysis of experimental C(T ) data for DMPC/DSPC 20/80 and 10/90 from [39] ((cid:46)· ), and [32] (×). Monte Carlo simulation data: present work ((cid:4)), [31] (+), [32] (∗). it is in the fluid state, would thus undergo two phase transitions: from the fluid state to the fluid -- gel coexistence, and, upon further cooling, another phase transition from the fluid -- gel coexistence to the gel state. To avoid confusion, we clarify that, wherever the expression "phase transition" in the present paper is applied to a two-lipid system, the phase transition from a single phase state into a two-phase coexistence state (or vice versa) is referred to. Studying the properties of these phase transitions, we believe, can reveal many details on the microscopic organization of the lipid membrane. Heat capacity curves C(T ) (both obtained experimentally and in MC simulations) were analyzed using the tangent method (see, e.g. [11]) to determine the onset and completion temperatures (i.e., estimates of the temperature of the transition from a single-phase state to two-phase coexistence). To do that, the outer slopes of the C(T ) profiles for a range of compositions (DMPC/DSPC = 0/100, 10/90, 20/80, ..., 90/10, 100/0) were fit with straight lines passing through the corresponding inflection points. The onset and completion temperatures in this case are defined as the respective intersections of the tangential lines with the zero-line. As a result, an experimental phase diagram for the binary lipid mixture can be constructed. We point out that in our work we apply the tangent method to analyze C(T ) data for all membrane compositions, inculding the single-lipid systems. This method results in a finite transition width for pure DMPC and DSPC membranes. The transition widths of single-lipid systems J Ehrig, E P Petrov and P Schwille 12 obtained in our MC simulations are in agreement with previous lattice-based MC simulation results [29], and as previously the simulated C(T ) peaks of individual lipids are wider than the ones observed experimentally -- for discussion see [29]. Notice that, in contrast to the present work, the onset and completion temperatures for the single lipids are frequently not reported in the literature, but rather a single temperature corresponding to the peak of the C(T ) dependence is given in the phase diagram. As is evident from figure 3(c), our MC simulation-based data are in perfect agreement with experimental results (both obtained in the present work and published earlier by other groups [31 -- 38]), as well as with results from the previous simulation studies [31, 32] carried out on the triangular lattice of lipid chains. This shows that the particular choice of a lattice geometry for lattice-based MC simulations should not be important, provided the lipid interaction parameters are chosen in a proper way.† It should be pointed out that generally there is neither proof nor guarantee that the onset and completion temperatures extracted from the C(T ) curves by the tangent method necessarily correspond to the actual temperatures for the phase transitions from the gel state to fluid -- gel coexistence and from fluid -- gel coexistence to the fluid state, respectively. Therefore, the empirical phase diagram constructed on the basis of heat capacity data may differ from the real phase diagram of the system and thus not provide an insight into the microscopic structure and the dynamics of the membrane. 3.2. Phase separation behaviour of the DMPC/DSPC system. Binodals and spinodals A closer look at the membrane snapshots from the MC simulation at different compositions and temperatures, as shown in figure 4, indeed reveals that the phase behaviour is, in fact, more complicated. In the region bounded by the curves of the onset and completion temperatures, as determined from the experimental and simulated C(T ) data (circles and crosses in figure 4), two types of phase behaviour can be observed: i) complete separation of the gel and fluid phases and ii) a highly dynamic intermixing of the two phases. To gain better understanding of this phenomenon, the radial autocorrelation function G(r) (2) and the structure function S(k) (4) were calculated for a wide range of membrane compositions and broad temperature intervals covering the whole area of the phase diagram. Importantly, our simulations allow us to study G(r) and S(k) for the spatial distributions of both lipids (DMPC or DSPC) and their states (fluid or gel) independently, thereby providing us with valuable details on the microscopic lipid organization in the membrane and its role in the dynamics of the phase separation. † While the arguments in favour of representing a lipid membrane as a triangular lattice of lipid chains may seem more or less sound in case of the gel state of the membrane, they do not look nearly as convincing in case of the fluid state. Moreover, there exists experimental evidence that the arrangement of lipid chains in the gel phase shows only a short-range positional order and thus represents the hexatic phase [55 -- 57], and therefore the triangular lattice of chains does not correctly represent the structure of the gel phase. This seriously weakens the arguments in favour of the triangular lattice of chains as the only correct choice in lattice-based lipid membrane simulations. J Ehrig, E P Petrov and P Schwille 13 Figure 4. Left-hand panel: Phase diagram of the DMPC/DSPC lipid mixture. Empirical results obtained from the analysis of excess heat capacity curves from differential scanning calorimetry measurements (◦ ) and MC simulations (+). Lipid state spinodal (- - - -), lipid state binodal ( -- -- -- ) and lipid demixing curves ( -- -- ) are shown. Right-hand panel: Membrane configurations corresponding to the compositions and temperatures marked by filled squares on the phase diagram. Lattice size: 600 × 600; scale bar: 200 lattice units ≈ 160 nm. We found that outside the region with a clear coexistence of two phases (the exact boundaries of this region will be discussed in what follows), structure functions of lipid states SS(k) are well described by the Ornstein -- Zernike (OZ) approximation [51] SS(k) = SOZ(k) = S0/[1 + (ξk)2], where k is a wavenumber, ξ is the correlation length of fluctuations, and the amplitude S0 is proportional to the order parameter susceptibility of the system.‡ The analysis of the structure functions in terms of the OZ approximation allowed us to study the temperature dependences of the correlation length ξS(T ) and the amplitude SS,0(T ) of the structure function, and thus to determine the lipid state spinodal and binodal. The binodal curves were determined from the condition d(1/ξS(T ))/dT = 0 [58]. The estimated accuracy of the binodal determination was found to be ±1 K. Notice that, when extended to DMPC/DSPC 0/100 and 100/0 compositions, the upper and lower binodal curves are expected to meet at the respective lipid melting transition temperatures, which is in agreement with the data we obtained (see figure 4). The origin of the nonmonotonic behavior of the upper binodal at DSPC mole fractions exceeding ‡ The order parameter susceptibility characterizes the sensitivity of the order parameter in the system to the external perturbation. For example, for a liquid -- gas transition the order parameter is the local density in the system, and the role of the order parameter susceptibility is played by the isothermal compressibility; in case of the Ising model consisting of magnetic spins, one deals, correspondingly, with the magnetic susceptibility. In the membrane system in question, the local order parameter corresponds to the lipid conformation, and the order parameter susceptibility will, among other things, characterize the efficiency of lipid domain formation around proteins preferentially wetted by one of the membrane phases (see [40] and refs. therein). J Ehrig, E P Petrov and P Schwille 85% still needs to be explained. 14 The spinodals can be determined by extrapolating the 1/SS,0(T ) dependence to zero crossing [58]. To determine the position of the upper (lower) spinodal from the results of our simulations, a set of SS,0(T ) data above the upper binodal (or below the lower binodal, respectively) was fit for every membrane composition to a dependence 1/SS,0(T ) = B1 − T /Tspp, where B is a constant prefactor, Tsp is the spinodal temperature, and p = O(1). To do that, typically 7 − 14 data points spanning a temperature range of 5 − 10 K were used. Note that for a system at criticality, the exponent p is nothing but the critical exponent γ (see section 3.8). The accuracy in determination of spinodals ranges from ±1 K in the regions where they approach close to binodals to ±3 K in the regions where the spinodals depart from the binodal curves. Surprisingly, we found that the phase diagram based on the binodal and spinodal curves differs significantly from the empirical phase diagram obtained from the heat capacity data (figure 4). What is the reason for such a strong difference? To answer this question, we analyzed the structure functions SL(k) characterizing the spatial distribution of lipids in the membrane outside the phase coexistence region. Quite unexpectedly, it turned out that the structure functions of lipids cannot be adequately described by the simple OZ approximation, and two OZ components are necessary to provide an adequate fit of the structure functions of lipids: SL(k) = SOZ L2 (k) (figure 5). L1 (k) + SOZ It should be noted that in order to provide a perfect fit of the SS(k) and SL(k) data within the whole range of wavenumbers, it was required that a small positive offset had to be included into the corresponding fit functions. This offset, being important only at high k-numbers, is clearly a consequence of carrying out simulations on a finite-size lattice, and was consistently reproduced in our simulations of the Ising model on the square lattices of the corresponding sizes (data not shown). It appears that parameters SOZ L1 (0) and ξL1 of the first OZ-component only weakly depend on temperature and correspond to demixing of lipids in the same state; SOZ L2 (0) and ξL2, on the other hand, show strong temperature dependences similar to those of SS(0) and ξS (figure 6(a, b)) and thus reflect the appearance of dynamic domains. L1 (0) = SOZ We found that for each lipid composition, the temperature at which the amplitudes of the two OZ components used to describe the structure function of the spatial distribution of lipids are equal, SOZ L2 (0), perfectly matches the temperature corresponding to the peak in the C(T ) profile. Physically, this means that at this temperature the susceptibilities responsible for the spontaneous clustering of molecules of the same lipid (irrespectively whether they are in the fluid or gel conformation) and their clustering into a particular membrane phase, become equal, and the peak in the heat capacity, in fact, marks the intense local lipid demixing due to transition from a single phase to a two-phase coexistence state. Therefore, we termed the curve defined by the condition SOZ L2 (0) as the lipid demixing curve. Notice that the lipid demixing curve very closely approximates the empirical phase diagram based on the heat capacity data (figure 4). Therefore, in what follows we will refer to the boundaries L1 (0) = SOZ J Ehrig, E P Petrov and P Schwille 15 Figure 5. Representative structure functions characterizing the spatial distribution of lipid states SS(k) (a) and lipids SL(k) (b) for a set of temperatures above the phase transition (top to bottom: T = 322, 322.5, 323, 324, 325, 326, 328, 331, and 335 K). Weighted least-squares fit of SS(k) at T = 322.5 K using the single-component OZ approximation (c); weighted least-squares fit of SL(k) at T = 322.5 K using the two-component OZ approximation (d). The grey dashed lines in (d) show the single components of the fit. For clarity, S(k) data are shown without the constant offset. of the empirical phase diagram as heat capacity-based lipid demixing curves. We point out that in the region of the phase diagram corresponding to the gap between the binodal and the lipid demixing curve, the membrane shows behaviour characteristic of neither the single-phase nor two-phase coexistence states. Namely, on the one hand, the membrane is not uniform anymore as in the fluid state, while on the other hand, the fluid and gel domains are very poorly defined, have sizes on a wide range of spatial scales, and show non-stop fluctuations without any signs of nucleation and growth. In fact, this behaviour is characteristic of near-critical fluctuations expected to take place in the vicinity of a critical point. This suggestion is further supported by the J Ehrig, E P Petrov and P Schwille 16 Figure 6. Temperature dependences of the amplitudes (a) and correlation lengths (b) of the OZ components for lipids and lipid states of the DMPC/DSPC 20/80 mixture. The arrows indicate the positions of the binodal (B), spinodal (S), and lipid demixing (L) temperatures. behaviour of the binodal and spinodal curves which, within the limits of our accuracy, touch in the corresponding region of the phase diagram, which should take place at the critical point. We will come again to the issue of the critical behaviour of the membrane in more detail later, in sections 3.4 and 3.8. Here we remark that in an earlier MC simulation-based study of a DMPC/DSPC membrane [43], a conclusion was made based on the analysis of size distributions of fluid and gel clusters, that the phase diagram of this lipid system is much more complicated than that based on heat capacity data (the work [43], however, did not discuss the possibility of the presence of a critical point in the system). Although the quantitative results of the above paper may suffer from a very small system size addressed (40 × 40 lipid chains), on the qualitative side, these conclusions are largely in agreement with the ones we draw in the present study using different arguments. 3.3. Transient lipid domains versus thermodynamically equilibrium phases It has been pointed out elsewhere [11] that, when discussing phase separation in lipid membranes, one should make a clear distinction between the thermodynamically stable phases and (usually, small) transient membrane domains statistically appearing and disappearing in an equilibrated membrane. In this context, we would like to check to what extent the concept of the thermodynamically equilibrium phase can be applied to results of our simulations. This is important in view of the apparent controversy regarding the applicability of the lever rule to the phase separation in two-component lipid membranes: on the one hand, the recent laser scanning microscopy experiments on two-component membranes (GUVs) [59] show that the gel and fluid domains in two-component membranes do indeed correspond to equilibrium phases, and therefore the lever rule applies to their J Ehrig, E P Petrov and P Schwille 17 description, on the other hand, in previously published MC simulation studies [31, 39] on the DMPC/DSPC systems the authors found out that the fluid and gel phase in the two-phase coexistence region do not follow the lever rule (this was interpreted in [31, 39] as being a consequence of the fact that the gel -- fluid transition in DMPC/DSPC mixtures is not a first-order phase transition). To resolve this contradiction, we studied whether the binodal curves which were determined in the previous section of the paper, can be successfully reconstructed by determining for different fixed membrane compositions the temperature dependence of the composition of the individual phases, which should exactly be the case if the lever rule applies to the system. Once one can clearly distinguish between the fluid and gel phases, the problem becomes trivial and is reduced to determining the relative amounts of the two lipids in each of the assigned membrane phases. In practice, of course, this should be carried out using only equilibrium membrane configurations, and averaging over a number of configurations is required to obtain reliable results. In our analysis, we first assumed, as it has been done in [31], that the fluid and gel phase in a particular membrane configuration are identified with the conformational state of the lipid: all lipids in the fluid conformation are counted as belonging to the fluid phase, and, correspondingly, all lipids in the gel conformation are counted as belonging to the gel phase. The results obtained in this way are presented in figure 7(a). Clearly, this analysis fails to consistently reproduce the binodals of the system, which is in qualitative agreement with previous results of [31]. Our interpretation of the reason why this analysis fails is different from the one previously proposed in [31]. A careful examination of membrane configurations allowed us to conclude that this analysis fails due to the presence of small short-lived fluid or gel domains, which, because of their transient nature, cannot be considered as a part of the thermodynamically stable phase. If counted as belonging to a particular phase, these domains can severely distort the phase ratio, as well as the lipid composition of the lipid and gel phases, and lead to a break-down of the lever rule. Therefore, one can hope that, if the small transient membrane domains are excluded from the analysis, and only the thermodynamically stable membrane phases are taken into account, the same analysis will reproduce the binodal curves of the system with a better accuracy. The problem, thus, boils down to selecting an appropriate rule to discriminate between small transient membrane domains and thermodynamically stable phases. We found that this could be successfully accomplished by low-pass spatial filtering of the lipid state configurations of the membrane. In particular, we found that a procedure consisting of applying a 2D Gaussian filter with the standard deviation of σ = 5 lattice units and subsequent thresholding the result at the 50% level resulted in consistent reproducible results.§ This procedure applied to results of MC simulations with lipid § An approach based on time averaging of membrane configurations over reasonably short simulation intervals was found to give similar results. J Ehrig, E P Petrov and P Schwille 18 Figure 7. DMPC/DSPC phase diagram reconstructed from the analysis of the temperature-dependent lipid compositions of the fluid and gel phases. Red and blue curves show the compositions (DSPC mole fraction) of the fluid and gel phases, respectively, as a function of temperature for different membrane compositions. (a) Direct calculation from equilibrium membrane configurations. (b) Calculation after selecting the gel and fluid phases by segmentation of the lipid state configurations (see text for details). Additionally, the lipid state binodal (long-dashed curves) and spinodal (short-dashed curves), and lipid demixing curves (grey solid curves) are shown (cf. figure 4). compositions DMPC/DSPC in the range from 20/80 to 80/20, consistently reproduced the upper and lower binodals of the system (figure 7(b)). The above procedure used for detection of fluid or gel phases in the membrane eliminates from the analysis small domains which contain below ∼ 50 lipid molecules, which is in agreement with the intuitive understanding that a portion of a membrane that can be classified into a thermodynamically stable phase should consist of a number of molecules much larger than one. Our present analysis shows that this number is of order of 102. Domains of this size ((cid:46) 10 × 10 nm2) are clearly below the optical resolution and are not observable in confocal fluorescence microscopy experiments, which additionally have a finite exposure time, thereby eliminating features fluctuating on a short time scale. Thus, the apparent controversy between the recent fluorescence microscopy data [59] and previously published results of MC simulations [31] is resolved if one takes into consideration only relatively large and stable membrane domains which constitute the thermodynamically stable phases, which is certainly the case for the domains detectable in fluorescence microscopy experiments. 3.4. Different scenarios of phase transition A careful examination of the membrane simulation snapshots at various temperatures and lipid compositions shows that, depending on the membrane composition, the phase transition from the fluid phase to the fluid -- gel phase coexistence can take place according J Ehrig, E P Petrov and P Schwille 19 to two difference scenarios. At compositions for which the lipid state binodal coincides with the lipid demixing curve (see figure 4), the transition has a quasi-abrupt character. This is exemplified in figure 8 by the corresponding temperature sequences of equilibrated membrane snapshots for DMPC/DSPC 80/20 and 50/50. There, lowering the temperature below the transition temperature Tt immediately leads to an abrupt fluid -- gel phase separation and formation of a circular-shaped, stable domain. Above Tt, the membrane stays in the all-fluid phase. On the other hand, in the region of the phase diagram where the lipid demixing curve strongly deviates from the lipid state binodal, a completely different scenario is observed. In particular, when the membrane is cooled down to temperatures below the lipid demixing curve, transient random-shaped fluctuating domains appear. As can be seen from the corresponding snapshots for DMPC/DSPC 20/80 in figure 8, with lowering the temperature, these domains become larger, but remain fractal-shaped and highly dynamic. The mean size of these gel and fluid domains, as well as the order-parameter susceptibility, which are determined from the analysis of the structure function SS(k), diverge when approaching the temperature of phase transition (see section 3.8). This means that, compared to the scenario discussed above, the character of the transition is changed, and we are dealing with a continuous phase transition. We find that within the corresponding range of lipid compositions, the lipid state spinodal closely approaches the binodal. Eventually these curves touch at a common (local) maximum, i.e. at the point where the compositions of the fluid and gel phase become indistinguishable, which should take place at a critical point. The shapes of the binodal and spinodal curves suggest that the critical composition of the membrane should be close to DMPC/DSPC 20/80. To determine the position of the critical point more exactly, we construct a curve in the phase diagram along which the amounts of lipids in the fluid and gel states are equal: Xfluid = Xgel = 0.5. At the critical point one should generally expect that not only the compositions of the fluid and gel phase become indistinguishable, but also their relative contributions become equal. Indeed, as can be clearly seen from figure 9, this curve crosses the binodal and spinodal lines exactly at the point where they touch, which confirms the presence of a critical point at this position. Thus, in the region of the phase diagram where the lipid demixing curve strongly deviates from the lipid state binodal, approaching the phase transition is accompanied by near-critical fluctuations of the fluid and gel phases. Cooling the membrane to temperatures below the phase transition leads to large-scale phase separation, and a single large domain of a fluid or gel phase is eventually formed in a fully equilibrated membrane. Notice that in the vicinity of the critical point, these domains show strong thermally induced shape fluctuations due to a decrease of the line tension in the vicinity of the critical point (see discussion in section 3.7). The existence of critical points in lipid membranes has been suggested in several recent experimental studies. In particular, (near-)critical behaviour was observed in giant unilamellar vesicles of ternary lipid mixtures [60] and in giant plasma membrane J Ehrig, E P Petrov and P Schwille 20 Figure 8. Representative equilibrium configurations obtained in MC simulations of the DMPC/DSPC system with three different compositions DMPC/DSPC = 80/20, 50/50, and 20/80 for a set of temperatures relative to the phase transition from the fluid state to fluid -- gel coexistence. For comparison, results for the pure DMPC and DSPC membranes depicting their fluid -- gel transitions are shown. Transition temperatures for DMPC/DSPC 100/0, 80/20, 50/50, 20/80 and 0/100 are Tt = 297.0, 309.7, 318.7, 320.5, and 328.0 K, respectively. Lattice size: 600 × 600; scale bar: 200 lattice units ≈ 160 nm. See text for a discussion of the different phase transition scenarios. TTTTTTTTTTJ Ehrig, E P Petrov and P Schwille 21 Figure 9. Phase diagram of the DMPC/DSPC system demonstrating the presence of a critical point. Lipid state spinodal (short-dashed lines), lipid state binodal (long- dashed lines) and lipid demixing curves (solid grey lines) are the same as in figure 4. The solid cyan line marks temperatures at which the DMPC/DSPC membrane shows equal amounts of the fluid and gel phase Xfluid = Xgel = 0.5. The critical point is marked with an open circle. Within the two-phase fluid -- gel coexistence region, areas marked as "gel in fluid" and "fluid in gel" correspond to the gel and fluid minority phases, respectively. Upon complete equilibration, the minority phase always forms a single circular-shaped domain surrounded by the majority phase. vesicles (with a lipid composition close to the one of the cell plasma membrane) [61]. Remarkably, the behaviour observed in [61] is very similar to the one in our MC simulation: depending on the membrane composition, the transition to the two-phase coexistence takes place either in an abrupt manner -- when the membrane does not pass through a critical point, or via critical fluctuations -- when the membrane does pass through a critical point (compare our figure 8 with figure 1(d, e) of [61]). Furthermore, there is experimental evidence of the existence of a critical point in the binary DMPC/DSPC lipid mixture. Atomic force microscopy experiments [62] provided structural evidence of criticality in DMPC/DSPC monolayers. More interesting conclusions were made [63 -- 65] for DMPC/DSPC bilayers, where the existence of a critical point hidden within the fluid -- gel coexistence region of the phase diagram has been suggested based on small-angle neutron scattering experiments. Notice that the critical point we find in our MC simulation is also located within the region of the phase diagram, where, based on the calorimetry data, the fluid -- gel coexistence would be generally expected. In addition, we arrive at quite an interesting conclusion: namely, that the binodals of the system are best reproduced via the analysis of the lipid compositions of the fluid and gel phases (see section 3.3) exactly at those membrane compositions and temperatures where the phase transition in the lipid system takes place in a quasi- abrupt manner, i.e., when it is close by its properties to the first-order transition, that is exactly under conditions where this analysis is expected to hold. J Ehrig, E P Petrov and P Schwille 22 In case of a single-lipid membrane, the phase transition also takes place via spatial fluctuations of the fluid and gel domains (see corresponding sequences of membrane snapshots in figure 8). For a single-lipid system, according to the Gibbs' phase rule, no phase coexistence can take place either above or below the transition temperature. In agreement with that, the results of MC simulations for single lipids in the vicinity of the phase transition show transient fluid or gel domains, similar to what was observed in a recent atomic-scale molecular dynamics simulation study [66]. Similar to [66] we find that at T = Tt + 8 K these small domains typically exist for no longer than ∼ 100 ns. Even at the transition temperature the domains are transient and their sizes are relatively small (∼ 100 lipids for DMPC and ∼ 20 lipids for DSPC) and, importantly, do not change with an increase of the simulation lattice size. Remarkably, this is in a very good agreement with results of a previous MC simulation study of a DPPC membrane using a ten-state Pink model [67], where the mean cluster size at the transition temperature was estimated to be ∼ 40 lipids, in agreement with experimental estimates [68]. Interestingly, this also very well agrees with our estimate of the size of transient lipid clusters we obtained above in section 3.3. Before concluding this section, three important points should be emphasized. First, equilibration of the membrane within the fluid -- gel phase coexistence region in our simulations always ultimately results in formation of a single domain of the minority phase, surrounded by the continuous majority phase. In fact, this should be expected in this system, because in the phase coexistence region of the phase diagram the system evolution is driven so as to minimize the line tension energy, and, since no explicit or implicit penalties on the domain growth are imposed in the present simulation, a single minority phase domain should eventually be formed. On the other hand, in cases where domains of the minority phase show a different curvature compared to the majority phase, the curvature mismatch can result in such a penalty, which at some point stops the growth of domains and prevents their coalescence (see, e.g., [69 -- 73]). Therefore, the presence of a number of minority phase domains in a two-phase coexistence region in a lattice-based simulation which does not penalize the domain growth and/or coalescence should generally mean that this membrane configuration is still far from the equilibrium one, and therefore, any results on diffusion of lipids (by either studying the time dependence of the mean-square displacement or simulating results of fluorescence correlation spectroscopy (FCS) experiments) obtained in this regime have nothing to do with the equilibrium properties of the membrane. Second, we remark on the shape of the domains of the minority phase formed as a result of phase separation after the membrane is abruptly quenched from T = ∞ to a temperature within the fluid -- gel coexistence. For the minority phase content X < 1/π the domains always have a circular shape. For larger fractions of the minority phase, a single circular domain is also initially formed. Notice, however, that if simulations are carried out, as in the present paper, on a 2D square area with periodic boundary conditions (toroidal boundary conditions), and the fraction of the minority phase is in the range of 1/π < X < 1/2, the complete equilibration of the system should eventually J Ehrig, E P Petrov and P Schwille 23 lead to the formation of a stripe-shaped domain, whereas a circular domain represents an extremely long-lived metastable state [74, 75]. For simulations on the L× L lattice, the time τ required for the conversion of a circular domain to a stripe-shaped domain in case of the Ising system strongly depends on the system size [74]: τ ∝ L2 exp(2sλL/kBT ), where s = 0.1346... and λ is the phase interface line tension (for determination of line tension in our system, see section 3.7). Our preliminary numerical experiments showed that the behaviour of our system is consistent with the above expression, in spite of the fact that our system is more complicated than the Ising model. It should be emphasized that for large enough lattices the time τ becomes extremely long: we estimate it as τ ∼ 1010 MC cycles for L = 400 and τ ∼ 1013 MC cycles for L = 600. Moreover, we believe that the stripe shapes of minority phase domains, which are expected at the minority phase content 1/π < X < 1/2 in lattice-based simulations with periodic (toroidal) boundary conditions, are, in fact, of little importance when simulations are directly compared with experimental data: according to recent results based on off-lattice simulations [75], the formation of a band-shaped domain is just an artifact of the toroidal boundary conditions, and is not observed on a surface of a sphere if the particle interactions are isotropic. The latter observation generally agrees with experiments on phase separation in giant unilamellar vesicles. In a similar way, circular domains are also expected for supported lipid bilayers (which obviously do not have periodic boundary conditions). Note that the present discussion is restricted exclusively to the effects of boundary conditions in simulation studies and is not related to experimental findings for two- component membranes, where, depending on the membrane tension, either circular- shaped or stripe-shaped domains can be observed [76]. There, the appearance of stripe- shaped domains is most likely related to the anisotropic orientational distribution of lipid molecules induced by the high membrane tension. Third, outside the fluid -- gel phase coexistence region, small transient domains can spontaneously form and afterwards spontaneously dissolve after very short time intervals. These domains have a typical diameter of ∼ 10 lipids or smaller, and, importantly, their average size does not change with an increase in the size of the simulation lattice. Because of their extremely small size and short lifetime, their presence should be of no importance for optical microscopy-based experiments, as well as for simulations of these experiments at the experimentally relevant spatial scales and time intervals. 3.5. (In)Sensitivity of the phase diagram to perturbation of lipid interaction parameters In this section we will demonstrate that the main features of the phase diagram obtained in our MC simulations are stable with respect to a specific choice of the lipid interaction parameters. In particular, the presence of the critical point in the phase diagram is not affected by reasonable perturbations of the interaction parameters. A straightforward approach to this issue would involve a systematic study of the J Ehrig, E P Petrov and P Schwille 24 phase diagram in full detail for a number of differently perturbed sets of the interaction parameters, i.e. determining binodals, spinodals and lipid demixing curves, as well as checking the presence of near-critical fluctuations and determining the position of the critical point. This approach, however, would involve immense computational efforts. As it was discussed in the previous section, away from the critical point, the phase transition takes place in an abrupt manner, whereas in the vicinity of the critical point, the approach to the phase transition takes place via near-critical fluctuations (see figure 8). Therefore, to study the stability of phase diagram with respect to perturbation of the set of interaction parameters on the qualitative level it should be enough to check whether the phase transition takes place abruptly or via the near-critical fluctuations, for respective membrane compositions, independent of reasonable perturbations of the lipid interaction parameters. One should generally expect that if the presence of the critical point is stable with respect to reasonable perturbations of the wmn , the membrane composition and temperature characterizing the critical point should not change much, and a behaviour similar to the one depicted in figure 8 for each of the lipid compositions should also be observed with the perturbed interaction parameters. ij As discussed in section 2.2, the adequate description of the experimental heat capacity data by the MC model requires to determine the lipid interactions parameters wmn ij with the accuracy of a few percent. Therefore, it seems reasonable to carry out the stability analysis by introducing 10% perturbations to the parameters wmn . ij 12 , wGF 11 , wGF 12 , wFF 22 , wGG To simplify the notation, we combine the parameters wmn in a single vector 21 }. Then results of simulations for the unperturbed W = {wGF parameter set W should be compared with the results obtained for several perturbed parameter sets W + ∆(k). The components of the perturbation vector ∆(k) were defined as follows: ∆(k) i Wi, where ζ(k) is a (random) vector whose components take values ±1, and δ = 0.1 is the relative perturbation level. i = δζ (k) 12 , wGF ij for three 50/50, carried out Simulations were and 20/80 for a set of lipid compositions DMPC/DPSC 80/20, temperatures Tt + ∆T , ∆T = −3,−2,−1,−0.5, +0.5, +1, +2, +3 K for the respective compositions, using four dif- ferent perturbation vectors ∆(k), k = 1, 2, 3, 4, characterized by the following ζ(1) = {+1,−1, +1,−1,−1, +1}, ζ(2) = {+1,−1,−1,−1, +1, +1}, ζ(3) = ζ(k): {−1, +1,−1, +1, +1,−1}, and ζ(4) = {−1, +1, +1, +1,−1,−1}. As it is clear from figure 10, although the 10% perturbations in the interaction parameters can shift the phase transition temperature up or down by up to 2 K and slightly change the relative amount of the fluid and gel phases, the character of the transitions at these three lipid compositions remains essentially the same. Namely, while for DMPC/DSPC 80/20 and 50/50 the transition remains abrupt, the near-critical fluctuations accompanying the phase transition for DMPC/DSPC 20/80, irrespectively of perturbation of the parameters, indicate that the phase diagram and its critical point are stable with respect to the reasonable perturbations of the lipid interaction parameters. We find that varying the single-lipid interaction parameters (within reasonable J Ehrig, E P Petrov and P Schwille 25 Figure 10. Equilibrium membrane configurations for three lipid compositions DMPC/DSPC 80/20, 50/50, and 20/80 obtained with the unperturbed set of lipid interaction parameters W, four different perturbed sets of lipid interaction parameters W + ∆(k), k = 1, 2, 3, 4, and the alternative set of parameters W(cid:48) (see text). The results are presented for the set of temperatures Tt + ∆T , ∆T = −3,−2,−1,−0.5, +0.5, +1, +2, +3 K for the respective compositions. Transition temperatures for DMPC/DSPC 80/20, 50/50, and 20/80 are Tt = 309.7, 318.7, and 320.5 K, respectively. See text for details. Lattice size: 400×400; scale bar: 150 lattice units ≈ 120 nm. limits) strongly affects the width of the C(T ) peak for single lipids; in particular, choosing these parameters lower than the optimum will lead to broadening of the single- lipid C(T ) peaks. On the other hand, even with a non-optimal choice on non-optimal 318.2 K319.2 K319.7 K320.7 K321.7 K317.7 K316.7 K315.7 KW+∆(2)W+∆(3)W+∆(4)DMPC/DSPC 50/50WW+∆(1)W´309.2 K310.2 K310.7 K311.7 K312.7 K308.7 K307.7 K306.7 KW+∆(1)W+∆(2)W+∆(3)W+∆(4)DMPC/DSPC 80/20WW´320.0 K321.0 K321.5 K322.5 K323.5 K319.5 K318.5 K317.5 KWW+∆(1)W+∆(2)W+∆(3)W+∆(4)DMPC/DSPC 20/80W´J Ehrig, E P Petrov and P Schwille 26 single-lipid interaction parameters, the rest of the parameter set can be chosen so that C(T ) profiles for mixtures of DMPC/DSPC within the range 10/90 to 90/10 are well reproduced, which allows one to reproduce well the main features of the empirical phase diagram. A thorough reanalysis of our results, that we carried out recently after the main body of our study was finished, allowed us to conclude that our original choice of singe-lipid interaction parameters was sub-optimal, which lead to broadening of single- lipid C(T ) peaks in our MC simulations. Nevertheless, as mentioned above, this set of parameters very well reproduces the phase diagram for membrane compositions ranging from 10/90 to 90/10. Adjusting the single-lipid interaction parameters to reproduce the experimentally observed widths of C(T ) peaks of single lipids along with readjustment of the two-lipid interaction parameters has resulted in the 21 } = following set of lipid interaction parameters W(cid:48) = {w(cid:48)GF {2030, 2211, 1356, 435, 3174, 3002} J mol−1. 22 , w(cid:48)GG 11 , w(cid:48)GF 12 , w(cid:48)FF 12 , w(cid:48)GF 12 , w(cid:48)GF We have found that with this choice of interaction parameters provides additionally excellent description of the empirical phase diagram also outside the range 10/90 to 90/10 (data not shown). On the other hand, we found that the characteristic behaviour of the system was not changed if the parameter set W(cid:48) was used instead of W. Similar to what is observed with the parameter set W, the use of the parameter set W(cid:48) leads to the near-critical fluctuations in the upper right part of the phase diagram and quasi-abrupt phase transitions elsewhere. This is demonstrated in figure 10, where the rightmost columns show the temperature behaviour of the membrane simulated using the parameter set W(cid:48). A preliminary analysis shows that changing from the lipid interaction parameter set W to W(cid:48), while not affecting the qualitative picture, does lead to certain quantitative changes: in particular, we believe that, if the parameter set W(cid:48) is used, the critical point is shifted closer to the lipid composition DMPC/DSPC 15/85 with the critical temperature close to 322.5 K. 3.6. Anomalous diffusion of lipid molecules due to near-critical fluctuations In our recent work [40], using the same MC simulation and lipid system as in the present study, we showed that near-critical fluctuations can lead to transient anomalous subdiffusion of lipid molecules in a membrane. We concluded there that this phenomenon should occur irrespectively of particular details of the system except for the proximity to the critical point. Here, we demonstrate that, indeed, also for the alternative set of parameters W(cid:48) discussed in section 3.5, the transient anomalous subdiffusion can be observed when the system is in a state of near-critical fluctuations. As already discussed above, the near-critical behaviour in the membrane is preserved when changing the parameter set from W to W(cid:48). It appears that the critical point is slightly shifted and is located close to DMPC/DSPC 15/85, T = 322.5 K. We studied whether the diffusion of DMPC molecules in a DMPC/DSPC 15/85 mixture at T = 323 K (very close to the critical point) and T = 328 K (away from the critical J Ehrig, E P Petrov and P Schwille 27 Figure 11. Transient anomalous diffusion of DMPC lipids in a DMPC/DSPC mem- brane close to the critical point for two different sets of lipid interaction parameters. Mean square displacement MSD(τ ) (a, d), local exponent d log MSD(τ )/d log τ (b, e), and FCS autocorrelation G(τ )/G(0) (c, f) for the two sets W (a -- c) and W(cid:48) (d -- f) of lipid interaction parameters (see text). Red curves show results of simulations carried out at DMPC/DSPC 20/80, T = 321 K (a-c) and DMPC/DSPC 15/85, T = 323 K (d-f) -- the temperatures and compositions close to the critical point for the respective parameter sets. Grey curves show results of simulations carried out at the same re- spective membrane compositions but at T = 328 K, away from the critical point. FCS results (c, f) are shown along with fits using the FCS diffusion model (6) ( -- -- -- ) -- for DMPC/DSPC 20/80 (c) this yields β = 1.01 for T = 328 K and β = 0.86 for T = 321 K; for DMPC/DSPC 15/85 (f) β = 1.02 for T = 328 K and β = 0.84 for T = 323 K. For comparison, fits with the fixed β = 1.0 are shown for DMPC/DSPC 20/80, T = 321 K (c) and DMPC/DSPC 15/85, T = 323 K (f)(······). point) shows the same behaviour as the one described in [40]. The MSD curves in figure 11(a, d) along with the plots of their local exponents d log MSD(τ )/d log τ (figure 11(b, e)) clearly show that the qualitative behavior is very similar for the two parameter sets. Namely, that near-critical fluctuations in the membrane can lead to transient anomalous subdiffusion that extends over several orders of magnitude in time while away from the critical point, diffusion is normal. This is also reflected by the FCS autocorrelation curves presented in figure 11(c, f). In agreement with the MSD data, at high temperatures, away from criticality, the diffusion is normal. When the system is approaches criticality the appearance of transient anomalous diffusion behaviour dominates the presented FCS curves, which can only be successfully described using (a)(b)(c)(d)(e)(f)J Ehrig, E P Petrov and P Schwille 28 the model for anomalous diffusion. In particular, fits using (6) result in β = 0.86 (figure 11(c), DMPC/DSPC 20/80, T = 321 K, parameter set W) and β = 0.84 (figure 11(f), DMPC/DSPC 15/85, T = 323 K, parameter set W(cid:48)). 3.7. Line tension The importance of line tension for phase separation in lipid membranes has been shown experimentally for both liquid -- liquid coexistence in ternary lipid mixtures [60, 69, 72, 77 -- 81] and for fluid -- gel coexistence in binary lipid mixtures [76, 82 -- 86]. The importance of line tension in fluid -- gel phase separation was pointed out already 15 years ago in a lattice-based simulation study [30], although the study did not address the line tension in a quantitative way. Recently, line tension between the gel and fluid phases has been addressed in non-lattice-based simulations [13, 14]. As discussed above, in our simulations the DMPC/DSPC mixtures at compositions and temperatures corresponding to the fluid -- gel phase coexistence region of the phase diagram always display circular-shaped domains, which can be attributed to the presence of an effective line tension between the fluid and the gel phase. After equilibration the system shows complete phase separation, and only a single domain of the minority phase remains surrounded by the majority phase, which corresponds to minimization of the perimeter of the phase interface, which is driven by minimization of the line tension energy. As already discussed in section 3.5, for membranes with the fraction of the minority phase in the range of 1/π < X < 1/2 the true equilibrium state is characterized by a stripe-shaped domain while the circular domain represents a metastable state. On the other hand, for the large lattice sizes we addressed in this study, this metastable state is extremely long-lived and thus the circular-shaped domains persist for the whole simulation run. These membrane domains show thermally-excited shape fluctuations, as it can be clearly seen in the corresponding membrane snapshots in figures 4, 9, and 12(a -- f). An analysis of fluctuating domain shapes should in principle allow one to determine the line tension between the fluid and gel phases in the membrane by analyzing a sequence of fluctuating domain contours. The line tension is then determined from the Fourier spectrum of domain contour fluctuations, using the approach originally introduced by Goldstein and Jackson [87] and later successfully applied to lipid bilayer membranes [69, 77] (for corrected expressions, see [78]). One of the important requirements of such an analysis is conservation of the domain area. Unlike the lipid composition of the membrane, which is kept strictly constant during an MC simulation, the relative amount of the gel and fluid phase at a fixed temperature and membrane composition in an equilibrated membrane is conserved only on average. As a result, the domain area is also subject to fluctuations. We have found, however, that the magnitude of these fluctuations is rather low (typically, below 3%), which allows one to consider the domain area as effectively constant and thus to extract the line tension from the analysis J Ehrig, E P Petrov and P Schwille 29 (a-f) Representative equilibrium membrane configurations of Figure 12. DMPC/DSPC 20/80 membrane at T = 310 (a), 314 (b), 315.5 (c), 317.5 (d), 318.5 (e), and 319.5 K (f) obtained in MC simulations. The fluid and gel phases are shown as white and black, respectively. The determined domain boundaries are shown in red (see text for details). Lattice size: 400 × 400; scale bar: 150 lattice units ≈ 120 nm. (• ) and a linear fit of the high-temperature part of these data (- - - -) giving an (g) Line tensions calculated from the Fourier spectrum of domain contour fluctuations estimate of the critical temperature Tc = (320.5 ± 0.4) K. of fluctuating domain contours. To determine the fluctuating contours of a membrane domain, the snapshots of the membrane (reflecting only the conformational state of lipids) were first segmented by applying low-pass filtering (convolution with a 2D Gaussian with σ = 5 lattice units) followed by thresholding of the result. A set of contours (typically, 200 − 500 contours) was analyzed using a procedure described in [77, 78] to extract the energies of the Fourier modes of domain fluctuations, and ultimately the values of the line tension corresponding to these modes. Since the low-pass filtering procedure employed for domain contour determination suppresses the contributions of the higher-order fluctuation modes, only the first m modes satisfying the condition m < π (cid:104)R(cid:105)/(6σ), where (cid:104)R(cid:105) is the mean domain radius, were used for determination of the line tension. This typically amounted to the line tension analysis based on the first 10 to 20 modes, depending on the domain size. We found that away from the critical point the line tension of membrane domains obtained from our simulation data is about 2 pN (figure 12(g)). This value is in agreement with experimental values of a few pN reported for membranes with various lipid compositions, including ternary lipid systems exhibiting fluid -- fluid phase coexistence [60, 69, 72, 77 -- 81] and binary lipid mixtures showing fluid -- gel coexistence [84]. Moreover, the line tension of 2 pN is in good agreement with the value expected [81] for the experimentally measured height mismatch of ∼ 0.9− 1.3 nm between the gel and fluid phases in DMPC/DSPC membranes [88, 89]; it also agrees well with the value obtained in another binary lipid system showing fluid -- gel coexistence with a similar J Ehrig, E P Petrov and P Schwille 30 hight mismatch [84]. When the membrane temperature and composition approach the critical point, the line tension approaches zero (figure 12), as it is generally expected [90] (see more details below). On the other hand, for lipid compositions displaying a quasi-abrupt phase transition (i.e., compositions with the DSPC fraction of ∼ 50% and below in the high- temperature part of the phase diagram) the line tension does not gradually approach zero as the phase transition temperature is approached (data not shown), which is in agreement with the behaviour illustrated in figure 8. 3.8. Behaviour of the membrane in the vicinity of the critical point When approaching the critical point, many parameters of a thermodynamic system will either tend to zero or diverge as some power of T − Tc [51, 91]. These powers are known as critical exponents. For example, the correlation length of fluctuations and order parameter susceptibility in a supercritical system diverge upon approaching the critical point with exponents ν and γ, respectively [51]. One therefore should expect that the inverse correlation length and inverse amplitude of the structure function (proportional to the inverse order parameter susceptibility) should vanish close to the phase transition as and ξ−1(T ) ∼ (T /Tc − 1)ν 0 (T ) ∼ (T /Tc − 1)γ. S−1 (7) (8) As it is exemplified in figure 13, the inverse correlation length of fluid and gel domains for the DMPC/DSPC 20/80 mixture indeed tends to zero when the system is gradually cooled down to approach the two-phase fluid -- gel coexistence region. A linear fit of this dependence yields an estimate of the critical temperature of Tc = (320.8 ± 0.1) K. In a similar manner, when approaching a critical point from the two-phase coexistence side, one should generally expect that the line tension vanishes according to the following law [90]: λ(T ) ∼ (1 − T /Tc)µ. (9) Indeed, in the case of the DMPC/DSPC 20/80 mixture (figure 12), which is very close to the critical composition, the line tension λ vanishes as the critical temperature Tc is approached and can be well described by a linear dependence λ(T ) ∼ (1 − T /Tc). The linear fit yields Tc = (320.5 ± 0.4) K. These two estimates of the critical temperatures are in good agreement with each other and reproduce well the position of the critical point estimated using the analysis depicted in figure 9. We can attempt a more precise determination of the critical temperature, as well as the critical exponents of our system by simultaneously fitting the temperature inverse amplitude of the structure dependences of the inverse correlation length, J Ehrig, E P Petrov and P Schwille 31 (a-f) Representative equilibrium membrane configurations of the Figure 13. DMPC/DSPC 20/80 lipid mixture at T = 320.7 (a), 321.0 (b), 321.5 (c), 322.5 (d), 324.0 (e), and 328.0 K (f) obtained in MC simulations. The fluid and gel phases are shown with white and black, respectively. Lattice size: 400 × 400; scale bar: 150 lattice units ≈ 120 nm. (g) Inverse correlation lengths ((cid:78)) and inverse amplitude of the structure function ((cid:4)). A linear fit of the temperature dependence of the inverse correlation length (- - - -) gives an estimate of the critical temperature Tc = (320.8 ± 0.1) K. function, and line tension in the vicinity of the critical point to expressions (7), (8) and (9) with a common critical temperature Tc. The results of this fit are shown in figure 14. The estimate of the critical temperature is again close to the above values: Tc = (320.5 ± 0.2) K. The estimates of the critical exponents obtained in this fit are as follows: µ = 1.17 ± 0.04, ν = 0.97 ± 0.05, and γ = 2.94 ± 0.05. It should be noted that different physical systems often show the same set of critical exponents, and in this case are said to belong to the same universality class. This means that close to the phase transition particular microscopic details of a system become unimportant, and its behaviour is governed by a small number of features, such as dimensionality and symmetry. Recently, it was suggested that three-component membranes close to the critical point might belong to the 2D Ising universality class [60, 61]. For 2D systems in the Ising universality class the critical exponents take the following values [51, 91]: µ = 1, ν = 1, and γ = 7/4. While the fit in figure 14 gives the estimates of the exponents µ and ν (1.17 ± 0.04 and 0.97 ± 0.05, respectively) close to those expected for a 2D Ising model, the estimate of the exponent γ (2.94 ± 0.05) absolutely does not fit the predictions for the 2D Ising model. We explain this strong discrepancy of the exponent γ from the Ising model- based expectations by the fact that our system is different from the Ising model in one important aspect. Namely, while the Ising system above the phase transition is characterized by a zero average magnetization (on average equal numbers of up and down spins, or black and white pixels in a graphic representation), our membrane, having an equal amount of lipids in the fluid and gel conformations at the critical point J Ehrig, E P Petrov and P Schwille 32 Figure 14. Temperature dependences of the domain line tension λ(T ) (◦ ) below the phase transition temperature, and of the inverse correlation length ξ−1(T ) ((cid:77)) and inverse amplitude of the structure functions S−1 0 (T ) ((cid:3)) above the phase transition obtained in MC simulations of the DMPC/DSPC 20/80 lipid system in the vicinity of the critical point. Curves show the global fit of these dependences to the expressions (7), (8) and (9) with a common Tc. The fit yields the estimates of the critical temperature Tc = (320.5±0.2) K and critical exponents µ = 1.17±0.04, ν = 0.97±0.05, and γ = 2.94 ± 0.05. (i.e., having a zero magnetization in terms of the Ising model), very fast transforms into the all-fluid state with increasing the temperature (i.e., in terms of the Ising model, its absolute average magnetization tends to 1). Our preliminary results show that, if, instead of keeping the membrane composition fixed, one would approach the critical point along the curve defined by the condition Xfluid = Xgel = 0.5 (cyan curve in figure 9), the exponent γ becomes close to the value of 7/4 expected for the 2D Ising system. Thus, based on this behaviour, we conclude that, generally, multicomponent membranes should not necessarily show the Ising criticality. On the other hand, if the amounts of lipids in the fluid and gel states both stay close to 0.5 within a wide temperature range around a critical point, one can indeed expect to observe critical behaviour close to that of the Ising model. It would be interesting to check experimentally whether this condition holds for three-component model membranes and giant plasma membrane vesicles whose behaviour close to the critical point was recently claimed to follow the Ising criticality [60, 61]. 3.9. Domain growth dynamics and dynamic scaling When the membrane, initially kept at a high temperature, where it is in the all-fluid state, is abruptly cooled down (quenched) to a temperature corresponding to fluid -- gel phase coexistence, phase separation takes place, and domains of the gel and fluid phase start to appear and coarsen with time. Studying the kinetics of membrane domain coarsening after a sudden temperature quench into the phase coexistence region can J Ehrig, E P Petrov and P Schwille 33 reveal information on the mechanisms involved in phase separation [92] and thus provide further understanding of the microscopic membrane organization and the interplay between the conformational (fluid -- gel) state of lipids and their lateral diffusion. Domain growth in lipid membranes upon a quench to the two-phase coexistence region has been addressed in several experimental works on three-component [81, 93, 94] and two-component [95] lipid systems and MC simulation studies of two-component lipid systems [30, 96]. We are not aware of any publication where a systematic study of the dependence of the domain growth and dynamic scaling would be carried out for a specific lipid mixture as a function of its composition and/or quench temperature. In this section, we briefly address this issue based on the results of our MC simulations. After the very early domain growth stage, during which small domains nucleate and become unstable, the stage with the power law growth of the mean size of domains R(t) ∼ tn sets in, with the growth exponent n being characteristic of the system type and growth mechanism [92, 97]. For systems with a non-conserved order parameter n = 1/2, whereas for systems with a conserved order parameter the growth exponent can take values of n = 1/3 or n = 1/4, depending on the particular mechanism of the domain growth. In case where the growth is controlled by evaporation of smaller domains, and larger domains grow due to diffusive transport of material through the medium from domain boundaries with larger curvature to domain boundaries of smaller curvature (Ostwald ripening), one should expect the domain growth with the exponent n = 1/3, known as the Lifshitz -- Slyozov -- Wagner growth [98, 99]. Originally, this law was derived for the growth of minority phase domains in a three-dimensional system in the limit of small minority phase concentration. Later it was argued that the same law also applies to two-dimensional systems, and, based on results of simulations, it was suggested that it should also apply for any concentration of the minority phase [100 -- 103]. When spinodal decomposition takes place in a system with a conserved order parameter one has to distinguish two possible scenarios. While for (close to) symmetric quenches, i.e. (close to) equal amounts of the two phases, the spinodal decomposition will necessarily produce a worm-like bicontinuos morphology, in case of strongly asymmetric quenches this situation is generally not expected - the minority phase occupies a too low area to form a network and no bicontinuous structure, usually believed to be characteristic for spinodal decomposition, will be observed. The notion of spinodal decomposition in this case will just mean that the system is instable with regard to small concentration fluctuations [58]. For both symmetric and asymmetric quenches, domains initially grow with n = 1/4 when spinodal decomposition takes place. For symmetric quenches this is the result of diffusion of particles along the interface boundaries of the bicontinuous phase morphology. If the quench is (strongly) asymmetric and thus the phase morphology after the quench is not bicontinuous, but rather appears as droplets of the minority phase embedded in the majority phase, the growth proceeds mainly via the Brownian motion and coalescence of these droplets for which in 2D n = 1/4 as well [97]. For J Ehrig, E P Petrov and P Schwille 34 both cases, with coarsening of domains, the diffusion through the medium becomes progressively more important and the domain growth crosses over to the asymptotic regime with n = 1/3 [104]. Notice that in some cases the observation of the asymptotic growth with n = 1/3 may require extremely long simulations and large lattice sizes [104]. To study the growth of domains after an abrupt quench of the membrane being originally in the fluid state to a temperature within the coexistence region of the phase diagram, we calculated the radial autocorrelation function G(r) (2) for a set of times t after the quench and extracted the time-dependent domain size R(t), which was defined as a distance r at which the first zero-crossing of G(r) occurs. As it is exemplified in figure 15, a power law growth with exponents ranging from n = 1/4 to n = 1/3 is observed in our simulations, in agreement with the theoretical expectations. We should remark here again that our system is different from the ones typically discussed in the literature on coarsening dynamics, where usually either conserved order parameter or non-conserved order parameter systems are studied. Our system, however, cannot be strictly classified into either conserved- or non-conserved order parameter case. While the lipid composition in our model system is conserved, the fraction of phases is free to change. On the other hand, the fact that we observe a power law growth of domains with exponents ranging from n = 1/4 to n = 1/3, i.e. the ones characteristic for the conserved order parameter systems, suggests that the lipid demixing plays an important part in the phase separation process. Remarkably, by comparing the domain growth shown in figure 15(b) with the experimental data on domain growth in double supported bilayers [93] we find that our simulations reproduce well the absolute rate of domain growth. In particular, extrapolation of the growth law to the later times shows that it takes of order of 100 s for the domains to grow to an average radius of ∼ 1 µm, in agreement with experiments [93]. We also studied whether dynamic scaling [105, 106] is observed for domain coarsening in the membrane after a quench, i.e. whether the G(r) corresponding to different time instants t after the quench would collapse onto one master curve if replotted as a function of the reduced radius r/R(t). It turned out that dynamic scaling is observed not in all cases, but only when the total fluid fraction Xfluid of the membrane is constant in time. The data presented in figure 15 demonstrate this observation. Figure 15 (a-d) shows an example for a close to symmetric quench to a point of the phase diagram located deep within the region between the two spinodal curves: the membrane with DMPC/DSPC 50/50 was quenched to T = 310 K. The fluid fraction Xfluid remains constant during domain coarsening, except for a very short initial growth stage. At the same time, the observed growth exponent crosses over from n ≈ 1/4 at early times to n ≈ 1/3 at later times. This suggests that the growth first proceeds via diffusion along the interface (n = 1/4) and later via bulk diffusion from smaller evaporating domains to larger ones (n = 1/3). This behaviour can indeed be observed from the time series of snapshots from the simulation. Dynamic scaling of G(r) is J Ehrig, E P Petrov and P Schwille 35 Figure 15. Domain coarsening in DMPC/DSPC 50/50 (a-d), 20/80 (e-h), 20/80 (i- l) and 70/30 (m-p) membranes after a sudden temperature quench from T = ∞ to T = 310, 317.5, 310, and 302 K, respectively. Panels (a, e, i, m) each show three corresponding non-equilibrium configurations of the membrane at times t = 104 , 105, and 106 MC cycles after the quench. Lattice size: 400 × 400; scale bar: 150 lattice units ≈ 120 nm. (b, f, j, n) Time dependence of the characteristic domain size R(t) (black curves); thin black lines represent power laws with the exponent of 1/3 and 1/4. Also shown are the time dependences of the total fluid fraction Xfluid(t); the time intervals where Xfluid(t) ≈ const are marked by the green and red colours, while the time intervals where Xfluid(t) (cid:54)= const are plotted in grey. (c, g, k, o) Radial autocorrelation functions G(r) for a set of times (stated in panels (d, h, l, p)) after the quench. (d, h, l, p) Same as (c, g, k, o), respectively, replotted as a function of the reduced distance r/R(t) to demonstrate that dynamic scaling is observed when Xfluid(t) ≈ const; the colours of the curves at different time instants are chosen to correspond to those of Xfluid(t). 317.5 KJ Ehrig, E P Petrov and P Schwille 36 observed for the whole time interval where Xfluid(t) = const. In figure 15 (e-h) an example for a DMPC/DSPC 20/80 membrane that was quenched to T = 317.5 K is shown. Here, the amounts of the fluid and the gel phase are clearly not equal. Therefore, in this situation we observe an asymmetric quench of the membrane and the phase morphology, unlike in the former example, is not expected to be bicontinuous. The mean domain size initially grows with an exponent of n ≈ 1/4, in agreement with the observed domain growth via the coalescence mechanism at early times, and later crosses over to n ≈ 1/3, due to the domain growth via the evaporation of smaller domains at later times. The total fluid fraction in the membrane is nearly constant for up to t ≈ 2 × 104 MC cycles and the G(r) shows dynamic scaling in this time range. At later times t ≈ (0.5 − 1) × 106 MC cycles the fluid fraction changes -- most probably due to the rearrangement of lipids between the fluid and gel phases -- and no dynamic scaling is observed. Finally, at even later times t > 2 × 106 MC cycles the fluid fraction becomes constant again, and dynamic scaling is again observed. A very similar behaviour is observed for a quench of a DMPC/DSPC 20/80 membrane to T = 310 K (figure 15 (i-l)): we observe two scaling regimes, both for time intervals where the fluid fraction of the membrane is constant, i.e. at times t (cid:46) 5 × 104 MC cycles and t (cid:38) 1×106 MC cycles. No dynamic scaling is observed at the intermediate times when the fluid fraction of the membrane is changing. Notice that, unlike in the former two examples, the mean domain size grows as R(t) ∼ t1/4 for the whole time range that was studied. Figure 15 (m-p) shows the results for domain coarsening in a DMPC/DSPC 70/30 membrane quenched to T = 302 K. The mean domain size R(t) shows a behaviour qualitatively similar to the first example of the DMPC/DSPC 50/50 membrane, quenched to T = 310 K, and grows with an exponent that crosses over from n ≈ 1/4 at early times to n ≈ 1/3 at later times, in agreement with the behaviour observed from a time series of membrane snapshots. Nevertheless, no dynamic scaling is observed in this case. Notice that the total fluid fraction under these conditions keeps changing over the whole time interval covered by the MC simulation. Remarkably, this change in the fluid fraction also changes the morphology of the phases. While the early stage is characterized by the appearance and growth of fluid droplets, at the later stage, when the fluid fraction grows to a value close to 0.5, the phase morphology changes to bicontinuous. We conclude that this intriguing behaviour reflects a complex interplay between fluid -- gel phase separation and lipid demixing. The equilibration of the total fluid fraction also needs lipid rearrangement via lateral diffusion of lipids, and the timescale of this process depends on the particular point in the phase diagram. This can serve as a potential explanation for the discrepancy in the experimental data on domain growth dynamics in lipid bilayers giving growth exponents from n ≈ 1/3 [81, 93] down to n ≈ 0.15 [94] (we should point out here that a careful analysis of the results published in [94] shows that that the late-stage growth kinetics exhibits the exponent close to 1/4, rather than 0.15 reported by the authors). The evolution of the total fluid fraction with J Ehrig, E P Petrov and P Schwille 37 time after the membrane quench to the two-phase coexistence region was previously observed in simulations [30, 96], and it was suggested on a qualitative level that non- equilibrium effects may influence the dynamics of phase separation on various length and time scales. Our results clearly show that, indeed, the dynamic scaling can be strongly affected by these non-equilibrium phenomena. In this context it is interesting to mention a recent simulation study [107] which showed that the presence of hydrodynamic interactions both within the membrane and in the surrounding solvent can lead to the breakdown of dynamic scaling upon spinodal decomposition in critical lipid mixtures. Our results thus point to an alternative mechanism which can lead to the breakdown of dynamic scaling in phase-separating membranes, which is related to the lateral redistribution of the membrane components. In experimentally realizable situations, we believe, both of the effects should be present and may enhance each other. We believe that further insight into these issues can be gained by applying the theory of non-equilibrium phase transitions and ageing [106] and by comparing the outcome of the simulation results with outcomes of appropriately designed experiments. 4. Conclusions In this paper, we have demonstrated that the properties of lipid membranes on the experimentally relevant spatial scales of order of a micrometer and time ranges of order of a second can be successfully addressed via lattice-based Monte Carlo simulations with very moderate computational efforts. Using the DMPC/DSPC mixture as an example of a two-component lipid membrane, our large-scale simulations allowed us to obtain the following important results, which, to the best of our knowledge, were previously not reported in simulation studies of lipid membranes: We found that within a certain range of lipid compositions, the phase transition from the fluid phase to the fluid -- gel phase coexistence proceeds via near-critical fluctuations, while for other lipid compositions this phase transition has a quasi-abrupt character. Qualitatively, this is in excellent agreement with recent experiments where these two scenarios of the phase transition were observed in three-component lipid membranes [61]. In the presence of near-critical fluctuations, close to the critical point of the membrane, lipids show transient subdiffusion, which is important for understanding the origins of anomalous diffusion in cell membranes, especially in view of the recent experimental results [61], showing the critical behaviour in giant plasma membrane vesicles isolated from living cells. In the fluid -- gel phase coexistence region, macroscopic phase separation takes place, and after full equilibration of the membrane, a single circular-shaped domain of the minority phase emerges. Analysis of the thermally-induced domain shape fluctuations allowed us to obtain the line tension between the fluid and gel phases. We found that in REFERENCES 38 the fluid -- gel coexistence region, away from the critical point, the line tension is ≈ 2 pN, in good agreement with the available literature data on line tension in lipid membranes [60, 69, 72, 77, 79 -- 81, 84]. When approaching the critical point, the line tension, as well as the inverse correlation length of fluid -- gel spatial fluctuations, and the corresponding inverse order parameter susceptibility of the membrane, approach zero. This is in agreement with recent experimental observations of the approach to criticality in three-component lipid bilayers and giant plasma membrane vesicles [60, 61]. On the other hand, contrary to the conclusions drawn in [60, 61], we find that our system is not in the Ising universality class and provide a tentative explanation for this discrepancy. An analysis of the domain coarsening dynamics after an abrupt quench of the membrane to the fluid -- gel coexistence region revealed that the domain growth law may depend on the lipid composition and temperature, which reflects a complex interplay between fluid -- gel phase separation and lipid demixing. We find that the domain growth exponent varies from 1/4 to 1/3, which is in a qualitative agreement with experimental data for lipid bilayers, for which a range of growth exponents has been reported as well [81, 93, 94]. We found out that the dynamic scaling of the radial autocorrelation function of the spatial distribution of the lipid state over the membrane during the domain coarsening is observed only when the fluid fraction of the membrane stays constant in time, which again shows the importance of lipid diffusion in the domain coarsening dynamics. Acknowledgements The authors acknowledge inspiring discussions with Herv´e Rigneault and Cyril Favard at the early stage of the project. Helpful comments by Richard L. C. Vink are gratefully appreciated. The work was supported by the Deutsche Forschungsgemeinschaft via Research Group FOR 877 'From Local Constraints to Macroscopic Transport'. References [1] Vereb G, Szollosi J, Matk´o J, Nagy P, Farkas T, V´ıgh L, M´atyus L, Waldmann T A and Damjanovich S 2003 Dynamic, yet structured: The cell membrane three decades after the Singer -- Nicolson model Proc. Natl. Acad. Sci. USA 100 8053 -- 8058 [2] Simons K and Ikonen E 1997 Functional rafts in cell membranes Nature 387 569 -- 572 [3] Garg S, Tang J X, Ruhe J and Naumann C A 2009 Actin-induced perturbation of PS lipid -- cholesterol interaction: A possible mechanism of cytoskeleton-based regulation of membrane organization J. Struct. Biol. 168 11 -- 20 REFERENCES 39 [4] Liu A P and Fletcher D A 2006 Actin polymerization serves as a membrane domain switch in model lipid bilayers Biophys. J. 91 4064 -- 4070 [5] Roux A, Cuvelier D, Nassoy P, Prost J, Bassereau P and Goud B 2005 Role of curvature and phase transition in lipid sorting and fission of membrane tubules EMBO J. 24 1537 -- 1545 [6] Sorre B, Callan-Jones A, Manneville J B, Nassoy P, Joanny J F, Prost J, Goud B and Bassereau P 2009 Curvature-driven lipid sorting needs proximity to a demixing point and is aided by proteins Proc. Natl. Acad. Sci. USA 106 5622 -- 5626 [7] Pike L J 2006 Rafts defined: A report on the Keystone symposium on lipid rafts and cell function J. Lipid Res. 47 1597 -- 1598 [8] Weissig V (ed) 2009 Liposomes: Methods and Protocols, Volume 2: Biological Membrane Models (Methods in Molecular Biology vol 606) (New York Dordrecht Heidelberg London: Springer) [9] Deserno M 2009 Mesoscopic membrane physics: Concepts, simulations, and selected applications Macromol. Rapid Commun. 30 752 -- 771 [10] Elson E L, Fried E, Dolbow J E and Genin G M 2010 Phase separation in biological membranes: Integration of theory and experiment Annu. Rev. Biophys. 39 207 -- 226 [11] Heimburg T 2007 Thermal Biophysics of Membranes (Weinheim: Wiley -- VCH) [12] Muller M, Katsov K and Schick M 2006 Biological and synthetic membranes: What can be learned from a coarse-grained description? Phys. Rep. 434 113 -- 176 [13] Marrink S J, Risselada J and Mark A E 2005 Simulation of gel phase formation and melting in lipid bilayers using a coarse grained model Chem. Phys. Lipids 135 223 -- 244 [14] Shi Q and Voth G A 2005 Multi-scale modeling of phase separation in mixed lipid bilayers Biophys. J. 89 2385 -- 2394 [15] Niemela P S, Ollila S, Hyvonen M T, Karttunen M and Vattulainen I 2007 Assessing the nature of lipid raft membranes PLoS Comput. Biol. 3 e34 [16] Bjelkmar P, Niemela P S, Vattulainen I and Lindahl E 2009 Conformational changes and slow dynamics through microsecond polarized atomistic molecular simulation of an integral Kv1.2 ion channel PLoS Comput. Biol. 5 e1000289 [17] Pandit S A and Scott H L 2009 Multiscale simulations of heterogeneous model membranes Biochim. Biophys. Acta 1788 136 -- 148 [18] Yuan H, Huang C, Li J, Lykotrafitis G and Zhang S 2010 One-particle-thick, solvent-free, coarse-grained model for biological and biomimetic fluid membranes Phys. Rev. E 82 011905 [19] Pasqua A, Maibaum L, Oster G, Fletcher D A and Geissler P L 2010 Large-scale simulations of fluctuating biological membranes J. Chem. Phys. 132 154107 REFERENCES 40 [20] Lowengrub J S, Ratz A and Voigt A 2009 Phase-field modeling of the dynamics of multicomponent vesicles: Spinodal decomposition, coarsening, budding, and fission Phys. Rev. E 79 031926 [21] Wang X and Du Q 2008 Modelling and simulations of multi-component lipid membranes and open membranes via diffuse interface approaches J. Math. Biol. 56 347 -- 371 [22] Reigada R, Buceta J, G´omez J, Sagu´es F and Lindenberg K 2008 Phase separation in three-component lipid membranes: From Monte Carlo simulations to Ginzburg -- Landau equations J. Chem. Phys. 128 025102 [23] Haataja M 2009 Critical dynamics in multicomponent lipid membranes Phys. Rev. E 80 020902(R) [24] Fan J, Sammalkorpi M and Haataja M 2010 Formation and regulation of lipid microdomains in cell membranes: Theory, modeling, and speculation FEBS Lett. 584 1678 -- 1684 [25] Tumaneng P W, Pandit S A, Zhao G and Scott H L 2010 Lateral organization of complex lipid mixtures from multiscale modeling J. Chem. Phys. 132 065104 [26] Mouritsen O G, Boothroyd A, Harris R, Jan N, Lookman T, MacDonald L, Pink D A and Zuckermann M J 1983 Computer simulation of the main gel -- fluid phase transition of lipid bilayers J. Chem. Phys. 79 2027 -- 2041 [27] Jørgensen K, Sperotto M M, Mouritsen O G, Ipsen J H and Zuckermann M J 1993 Phase equilibria and local structure in binary lipid bilayers Biochim. Biophys. Acta 1152 135 -- 145 [28] Zhang Z, Sperotto M M, Zuckermann M J and Mouritsen O G 1993 A microscopic model for lipid/protein bilayers with critical mixing Biochim. Biophys. Acta 1147 154 -- 160 [29] Sug´ar I P, Biltonen R L and Mitchard N 1994 Monte Carlo simulations of membranes: Phase transition of small unilamellar dipalmitoylphosphatidylcholine vesicles Methods Enzymol. 240 569 -- 593 [30] Jørgensen K and Mouritsen O G 1995 Phase separation dynamics and lateral organization of two-component lipid membranes Biophys. J. 69 942 -- 954 [31] Sug´ar I P, Thompson T E and Biltonen R L 1999 Monte Carlo simulation of two-component bilayers: DMPC/DSPC mixtures Biophys. J. 76 2099 -- 2110 [32] Hac A E, Seeger H M, Fidorra M and Heimburg T 2005 Diffusion in two- component lipid membranes -- a fluorescence correlation spectroscopy and Monte Carlo simulation study Biophys. J. 88 317 -- 333 [33] Shimshick E J and McConnell H M 1973 Lateral phase separation in phospholipid membranes Biochemistry 12 2351 -- 2360 [34] Mabrey S and Sturtevant J M 1976 Investigation of phase transitions of lipids and lipid mixtures by high sensitivity differential scanning calorimetry Proc. Natl. Acad. Sci. USA 73 3862 -- 3866 REFERENCES 41 [35] Lentz B R, Barenholz Y and Thompson T E 1976 Fluorescence depolarization studies of phase transitions and fluidity in phospholipid bilayers. 2. Two- component phosphatidylcholine liposomes Biochemistry 15 4529 -- 4537 [36] van Dijck P W M, Kaper A J, Oonk H A J and de Gier J 1977 Miscibility properties of binary phosphatidylcholine mixtures. A calorimetric study Biochim. Biophys. Acta 470 58 -- 69 [37] Wilkinson D A and Nagle J F 1979 Dilatometric study of binary mixtures of phosphatidylcholines Biochemistry 18 4244 -- 4249 [38] Nibu Y, Inoue T and Motoda I 1995 Effect of headgroup type on the miscibility of homologous phospholipids with different acyl chain lengths in hydrated bilayer Biophys. Chem. 56 273 -- 280 [39] Sug´ar I P and Biltonen R L 2000 Structure-function relationships in two- component phospholipid bilayers: Monte Carlo simulation approach using a two- state model Methods Enzymol. 323 340 -- 372 [40] Ehrig J, Petrov E P and Schwille P 2011 Near-critical fluctuations and cytoskeleton-assisted phase separation lead to subdiffusion in cell membranes Biophys. J. 100 80 -- 89 [41] Angelova M I and Dimitrov D S 1986 Liposome electroformation Faraday Discuss. 81 303 -- 311 [42] Sun H and Sug´ar I P 1997 Acceleration of convergence to the thermodynamic equilibrium by introducing shuffling operations to the Metropolis algorithm of Monte Carlo simulations J. Phys. Chem. B 101 3221 -- 3227 [43] Michonova-Alexova E I and Sug´ar I P 2002 Component and state separation in DMPC/DSPC lipid bilayers: A Monte Carlo simulation study Biophys. J. 83 1820 -- 1833 [44] Kuo A L and Wade C G 1979 Lipid lateral diffusion by pulsed nuclear magnetic resonance Biochemistry 18 2300 -- 2308 [45] Vaz W L C, Clegg R M and Hallmann D 1985 Translational diffusion of lipids in liquid crystalline phase phosphatidylcholine multibilayers. A comparison of experiment with theory Biochemistry 24 781 -- 786 [46] Dolainsky C, Karakatsanis P and Bayerl T M 1997 Lipid domains as obstacles for lateral diffusion in supported bilayers probed at different time and length scales by two-dimensional exchange and field gradient solid state NMR Phys. Rev. E 55 4512 -- 4521 [47] Matsumoto M and Nishimura T 1998 Mersenne Twister: A 623-dimensionally equidistributed uniform pseudo-random number generator ACM Trans. Model. Comput. Sim. 8 3 -- 30 [48] Preis T, Virnau P, Paul W and Schneider J J 2009 GPU accelerated Monte Carlo simulation of the 2D and 3D Ising model J. Comput. Phys. 228 4468 -- 4477 REFERENCES 42 [49] Weigel M 2011 Simulating spin models on GPU Comput. Phys. Commun. in press; doi:10.1016/j.cpc.2010.10.031 [50] Hansen J P and McDonald I R 2006 Theory of Simple Liquids 3rd ed (London: Academic Press) [51] Fisher M E 1964 Correlation functions and the critical region of simple fluids J. Math. Phys. 5 944 -- 962 [52] Allen M P and Tildesley D J 1991 Computer Simulations of Liquids (New York: Oxford University Press) [53] Petrov E P and Schwille P 2008 State of the art and novel trends in fluorescence correlation spectroscopy Standardization and Quality Assurance in Fluorescence Measurements II (Springer Series on Fluorescence vol 6) Resch-Genger U (ed) (Berlin Heidelberg New York: Springer) pp 145 -- 197 [54] Kastrup L, Blom H, Eggeling C and Hell S W 2005 Fluorescence fluctuation spectroscopy in subdiffraction focal volumes Phys. Rev. Lett. 94 178104 [55] Kjaer K, Als-Nielsen J, Helm C A, Laxhuber L A and Mohwald H 1987 Ordering in lipid monolayers studied by synchrotron X-ray diffraction and fluorescence microscopy Phys. Rev. Lett. 58 2224 -- 2227 [56] Smith G S, Sirota E B, Safinya C R, Plano R J and Clark N A 1990 X-ray structural studies of freely suspended ordered hydrated DMPC multimembrane films J. Chem. Phys. 92 4519 -- 4529 [57] Bernchou U, Brewer J, Midtiby H S, Ipsen J H, Bagatolli L A and Simonsen A C 2009 Texture of lipid bilayer domains J. Am. Chem. Soc. 131 14130 -- 14131 [58] Strobl G 1997 The Physics of Polymers 2nd ed (Berlin Heidelberg New York: Springer) [59] Fidorra M, Garcia A, Ipsen J H, Hartel S and Bagatolli L A 2009 Lipid domains in giant unilamellar vesicles and their correspondence with equilibrium thermodynamic phases: A quantitative fluorescence microscopy imaging approach Biochim. Biophys. Acta 1788 2142 -- 2149 [60] Honerkamp-Smith A R, Cicuta P, Collins M D, Veatch S L, den Nijs M, Schick M and Keller S L 2008 Line tensions, correlation lengths, and critical exponents in lipid membranes near critical points Biophys. J. 95 236 -- 246 [61] Veatch S L, Cicuta P, Sengupta P, Honerkamp-Smith A, Holowka D and Baird B 2008 Critical fluctuations in plasma membrane vesicles ACS Chem. Biol. 3 287 -- 293 [62] Nielsen L K, Bjørnholm T and Mouritsen O G 2007 Thermodynamic and real- space structural evidence of a 2D critical point in phospholipid monolayers Langmuir 23 11684 -- 11692 [63] Knoll W, Ibel K and Sackmann E 1981 Small-angle neutron scattering study of lipid phase diagrams by the contrast variation method Biochemistry 20 6379 -- 6383 REFERENCES 43 [64] Knoll W, Schmidt G and Sackmann E 1983 Critical demixing in fluid bilayers of phospholipid mixtures. A neutron diffraction study J. Chem. Phys. 79 3439 -- 3442 [65] Sackmann E 1995 Physical basis of self-organization and function of membranes: Physics of vesicles Structure and dynamics of membranes: From cells to vesicles (Handbook of Biological Physics vol 1A) Lipowsky R and Sackmann E (eds) (Amsterdam: Elsevier) pp 213 -- 304 [66] Murtola T, Rg T, Falck E, Karttunen M and Vattulainen I 2006 Transient ordered domains in single-component phospholipid bilayers Phys. Rev. Lett. 97 238102 [67] Cruzeiro-Hansson L and Mouritsen O G 1988 Passive ion permeability of lipid membranes modelled via lipid-domain interfacial area Biochim. Biophys. Acta 944 63 -- 72 [68] Freire E and Biltonen R 1978 Estimation of molecular averages and equilibrium fluctuations in lipid bilayer systems from the excess heat capacity function Biochim. Biophys. Acta 514 54 -- 68 [69] Baumgart T, Hess S T and Webb W W 2003 Imaging coexisting fluid domains in biomembrane models coupling curvature and line tension Nature 425 821 -- 824 [70] Auth T and Gompper G 2009 Budding and vesiculation induced by conical membrane inclusions Phys. Rev. E 80 031901 [71] Ursell T S, Klug W S and Phillips R 2009 Morphology and interaction between lipid domains Proc. Natl. Acad. Sci. USA 106 13301 -- 13306 [72] Semrau S, Idema T, Schmidt T and Storm C 2009 Membrane-mediated interactions measured using membrane domains. Biophys J 96 4906 -- 4915 [73] Idema T, Semrau S, Storm C and Schmidt T 2010 Membrane mediated sorting Phys. Rev. Lett. 104 198102 [74] Neuhaus T and Hager J S 2003 2D crystal shapes, droplet condensation, and exponential slowing down in simulations of first-order phase transitions J Stat Phys 113 47 -- 83 [75] Fischer T and Vink R L C 2010 The Widom -- Rowlinson mixture on a sphere: Elimination of exponential slowing down at first-order phase transitions J. Phys.: Condens. Matter 22 104123 [76] Li L and Cheng J X 2006 Coexisting stripe- and patch-shaped domains in giant unilamellar vesicles Biochemistry 45 11819 -- 11826 [77] Esposito C, Tian A, Melamed S, Johnson C, Tee S Y and Baumgart T 2007 Flicker spectroscopy of thermal lipid bilayer domain boundary fluctuations Biophys. J. 93 3169 -- 3181 [78] Camley B A, Esposito C, Baumgart T and Brown F L H 2010 Lipid bilayer domain fluctuations as a probe of membrane viscosity Biophys. J. 99 L44 -- L46 [79] Baumgart T, Das S, Webb W W and Jenkins J T 2005 Membrane elasticity in giant vesicles with fluid phase coexistence Biophys. J. 89 1067 -- 1080 REFERENCES 44 [80] Tian A, Johnson C, Wang W and Baumgart T 2007 Line tension at fluid membrane domain boundaries measured by micropipette aspiration Phys. Rev. Lett. 98 208102 [81] Garc´ıa-S´aez A J, Chiantia S and Schwille P 2007 Effect of line tension on the lateral organization of lipid membranes J. Biol. Chem. 282 33537 -- 33544 [82] de Almeida R F M, Loura L M S, Fedorov A and Prieto M 2002 Nonequilibrium phenomena in the phase separation of a two-component lipid bilayer Biophys. J. 82 823 -- 834 [83] Leidy C, Kaasgaard T, Crowe J H, Mouritsen O G and Jørgensen K 2002 Ripples and the formation of anisotropic lipid domains: Imaging two-component supported double bilayers by atomic force microscopy Biophys. J. 83 2625 -- 2633 [84] Blanchette C D, Lin W C, Orme C A, Ratto T V and Longo M L 2007 Using nucleation rates to determine the interfacial line tension of symmetric and asymmetric lipid bilayer domains Langmuir 23 5875 -- 5877 [85] Connell S D and Smith D A 2006 The atomic force microscope as a tool for studying phase separation in lipid membranes Mol. Membr. Biol. 23 17 -- 28 [86] Gordon V D, Beales P A, Zhao Z, Blake C, MacKintosh F C, Olmsted P D, Cates M E, Egelhaaf S U and Poon W C K 2006 Lipid organization and the morphology of solid-like domains in phase-separating binary lipid membranes J. Phys. Condens. Matter 18 L415L420 [87] Goldstein R E and Jackson D P 1994 Domain shape relaxation and the spectrum of thermal fluctuations in Langmuir monolayers J. Phys. Chem. 98 9626 -- 9636 [88] Giocondi M C and Le Grimellec C 2004 Temperature dependence of the sur- face topography in dimyristoylphosphatidylcholine/distearoylphosphatidylcholine multibilayers Biophys. J. 86 2218 -- 2230 [89] Giocondi M C, Pacheco L, Milhiet P E and Le Grimellec C 2001 Temperature dependence of the topology of supported dimirystoyl-distearoyl phosphatidylcholine bilayers Ultramicroscopy 86 151 -- 157 [90] Widom B 1965 Surface tension and molecular correlations near the critical point J. Chem. Phys. 43 3892 -- 3897 [91] Hohenberg P C and Halperin B I 1977 Theory of dynamic critical phenomena Rev. Mod. Phys. 49 435 -- 479 [92] Bray A J 1994 Theory of phase-ordering kinetics Adv. Phys. 43 357 -- 459 [93] Jensen M H, Morris E J and Simonsen A C 2007 Domain shapes, coarsening, and random patterns in ternary membranes Langmuir 23 8135 -- 8141 [94] Saeki D, Hamada T and Yoshikawa K 2006 Domain-growth kinetics in a cell-sized liposome J. Phys. Soc. Jpn. 75 013602 [95] Hu Y, Meleson K and Israelachvili J 2006 Thermodynamic equilibrium of domains in a two-component Langmuir monolayer Biophys. J. 91 444 -- 453 REFERENCES 45 [96] Jørgensen K, Klinger A and Biltonen R L 2000 Nonequilibrium lipid domain growth in the gel-fluid two-phase region of a DC16PC − DC22PC lipid mixture investigated by Monte Carlo computer simulation, FT-IR, and fluorescence spectroscopy J. Phys. Chem. B 104 11763 -- 11773 [97] Furukawa H 1985 A dynamic scaling assumption for phase-separation Adv. Phys. 34 703 -- 750 [98] Lifshitz I M and Slyozov V V 1961 The kinetics of precipitation from supersaturated solid solutions J. Phys. Chem. Solids 19 35 -- 50 [99] Wagner C 1961 Theorie der Alterung von Niederschlagen durch Umlosen (Ostwald- Reifung) Z. Elektrochem. 65 581 -- 591 [100] Huse D A 1986 Corrections to late-stage behavior in spinodal decomposition: Lifshitz-Slyozov scaling and Monte Carlo simulations Phys. Rev. B 34 7845 -- 7850 [101] Amar J G, Sullivan F E and Mountain R D 1988 Monte Carlo study of growth in the two-dimensional spin-exchange kinetic Ising model Phys. Rev. B 37 196 -- 208 [102] Rogers T M, Elder K R and Desai R C 1988 Numerical study of the late stages of spinodal decomposition Phys. Rev. B 37 9638 -- 9649 [103] Rogers T M and Desai R C 1989 Numerical study of late-stage coarsening for off-critical quenches in the Cahn -- Hilliard equation of phase separation Phys. Rev. B 39 11956 -- 11964 [104] van Gemmert S, Barkema G T and Puri S 2005 Phase separation driven by surface diffusion: A Monte Carlo study Phys. Rev. E 72 046131 [105] Gunton J, San Miguel M and Sahni P S 1983 The dynamics of first-order phase transitions Phase Transitions and Critical Phenomena vol 8 Domb C and Lebowitz J (eds) (New York: Academic Press) pp 267 -- 482 [106] Henkel M and Pleimling M 2010 Non-Equilibrium Phase Transitions. Volume 2: Ageing and Dynamical Scaling Far from Equilibrium (Berlin Heidelberg New York: Springer) [107] Fan J, Han T and Haataja M 2010 Hydrodynamic effects on spinodal decomposition kinetics in planar lipid bilayer membranes J. Chem. Phys. 133 235101
1305.1049	1	1305	2013-05-05T20:03:38	Elasticity of cross-linked semiflexible biopolymers under tension	[ "physics.bio-ph", "cond-mat.soft", "cond-mat.stat-mech" ]	Aiming at the mechanical properties of cross-linked biopolymers, we set up and analyze a model of two weakly bending wormlike chains subjected to a tensile force, with regularly spaced inter-chain bonds (cross-links) represented by harmonic springs. Within this model, we compute the force-extension curve and the differential stiffness exactly and discuss several limiting cases. Cross-links effectively stiffen the chain pair by reducing thermal fluctuations transverse to the force and alignment direction. The extra alignment due to cross-links increases both with growing number and with growing strength of the cross-links, and is most prominent for small force f. For large f, the additional, cross-link-induced extension is subdominant except for the case of linking the chains rigidly and continuously along their contour. In this combined limit, we recover asymptotically the elasticity of a weakly bending wormlike chain without constraints, stiffened by a factor four. The increase in differential stiffness can be as large as 100% for small f or large numbers of cross-links.	physics.bio-ph	physics	Elasticity of cross-linked semiflexible biopolymers under tension Alice von der Heydt,1, ∗ Daniel Wilkin,1 Panayotis Benetatos,2 and Annette Zippelius1, 3 1Institute for Theoretical Physics, Georg-August University of Gottingen, Friedrich-Hund-Platz 1, 37077 Gottingen, Germany 2Department of Physics, Kyungpook National University, 80 Daehak-ro, Buk-gu, Daegu, 702-701, Korea 3Max Planck Institute for Dynamics and Self-Organization, Am Fassberg 17, 37077 Gottingen, Germany (Dated: October 30, 2018) Aiming at the mechanical properties of cross-linked biopolymers, we set up and analyze a model of two weakly bending wormlike chains subjected to a tensile force, with regularly spaced inter- chain bonds (cross-links) represented by harmonic springs. Within this model, we compute the force-extension curve and the differential stiffness exactly and discuss several limiting cases. Cross- links effectively stiffen the chain pair by reducing thermal fluctuations transverse to the force and alignment direction. The extra alignment due to cross-links increases both with growing number and with growing strength of the cross-links, and is most prominent for small force f . For large f , the additional, cross-link-induced extension is subdominant except for the case of linking the chains rigidly and continuously along their contour. In this combined limit, we recover asymptotically the elasticity of a weakly bending wormlike chain without constraints, stiffened by a factor four. The increase in differential stiffness can be as large as 100% for small f or large numbers of cross-links. PACS numbers: 87.10.Pq, 87.15.La, 82.37.Rs, 36.20.Ey I. INTRODUCTION Many important biopolymers, such as DNA, the cy- toskeletal filaments (filamentous (F-)actin, microtubules, intermediate filaments), as well as collagen in the ex- tracellular matrix are fluctuating macromolecules with a bending stiffness intermediate between that of a ran- dom coil (Gaussian chain) and a rigid rod. Polymers whose elastic behavior is dominated by their bending rigidity are known as semiflexible. Numerous experi- ments probing their elasticity have become available [1, 2] with the advances in single-molecule manipulation, par- ticularly for DNA. Intriguing and qualitatively novel me- chanical behavior arises if semiflexible polymers are pair- wise permanently cross-linked. The elasticity of cross- linked biopolymers is widely studied experimentally via force-extension measurements. In this article, we study analytically the force-extension relation of an irreversibly cross-linked pair of semiflexible polymers within a meso- scopic theoretical model. Ubiquitous as extracellular mechanical support is the connective-tissue protein collagen, whose fibrils achieve their strength via covalent intermolecular cross-links be- tween triple-helical molecules [3, 4]. Atomic force mi- croscopy [5 -- 7] has been used to analyze single collagen fibrils, which themselves consist of many microfibrils and hence can be modeled as anisotropic networks of irre- versibly cross-linked semiflexible polymers [8]. Cell shape and stability is provided by the actin cytoskeleton, a net- work of cross-linked F-actin ranging in morphology from a dilute mesh to bundles of parallel filaments [9]. The special elastic properties due to cross-linking, closely re- lated to biological function, thus have become a subject ∗ [email protected] of increasing interest and vigorous research activity [10]. Yet, theoretical understanding is incomplete and expla- nations based on semi-microscopic descriptions are rare. Crucial experimental results such as the strong stretch- ing of double-stranded DNA [11] have been successfully explained by the theoretical force-extension relation for a weakly bending wormlike chain [12]. The wormlike chain (WLC) [13 -- 15] maps the conformations of an in- extensible semiflexible polymer to one-dimensional paths whose statistical weight penalizes curvature and is de- termined by two length scales only: the total contour length L and the directional correlation or persistence length Lp, proportional to the bending rigidity κ. The weakly bending approximation of a WLC [12] simplifies analytical treatment by assuming that the tangent vec- tor at any arc-length position and the end-to-end vector make a small angle. This approximation applies to poly- mers with a large persistence length Lp (compared to L) or subjected to strong stretching. Inhomogeneities or inter-molecular interactions have been the subject of sev- eral modifications and extensions of the weakly bending WLC: In [16], the force-extension relation of two par- allel aligned, weakly bending WLCs with a single irre- versible cross-link and the elasticity of an anisotropic network of aligned chains have been analyzed. In the wormlike bundle model, an arbitrary number of regularly arranged parallel filaments is effectively cross-linked by a coarse-grained, continuous interaction [17]. The effect of spontaneous polymer curvature has been studied in [18, 19]. Weak extensibility of semiflexible polymers at strong stretching has been addressed with a combination of a WLC and a Gaussian chain, the semiflexible har- monic chain (SHC), in [20]. In this work, we consider the elasticity of two iden- tical weakly bending WLCs connected by an arbitrary number of cross-links regularly spaced along the poly- mer contour. The cross-links are represented by entropic harmonic springs, which allow for a finite extent of the inter-polymer distance at the cross-link sites. In the case of infinite spring strength, we obtain the limit of hard cross-links (strong topological constraints). By introduc- ing infinitely many cross-links at fixed contour length, we can also model a continuous cross-linking or an attrac- tive inter-molecular interaction. The ladder structure of our system is reminiscent of the base-pair sequence of double-stranded DNA, but for the reversible hydrogen bonding. This obvious modification of our model to re- versible and/or sectional cross-linking may prove versa- tile for future studies of, e.g., the denaturation of DNA. The paper is organized as follows: In Section II, we in- troduce the model and the observable. In Section III, we present the main steps in calculating the canonical parti- tion function, from which all equilibrium quantities can be derived. Details of this calculation are given in Ap- pendices A and B. In Section IV, we present the central result which is the force-extension relation. After pre- senting the general result, we particularly focus on the limit of hard cross-links, of continuous cross-linking, the linear elasticity for small forces, and the strong stretching limit. For the limit of continuous cross-linking, the gen- eral result and a short comparative discussion are given in Appendix C. We conclude and discuss further exten- sions of this work in Section V. We consider two weakly bending, semiflexible chains of equal bending rigidity and contour length, aligned parallel along a preferential direction x and cross-linked at equidistant arc-length positions specified below. The chain configurations are described by paths r1(s), r2(s), with s ∈ [0, L] the arc-length parameter, and tangential vectors tj(s) = ∂srj := ∂rj/∂s. Our setup, taking into account space dimension d = 2 only, is sketched in Fig. 1 for 3 cross-links. As indicated, we assume hinged-hinged FIG. 1. (Color online) Stretched, weakly bending (see main text for explanation) chain pair connected by 3 cross-links. boundary conditions, implying confinement of the verti- cal positions (here, to y = 0) and vanishing curvature at the ends. These boundary conditions are motivated by the following situation: Experimentally, a tensile force can be applied via optical or magnetic tweezers that con- trol the position of beads attached to the polymers' ends, cf., e.g., [21]. Optical tweezers usually restrict the bead's H = II. MODEL = −f j=1 0 2 transverse motion, but not the rotation, so that no mo- ments are exerted at the ends. Additionally, we assume x1(0) = x2(0), in order to exclude an overall x shift be- tween the chains. The effective bending potential of semiflexible chains (without taking into account torsions [21]) is Hbend = κ 2 ds 0 j=1 ∂stj(s)2 , (1) with the bending rigidity κ, related to the persistence length Lp in d = 2 via (cid:90) L 2(cid:88) κ = 1 2 kBT Lp, and with the local inextensibility constraint tj(s) ≡ 1, s ∈ [0, L], j = 1, 2. In order to account for inextensibility in a mathemat- ically tractable way, we consider the weakly bending ap- proximation: The chains' tangents preferentially align with a given direction, here x. A stretching force of strength f , acting on both ends of the chains (cf. Fig. 1), is described by the potential (2) (3) (4) Hstretch = −f ex · 2(cid:88) (cid:0)rj(L) − rj(0)(cid:1) (cid:90) L j=1 ds ∂sxj(s). 2(cid:88) For sufficiently large stretching forces or bending rigidi- ties, the tilt of the tangent vector away from the x axis is small, so that the condition Eq. (3) reads approximately (cid:16) (∂syj(s))4(cid:17) . (5) ∂sxj(s) = 1 − 1 2 (∂syj(s))2 + O Inserting this expansion into Eqs. (1) and (4) and dis- carding all but quadratic terms in derivatives of y, we arrive at the weakly bending approximations of the bend- ing and stretching potentials. Cross-links between the two chains are introduced at N − 1 sites regularly spaced along the contours, sb = bL N , b = 1, 2, . . . , N − 1, (6) dividing the contour length L into N sections, cf. Fig. 1. Explicitly, we model cross-links as entropic, harmonic springs of strength g = 2 kBT /a2 is the temperature-independent squared equilibrium length of one cross-link. c, where a2 c Finally, the total effective Hamiltonian is (cid:19) (∂syj)2 − 2f L (cid:125) 2(cid:88) (cid:124) j=1 + g 2 ds s yj (cid:0)∂2 (cid:18) κ (cid:90) L (cid:1)2 (cid:123)(cid:122) N−1(cid:88) (cid:0)y1(sb) − y2(sb)(cid:1)2 H0 2 + f 2 0 b=1 , (7) yx−f−fffr1(s)t1(s)r2(s)s=0s=Lcross-links13 we impose hinged-hinged boundary conditions, which for our system translate into [y(cid:48)(cid:48) yj(0) = yj(L) = 0 y(cid:48)(cid:48) j = 1, 2. (13) j (s) := ∂2 s y(s)] j (L) = 0, According to these boundary conditions, our Fourier- series ansatz is y1(s) = Am sin(qms), j (0) = y(cid:48)(cid:48) M(cid:88) M(cid:88) m=1 The first step is to expand the chain configurations yj(s) in appropriate eigenfunctions. As mentioned above, due to thermal fluc- tuations of uncross-linked weakly bending WLCs only is H0 where H0 is the Hamiltonian of the system without cross- linking, and the last term ∝ kBT describes the entropic cross-links in weakly bending approximation. Starting from the concept of harmonic cross-links at discrete sites sb = bL/N , we will also consider the limit of continuous cross-linking, achieved by taking N → ∞ and ∆s := L/N → 0. In this case, the strength g of a single cross-link has to go to zero, such that the total strength 0 ds in the cross-link part of Eq. (7) gives a continuous, harmonic inter-chain attraction of strength g/L, g := N g remains finite. Replacing(cid:80)N−1 ds(cid:0)y1(s) − y2(s)(cid:1)2 b=1 → N (cid:90) L H(c) = (cid:82) L g 2L L 0 It is our aim to study the effect of cross-links on the chain elasticity and hence to compute the force-extension relation exactly for an arbitrary number of irreversible cross-links. Thus, the relevant quantity is the average end-to-end extension of one chain in force direction x, (cid:42)(cid:90) L (cid:43) x(L)− x(0) H = L− 1 2 ds (∂sy)2 , (9) H 0 (cid:69) H denotes the canonical average with the Hamil- tonian of Eq. (7). The force-extension relation of one weakly bending WLC without cross-links (Hamiltonian H0), first ad- dressed by Marko and Siggia [12], is (cid:10)x(cid:11) := where(cid:10) (cid:68) (cid:11) · (cid:10)x(cid:11) H0 L = 1 − L 2Lp (cid:26) coth√fr√fr − 1 fr (cid:27) , (10) in terms of the dimensionless variable fr := f L2/κ, (11) which is the ratio of stretching energy, f L, and bending energy, κ/L. The force-extension curve displays a lin- ear regime for small forces f and in the limit of strong stretching approaches the maximal end-to-end extension L with a characteristic saturation ∝ f−1/2. III. PARTITION FUNCTION cal partition function, Z =(cid:82) In this section, we detail the calculation of the canoni- D[y(s)] e−βH[y(s)] (the con- figurational integral for both chains denoted by D[y(s)]), which provides access to all equilibrium observables. For the purpose of this work, the end-to-end extension de- fined in Eq. (9) is obtained from lnZ or the free energy [22] by differentiation with respect to the force f : (cid:10)x(cid:11) = kBT 2 ∂ lnZ ∂f . (12) . (8) m=1 y2(s) = Bm sin(qms), (14) with wave numbers qm := mπ L , m the mode number, (15) and M the largest undulation mode considered within our continuum model (roughly, the wave-length reso- lution is bounded by molecular distances). With this ansatz, the Hamiltonian H can be written as a quadratic form in the coefficient vector Γ := (A1, B1, A2, B2, . . .)T . (16) Omitting the constant −2f L, M(cid:88) (cid:16) C(cid:96)(cid:96)(cid:48) +(cid:0)U U T(cid:1) (cid:17) (cid:96)(cid:96)(cid:48) Γ(cid:96)(cid:48), (17) H[Γ] = Γ(cid:96) (cid:96),(cid:96)(cid:48)=1 where C, due to H0 of the uncross-linked system, is a diagonal matrix (⊗ denotes the Kronecker product), (cid:19) (cid:18) 1 0 0 1 C = diag(c1, c2, . . .) ⊗ cm := L 4 (f + κq2 m)q2 m, , (18) and U U T the matrix due to the cross-link Hamiltonian, cf. the second line of Eq. (7), or Eq. (8). The partition function follows as a generalized Gaus- sian integral over the mode coefficients normalized by L, (cid:90) M(cid:89) Z = D[Γ] := D[Γ] e −βH[Γ], (cid:18) Am (cid:19) d L (19) (cid:18) Bm (cid:19) L . d m=1 Returning to the end-to-end x-extension introduced in Eqs. (9) and (12), we wish to focus primarily on the cross- link contribution, i.e., (cid:69) (cid:68) (cid:68) (cid:10)∆x(cid:11) := since the extension(cid:10)x(L) − x(0)(cid:11) x(L) − x(0) H − x(L) − x(0) , H0 (20) (cid:69) known [12]. To that end, we write the partition function as Z = ZrelZ0, where Z0 is the partition function of the uncross-linked system and address the relative partition function [cf. Eq. (17)], Zrel := (cid:82) (cid:82) D[Γ] e−βH[Γ] =(cid:0)det(1 + C−1U U T )(cid:1)−1/2 D[Γ] e−βH0[Γ] (cid:10)∆x(cid:11) = ∂ lnZrel . kBT 2 ∂f (21) . (22) This yields the cross-link-induced extra displacement 4 the occupation structure of the N − 1 eigenvectors (la- beled by l in the following) in the Fourier basis, each with nonzero amplitudes only for a subset of modes, which at all cross-link sites are pairwise in-phase or phase-shifted by π. Modes indexed by multiples of N have nodes at all cross-link sites, thus do not contribute to the cross-link energy, and constitute the kernel of P . Due to this special form of the matrix C−1U U T , we are able to derive a closed expression for the trace of any power of C−1U U T (cf. Appendix B), viz. (cid:18) gN (cid:19)−k tr(cid:0)C−1U U T(cid:1)k (cid:32) ∞(cid:88) (cid:16) N−1(cid:88) 2 = c−1 2(µ−1)N +l + c−1 2µN−l (26) (cid:17)(cid:33)k . A. Finite number of cross-links l=1 µ=1 Zrel = exp − 1 2 tr (−1)k+1 k (C−1U U T )k . (24) By inserting the c−1 factors of the partition function, Eq. (27), m into the series, we obtain for the First, we address a finite number N − 1 of equidistant harmonic cross-links, for which the cross-link Hamilto- nian is quadratic, but not diagonal in the modes. The matrix U U T is a sum of N − 1 projectors, N−1(cid:88) (cid:0)ub ⊗ uT b (cid:1) ⊗ g 2 (cid:19) (cid:18) 1 −1 1 , (23) U U T = −1 b := (sin(q1sb), sin(q2sb), . . .). b=1 uT Using the identity det exp A = exp tr A to expand the determinant in Eq. (21), Zrel is accessible via traces of powers of the matrix C−1U U T , (cid:41) (cid:40) ∞(cid:88) k=1 N−1(cid:88) In the trace of a power k of C−1U U T , the 2×2 matrices in the Kronecker products, Eqs. (18) and (23), merely pro- duce a factor 2k, which can be computed separately be- fore performing the trace operation over the mode indices mj. Hence, we are left with handling the non-diagonal mode-index structure of the projector sum Eq. (23). The corresponding matrix of rank (N − 1), with mode indices m1, m2 has entries P := 2 N N−1(cid:88) 2 N b=1 ub ⊗ uT b , (cid:19) (cid:18) bm1π sin (cid:18) bm2π (cid:19) Pm1m2 = sin N b=1 N = δm1−m2,2ZN − δm1+m2,2ZN , (25a) (25b) where Z denotes the set of integers, such that P is a sparse matrix of block form: In each quadratic block of dimension 2N , nonzero entries appear on the diagonal (+1) and on one anti-diagonal (−1) only. For the explicit form of P , see Appendix A. The rows/columns display Here, due to the fast decay with mode number of the inverse coefficients c−1 m from Eq. (18) -- basically inverse elastic constants for the undulation modes -- we have ex- tended the summation over modes to a series. Combin- ing Eqs. (24) and (26), we find that the relative partition function Zrel factorizes into N − 1 different "eigenvector" factors, or equivalently, (27) (cid:17)(cid:27) c−1 2(µ−1)N +l + c−1 2µN−l (cid:16) ∞(cid:88) µ=1 gN 2 (cid:26) lnZrel 1 = − 2 N−1(cid:88) N−1(cid:88) l=1 =: l=1 ln Zl. ln 1 + (cid:26) (cid:16) gL N f (cid:17)(cid:27)−1/2 Zl = 1 + ψl(0) − ψl(δf ) , (28) ψl(δf ) = sinh δf δf cosh δf − cos φl . Herein, we employ the dimensionless variable δf := , (29) L√f N√κ (cid:112)κ/f of the stretched WLC, or the penetration which is the ratio of two lengths: The arc-length spacing L/N between cross-links and the directional 'memory' length depth of boundary conditions [18]. The dependence of the partition function on the bending rigidity κ is via this ratio only. Additionally, we define the phases specific to the N − 1 eigenvectors, φl := . (30) πl N We note that the Gaussian statistical weights and the regular cross-link spacing simplify enormously, if not en- able at all, analytical calculations. In fact, Eqs. (27) and (28) are a central result of our paper, yielding the exact free energy of two cross-linked, weakly bending WLCs as F = F0 + ∆F , where F0 is the free energy of the chain pair without cross-links, F0 = −kBT ln Z0 = kBT ln L2cm πkBT , (31) M(cid:88) m=1 (leading to the force-extension relation Eq. (10)), and the free-energy increment ∆F due to cross-links is the sum of the Fl = −kBT ln Zl from Eq. (28). Via δf , this free energy depends on the dimensionless energy ratio fr = N 2δ2 f introduced in Eq. (11). As already mentioned, use of the weakly bending approximation requires that either the work done by the external force f or the bend- ing energy is large compared to the thermal energy, in order to restrict the transverse fluctuations to be small. The dependence on the free parameter fr will be fur- ther discussed for the cross-link contribution to the force- extension relation. In addition, the free energy depends on the ratio of cross-link energy to stretching energy, g := gL N f (32) and, of course, on the number of cross-links, N − 1. B. Continuous cross-linking via harmonic inter-chain attraction Here, we sketch the derivation for an infinite number of regularly spaced cross-links, N → ∞, at finite total strength g := N g, and for finite contour length L. With the continuous cross-link Hamiltonian Eq. (8) in normal- mode representation, (Am1 − Bm1) (Am2 − Bm2 ) (33) M(cid:88) H(c) = g 2L m1,m2=1 (cid:90) L M(cid:88) 0 m=1 × g 4 = ds sin(qm1s) sin(qm2s) (cid:0)Am − Bm (cid:1)2 , the total Hamiltonian is diagonal with respect to the mode indices. Thus, we find for the excess free energy due to the inter-chain attraction a closed expression, again extending the sum over the modes to a series, (cid:18) ∞(cid:88) ∆F (c) = −kBT lnZ (c) rel = kBT 2 ln m=1 (cid:19) , 1 + c−1 m g 2 (34) in agreement with performing the limit N → ∞ at finite g in Eq. (27). IV. FORCE-EXTENSION RELATION 5 Using Eqs. (27) and (28), straightforward yet tedious differentiation with respect to f yields the force-extension relation (cid:10)∆x(cid:11) N−1(cid:88) l=1 = kBT g 8N f 2 nl(δf ) dl(δf , g) , (35) L with numerator (cid:20) nl(δf ) = 2 (cosh δf − cos φl) − (1 − cos φl) 3 and denominator (cid:21) (36a) sinh δf δf − 1 − cos φl cosh δf cosh δf − cos φl dl(δf , g) = (cid:20) (1 − cos φl) (cosh δf − cos φl) + g cosh δf − cos φl − (1 − cos φl) (cid:21) (36b) sinh δf δf in terms of the length ratio δf from Eq. (29), the ratio g from Eq. (32), and the phases φl from Eq. (30) [23]. Since a direct interpretation of the expressions in Eqs. (35) and (36) is difficult, in Fig. 2 we show the cal- FIG. 2. (Color online) Force-extension relation at a relative persistence length Lp/L = 10: for different numbers of cross- links at finite cross-link strength ga = 50 (top) and for differ- ent cross-link strengths at 10 cross-links (bottom). 0.984 0.986 0.988 0.99 0.992 0.994 0.996 0 10 20 30 40 50rel. extension AExae/Ldim.-less force fr = fL2/kX-link strength ga = 50, persist. length Lp = 10LX-links:012101001000 0.984 0.986 0.988 0.99 0.992 0.994 0.996 0 10 20 30 40 50rel. extension AExae/Lforce fr = fL2/k10 X-links, persist. length Lp = 10LX-link strength ga:110505005000culated force-extension relation for several numbers and strengths of cross-links, as a function of the dimensionless force variable fr, Eq. (11). The dimensionless parameter for the cross-link strength is ga := gL2 kBT = 2 L2 a2 c , (37) which, by virtue of our entropic-spring model for the cross-links, can be expressed as the squared ratio of the WLC contour length and one cross-link's length at rest. First, in the upper part of Fig. 2, the force-extension curve is plotted for different numbers of cross-links, at constant strength of a single cross-link. The overall form of the saturation curve is reminiscent of an unconstrained weakly bending WLC [12]. Evidently, the general effect of cross-linking is to increase the extension in force direc- tion relative to an uncross-linked weakly bending chain, because cross-links effectively suppress thermal fluctua- tions perpendicular to the aligning force. The growth of the extra alignment with the number of cross-links is nonlinear, the increase relative to a chain pair with less cross-links being largest for a few cross-links, and for weak stretching. The limit of continuous cross-linking, cf. Sec. (III B), is discussed in Sec. IV B. Enforcing a smaller and smaller cross-link length (in- creasing the cross-link strength) enhances the alignment or effective stiffness, too, as visible in the lower part of Fig. 2, in which the cross-link strength is varied at a fixed number of cross-links. The limit of strong topologi- cal constraints at the cross-link sites (hard or inextensible cross-links) is presented in Sec. IV A. Cross-links are most effective in suppressing transverse fluctuations and aligning the chain pair at small reduced stretching forces, at which the directional memory length (cid:112)κ/f is still large compared to the cross-link spacing L/N . For these relatively weak pulling forces, there is a regime of linear elasticity for all numbers and strengths of cross-links, taken a closer look upon in Sec. IV C. For increasing force, the incremental extension due to cross-links decreases, since at strong stretching, the dom- inant contribution to the saturating extension arises from "pulling out" the remaining length reserves stored in thermal undulations. The asymptotic decay of the cross- link contribution with force is computed in Sec. IV D. In Fig. 2, we have chosen a ratio of persistence to con- tour length Lp/L = 10 sufficiently large as to give for all fr relative extensions close to 1, in order to explore the entire range of stretching forces and yet keep the weakly bending approximation. A ratio Lp/L of this order would apply to long microtubules [24]. The persistence length of actin is about 15 µm [24, 25], thus for typical lengths of actin filaments in solution, the ratio Lp/L is of or- der 1. For smaller ratios, e.g., Lp/L ∼ 0.1 for type I collagen fibrils [26, 27], or Lp/L ∼ 0.01 for 10 µm of double-stranded DNA [11], our predictions are reason- able at strong stretching only. A. Limit of hard cross-links 6 In the limit of infinite cross-link strength or vanishing ratio of cross-link to contour length, ac/L → 0, we have (cid:10)∆x(h)(cid:11) L N−1(cid:88) l=1 = kBT 8f L nl(δf ) d(h) (δf ) l , (38) with nl(δf ) from Eq. (36a), and d(h) l (δf ) = cosh δf − cos φl − (1 − cos φl) sinh δf δf . (39) The corresponding force-extension curves are shown in Fig. 3, for more flexible weakly bending chains (with FIG. 3. (Color online) Force-extension relation for different numbers of hard cross-links and Lp = L. Lp = L) than the rather rod-like chains in Figs. 2 and 4. Since hard cross-links completely eliminate relative motion of the filaments transverse to stretching at the cross-linking sites, the relative alignment effect due to cross-links is seen to be stronger. B. Continuous cross-linking In this Section, we discuss the limit of continuous cross- linking from Sec. III B, N → ∞ at total strength g := N g. The incremental extension of one chain due to con- tinuous cross-linking is computed from ∆F (c), Eq. (34), by differentiation, cf. the general expression for all values of g and further remarks in Appendix C. In the case of continuous and rigid cross-linking (ac → 0), the force- extension relation is (cid:10)x(c, h)(cid:11) L = 1 − L 4Lp (cid:26) coth√fr√fr − 1 fr (cid:27) . (40) Comparing this result to Eq. (10), we observe that the squared thermal y-fluctuations, cf. Eq. (9), are reduced by 1/2 relative to the uncross-linked case. This does, however, not imply that the two chains attached to each other rigidly can be treated as unconstrained weakly 0.82 0.84 0.86 0.88 0.9 0.92 0.94 0.96 0 10 20 30 40 50rel. extension AExae/Lforce fr = fL2/khard X-links, persist. length Lp = LX-links:01210contin.bending WLCs with just one effective persistence length or bending stiffness κeff, since fr itself depends on the bending stiffness κ. The different apparent persistence lengths in the force regimes of linear elasticity and of strong stretching are discussed in the next two Sections. The extension for rigid, continuous cross-linking for ar- bitrary force, Eq. (40), is shown as the topmost curve in Fig. 3, corresponding in Fig. 2 to the asymptotic case of both infinite number and strength of cross-links. C. Force-free extension and linear response regime Knowing the exact extension curve for all values of the force f allows us to address the equilibrium extension in the limit f → 0 and the linear elasticity for small f -- assuming a large persistence length Lp, so that weakly bending holds. This linear response or weak-perturbation regime may be the best accessible for stretching exper- iments on sensitive biopolymers. Moreover, the force- extension curves computed within our model suggest that the chain pair's extension at moderate or zero force, cf. Fig. 2, is most indicative of the degree of cross- linking. Without cross-links, the equilibrium extension of a weakly bending WLC parallel to alignment for f → 0 is, following Eq. (10), (cid:10)x0 (cid:11) L H0 (cid:10)x(cid:11) := lim f→0 H0 L = 1 − L 6Lp , (41) cf. [21]. The deviation from the maximal extension is inversely proportional to the persistence length. Cross- links increase the equilibrium extension according to = N−1(cid:88) (cid:2)x2 l + 13xl + 16(cid:3) / (1 − xl) 6 (1 − xl)2 + 2gaL N 3Lp (2 + xl) (42) , 20N 5L2 p l=1 with xl := cos φl. For large Lp and finite ga, this expres- sion is O (L/Lp)2 and hence small. For hard cross-links, Eq. (42) is linear in L/Lp, so that an effective persis- tence length Lp,eff > Lp of the cross-linked chains can be defined, viz., Lp Lp,eff = 1 − 3 20N 2 x2 l + 13xl + 16 (1 − xl) (2 + xl) . (43) In the limit of continuous, rigid cross-linking discussed in Sec. IV B, the increase in equilibrium extension is 7 FIG. 4. (Color online) Linear elastic constant as a function of the number of cross-links, N − 1, for Lp/L = 10. In Fig. 4, we show the dependence of the linear elastic constant, computed in dimensionless form as (cid:32) (cid:10)x(cid:11) L (cid:33)−1(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)fr=0 E0 := ∂fr , (45) on the number of cross-links. The elastic constant of an uncross-linked weakly bending WLC is given by E0 = 90Lp/L. For extensible cross-links, a large increase with cross-link number up to N ≈ 10 is followed by a satu- ration to twice the elastic constant of an uncross-linked weakly bending WLC. For hard cross-links, the increase in the elastic constant caused by introducing only a few cross-links is even more drastic, and the curve approaches a step function. D. Strong stretching limit Here, we consider the limit of strong stretching, i.e., fr (cid:29) 1, or for finite N , √fr/N (cid:29) 1, which means that the directional memory length introduced after Eq. (29) is much smaller than the cross-link spacing, (cid:114) κ f (cid:28) L N . (46) At finite cross-link strength and for a finite number of cross-links, the dependence on the individual eigenvector phases remains in the limit of strong stretching, yet the asymptotic scaling with fr is the same for all summands, (cid:10)∆x(cid:11) L gaL2 N L2 p = f−2 r N−1(cid:88) l=1 1 1 − cos φl + O (47) (cid:16) (cid:17) . f−5/2 r (cid:10)∆x0 (cid:11) L gaL2 N−1(cid:88) l=1 (cid:10)∆x(c, h) (cid:11) 0 L Upon comparison with Eq. (41), we thus find the zero- force extension of one weakly bending WLC with twice the original persistence length or κeff = 2κ. = L 12Lp . (44) The same asymptotic decay ensues for continuous cross- linking at finite total strength ga := 2N L2/a2 c, cf. Eq. (C1), but with a simpler coefficient, (cid:10)∆x(c)(cid:11) L (cid:16) (cid:17) = gaL2 3L2 p f−2 r + O f−5/2 r . (48) 800 1000 1200 1400 1600 1800 0 10 20 30 40 50linear elast. const. E0number of cross-links N-1persist. length Lp = 10LX-link strength:ga = 50hard X-linksIn the presence of a finite number of hard cross-links, strong stretching asymptotically results in an extension increment, which is independent of the individual eigen- vector phases and decays proportional to f−1 , r (cid:16) (cid:17) = (N − 1)L 2Lp f−1 r + O f−3/2 r . (49) (cid:10)∆x(h)(cid:11) L Hence the asymptotic decay of the extra alignment with force is markedly slower than for extensible cross-links. In all cases mentioned so far, the impact of cross-linking diminishes fast for strong stretching, and the extension curve displays the saturation ∝ f−1/2 of the uncross- linked chain's extension to the contour length. Obviously, the asymptotic behavior computed for hard cross-links in Eq. (49) cannot apply in the limit N → ∞ of continuous, rigid cross-linking, due to the diverging prefactor ∝ (N − 1). Indeed, in this case, the asymptotic decay of the inter-chain contribution to the extension is even slower, viz. (cid:10)∆x(c, h)(cid:11) L (cid:0)f−1 r (cid:1) . = f−1/2 r L 4Lp + O (50) This is the same scaling as the saturation of the uncross- linked chain's extension, hence the stabilizing effect of continuous, rigid cross-linking is manifest even for large stretching forces. Moreover, in the strong stretching limit, a continuously and rigidly linked chain behaves ef- fectively like a weakly bending WLC with fourfold origi- nal persistence length or κeff = 4κ. Of course, the WLC picture is oversimplified at very strong stretching, at which inextensibility is clearly vio- lated for many semiflexible biopolymers [2, 28]. E. Differential stiffness A quantity of interest related to the x extension is the differential stiffness: In the corresponding experiment for our system, both fibers are pre-stressed by longitudinal stretching, subsequently, the strain response to a small change in the applied stress is measured. From the force- extension relation, we can readily compute the differen- tial stiffness as the quotient of force and extension in- crement at a given pre-stretching force. More precisely, we consider the dimensionless differential stiffness as a function of the dimensionless force fr, (cid:32) (cid:33)−1 (cid:10)x(cid:11) L E(fr) := ∂fr , (51) generalizing the elastic constant discussed in Sec. IV C. In an effort to highlight the effective stiffening due to cross-links, we show in Fig. 5 the increase in differential stiffness relative to uncross-linked weakly bending WLCs for the case of hard cross-links. Again, for a few cross- links, the differential stiffness is particularly enhanced 8 FIG. 5. (Color online) Increase in differential stiffness relative to a weakly bending WLC without cross-links. at weak stretching. Already for a single, hard cross-link, the linear elastic constant is increased by more than 80 % compared to the uncross-linked case. Upon approaching the limit of rigid, continuous cross-linking, cf. Eq. (40), the differential stiffness is increased by a factor two for all values of fr. V. DISCUSSION AND OUTLOOK Within a transparent mesoscopic model of cross-linked polymers, the elasticity of two irreversibly cross-linked WLCs subjected to a tensile force has been studied in the weakly bending approximation. The validity of the latter is granted by assuming either a large tensile force or a large bending rigidity. For an arbitrary number of cross-links with given strength, we have calculated the free energy and, derived thereof, the force-extension re- lation exactly. Both with increasing number N and with increasing strength g of the cross-links, the effective stiff- ness of the chain pair increases, since cross-links stabilize the chains against thermal undulations. Particularly for weak stretching, the enhancement in alignment is con- siderable, such that in corresponding weak-perturbation experiments on biopolymers, the increase in the linear elastic constant may be a useful indicator of (partial) cross-linking. As expected, the effect is most pronounced for hard cross-links. In the limit of strong pulling forces, the additional extension (cid:104)∆x(cid:105) due to cross-linking de- creases, and the elasticity of an uncross-linked WLC [12] dominates. However, the asymptotic behavior for large stretching forces is different for hard and extensible cross- links, as well as for discrete and continuous cross-linking, and is summarized in Table I. For extensible cross-links, the cross-link contribution decays as f−2, for a finite number of hard cross-links, as f−1. A slower decay is found in the limit of both cross-link number N → ∞ and cross-link strength g → ∞, in which the two chains are linked continuously and rigidly along their contour: For strong stretching, the asymptotic form of the force- 0 20 40 60 80 100 0 10 20 30 40 50rel. increase diff. stiffness [%]force fr = fL2/khard X-links, persist. length Lp = 10LX-links:124101009 extension relation reflects the behavior of one uncross- linked weakly bending chain with effective persistence length 4Lp and with the known f−1/2 scaling. TABLE I. Exponents of the asymptotic scaling of (cid:104)∆x(cid:105) with force fr = f L2/κ in the strong stretching limit, fr (cid:29) 1. number of cross-links extensible hard finite −2 −1 infinite −2 −1/2 method two parallel aligned WLCs whose arc-length is sectioned into cross-linked and disconnected parts. In the set-up we have considered here, all ingredients, viz., cross-linking, bending stiffness, and longitudinal forcing, act to decrease transverse fluctuations of the chains. An unzipping transition could presumably be studied in an altered situation, e.g., one, in which the cross-linked fil- aments are teared apart at one end. More complicated refinement of our model might account for twist and over- stretching, effects shown to be essential for the elasticity of DNA [37]. From the exact extension for all stretching forces, we have computed another experimentally relevant observ- able, viz., the differential stiffness of the (pre-stretched) cross-linked chains. Even a small number of cross-links enhances the differential stiffness dramatically. Again, the impact is largest for small stretching forces, which can be considered within the weakly bending approxima- tion for WLCs with a large persistence length Lp/L. Several generalizations of our approach are possible: As alluded to in [16], our model is not in principle re- stricted to a pair of cross-linked filaments, but should be generalizable to describe the tensile elasticity of a stretched, weakly bending WLC bundle, possibly with random and/or reversible cross-links. In order to take into account non-affine deformation of cross-linked bun- dles, we may have to consider also the shearing of cross- links. Detailed analysis of bundles exists, due to the relevance for actin networks, mostly for reversibly cross- linked and extensible, semiflexible polymers [29, 30]. Par- ticularly for bundles, the effect of excluded volume inter- action, neglected in this work, remains to be explored. Apart from activities in this realm, major recent re- search efforts are devoted to the impact of structural inhomogeneities caused by the (local) breaking of com- plimentary base-pair bonds ("unzipping" or denatured "bubbles") on the elasticity of double-stranded DNA [31, 32]. A class of semi-microscopic models convenient for analyzing the thermal denaturation transition, as well as the "bubble" statistics and dynamics [33], focuses on the form of the base-pairing interaction, but does not account for the polymers' conformational degrees of free- dom, which determine certain DNA properties [34]. In the breathing DNA model [35], two discrete chains (con- sisting of interacting "beads") with bending and stretch- ing rigidity interact via the pairing energy of compli- mentary bases, represented by a Morse potential. An- other semi-microscopic model, amenable to a transfer matrix method, considers a discrete WLC model for the chain conformations, coupled to an one-dimensional Ising model describing the internal base-pair states [36]. In the context of denaturation of DNA, it would be interesting to extend our model to reversible cross-linking, in order to study the coexistence of "ladders" (cross-linked strand sections) and "bubbles" (open sections). A first, obvious step towards this direction will be to address with our ACKNOWLEDGMENTS We thank W. T. Kranz for valuable discussions and careful reading of the manuscript. Financial support by the Deutsche Forschungsgemeinschaft through grants SFB-937/A1 and SFB-937/A4 is gratefully acknowl- edged. P. B. acknowledges support by Kyungpook Na- tional University Research Fund, 2012. Appendix A: Structure of the projector sum P The matrix representation of the cross-link projector sum P , cf. Eqs. (25), has the following structure:  1 1 . . . 1 1 ... . . . ⊗ P =   . 0 ... 0 ... 0 1 1 0 . . . . . . 0 . . . ... 0 −1 0 ... . . . 1 0 ... 0 0 . . . ... 0 −1 . . . 0 −1 0 0 . . . −1 1 0 ... 0 . . . 0 . . . 0 . . . 0 . . . 2N×2N 0 (A1) Appendix B: Trace of powers of the projector sum By decomposing mode indices m ∈ {1, 2, . . .} according to the block structure into :=(cid:6) m (cid:7) ρ(r) :=(cid:4) r (cid:5) 2N m = 2µN − r with the definitions µ r ∈ {1, 2, . . .} (block index), ∈ {0, . . . , 2N − 1} (index within a block), ∈ {0, 1} (quadrant within a block), N (B1) the entries of P , cf. Eqs. (25) and (A1), can be encoded in product form, the first two factors indicating the location of nonzero entries, the last factor the sign, Pm1m2 = δm1−m2,2ZN − δm1+m2,2ZN = (δr1,r2 + δr1,2N−r2)(1 − δr1,ZN )(−1)ρ1+ρ2 . (B2) Then, with the diagonal matrix C, cf. Eq. (18), we write the trace of a power k of C−1U U T , cf. Eq. (23), as linking computed from Eq. (34) is (cid:10)∆x(cid:11) L = L 4Lp  sinh g+ sin g− g− − cosh g+ − cos g− + g+ 10 (C1)  , coth√fr√fr − 1 fr c−1 m1 (δm1−m2,2ZN − δm1+m2,2ZN ) · . . . 2N−1(cid:88) (cid:18) gN = = 2 m1,...,mk=1 tr(cid:0)C−1U U T(cid:1)k (cid:19)−k ∞(cid:88) ∞(cid:88) × c−1 mk (δmk−m1,2ZN − δmk+m1,2ZN ) × c−1 × c−1 × . . . × c−1 µ1,...,µk=1 r1,...,rk=0 2µ1N−r1(δr1,r2 + δr1,2N−r2)(1 − δr1,ZN )(−1)ρ1+ρ2 2µ2N−r2(δr2,r3 + δr2,2N−r3)(1 − δr2,ZN )(−1)ρ2+ρ3 2µkN−rk (δrk,r1 + δrk,2N−r1)(1 − δrk,ZN )(−1)ρk+ρ1. Due to the symmetry of the constraints, we are left with the summation over one of the rj without 0 and N . If we split this sum according to the constraints and to the eigenvector structure mentioned above as 2N−1(cid:88) r=0 fr(1 − δr,ZN ) = N−1(cid:88) l=1 (fl + f2N−l) , (B4) we finally arrive at (cid:18) gN 2 (cid:16) tr(cid:0)C−1U U T(cid:1)k N−1(cid:88) (cid:19)−k ∞(cid:88) (cid:16) (cid:32) ∞(cid:88) c−1 2(µk−1)N +l + c−1 N−1(cid:88) µ1,...,µk=1 (cid:16) l=1 × = = c−1 2(µ1−1)N +l + c−1 2µ1N−l (cid:17) 2µkN−l (cid:17)(cid:33)k . c−1 2(µ−1)N +l + c−1 2µN−l (B5) (cid:17) · . . . l=1 µ=1 Appendix C: Continuous cross-linking Here, we present the general result for the limit of continuous cross-linking dealt with in Secs. III B and IV B. The incremental extension due to this kind of cross- as a function of the dimensionless force fr, Eq. (11), and the dimensionless parameters (B3) g− := g+ := 2 2 2gL3/κ − fr, 2gL3/κ + fr, (C2) (cid:113) (cid:113) (cid:112) (cid:112) which apart from fr contain the ratio of total inter-chain attraction gL2 = 2N kBT L2/a2 c to bending energy κ/L. (The function in Eq. (C1) is a continuous, real-valued function of fr independently of the sign of the outer square root's argument.) Comparing the force-extension curves for finite num- bers of cross-links to each other and to those for con- tinuous cross-linking, we find an approximate collapse of all curves with the same total inter-chain attraction ga = 2N L2/a2 c, as shown in Fig. 6. Except for the FIG. 6. ment at total cross-link strength ga = 500 and Lp/L = 10. (Color online) Cross-link-induced extension incre- case of a single cross-link, the force-extension relation is over a large range of forces to high numerical preci- sion determined by the product of cross-link number +1 and strength of a single cross-link only. On the basis of Eqs. (34) and (27), this apparent scaling can be traced back to the rapid decay of the coefficients c−1 m with mode index m. The sum over eigenvectors l in Eq. (27) for finite N is dominated by those with entries at the low- est modes, and of the set of modes represented by one eigenvector, only the lowest mode gives an appreciable contribution. Thereby, the "missing" modes m = ZN and, finally, the deviation from the series in Eq. (34) are negligible but for very small N . According to the slower decay ∝ m−2 of the stretching contribution to the coeffi- cients c−1 m , the approximate scaling must break down for strong stretching (and small persistence lengths). In this regime, however, the additional extension due to cross- links is a small quantity anyway. 0 0.0005 0.001 0.0015 0.002 0.0025 0.003 0.0035 0 10 20 30 40 50extension increm. AEDxae/Lforce fr = fL2/ktotal X-link strength 2N (L/ac)2 = 500, Lp = 10LX-links:1210contin.11 [1] C. Bustamante, J. F. Marko, E. D. Siggia, and S. Smith, 022801 (2011). Science 265, 1599 (1994). [2] X. Liu and G. H. Pollack, Biophys. J. 83, 2705 (2002). [3] M. L. Tanzer, Science 11, 561 (1973). [4] D. R. Eyre and J.-J. Wu, in Collagen, Top. Curr. Chem., Vol. 247, edited by J. Brinckmannn, H. Notbohm, and P. K. Muller (Springer, Berlin Heidelberg, 2005) pp. 207 -- 229. [5] J. S. Graham, A. N. Vomund, C. L. Phillips, and M. Grandbois, Exp. Cell Res. 299, 335 (2004). [6] J. A. J. van der Rijt, K. O. van der Werf, M. L. Bennink, P. J. Dijkstra, and J. Feijen, Macromol. Biosci. 6, 697 (2006). [7] L. Yang, K. O. van der Werf, B. F. Koopman, V. Sub- ramaniam, M. L. Bennink, P. J. Dijkstra, and J. Feijen, J. Biomed. Mater. Res. 82A, 160 (2007). [20] J. Kierfeld, O. Niamploy, V. Sa-yakanit, and R. Lipowsky, Eur. Phys. J. E 14, 17 (2004). [21] P. K. Purohit, M. E. Arsenault, Y. Goldman, and H. H. Bau, Int. Journal Non-Linear Mech. 43, 1056 (2008). [22] There are several possibilities to compute the force- extension relation of this model with quadratic-form Hamiltonian. In [16], for the case of a single cross-link, the matrix was inverted using the Sherman-Morrison for- mula, an approach which becomes increasingly involved for larger numbers of cross-links. [23] For a single cross-link, cos φ1 = 0, and the expressions in Eqs. (36) simplify considerably, cf. Eq. (9) in [16]. [24] F. Gittes, B. Mickey, J. Nettleton, and J. Howard, J. Cell. Biol. 120, 923 (1993). [25] A. Ott, M. Magnasco, A. Simon, and A. Libchaber, [8] P. Benetatos and A. Zippelius, Phys. Rev. Lett. 99, Phys. Rev. E 48, R1642 (1993). 198301 (2007). [26] Y.-L. Sun, Z.-P. Luo, A. Fertala, and K.-N. An, Biochem. [9] O. Lieleg, M. M. A. E. Claessens, and A. R. Bausch, Biophys. Res. Comm. 295, 382 (2002). Soft Matter 6, 218 (2010). [10] M. L. Gardel, J. H. Shin, F. C. MacKintosh, L. Mahade- van, P. Matsudaira, and D. A. Weitz, Science 304, 1301 (2004). [27] L. Bozec and M. Horton, Biophys. J. 88, 4223 (2005). [28] S. B. Smith, Y. Cui, and C. Bustamante, Science 271, 795 (1996). [29] C. Heussinger, M. Bathe, and E. Frey, Phys. Rev. Lett. [11] C. Bustamante, S. B. Smith, J. Liphardt, and D. Smith, 99, 048101 (2007). Curr. Opin. Struct. Biol. 10, 279 (2000). [30] O. Lieleg, M. M. A. E. Claessens, C. Heussinger, E. Frey, [12] J. F. Marko and E. D. Siggia, Macromolecules 28, 8759 and A. R. Bausch, Phys. Rev. Lett. 99, 088102 (2007). (1995). [31] A. Hanke, M. G. Ochoa, and R. Metzler, Phys. Rev. [13] O. Kratky and G. Porod, Recl. Trav. Chim. Pays-Bas 68, Lett. 100, 018106 (2008). 1106 (1949). [32] D. Marenduzzo, A. Maritan, E. Orlandini, F. Seno, and [14] N. Saito, K. Takahashi, and Y. Yunoki, J. Phys. Soc. A. Trovato, J. Stat. Mech. 2009, L04001 (2009). Jpn. 22, 219 (1967). [33] N. Theodorakopoulos, J. Nonlin. Math. Phys. 18, 419 [15] A. Y. Grosberg and A. R. Khokhlov, Statistical Physics of Macromolecules (AIP Series in Polymers and Complex Materials, New York, 1994). [16] P. Benetatos, S. Ulrich, and A. Zippelius, New J. Phys. 14, 115011 (2012). (2011). [34] N. Theodorakopoulos and M. Peyrard, Phys. Rev. Lett. 108, 078104 (2012). [35] O.-c. Lee and W. Sung, Phys. Rev. E 85, 021902 (2012). [36] J. Palmeri, M. Manghi, and N. Destainville, Phys. Rev. [17] C. Heussinger, F. Schuller, and E. Frey, Phys. Rev. E E 77, 011913 (2008). 81, 021904 (2010). [18] P. Benetatos and E. M. Terentjev, Phys. Rev. E 81, 031802 (2010). [19] P. Benetatos and E. M. Terentjev, Phys. Rev. E 84, [37] P. Gross, N. Laurens, L. B. Oddershede, U. Bockelmann, E. J. G. Petermann, and G. J. L. Wuite, Nature Phys. 7, 731 (2011).
1804.01173	2	1804	2018-06-27T13:28:29	Comment on "Minimum Action Path Theory Reveals the Details of Stochastic Transitions Out of Oscillatory States"	[ "physics.bio-ph", "cond-mat.stat-mech" ]	De la Cruz et al. [Phys. Rev. Lett. 120, 128102 (2018); arXiv:1705.08683] studied a noise-induced transition in an oscillating stochastic population undergoing birth- and death-type reactions. They applied the Freidlin-Wentzell WKB formalism to determine the most probable path to the noise-induced escape from a limit cycle predicted by deterministic theory, and to find the probability distribution of escape time. Here we raise a number of objections to their calculations.	physics.bio-ph	physics	Comment on "Minimum Action Path Theory Reveals the Details of Stochastic Transitions Out of Oscillatory States" Racah Institute of Physics, Hebrew University of Jerusalem, Jerusalem 91904, Israel Baruch Meerson∗ and Naftali R. Smith† 8 1 0 2 n u J 7 2 ] h p - o i b . s c i s y h p [ 2 v 3 7 1 1 0 . 4 0 8 1 : v i X r a In a recent Letter [1] de la Cruz et al. studied a noise- induced transition in an oscillating stochastic popula- tion undergoing birth- and death-type reactions. When described by deterministic rate equations, the popula- tion approaches a stable limit cycle. The intrinsic noise, caused by the discreteness of molecules and randomness of their interactions, leads to escape from this limit cycle through an adjacent unstable limit cycle, and de la Cruz et al. attempted to evaluate the mean first passage time (MFPT) to escape. A crucial approximation, made in the Letter, was to replace the original Master equation by the "chemical Langevin equation" (CLE), their Eq. (2). Unfortunately, this standard procedure, based on the van Kampen ex- pansion in the inverse population size 1/Ω ≪ 1 [2], ap- plies only for typical, small fluctuations around the stable limit cycle. It fails in the tails of the metastable quasi- stationary distribution of the population size around the limit cycle. One of these tails determines the escape rate of the population through the unstable limit cycle. As a result, the MFPT, predicted by de la Cruz et al, involves an error which grows exponentially with the population size Ω ≫ 1, due to an error in the calculation of S. In this situation their study of a pre-exponential factor in the MFPT does not have much meaning. The inadequacy of the van Kampen system-size expan- sions for a description of large fluctuations in Markov jump processes is by now well documented [3–9]. The only general exception appears when the system is close to the proper bifurcation of the underlying deterministic model [5, 9–12]. In the present case it is the saddle-node bifurcation of the stable and unstable limit cycles. Fortunately, there is no need for uncontrolled approx- imations. The Freidlin-Wentzell WKB theory was ex- tended to stochastic populations quite some time ago [13–15]. The corresponding WKB technique employs the same large parameter Ω ≫ 1 but circumvents the van Kampen system-size expansion, see e.g. Ref. [9] for a re- cent review. Moreover, this WKB technique was already applied to escape from a limit cycle, in the context of extinction of long-lived oscillating populations [16]. Even within the framework of the CLE, much of the Letter is devoted to a rediscovery of known results, as de la Cruz et al. seem to be unaware of a body of im- portant previous analytical, numerical and experimental work on noise-induced escape from limit cycles and from attractors of dynamical systems in general [17–22]. A proper formulation of the Freidlin-Wentzell escape opti- mization problem, which was put forward in these works, and which is lacking in the Letter, involves the time in- terval −∞ < t < ∞. A minimum action path – an instanton – exits the limit cycle at t = −∞ while per- forming an infinite number of loops. There is a whole one-parameter family of instanton solutions, linked to one another through the time translations t → t + const, and each instanton yields the same classical action. Any evidence to the contrary results from finite-time numeri- cal artifacts. We acknowledge discussions with M.I. Dykman and support from the Israel Science Foundation (Grant No. 807/16). ∗ [email protected] † [email protected] [1] R. de la Cruz, R. Perez-Carrasco, P. Guerrero, T. Alar- con, and K. M. Page, Phys. Rev. Lett. 120, 128102 (2018); arXiv:1705.08683. [2] N. van Kampen, Stochastic Processes in Physics and Chemistry (Elsevier, New York, 2007). [3] B. Gaveau, M. Moreau, and J. Toth, Lett. Math. Phys. 37, 285 (1996). [4] V. Elgart and A. Kamenev, Phys. Rev. E 70, 041106 (2004). [5] C. R. Doering, K.V. Sargsyan, and L. M. Sander, Multi- scale Model. Simul. 3, 283 (2005). [6] D. A. Kessler and N. Shnerb, J. Stat. Phys. 127, 861 (2007). [7] M. Assaf and B. Meerson, Phys. Rev E 75, 031122 (2007). [8] O. Ovaskainen and B. Meerson, Trends Ecol. Evol. 25, 643 (2010). [9] M. Assaf and B. Meerson, J. Phys. A: Math. Theor. 50, 263001 (2017). [10] B. Meerson and P.V. Sasorov, Phys. Rev. E, 78, 060103(R) (2008). [11] C. Escudero and A. Kamenev, Phys. Rev. E 79, 041149 (2009). [12] M. Assaf and B. Meerson, Phys. Rev E 81, 021116 (2010). [13] R. Kubo, K. Matsuo, and K. Kitahara K J. Stat. Phys. 9, 51 (1973). [14] H. Gang, Phys. Rev. A 36, 5782 (1987). [15] M. I. Dykman, E. Mori, J. Ross, and P. M. Hunt, J. Chem. Phys 100, 5735 (1994). [16] N.R. Smith and B. Meerson, Phys. Rev. E 93, 032109 (2016). [17] R. Kautz, Phys. Lett. A 125, 315 (1987). [18] R. Kautz, Phys. Rev. A 38, 2066 (1988). [19] P. Grassberger, J. Phys. A: Math. Gen. 22, 3283 (1989). [20] R.L. Kautz, Rep. Prog. Phys. 59, 935 (1996). [21] V. N. Smelyanskiy, M. I. Dykman, and R. S. Maier, Phys. Rev. E 55, 2369 (1997); Phys. Rev. E 56, 2332 (1997). [22] I. A. Khovanov, D. G. Luchinsky, R. Mannella, and P. V.E. McClintock, in "Stochastic Processes in Physics, Chemistry and Biology", edited by J. A. Freund and T. Poschel (Springer, Berlin, 2000), p. 378. 2
1302.1615	1	1302	2013-02-06T23:33:10	Combined evanescent-wave excitation and supercritical-angle fluorescence detection improves optical sectioning	[ "physics.bio-ph", "physics.optics" ]	Evanescent-wave microscopy achieves sub-diffraction axial sectioning by confining fluorescence excitation to a thin layer close to the cell/substrate interface. How thin this light sheet exactly is, however, is often unknown. Particularly in the popular objective-type total internal reflection fluorescence microscopy (TIRFM) configuration large deviations from the expected exponential intensity decay of the evanescent wave have been reported. Propagating, i.e., non-evanescent, excitation light diminishes the optical sectioning effect, reduces contrast and renders the quantification of TIRFM images uncertain. Here, we use a combination of azimuthal- and polar-angle beam scanning, dark-field scatter imaging, and atomic force microscopy to identify the sources of this unwanted background fluorescence excitation. We identify stray light originating from the microscope optics and the objective lens itself as the major sources of background, with minor contributions due to evanescent-wave scattering at the reflecting interface and at refractive-index boundaries in the sample. Apart from evanescence in excitation light, light emitted from a fluorophore can also show observable effects of evanescence. Only fluorophores located close to the coverslip can couple their near-field radiation into propagating waves detectable at supercritical angles. We show that selectively detecting this supercritical-angle fluorescence (SAF) through a high-numerical aperture objective effectively rejects fluorescence from deeper sample regions and improves optical sectioning. The microscopy scheme presented here merges the benefits of TIRF excitation and SAF detection and provides the conditions for quantitative wide-field imaging of fluorophore dynamics at or near the plasma membrane.	physics.bio-ph	physics	Combined evanescent-wave excitation and supercritical-angle fluorescence detection improves optical sectioning Maia Brunsteina,b,c, Maxime Teremetza,b,c,d,1, Christophe Touraina,b,c,e, Martin Oheima,b,c,2 aCNRS, UMR 8154, Paris, F-75006 France; bINSERM, U603, Paris, F-75006 France; cLaboratoire de Neurophysiologie et Nouvelles Microscopies, 45 rue des Saints Pères, Université Paris Descartes, PRES Sorbonne Paris Cité, Paris, F-75006 France. dMaster Programme: Biologie Cellulaire, Physiologie et Pathologies (BCPP), Université Paris Diderot, PRES Sorbonne Paris Cité, Paris, France. eService Commun de Microscopie (SCM), Institut Fédératif de Recherche en Neurosciences, 45 rue des Saints Pères, Paris, F-75006 France. [email protected] Evanescent-wave microscopy achieves sub-diffraction axial sectioning by confining fluorescence excitation to a thin layer close to the cell/substrate interface. How thin this light sheet exactly is, however, is often unknown. Particularly in the popular objective-type total internal reflection fluorescence microscopy (TIRFM) configuration large deviations from the expected exponential intensity decay of the evanescent wave have been reported. Propagating, i.e., non -evanescent, excitation light diminishes the optical sectioning effect, reduces contrast and renders the quantification of TIRFM images uncertain. Here, we use a combination of azimuthal - and polar-angle beam scanning, dark-field scatter imaging, and atomic force microscopy to identify the sources of this unwanted background fluorescence excitation. We identify stray light originating from the microscope optics and the objective lens itself as the major sources of background, with minor contributions due to evanescent-wave scattering at the reflecting interface and at refractive -index boundaries in the sample. Apart from evanescence in excitation light, light emitted from a fluorophore can also show observable effects of evanescence. Only fluorophores located close to the coverslip can couple their near-field radiation into propagating waves detectable at supercritical angles. We show that selectively detecting this supercritical -angle fluorescence (SAF) through a high-numerical aperture objective effectively rejects fluorescence from deeper sample regions and improves optical sectioning. The microscopy scheme presented here merges the benefits of TIRF excitation and SAF detection and provides the conditions for quantitative wide -field imaging of fluorophore dynamics at or near the plasma membrane. It is estimated that 30 to 40% of all cellular proteins reside in the non-aqueous environment of lipid membranes where they perform important metabolic and signaling functions and regulate the transfer of information and material in and out of the cell. Among the techniques study membrane dynamics and used to organization, total internal reflection fluorescence microscopy (TIRFM) occupies a central place. TIRFM is a wide-field technique that confines fluorescence to a thin layer defined by the intensity decay of the evanescent wave set up by the internal reflection of total excitation beam at the cell/substrate boundary. This confinement reduces fluorescence background and photobleaching and is the basis super-resolution single-molecule for and near-membrane imaging, fluorescence correlation spectroscopy, and membrane- selective photoactivation/ photobleaching assays (1-5) all of which rely on an axially well- defined probe volume. For TIR to occur, light must be directed to the interface at supercritical angles  > c = asin(n1/n2). Here, n1 and n2 are, respectively, the refractive indices of the sample and substrate at the excitation wavelength . In the popular prism-less ‘objective-type’ configuration (6) a laser beam is focused in the periphery of the back-focal plane (BFP) of a high-numerical aperture objective (NA > n2) and its radial displacement controls the beam angle . However, because of local variations of cell adhesion and refractive index (n1) the exact penetration depth  = /[4(n2²sin²- n1²)1/2] of the evanescent wave is often unknown and the interpretation of biological TIRFM images difficult (7-9). Most authors therefore report the calculated values of . Additional problems specific are objective-type TIRFM to interference fringes, uneven illumination and contrast degradation due to propagating non- evanescent light that excites fluorescence in deeper The (10-11). regions sample development of techniques that produce better TIRFM images has been the topic of active research in recent years. Azimuthal scanning 1 TIRFM (12-14) appears well suited to provide more evenly lit TIRFM images, but its actual impact on biological images has not been demonstrated. Here, we combine rapid acousto-optic polar- () and azimuthal-angle () beam scanning TIRFM (13), dark-field scatter imaging and atomic force microscopy (AFM) to identify sample, coverslip and instrument parameters that contribute to the loss of excitation confinement in objective-type TIRFM and devise strategies for improvement. We show that, if azimuthal beam spinning results in greater image homogeneity, it fails to better confine excitation. The reason is that most diffuse background excitation originates from the beam delivery optics and the objective itself (stray light and high-NA aberrations) rather than from the nanometric roughness of the reflecting interface or scattering at intracellular high-index organelles. In an attempt this fluorescence to reject background, we combined TIRFM with the selective detection of the directional emission of supercritical angle fluorescence (SAF) (15, 16). Some of the near-field light emitted from near- interface fluorophores converts into light propagating at supercritical angles in a nearby glass substrate. That hollow cone can be captured by the high-NA objectives used in TIRFM, as NA ≥ n2. Importantly, none of the far-field radiation is cast into such high angles so that SAF detection suppresses signal from fluorophores located in deeper cytosolic regions. By combining azimuthal beam-scanning EW excitation with SAF detection, we obtain evenly lit images of cultured cortical astrocytes with an improved optical sectioning compared to standard TIRFM images. Results Beam spinning abolishes image non- homogeneity but does not affect contrast cellular Optically dense structures, like chromaffin granules (17), or protein-rich adhesion sites spread light by scattering. Evanescent-wave scattering produces a flare of light in the sense of evanescent-wave propagation (17, 18). ‘Negative staining’ (19) images of an unlabeled BON cell bathed in a fluorescein-containing extracellular solution displayed irregular intensity bands co-linear with evanescent-wave of direction the propagation, Fig. 1A. Changing the azimuthal angle  rotated the propagation direction and also changed the stripe orientation. Restoring the illumination symmetry by scanning the spot on a circular orbit with kHz frequency during image acquisition (13) resulted in a more homogenously lit field of view (center image of Fig. 1A). Experimental details are given in Fig. S1 of the Supporting Information (SI). Uneven illumination adversely affects image quality and alters the conclusions drawn from the images. To illustrate this effect, we observed the same cultured cortical astrocytes labeled with FM2-10, a lysosomal marker in these cells (20), with conventional eccentric-spot and spinning TIRFM (spTIRFM), Fig. 2A. While Weber (CW) or Michelson contrast (CM) were unaffected (CW = 13.6 ± 3.4 for unidirectional vs. 12.3 ± 3.6 for spTIRF; CM = 0.985 ± 0.006 vs. 0.988 ± 0.008, n = 9 cells, n.s.), Fig.2B, more organelles were detected with spTIRFM than with conventional TIRF illumination (66 ± 7 vs. 40 ± 16 spots; 0.018 ± 0.002 µm-² vs. 0.010 ± 0.004 µm-²; n = 9 cells, p < 0.01). Individual spots were brighter (4522 ± 1910 cts vs. 2871 ± 2750 cts, mean ± SD, n = 21, p = 0.01) and less variable with spTIRFM than with unidirectional TIRF (42% coefficient of variation, CV, vs. 95%), for which lysosomes localized on or alongside the bright excitation bands showed markedly distinct fluorescence, Fig. 2C. With conventional eccentric-spot illumination, such intensity differences would erroneously be interpreted as organelles being located at different axial distances, having unequal dye content or different refractive index. In addition to making intensity measurements more reliable, beam spinning abolished aberrant directional features like a detection bias for mitochondria the parallel with aligned evanescent-wave propagation direction, Fig. S2. Denser labeling aggravated the deleterious effects of unidirectional TIRF excitation but independent of the vesicular, mitochondrial, fluorophore plasma-membrane or cytosolic targeting, spTIRFM produced more homogenous images, Fig. S3. Restoring the illumination symmetry produces evenly lit TIRFM images, improves the visibility of 2 the facilitates fluorescent organelles and accurate quantification of biological processes at or near the plasma membrane. A B BF *** Fig. 1. Azimuthal beam-scanning produces evenly lit TIRFM images. (A), unlabeled BON cell in dye containing extracellular solution. Cardinal four images show standard unidirectional TIRFM images, symbols show position of the focused spot in the objective BFP. Note excitation patterns co-linear with evanescent-wave propagation direction. Rapid azimuthal beam spinning (spTIRFM, center image) evens out the excitation pattern. Polar beam angle was 68°. Scale bar, 10 µm. (B), evolution of intensity along a 2-µm wide circular region (see bright-field image, BF) for different evanescent-wave propagation directions and spTIRFM. Thin lines are individual measurements, solid trace ensemble average over n = 4 cells. Inset shows reduction of the coefficient of variation (CV) of the measured intensity upon spTIRF (0.12 ± 0.05), red, compared to unidirectional TIRFM (black, 0.39 ± 0.15, 0.29 ± 0.05, 0.27 ± 0.03, 0.28 ± 0.14 for NWSE cardinal images, 0.31 ± 0.11 mean ± SD over all directional TIRF images, p < 0.001). Beam spinning does not alter contrast Compared to the four cardinal images obtained with eccentric-spot illumination, spTIRF images appeared hazier, Fig. 2A. We calculated Weber (CW) and Michelson contrast (CM), two sensitive reporters of image noise and background, respectively. Despite the nominally identical beam angle, CW and CM varied among EW propagation directions which was not due to image beam misalignment or polarization effects but resulted from the coverslip surface not being perfectly perpendicular to the optical axis. Due to the steep angular dependence of the EW intensity on the polar beam angle , even slight (<0.5°) intensity large to translates tilt differences for different azimuthal angles . Mounting the sample on a tip-tilt stage allowed us a perfect alignment. Averaged over n = 9 cells, spTIRF did not measurably degrade contrast compared to the average of the four eccentric-spot excitation images but rather eliminated contrast outliers, Fig. 2B, facilitating a cell-to-cell comparison in population studies typical for biological imaging. A B Fig. 2. Uneven illumination affects interpretation of TIRFM images. (A) Unidirectional and spTIRFM images of a FM2-10-labeled mouse cortical astrocyte in culture. Beam angle was 73°. Symbols indicate position of the focused spot in the objective BFP. Note the flare in direction of evanescent-wave propagation absent on the azimuthal beam-spinning image. Contrast inverted for display. Scale bar, 20 µm. (B) Left, detail from a spTIRFM (top) and unidirectional image (middle) and pseudo-color overlay (bottom). Note the intense beam of excitation light propagating across the image (arrowheads).  = 70°, scale bar, 5 µm. Right, single- lysosomes intensity profiles upon spTIRFM (red) and unidirectional (green) illumination. Bottom, cross- sectional intensity profile along the dashed line in panel B. 3 1.00.50.0F/Fmax1.00.50.0F/Fmax-101x (µm)1.00.5F/Fmax86420distance (µm)012120459013518022527031501212045901351802252703150121204590135180225270315012120459013518022527031501212045901351802252703150.40.20.0CV Evanescent-wave scattering by the sample plays a minor role If the observed non-uniformities were primarily sample-induced, shallower penetration depths should result in less scattering (17). Hence, the fluorescence Fs’ measured in a circular band around an unlabeled BON cell upon spinning excitation (to average out local effects) should decrease with larger beam angles. While Fs’ decreased, Fig. 3A, the normalized intensity [Fs’()]norm, corrected for the angle-dependence of the evanescent-field intensity I() itself, was unchanged, Fig. 3B, suggesting that a constant offset rather than scattering at intracellular high- index organelles was the major source of propagated excitation light. To better separate evanescent and propagating excitation components, we directly quantified scattering by dark-field imaging. A second 60/NA1.1 water-immersion objective was positioned above the reflecting interface to collect propagating excitation light. No light should enter this objective in the absence of scattering because of total internal reflection. However, even with bare coverslips the mean scattered intensity Is’ was non-zero. Dark-field images showed characteristic ‘fingerprints’ for each objective and beam scanning evened out these patterns without changing the mean intensity, Fig. 3C. Addition of polystyrene latex beads (Table S3) changed Is’ less than ~10%, although we varied the scattering coefficient µs’ over five orders of magnitude, Fig. 3D. Is’was slightly larger with bigger beads and at greater penetration depths but overall, we obtained a similar offset for different bead diameters (0.7, 2.8 and 90 µm), at different beam angles ( = 64, 70 and 78°), with unidirectional or spinning excitation and for a monolayer of scattering beads on the coverslip surface instead of bead suspensions, Fig. S4. The only slight dependence of non-evanescent excitation light intensities on either the probe volume or scattering coefficient µs’ confirm that other sources of background dominate over scattering due to sample irregularities. B A BF spTIRF 11 5 87 73 65 nm D d = 0.7 C D TL obj2 obj 1 2.8 90µm Fig. 3. Sample-induced scattering is not a major source of background. (A), left, bright-field (BF, top) and spinning TIRF (spTIRF, bottom) negative-staining fluorescence images of an unlabeled BON cell for different penetration depths. Images have same grey scale. The footprint region where the cell adheres is blocked out for better clarity. Focus is at the near the glass coverslip. Scale bar, 10 µm. (B), top, dependence on beam angle  of the calculated evanescent-wave penetration depth δ (black dots) and measured near-surface evanescent-wave intensity I0 (red circles). Bottom, scattered intensity measured in a ring around the cell (open symbols) and normalized with the excitation intensity measured (filled symbols), as a function of . Symbols and error bars show mean ± SD from 5 cells. Objective was x60/1.49NA. (C), optical scheme for dark-field imaging. TIRF is set up as before (obj1 – x60/1.45NA or 1.49NA) and scattered light imaged with a second water-immersion objective (obj2 – x60/1.1w, TL – tube lens, D - detector). Dark-field images were obtained a bare BK-7 coverslip upon unidirectional and spinning excitation. (D) Scattered intensity Is’ = Is/I0() (corrected for the angle-dependent EW intensity I0()) vs. reduced scattering coefficient µs’ for three different beam angles  (blue - 60°, red - 70° and black - 78°) and three different diameters of scattering beads (from left to right: 0.7, 2.8 and 90 µm) upon 488-nm spinning excitation. Light lines are individual measurements, dark lines, symbols and error bars show means ± SD from triplicate experiments. Inner filtering due to multiple scattering results in a drop of Is’ at high µs’. 4 600400200010-12 10-8 10-4 10010080 (nm)7268 (°)1.00.5I0() (AU)1.00.5AU7268 (°)600400200010-12 10-8 10-4 1006004002000Is' (counts/px)10-12 10-8 10-4 100µs' (mm-1)spTIRF Coverslips are rough on the length-scale probed by TIRF TIRF produces an excitation maximum at the reflecting interface and therefore is very sensitive to surface irregularities. Perhaps these are more important than volume scattering. Direct evidence for the relevance of surface scattering comes from the observation of non- specular in atoms slow reflections of evanescent-wave mirrors that were abolished by flame-polishing the substrate (21). AFM images revealed scratches, dimples, and holes on some borosilicate coverslips but no obvious defects on others; quartz had a ‘rolling hill’ aspect, Fig. 4A. Rejecting coverslips with large irregularities, the remainder had sub- nanometric RMS roughness (Rq), including after coating with polyornithine or collagen, Fig. 4B. However, Rq increased to ~2 nm when wetting coverslips, with peak roughness Rp of tens of nanometers, and peak-to-peak heights up to 60 nm, Table 1. Thus, under biological recording conditions, cell adhesion molecules produce height features comparable with the penetration depth of the evanescent wave. Table 1: Coverslip surface roughness a quartz Rq (nm) a Rq (nm) a collagen, dry (n = 5) 0.44 ± 0.27 BK-7 bare, dry bare, dry (n = 4) (n = 8) 0.37 ± 0.05 0.24 ± 0.04 collagen, dry polyornithine, dry polyornithine, dry (n = 4) (n = 6) (n = 5) 0.69 ± 0.06 0.65 ± 0.31 0.44 ± 0.12 collagen, wet polyornithine, wet polyornithine, wet (n = 5) (n = 5) (n = 10) Rq (nm) a 2.6 ± 0.3 2.8 ± 1.2 5.4 ± 2.2 Rq (nm) b 2.0 ± 0.4 2.8 ± 1.2 1.4 ± 0.5 Ra (nm) b 1.1 ± 0.2 1.6 ± 0.8 1.0 ± 0.3 Rt (nm) b 49.8 ± 9.1 27.8 ± 14.6 24.1 ± 8.4 pk-pk height (nm) b 16 (min); 60 (max) 13 (min); 48 (max) 41 (min); 62 (max) RMS roughness over 5µm × 5µm of non-selected commercial coverslips. b RMS and absolute roughness as well as peak height, calculated over 2µm × 2µm regions of interest from selected coverslips devoid of large surface defects. Glare and aberrations of peripheral beams limit excitation confinement Variable-angle or beam-spinning TIRFM uses ‘critical illumination’ in which the surface of a rotating wedge (12), a scan mirror (14) or a pair of AODs (2, 13) is imaged into the sample plane, Fig. 5A. As a consequence, any irregularities, dust or scratches on the scanning device or a conjugate plane are imaged into the sample plane and produce stray light. Microscope- induced glare was measured in dark-field by moving the upper objective: a defocus by dz displaces the conjugate image plane located inside the microsope by dz multiplied by the longitudinal magnification (i.e., lateral the magnification squared). On the plot of Is’ vs. dz the coverslip and AOD surfaces are easily recognized as peaks of the dark-field signal, Fig. 5B (black trace). Placing an appropriately-sized disk in a conjugate aperture plane of the excitation path blocked most of this stray excitation (red trace). A second source of non-evanescent excitation light are stray reflections from inside the objective. TIRFM uses its extreme periphery for guiding the excitation beam at supercritical angles to the reflecting interface. To assess beam quality beyond beam angles otherwise obscured by TIR, we imaged the beam with an oil-coupled high-index solid immersion lens onto a wave front analyzer and calculated the beam parameter product (M²) and Strehl ratio as a function of , Fig. 5C Two objectives from different manufacturers showed stable beam 5 quality up to  ~ 50° that then deteriorated abruptly, phase off-axis that indicating aberrations or partial obstruction degrade objective performance. These effects could be due to a NA smaller than specified. Early TIRF microscopists remember that some lenses sold as 1.4-NA actually had an effective NA (NAeff) to 1.38. Measuring NAeff using a closer technique angle supercritical on based fluorescence (SAF) detection (22), Fig. S5, we found that the1.45–1.46 objectives generally fulfilled their specification, but a nominal 1.49- NA lens only had NAeff = 1.47, table 2. A BK-7 bare quartz A obj FP EBFP AODs B 255 i ii iii 20 1 µm 0 nm B BK-7 + collagen BK-7 + polyornithine 0 AU C  = 0° 64° dry wet 0.5 nm Fig. 4. Coverslips display height variations relevant on the length-scale probed by TIRFM. (A), examples of contact-mode atomic-force microscopy (AFM) images showing the surface roughness of bare BK-7 (left) and quartz coverslips (right). Vertical tip displacement (i.e., surface height) is pseudo-color coded from 0 to 20 nm. The RMS roughness Rq was 0.43 and 0.58 nm, respectively, for the images shown. Insets show scratches and surface defects of coverslips that were rejected. Curves show surface profiles along the dotted line as shown on the left image. (B) Same for BK-7 after collagen (left) and polyornithine treatment (right), seen as a fibrous deposit. Rq were 0.49 and 0.38 for the images shown respectively. Inset, tapping-mode AFM image illustrating the marked morphology change of collagen upon hydration. Note that height scale is 0 to 200 nm for inset. See table 1 for quantifications. Fig. 5. Glare and off-axis aberrations limit excitation confinement. (A), dark-field detection of microscope- induced stray light Is’. Scanning (white arrow) the focal plane (FP, solid line) of the upper objective (obj) moves its conjugate focal plane (dashed) across the excitation optical path (grey filled arrow), allowing to identify the sources of stray light. (B), top, Peaks of Is’ were detected (black trace) at the coverslip surface (i), at consecutive AOD surfaces (see ii as an example), and - to a lesser degree - within the AOD crystal, (iii), and are reduced when placing an opaque disk [red on (A)] in the center of an aperture plane (equivalent back-focal plane of the objective, EBFP), red trace. Bottom, representative dark - field images taken at planes indicated. Dashed line identifies region of intensity measurements. Scale bar, 25 µm. (C), left, example cross-sectional intensity images (top) taken at a beam angle  of 0° and 64°, with M² and Strehl ratio values of 2.5 and 0.8 (left) and 3 and 0.6 (right), respectively. Bottom, evolution of M² (red) and Strehl ratio (black) with , for the x60/NA1.45 (solid symbols) and x100/NA1.46-objective (circles). Symbols and error bars represent mean ± SD from triplicate measurements. Note the degradation of beam quality starting well before the critical angle c for the BK- 7/water interface (dashed, blue). 6 15010050Is' (AU) 151050-5defocus (µm)1.00.50.0z (nm)420x (µm)1.00.50.0z (nm)420x (µm)210z (nm)420x (µm)10z (nm)420x (µm) Table 2:Effective NAs of TIRFM objectives 1.457 ± 0.007 NAeff 1.468 ± 0.004 a 1.456 ± 0.003 1.453 ± 0.004 PlanApo ×60/NA1.45 (objective 1) PlanApo ×60/NA1.45 (objective2) Plan-Apochromat ×100/NA1.46(objective 1) Plan-Apochromat ×100/NA1.46(objective 2) ApoN ×60/NA1.49 1.474 ± 0.007 a Measurements (ref. (22)) were robust against re-aligning the Bertrand lens, re-focusing or moving laterally in the fluorophore layer. Absolute variations are given as uncertainties. Taken together, our experiments identify the objective as a non-negligible contributor to non- evanescent excitation light in prismless TIRFM. While permitting near-membrane imaging on a slightly modified epifluorescence microscope, through-the-objective TIRFM inevitably comes the price of a degraded excitation at confinement. Objective-induced glare is a consequence of the extreme off-axis use of the objective and, as such, it is the same for unidirectional vs. spTIRF. Thus, other strategies than beam spinning must be devised to abolish the long-range excitation component introduced by the microscope objective. Supercritical angle fluorescence (SAF) detection rejects nonevanescent background Fluorophores closer than a light wavelength from the coverslip change their radiation pattern because components emission evanescent couple to the surface and become propagative. These waves are directed exclusively into NAs > n2, i.e., angles beyond the critical angle, which is the same as it would be for TIR excitation at the same wavelength. None of the fluorescence originating from deeper within the sample can be emitted in this hollow cone. SAF microscopy discriminates between surface-proximal and distant fluorophores by selectively collecting this high-NA information and leads to a similar axial confinement as EW excitation (23, 24). Undercritical angle fluorescence (UAF) is typically rejected by obstruction of the central part of the objective pupil. This, together with high-NA degrades effects polarization resolution (15, 16). An elegant solution is the subtraction of an image acquired at a slightly lower NA (collecting only UAF, Fig. 6A), from the image taken at full aperture (collecting UAF + SAF) that creates a ‘virtual’ SAF image IvSAF = IUAF+SAF - IUAF (25), Fig. 6B. Because high NAs are used for the acquisition of both images, resolution is preserved, Fig.6C and Table S1. We then compared spTIRFM images of cortical astrocytes transfected with vinculin-GFP, a membrane-cytoskeletal protein involved in the linkage of integrin to actin. We adjusted the beam angle to 74°, corresponding to a calculated  of 72 nm. Both UAF and UAF + taken with a ×100/1.46-NA images SAF objective showed the characteristic focal adhesion sites on the substrate (Fig. 6C) but the vSAF image was devoid about half of the vinculin-bearing vesicles and tubules seen on the conventional TIRFM image, indicating that these were located close enough to the reflecting interface to be reached by excitation light, but too far from the reflecting interface to emit SAF. Compared classical full-NA collection, SAF detection captures about one third of the fluorescence. Nevertheless, the suppression of background fluorescence outweighs the signal loss, because the SAF image contains information otherwise obscured by background. An example is shown in Fig. 6D, where SAF reveals an alternate sequence of membrane- proximal and distant stretches rather than a single flat adhesion site as seen in TIRFM, Fig. 6D. We conclude that the combined use of evanescence in excitation and emission presents clear advantages for imaging fluorophores near or at the basal plasma membrane compared to TIRFM alone. vSAF detection has the extra benefit of abolishing the background resulting from non-evanescent excitation that is notorious with objective-type TIRFM. Discussion The use of supercritical-angle fluorescence excitation and emission collection both date back to the 1960-70’s. Their combination in confocal-spot TIRFM with a parabolic mirror (26), or the combination of prism-based excitation and SAF-detection through a high- NA objective (27) produces important sensitivity gains in single-molecule detection. We now show that simultaneous supercritical 7 excitation and emission collection overcomes the major inconvenience of objective-type TIRFM and allows more reliable fluorophore localization. A C BFP EBFP (iris) 0.5 µm TIRF TIR-vSAF obj L2 L3 TL LBL D B UAF + SAF UAF vSAF - = D UAF / vSAF overlay Fig. 6. Virtual super-critical angle fluorescence (vSAF) detection improves axial confinement in objective-type TIRFM. (A) Schematic layout of the emission optical path. Sample fluorescence (green) is imaged via the objective (obj) and tube lens (TL) onto the detector (D). Lenses L2 and L3 create an intermediate aperture plane (equivalent back focal plane, EBFP) permitting the selection of emission angles (Fourier-plane filtering) by an adjustable iris. A removable Bertrand lens (LBL) permits imaging the EBFP for alignment. (B), spTIRFM images of 93 -nm red-fluorescent microspheres (inset shows single bead) and intensity profile upon conventional full-aperture detection (TIRF) and after vSAF detection, i.e., subtraction of an image taken with the iris adjusted to collect only undercritical angle fluorescence (UAF). Lateral resolution is identical (see Table S1) because high NAs are used for the acquisition of both images. (C), field- and aperture-plane (insets) images of a cultured cortical astrocyte expressing vinculin-EGFP. Autoscaled images show, from left to right, spTIRFM image acquired at full objective NA (i.e., collecting both UAF and SAF), the UAF image with the iris partially stopped down, and the resulting ‘virtual’ SAF (difference) image, containing only SAF. (D), green/red pseudocolor overlay of the UAF and vSAF images as in (C). Inset shows boxed region and line profile along dotted line, showing fine detail on the vSAF image. Conventional eccentric-spot TIRF produces non-homogeneities and directionality that can be avoided by beam spinning (Figs 1, 2, S2 and S3), but this does not fundamentally address the problem of spurious far-field excitation (11). The reason the diffuse that most of is background in prismless TIRFM stems from the objective and beam-delivery optics and is independent of the evanescent-wave penetration depth (Fig. 3) or propagation direction (Fig. S4, 5). A minor contribution is due to scattering at refractive-index heterogeneities in the sample (Fig.3) and the reflecting irregularities of interface (Fig.4). By selecting for surface- proximal fluorophores SAF detection suppresses this background, improving optical sectioning and image contrast. Concurrent spTIR-vSAF imaging outperforms conventional TIRF for near-membrane fluorophore detection (Fig.6). Compared trans-prism alternative the to geometry, objective-type TIRFM (6) produces brighter but less crisp images (10, 23, 28). Brightness is a result of high-NA SAF collection (23, 29). Reduced contrast has been attributed to stray reflections (4, 11, 13), interference (10, 30-32), or propagating shafts of light emerging from the margin of the objective (31). Moerner and colleagues estimated that as little as 26% of the initial power is returned from the objective in the ‘totally’ suggesting (33), reflected beam important losses along the excitation path (34). Identifying their origin has been difficult because of the restricted access inside the microscope. We now identify glare and partial obstruction of the beam at high NAs as the major sources of non-evanescent excitation. Two strategies have been used to improve confinement. excitation Two-photon fluorescence excitation in TIRFM reduces background because scattered photons are too dilute to result in appreciable fluorescence generation. Although effective, the technique has been little used (20, 36). Alternatively, combining one-photon excitation with a continuous change of the direction of EW propagation (11-13, 32) results in a time- averaging over non-homogeneities. Both approaches are analogous to those used in selective-plane illumination microscopy (SPIM) (37, 38). 8 -1,0-0,50,00,51,00,00,51,0 Normalized IntensityDistance (µm)80604020cts3210µm We show that spTIRFM is superior to classical TIRFM for quantitative near-membrane imaging, but it is no cure-all against non-evanescent excitation light. Because the scanning device is imaged into the sample plane, any imperfection appears the field-of-view. Fourier-plane in filtering reduces background, but does not affect stray light generated in the excitation path downstream of the aperture mask, arising from the passage through the periphery and clipping of the excitation beam inside the microscope objective. We show that this stray light is efficiently rejected by SAF detection. Again, an analogy can be drawn with SPIM for which a combination of scanning light-sheet illumination and confocal slit-scanning detection produces crisper images (38). Our study underpins the importance of testing high-NA objectives individually prior to order. Not all objectives met their specifications suggesting that the commercial race to higher NAs has led to somewhat overly optimistic statements of performance. Also, all objectives displayed measurable autofluorescence that interfered with the detection of faint green and yellow fluorescence (Fig. S5), which is the reason why we used red-emitting beads for PSF measurements. How does TIR-vSAF compare to alternative approaches that increase the signal-to- background ratio? With a NA-1.46 objective, vSAF collects about one third of the emission of a fluorophore located within 100 nm from the interface, compared to conventional full- aperture detection. The calculation of the vSAF image requires N+1 planes and thus produces lower phototoxicity/bleaching compared to TIRF deconvolution, as soon as three or more z- planes are acquired. Similar to surface-plasmon enhanced fluorescence (39, 40), spTIR-vSAF achieves optical sectioning through its distance- dependent efficiency, however, collection neither does it rely on special silver-coated coverslips, nor does it suffer from metal- induced quenching of near-surface fluorophores. vSAF detection requires only a minor modification of the collection optical path. It can be implemented on any standard microscope where it will improve single-molecule detection, TIR-FCS or TIRF-FRAP measurements, and the growing number of super-resolution techniques including standing-evanescent-wave structured illumination, TIRF-STED and localization- based (PALM/STORM) microscopies. The association of TIRFM and vSAF is of particular interest with the recent 100 and 150 objectives that, as a consequence of their small back pupils, impose a very precise control and tight focusing of the excitation beam in the BFP. Materials and Methods Cell culture. Astrocytes and BON cells were cultured as described in the SI. FM2-10 and FM4-64 labeled astrocyte lysosomes (18). Cells were transfected with plasmids encoding VAMP2-EGFP, mito-EGFP, Lck- EGFP, vinculin-GFP or CD63-GFP. Experiments were performed at 22-23°C. spTIRF microscopy. A 488-nm laser spot was scanned in the objective BFP (13) under LABVIEW control azimuthal angle ( ) of the totally reflected beam, (Fig. S1). Objectives (Table 2) were piezo-mounted for accurate focusing (PIFOC, Physik Instrumente). Fluorescence was detected through filters listed in Table S4 on electron- multiplying charge-coupled device cameras (EMCCD, Photometrics). Total magnifications were 80, 103 and 120 nm/pixel, as indicated. We used low µW powers in the sample plane. SAF detection and effective-NA measurements. For SAF, an iris was positioned in an EBFP of the objective while the sample was imaged on the EMCCD. An equally removable Bertrand lens permitted EBFP imaging. Both the aperture mask and lens could be centered with precision translation tables. NAeff was measured using a FITC (500 µM) spin-coated coverslip (Fig. S6). A MATLAB (Mathworks) routine allowed determining rc and rNA from the circularly averaged steepest intensity change. rNA and rc = f · sin( c) are, respectively, the measured intensity cut-offs due to the upper limit of the objective effective NA collection angle and the radius for which TIR at the air/substrate interface occurs. NAeff = rNA/rc. Darkfield scatter imaging. Scattering coefficients of bead suspensions (Polysciences, Table S3) were calculated using the interactive Mie Scattering Calculator, http://omlc.ogi.edu/calc/mie_calc.html written by Scott Prahl. Scattered excitation light was detected through a dipping objective (LUMFl 60/NA1.1, Olympus). AFM. Images were obtained on a Bioscope (Veeco) with roughness was IIIa controller. Surface Nanoscope Rq evaluated by the calculation of = and Rt = , Ra = . M is the number of 9 MjNijiyxzMN112),(1MjNijiyxzMN11),(1),(minmaxmaxyxzz points per scan line and N is the number of lines, z(x,y) is the vertical tip displacement at point (x,y). Tapping-mode AFM was used for wet coverslips. Image analysis. Dark images were subtracted from all fluorescence images. Weber contrast was calculated as CW=(I–Ib)/Ib ,where I and Ib are the fluorescence intensity of image features and background, respectively. Ib was measured in a large cell-free polygonal ROI identified on the bright-field/fluorescence images. Signal intensities I were measured in ROIs outlined by a border exceeding two times the SD of Ib. Michelson contrast (visibility) was calculated as contrast as , RMS , where Iij is the intensity of pixel (i, j) of a M × N pixel image. is the average fluorescence of all pixels. Line-profiles across single lysosomes measured on spT IRF and unidirectional TIRF images were compared by fitting the central intensity peak with a Gaussian distribution and noting peak and local background fluorescence. Image analysis was performed with MetaMorph (Molecular Devices) and IGOR (Wavemetrics) using custom macros. Acknowledgment. We thank K Hérault and C Debaecker for cell culture, D Li, M van ’t Hoff and C Ventalon for help with experiments/programming, P Jegouzo for custom mechanics, F Jean for administrative support and V Emiliani, O Faklaris (IJM Paris), S Konzack (Olympus), G Louis, H Schrader (Zeiss) for the loan of equipment. We thank T Barroca, F Lison, C Pouzat and R Uhl for discussion, RH Chow, D Li, N Ropert and C Ventalon for comments on earlier versions of the manuscript and JS Kehoe for careful proofreading. Financed by the EU (FP6-STRP-n°037897-AUTOSCREEN, FP7-ERA-NET la the Agence Nationale de n°006-03-NANOSYN), Recherche (ANR P3N 09-044-02 nanoFRET²), the FranceBioImaging (FBI) initiative and mobility support from the Franco-Bavarian University Cooperation Centre (BFHZ-CCUFB). AFM imaging was performed on the Paris Descartes St Pères core imaging facility, Service Commun de Microscopie (SCM). REFERENCES 1. Chung E, Kim D, So PT (2006) Extended resolution wide-field optical imaging: objective- launched standing-wave total internal reflection fluorescence microscopy. Opt Lett 31(7): 945-7 2. Gilko O, Reddy GD, Anvari B, Brownell WE, Saggau P. (2006) Standing wave total internal reflection fluorescence microscopy to measure the size of nanostructures in living cells. J Biomed Opt 11: 064013 3. Kner P, Chun BB, Griffis ER, Winoto L, Gustafsson MG. (2009) Super-resolution video microscopy of live cells by structured illumination. Nat Methods 6(5): 339-42 4. Gould TJ, Myers JR, Bewersdorf J. (2011) Total internal reflection STED microscopy. Opt Express 19(14): 13351-7 5. Leutenegger M, Ringemann C, Lasser T, Hell SW, Eggeling C (2012) Fluorescence correlation spectroscopy with a total internal reflection fluorescence STED microscope (TIRF-STED- FCS). Opt Express 20(5): 5243-63 6. Stout AL Axelrod D (1989) Evanescent field excitation of fluorescence by epi-illumination microscopy. Appl Opt 28: 5237–42 7. Simon SM. (2009) Partial internal reflections on total internal reflection fluorescent microscopy. Trends Cell Biol 19: 661-8 8. Toomre D, Bewersdorf J (2010) A new wave of cellular imaging. Annu Rev Cell Dev Biol 26: 285-314 9. Schwarz JP, König I, Anderson KI (2011) Characterizing system performance in total internal fluorescence microscopy. reflection Meth Mol Biol 769: 373-86 10. Conibear PB, Bagshaw CR. (2000) A comparison of optical geometries for combined flash photolysis and total internal reflection fluorescence microscopy. J Microsc 200(3): 218- 29 11. Mattheyses A, Axelrod D. (2006) Direct measurement of the evanescent field profile produced by objective-based total internal reflection fluorescence. J Biomed Opt 11: 014006A 12. Mattheyses AL, Shaw K, Axelrod D (2006) Effective elimination of interference laser fringing in fluorescence microscopy by spinning azimuthal incidence angle. Microsc Res Tech 69(8): 642-7 13. van 't Hoff M, de Sars V, Oheim M. A programmable light engine for quantitative single molecule TIRF and HILO imaging. Opt Express, 16(22): 18495-504 14. Fiolka R, Belyaev Y, Ewers H, Stemmer A (2007) Even illumination in total internal reflection fluorescence microscopy using laser light. Microsc Res Tech 71: 45–50 15. Enderlein J, Gregor I, Ruckstuhl T (2011) Imaging properties of angle supercritical fluorescence optics. Opt Express 19(9): 8011-8 16. Barroca T, Balaa K, Delahaye J, Lévêque-Fort S, Fort E (2011) Full-field supercritical angle fluorescence microscopy for live cell imaging. Opt Letters 36(16): 3051-3 17. Rohrbach A. secretory (2000) Observing granules with a multiangle evanescent wave microscope. Biophys J 78(5): 2641-54. 18. Schapper F, Gonçalves JT, Oheim M (2003) Fluorescence imaging with two-photon evanescent wave excitation. Eur Biophys J 32(7): 635-43 19. Gingell D, Todd I, Bailey J (1985) Topography of cell-glass apposition revealed by total internal 10 minmaxminmaxIIIICMMjNiijRMSIIMNC0021I reflection fluorescence of volume markers. J Cell Biol 100(4): 1334-8 20. Li D, Hérault K, Oheim M, Ropert N. FM dyes enter via a store-operated calcium channel and modify calcium signaling of cultured astrocytes. Proc Natl Acad Sci USA 106(51): 21960-5 21. Landragin A et al. (1996) Specular versus diffuse reflection of atoms from an evanescent-wave mirror. Opt Lett 21(19): 1591-1593 22. Dai L, Gregor I, von der Hocht I, Ruckstuhl T, Enderlein J (2005) Measuring large numerical apertures by imaging the angular distribution of radiation of fluorescing molecules. Opt Express 13(23): 9409-14 23. Axelrod D (2001) Selective imaging of surface fluorescence with very high aperture microscope objectives. J Biomed Opt 6(1): 6-13 24. Ruckstuhl T, Verdes D (2004) Supercritical angle fluorescence (SAF) microscopy. Opt Express 12(18): 4246-54 25. Barroca T, Balaa K, Lévêque-Fort S, Fort E (2012) Full-field near-field optical microscope for cell imaging. Phys Rev Lett 108(21): 218101 26. Välimäki H, Kappura K (2009) A novel platform fluorescent surface-sensitive for highly measurements applying simultaneous total internal reflection excitation and super critical angle detection. Chem Phys Lett 473(4–6): 358- 62 27. Burghardt TP, Hipp AD, Ajtai K (2009) Around- the-objective total internal reflection fluorescence microscopy. Appl Opt 48(32): 6120-31 28. Ambrose WP, Goodwin PM, Nolan JP (1999) Single-molecule detection with total internal reflection excitation: comparing signal-to- background and in different total signals geometries. Cytometry 36: 224-31 29. Enderlein J, Böhmer M. (2003) Influence of interface-dipole interactions on the efficiency of fluorescence light collection near surfaces. Opt Lett 28(11): 941-3 30. Thompson NL, Steele BL. (2007) Total internal reflection with fluorescence correlation spectroscopy. Nat Protoc 2(4): 878-90 31. Axelrod reflection internal (2007) Total fluorescence microscopy. In “Optical Imaging and Microscopy: Techniques and Advanced Systems (2nd ed)”, Springer Series in Optical Sciences, Török P, Kao FJ (Eds.), pp. 195-236 32. Johnson DS, Jaiswal JK, Simon S. (2012) Total internal reflection fluorescence (TIRF) microscopy illuminator for improved imaging of cell surface events. Curr Protoc Cytom Ch12: Unit 12.29 33. Paige MF, Bjerneld EJ, Moerner WE (2001) A comparison of through-the-objective total internal epi- and reflection microscopy fluorescence microscopy for single-molecule fluorescence imaging. Single Mol 3: 191-201 34. Lang E, Baier J, Köhler J. (2006) Epifluorescence, confocal and internal total single-molecule reflection microscopy for experiments: a quantitative comparison. J Microsc 222(2):118-23 35. Engelhardt S. (2007) Characterization of the cell - sensor contact with total internal reflection der fluorescence microscopy. Berichte Forschungsanstalt Jülich, http://hdl.handle.net/2128/3088 36. Chon (2003) Two-photon JWM, Gu M fluorescence scanning near-field microscopy based on a focused evanescent field under total internal reflection. Opt Lett 28: 1930-2 37. Keller PJ et al. (2010) Fast, high-contrast imaging of animal development with scanned light sheet–based structured-illumination microscopy. Nat Methods 7, 637–42. 38. Baumgart E, Kubitscheck U (2012) Scanned light sheet microscopy with confocal slit detection. Opt Express 20(19): 21805-14 39. Yokota H, Saito K, Yaganida T (1998) Single Molecule Imaging of Fluorescently Labeled Proteins on Metal by Surface Plasmons in Aqueous Solution. Phys Rev Lett 80, 4606-9 40. He RY et al. (2009) Surface plasmon-enhanced two-photon fluorescence microscopy for live cell membrane imaging. Opt Express 17(8): 5987-97 11 SUPPORTING INFORMATION List of abbreviations AFM AOD AOTF aq. BF BK BON CNRS DAC (E)BFP - (EM)CCD EtOH EU EW FITC FOV IMA IQR NA n.s. OPSL PALM pk-pk PSF RMS ROI RT SAF SIL spTIRF - STED STORM - TIR[F(M)] UAF VCO vSAF atomic force microscopy - acousto-optical deflector - acousto-optical tunable filter - aqueous - bright field - borosilicate (glass) - A cell line derived from a metastatic pancreatic human carcinoid tumor - Centre National de la Recherche Scientifique - digital-to-analog card - (equivalent) back focal plane (electron-multiplying) charge-coupled device - ethyl alcohol, ethanol - European Union - - evanescent wave fluorescein isothiocyanate - field of view - Integrated Morphometry Analysis - - interquartile range numerical aperture - not significant - optically pumped solid-state laser - - photoactivated localization microscopy peak-to-peak - point spread function - root mean square - region of interest - - room temperature, 22-23°C supercritical angle fluorescence - - solid immersion lens spinning TIRF - stimulated emission depletion microscopy stochastic optical reconstruction microscopy total internal reflection [fluorescence (microscopy)] - undercritical angle fluorescence - - voltage-controlled oscillator virtual supercritical angle fluorescence - SI Methods Coverslips, dyes and bead suspensions. 25-mm diameter #1 and #1.5 coverslips (147.8  2.8 µm and 173.6  2.8 µm thick, respectively, n = 35 each, SchottDesag D263M, Thermo Fisher Menzel, Braunschweig, Germany) were sequentially passed twice through baths of 70% EtOH and sterile water, respectively, and were used thereafter or else treated with poly- ornithine (1.5 µg/ml, 30 min, 37°C, 5% CO2) or collagen soluble tail acid rat (Glassand Bornstein Traub type I collagen, Sigma, Lyon, France). Collagen was prepared at 1 mg/ml in aq. 1% acetic acid. This stock was diluted 1:200 in 30% EtOH and the coverslips incubated 3h at room temperature (RT, 22-23°C), the excess liquid removed and the coverslips dried (30 min, RT) under dust-free air flow before use. Coverslips were stored individually in sealed 6- well plates. Surface roughness of bare and treated BK-7 and fused silica substrates (TGP Inc., Painesville, OH) was measured with AFM (Bioscope, Veeco, Plainview, NY). 12 Submicron layers of rhodamine 6G (R6G, Sigma) or fluorescein isothiocyanate (FTIC, Fluka, Buchs, Switzerland) were deposited on coverslips with a spin coater (KW-4A, SPI supplies, West Chester, PA) at 3,000 rpm. Dilute solutions of red-emitting polystyrene latex microspheres (488/685 nm TransFluoSpheres, 93-nm diameter, Invitrogen, Saint Aubin, France) were drop-cast onto a clean coverslip and immobilized by solvent evaporation and used for the measurement of point-spread functions (PSFs). The pixel size for PSF measurements was 77 nm. Refractive of indices immersion liquids were measured (589 nm, RT) with an Abbe refractometer (WYA, Shanghai, China; Table S2). Suspensions of non-fluorescent PS latex beads (Polysciences, Inc., Warrington, PA, Table S3) were used as scattering samples. Scattering properties were calculated using the Mie Scattering Calculator at http://omlc.ogi.edu/calc/mie_calc.html written by Scott Prahl, assuming nPS = 1.6053 at 488 nm, from http://refractiveindex.info by Mikhail Polyanskiy. Cell preparation. Experiments followed EU and institutional guidelines for the care and use of laboratory animals (Council directive 86/609EEC). Astrocytes were prepared from P0-1 (P0 being the day of birth) NMRI mice (Janvier, Montpellier, France) as described (3). Briefly, neocortices were dissected and mechanically dissociated. Cells were plated and maintained in Petri dishes for one week to reach confluence before their transfer onto poly-L- ornithine-coated cover slips. Secondary cultures were maintained in DMEM, supplemented with 5% FCS, penicillin (5 U/ml), and streptomycin (5 μg/ml) at 37 °C in a humidified 5% CO2 atmosphere. Astrocytes kept for one more week in secondary culture and incubated with the fluorescent lysosomal marker FM2-10 (50 µM, 30 min) (4) or transfected with the plasmids listed in Table S4 were used for imaging, during which they were continuously perfused at 1 ml/min with extracellular saline containing, in mM: 140 NaCl, 5.5 KCl, 1.8 CaCl2, 1 MgCl2, 20 glucose, 10 HEPES (pH 7.3, adjusted with NaOH). Isolated astrocytes or small islets of astrocytes were imaged after 20 min wash (FM dyes) or 24-36 h following transfection. BON cells were a gift from Dr C. Desnos (CNRS UMR8192, University Paris Descartes) and were cultured and plated as described (5). The BON cell line was established from a lymph node metastasis of a human pancreatic carcinoid tumor (6) and provided to CNRS by Dr C. M. Townsend (University of Texas, Medical Branch, Galveston, TX). Cells were cultured at 37°C in a humidified 5% CO2 atmosphere in Ham’s F-12/DMEM with 10% fetal bovine serum and seeded onto collagen- coated glass coverslips for imaging, 1-5 days after plating. Azimuthal beam-spinning TIRFM. In objective-type TIRFM, a laser spot is focused in an eccentric position in the BFP of an objective with NA= , producing an oblique collimated beam emerging from the objective at an angle . (eq.1) The reflection is total for radii r , . Here, r, fFL and M are implying that the spot’s radius from the optical axis, the focal length of the focusing lens and the objective transverse magnification, respectively. For an aplanatic objective, the lateral magnification is , where fTL is the focal length of the used tube lens. We modified an earlier described microscope (7), Fig.S1A. Briefly, the 488-nm line of an Ar+- ion laser (Reliant 150, LaserPhysics, West Jordan, UT) or 488-nm OPSL (Sapphire 488, 20 mW, Coherent, Santa Clara, CA) was isolated and AOTF an with shuttered (AA.Optoélectronique, St.Rèmy-en-Chèvreuse, France), spatially filtered by passing it through a mono-mode optical fiber (kineFLEX, Qioptiq, formerly Point Source, Hamble, Southhampton, UK) and expanded by an achromatic 2.8 telescope (f1 = 50 mm, Thorlabs, Dachau, Germany; f2= 140 mm, Qioptiq, formerly Linos, Göttingen, Germany). Polarization was linear. 13 2322DnNAnsin2FLfnrM2/arcsinfnrc21nNAobjTLffM Two crossed AODs (AA.DTS XY-250@405nm, AA.Optoélectronique) positioned were (x,y,z,,) in a conjugate field plane. A compressing telescope (0.25, f1 =200 mm, f2 = 50 mm, Linos) increased the scan angle. An appropriately sized iris and opaque disc were placed in the conjugate aperture plane (EBFP of the objective) to reduce stray light from the AOD surfaces. A high-quality focusing lens f = 135 mm, Qiopiq, formerly (Rodagon Rodenstock, Feldkirchen, Germany) and high- NA oil-immersion objective (f=3 mm, PlanApo 60/NA1.45oil or APO N TIRFM, 60/NA1.49oil (Olympus, Hamburg, Germany) formed another telescope (0.022), resulting in a circular on-axis Gaussian beam of ~ 21 ± 2 µm (30 ± 1 µm) FWHM (1/e²) diameter in the sample plane and 0.77° ± 0.07° (13 ± 1 mrad) far-field divergence (half angle of the FWHM intensity) emerging from the objective. This optical path allowed us to freely position within µs the focused spot in the BFP up to 5 mm off the optical axis, a radius larger than the pupil diameter of the NA1.49 objective rNA= NAf = 4.47 mm). For SAF experiments, we used a Plan-Apochromat 100/NA1.46oil (Zeiss, Oberkochen, Germany). The objective position could be adjusted with a piezo-electric focus drive (P721.PLQ, Physik Instrumente, Karlsruhe, Germany). Scheme 1: electrical layout of the buffer circuit The polar and azimuthal beam angles (, ) were software controlled under LABVIEW (National Instruments, Austin, TX) using the amplified output of a 16-bit DAC board (NI U2B-6211, National Instruments) connected to the frequency modulation input of the VCO driver (AA.DRF.10Y2X, AA.Optoélectronique). We increased the output power of the DAC board by a custom buffer circuit with unity gain (scheme 1). The excitation optical path was aligned using the BFP image taken on CCD camera ‘2’, Fig.S1A. For a perfectly symmetric illumination, the coverslip could finally be aligned orthogonally to the optical axis with a two-axes tip-tilt platform (M-TTN80, Newport, Irvine, CA). Set-up characterization. Because no light is transmitted beyond the critical angle TIR limits the range within which the beam angles  and can be calibrated. We therefore employed an oil-coupled solid-immersion half-ball lens (SIL, 8 mm , S-LAH79, Edmund Optics, Barrington, NJ) that allowed us to project the beam beyond angles otherwise obscured by TIR (8, 9), inset in Fig.S1A. At 488 nm, the S-LAH79 glass (Ohara, Branchburg, NJ) of the SIL has a refractive index of n3 = 2.027  0.002, refracting the beam to angles closer the optical axis. The to transmitted beam could then be projected onto a screen and the beam angle  estimated and related to the measured radial and axial distances a and b as , (eq.2) (and likewise for in orthogonal direction). Here, n2 = 1.521 ± 0.006 is the refractive index at 488 nm of the BK-7 front lens of the TIRF objective. Multi-layer effects (10, 11) were neglected. Phase aberrations were measured with a lateral shearing interferometric wave-front sensor (12) (SID4, Phasics, Palaiseau, France). M², the beam parameter product relative to that of an ideal Gaussian beam, and the Strehl ratio ( being the RMS deviation of the wave front and  the wavelength) calculated and plotted as a function of . Due diffraction angle-dependent the to efficiency of the AODs the transmitted light intensity varied with spot position in the BFP, Fig.S1B, and was further modified by the dichroic and objective, Fig.S1C. These effects 14 )/arctan(sinarcsin23abnn2/2eS automatically be principle, in could, compensated for by feeding a correction signal into the intensity modulation input of the VCO, however, this possibility was not used here. Rather, we simply increased the laser power to have sufficient signal at high beam angles, where the diffraction efficiency of the AOD scanner is low. Typical laser powers in the sample plane were of the order of a few µW. We used a 3-mm thick zt491 RDCXT (AHF, Tübingen, Germany) to direct the scanned 488- nm beam to the objective. Optical flatness (/10) and strain-free mount of the dichroic were crucial for optimal spTIRF. A small amount of the excitation light transmitted by the dichroic formed a scaled image (f1 = 80 mm, f2 = 40 mm, both from Linos) on an inexpensive CCD camera (Foculus FO432SB, Elvitec, Pertuis, France, 4.65 µm pixel size) permitting the simultaneous visualization of the excitation light distribution in the objective BFP (7) and of the fluorescence on a second, electron- multiplying charge-coupled device camera (EMCCD, Cascade 128+ or QuantEM512C, both from Photometrics, Tucson, AZ). Pixel sizes were 194 nm or 120 nm in the sample plane –calibrated with a G300-Cu copper grid (EMS, Hatfield, PA) or the 5-µm divisions of a reticle (Qioptiq-Linos), corresponding to an effective field-of-view of 25×25 µm² (61×61 µm²) for the Cascade 128+ (QuantEM512C). Combinations of the optical filters used are listed in Table S5. Filters were from AHF Analysentechnik (Tübingen, Germany), Barr Assoc. (Westford, MA), Chroma Technology (Bellows Falls, VT) and Semrock (Rochester, NY). Automatic image thresholding, segmentation and statistics. EMCCD images were analyzed after subtraction of a dark image time, the same exposure taken at hardware- and amplifier-gain. Foculus CCD images are shown as raw images. Images are shown on an inverted look-up-table for better display in print with intensity displayed in grey value autoscaled. Circular line regions of interest (Fig. 1A) were centered on the cell identified from bright-field (BF) images and the intensity profile plotted as a polar graph I( )/Imean to allow cell-to-cell comparison, CV = ISD /Imean. Image segmentation and analysis (Fig. 2A, Fig. S2) was performed using Metamorph’s IMA tool (v7.5.4, Molecular Devices). To isolate individual lysosomes or mitochondria the local background was suppressed by subtraction of a corresponding low-pass filtered image (3.9 µm) and the result segmented using isodata histogram thresholding (13). Briefly, in this procedure the image histogram is initially segmented into two parts using a starting threshold intensity 0 = 2B-1, half of the maximum dynamic range. The sample mean (mf,0) of the grey values associated with the foreground pixels and the sample mean (mb,0) of the grey values associated with the background pixels are computed. A new threshold value 1 is computed as the average of these two sample means, (mf,0 + mb,0)/2 . The process is repeated, based upon the new threshold, until k converges, k = (mf,k-1 + mb,k-1)/2 until k = k-1 (eq.3) Individual mitochondria were identified using {area ≥ 5px (~ 1µm²) AND length ≥ 2px (0.38 µm) AND breadth ≥ 2px)} as classifiers and their area A (pixels above ∞), orientation (angle between longest chord of the object and the horizontal, i.e., -90° = downward), shape factor ( = 4A/P, P = perimeter, i.e., flat = 0, ..., 1 = circle), and elliptical form factor ( = length/breadth) measured. For the mitochondrial data set in Fig. S2 (imaged at 78° beam angle and 194-nm pixel size), this corresponded to 14.4 ± 2.8% (n = 24 images) of the maximal intensity. The means of normally and log- normally distributed data sets having the same variance were compared with Student’s t-test. The non-parametric and distribution-free KS- test was used for comparing non-normally distributed data. Points ≥1.5× IQR above the third quartile or below the first quartile were considered as outliers (Turkey). Differences were considered significant for p < 0.05. On figures, , , and are shorthand for p < 15  0.05, p < 0.01 and p < 0.001, respectively. n.s. means not significant. SUPPLEMENTARY TABLES Table S1: Measured microscope PSFs TIR-vSAF 374 ± 20 ×100/1.46 ×60/1.45 TIRF a spTIRF 388 ± 19 c 366 ±19 b 336 ± 21 332 ± 25 314 ± 26 c 312 ± 14 a reported as FWHM (mean ± 1 SD) of a Gaussian fit (14) with the measured radial intensity profile of n = 10 isolated in- focus red-fluorescent ( = 488/685 nm) 93-nm diameter beads. b The PSF for the Zeiss ×100/1.46 lens is an overestimate because it was measured without the matched Zeiss tube lens that provides part of the aberration correction. c same, for a second, nominally identical, PlanApo ×60/NA1.45oil TIRFM objective (Olympus). spTIR-vSAF 375 ± 15 Table S2: Measured refractive indices of used immersion liquids (measured) a Olympus Zeiss Immersol 518F 1.5122 ± 0.0009 1.5156 ± 0.0002 (1) 1.5157 ± 0.0002 (2) 1.5158 ± 0.0002 (3) 1.479 1.4804 ± 0.0003 Cargille FF a measured at 22-23°C on a WYA Abbe refractometer using the sodium D line (589 nm). Values are mean ± 1 SD of five independent measurements for each immersion oil. Three different batches (1) -(3) of the Zeiss oil were tested. 1.516 1.518 n (nominal) Table S3: Characteristics of polystyrene (PS) microsphere suspensions (a) d (µm) 0.7 2.8 90 x b c g (µm-3) (µl-1) 10-10 0.912 0.1 6 10-8 10 10-7 100 1.5 ×10-6 1,500 1.5 ×10-5 15,000 1.5×105 1.5 ×10-4 1.5×106 1.5 ×10-3 10-10 0.813 0.1 20.6 10-8 10 10-7 100 1.5 ×10-6 1,500 1.5 ×10-5 15,000 1.5×105 1.5 ×10-4 10-11 579.4 0.929 0.01 6 ×10-10 0.6 6 ×10-9 6 6 ×10-8 60 µs’ µs (mm-1) (mm-1) 8.8 ×10-12 10-10 8.362 ×10-7 9.5 ×10-6 9.5 ×10-5 8.362 ×10-6 1.43 ×10-3 1.258 ×10-4 1.258 ×10-3 0.01425 0.0125 0.1425 0.1254 1.425 10-10 1.87 ×10-11 1.58 ×10-4 2.955 ×10-5 1.58 ×10-3 2.955 ×10-4 4.416 ×10-3 0.02362 0.0442 0.2362 0.4416 2.362 7.1 ×10-12 10-10 5.68 ×10-4 0.008 0.0055 0.078 0.778 0.0552 16 (max)(max)/emex2322Dn a input parameters: mPS = nPS - ikPS = 1.6053 + 0 (the imaginary part is negligible at visible wavelengths); nmedium = 1.33;  = 488 nm; d – sphere diameter; c – concentration in in µl-1 ( mm-3) b output parameters calculated with the Mie scattering calculator http://omlc.ogi.edu/calc/mie_calc.html size parameter x = 2d/, where=nmedium x - g average cosine of the phase function (scattering anisotropy) g = <cos> - µs scattering coefficient - µs’ - reduced scattering coefficient µs’ = µs · (1-g) Table S4: Plasmids transfected into astrocytes (a) Source (b) Plasmid Concentration (µg/µl) Frank Kirchhoff (Univ. des Saarlands, Homburg; Germany) 4.06 VAMP2-EGFP idem 3 YFP Johannes Hirrlinger (Univ. Leipzig, Germany) 0.25 Mito-GFP Steven Green (Univ. Iowa, USA) 2 Lck-EGFP Thierry Galli (Inst. Jacques Monod, Paris, France) 1 CD63-GFP Vinculin-GFP Maité Coppey (Inst. Jacques Monod, Paris, France) 2.4 a using lipofectamine 2000 following standard protocols b the kind gift of plasmids is gratefully acknowledged Table S5: Used filter combinations (a) emission dichroic Fluorophore/experiment E580/40 Q515LP FM2-10 E580/40 zt491 RDCXT 600LP zt491 RDCXT FM4-64 520LP Q515LP EYFP/EGFP BP530/40 (676/29BP) zt491 RDCXT 488/560nm (488/685nm)TransFluoSpheres HC488/10 or HC MaxLine 488/1.9 zt491 RDCXT Dark-field a 488-nm excitation was used in all experiments. Returned excitation light was stopped by a 488-nm rugate notch filter (Barr Assoc., Westford, MA) and z488rdc RazorEdge (Semrock, Rochester, NY) having a combined suppression of >10-9. 17 SUPPLEMENTARY FIGURES A x   obj  L3 L2 laser in r   y ebfp hbl bfp hb l 2 cm B Fig. S1. Spinning TIRF (spTIRF) excitation. (A), simplified optical layout of the excitation optical path. Solid and dashed lines are conjugate field and aperture planes, respectively. Optical elements are identified in the SI materials section. Insets show, on the left, photograph of the half ball lens (HBL) in a custom holder on top of the objective that allowed us to measure beam angles otherwise obscured by TIR. Right, definition of variables in the objective BFP. r and dashed line – radius; θ azimuthal angle; grey area – radii below rc corresponding subcritical angles at which light is refracted and propagated into the sample (epi); white annulus – radii corresponding to supercritical angles for which total internal reflection occurs; turquoise spot is focused excitation beam (not to size); solid line – radius rNA the objective. (B), calibration of command voltage vs. beam angle limiting NA of to corresponding C L1 AODs D is the measured angle of the for the 60/NA1.45, red, and 100/NA1.46 objective, blue. beam exiting the HBL, n3 = 2.03 its refractive index at 488 nm. Symbols and error bars represent center and diameter of the beam projected against a small screen. Fits with the objective transverse magnification M, n 3 and fL3 as free parameters yielded (M, n3, fL3) = (58.83 ± 3.24, 2.04 ± 0.04, 137.82 ± 2.12, red) and (101.31 ± 2.2, 2.07 ± 0.05, 136 ± 3.56, blue), respectively, and compare favorably to the real values (60, 2.03, 135) and (100, 2.03, 135) for either objective. (C), measured laser power diffracted into order (1, 1) as a function of the command voltage of the AODs, for the four cardinal beam positions, symbols, their mean, dashed, and upon beam spinning, solid line. (D) Fractional intensity after the objective, as a function of beam angle θ for the four cardinal directions (symbols as in left panel) and their mean, solid. For all objectives, transmitted intensities were reduced at high NAs. At a given beam angle, laser power was adjusted to give the desired signal-to-noise ratio. 18 'sinarcsin23nn'25201510power (µW)543210voltage (V) N E W S meanNESW spinning0.80.70.60.50.4relative intensity6040200beam angle (°) x60/1.45NAx100/1.46NA806040200beam angle (°)43210voltage (V) x60/NA1.45 x100/NA1.46' A B spTIRF FIG. S2. Eccentric-spot excitation introduces a directional detection bias. (A), conventional unidirectional TIRFM images (cardinal images) of astrocytes expressing enhanced green fluorescent protein linked to a mitochondrial targeting sequence (mito-EGFP) show preferentially those mitochondria that are aligned parallel to the EW propagation direction, whereas those perpendicular to it are less visible. Symbols indicate position of the focused exciation spot in the objective BFP. (B), quantitative morphometry reveal that mitochondria detected upon illumination with a horizontally propagating EW (grey open triangles) had, on average, orientations closer to the horizontal axis, (33.7° ± 16.6°, median ± abs. deviation, n = 268 mitochondria in 6 cells), whereas those detected with a 90° -rotated EW field (black filled triangles) were orientated closer to the vertical axis (53.2° ± 18.6°, n = 357, p < 0.001). Triangles point in direction of EW propagation. Restoring a symmetric illumination by azimuthal beam scanning abolished this directional detection bias (red circles), and the cumulative distribution of absolute orientations (b etween 0 and 90°) now had a median at the expected 45° (p < 0.001 vs. each of the four unidirectional images, n = 155 mitochondria in 6 cells). Color code as on panels of orientation vs. shape factor that is 4πA/P², where A and P are the organelle area and perimeter, respectively. 19 -80-4004080orientation (°)0.80.4shape factor (AU)0.80.4shape factor (AU)-80-4004080orientation (°)1.00.80.60.40.2fraction806040200orientation (°)**** A B 65.3° 73° 65.3° 68° 68° 71.2° 65.3° Fig. S3. High fluorophore density and large penetration depth aggravate excitation non -homogeneity. Image pairs show cultured cortical astrocytes, (A), labeled with FM2-10, after transfection with various fluorescent-protein constructs, upon azimuthal beam spinning (left, spTIRF) and unidirectional 488 -nm evanescent-wave excitation (right row). Denser fluorophore distributions like membrane (Lck-EGFP), cytoplasmic (YFP) or dense vesicular labeling (vesicular - membrane associated protein-2, VAMP-2) are more sensitive reporters of non-evanescent excitation light than are sparser labels like lysosomes (FM2-10). Uneven illumination is less perceptible at higher beam angles (smaller penetration depths of the evanescent wave). In all cases, azimuthal beam spinning effectively eliminates excitation patterns. Objective was ×60/NA1.45. Fields are 25×25 µm², pixel size 194 nm in the sample plane. S ee table S4 for used filter combinations. 20 d = 0.7 µm 2.8 µm FIG. S4. Effect of evanescent-field scattering. (A), dependence of the scattered light intensity I s’ on the reduced scattering coefficient µs’ of dilute solutions of non-fluorescent polystyrene beads, for three different beam angles θ (black, 64°; red, 70° and blue, 78°) and three different bead diameters, from left to right: 0.7, 2.8 and 90 µm. For each color, solid lines and symbols show mean ± SD from triplicate experiments, thin faint lines are individual measurements. Drop at high values of µs’ is due to inner-filter effects due to multiple scattering. Scattered light was detected in dark - field through a second upright ×60/1.1 water-immersion objective the focus of which was at the reflecting interface (see Fig. 3A). The EW was set up by focusing a 488-nm laser beam in the periphery of the lower ×60/1.45 oil-immersion objective. (B), Dependence of Is’ on the surface density of scattering beads (mm-2), for a monolayer of 0.7- (left) and 2.8- µm beads (right) drop-cast on the coverslip. Color code as before. Compare with Fig. 3B in the main text. 21 10008006004002000Is' (counts/px)10-12 10-6 100µs' (mm-1)0.7 µm1000800600400200010-12 10-6 1002.8 µm1000800600400200010-12 10-6 10090 µm Fig. S5. Properties of high-NA objectives used. (A), schematic optical layout. L1 – AC250 mm (Linos), L2 – AC110 mm (TILL). EMCCD – QuantEM512C. (B), measured intensity distribution at the BFP for, from left to right, two nominally identical Olympus ×60/NA1.45 objectives, the Olympus ×60/NA1.49 and the Zeiss ×100/NA1.46 objective, r c and rNA indicated by black solid lines are, respectively, the radii corresponding to the critical and maximal emission angle supported by the objective and are related to θNA by r = f sin(θNA), where f = fTL/M and θNA = arcsin(n2/NA) are the objective focal length and aperture angle, respectively. Sample was a 100 -nm thick FITC layer (500 µM). Laser powers in the back pupil were 226, 104, 117 and 100 µW, respectively. Measured effective NAs (NA eff) are shown below the corresponding images, the error giving the SD of independent measurements (re-aligning the Bertrand lens, re-focusing the objective or moving laterally in the fluorophore layer). (C), (left) sketch of BFP upon focusing a low -µW 488-nm beam in the BFP of a Zeiss x100/NA1.46 objective. The beam wa s spun at constant a radius, producing in several lenses measureable yellow-green autofluorescence (right image), the most intense of which originating from glass close to the BFP (arrowhead). 22 bfpobjEMCCDL1L21.457 ±0.0071mm1.468 ±0.0041.456 ±0.0031.474 ±0.0071.453 ±0.004effective NAsNA1.45NA1.45NA1.49NA1.46ABCautofluorescencebfpobjEMCCDL1L2bfpobjEMCCDL1L21.457 ±0.0071mm1mm1mm1.468 ±0.0041.456 ±0.0031.474 ±0.0071.453 ±0.004effective NAsNA1.45NA1.45NA1.49NA1.46ABCautofluorescence SUPPORTING REFERENCES Brunstein M, Tourain C, & Oheim M (2012) Glare and other sources of non-evanescent excitation light in objective-type TIRF microscopy. In: 12th European Light Microscopy Initiative (ELMI) Meeting (RMS, Leuven). Teremetz M (2011) Advantages of spinning TIRF illumination for subcellular fluorescence imaging. In: Master in Cell Biology and Pathology (BCPP) (Université Paris Diderot, Paris), p. 28. Li D, Hérault K, Oheim M, & Ropert N (2009) FM dyes enter via a store-operated calcium channel and modify calcium signaling of cultured astrocytes. Proc. Natl. Acad. Sci. U.S.A. 106, 21960-21965. Li D, Ropert N, Koulakoff A, Giaume C, & Oheim M (2008) Lysosomes are the major vesicular compartment undergoing Ca2+-regulated exocytosis from cortical astrocytes. J. Neurosci. 28, 7648-7658. Huet S, Fanget I, Jouannot O, Meireles P, Zeiske T, Larochette N, Darchen F, & Desnos C (2012) Myrip couples the capture of secretory granules by the actin-rich cell cortex and their attachment to the plasma membrane. J. Neurosci. 32, 2564-2577. Evers BM, Townsend CMJ, Upp JR, Allen E, Hurlbut SC, Kim SW, Rajaraman S, Singh P, Reubi JC, & Thompson JC (1991) Establishment and characterization of a human carcinoid in nude mice and effect of various agents on tumor growth. Gastroenterology 101, 303-311. van 't Hoff M, de Sars V, & Oheim M (2008) A programmable light engine for quantitative single molecule TIRF and HILO imaging. Optics Express 16, 18495-18504. Matsuo S & Misawa H (2002) Direct measurement of laser power through a high numerical aperture oil immersion objective lens using a solid immersion lens. Rev. Sci. Instr. 73, 2011- 2015. Schwarz JP, König I, & Anderson KI (2011) Characterizing system performance in total internal reflection fluorescence microscopy. In: Cell Migration: Developmental Methods and Protocols, eds. Wells CM & Parsons M (Springer, Berlin Heidelberg New York), pp. 373- 386. Gingell D, Heavens OS, & Mellor JS (1987) General electromagnetic theory of total internal reflection fluorescence: the quantitative basis for mapping cell-substratum topography. J. Cell Sci. 87, 677-693. Heavens OS (1995) Use of the approximations in cell studies by total internal reflection fluorescence microscopy (TIRF). J. Microsc. 180, 106-108. Velghe S, Primot J, Guérineau N, Cohen M, & Wattellier B (2005) Wave-front reconstruction from multidirectional phase derivatives generated by multilateral shearing interferometers. Opt. Lett. 30, 245-247. Ridler TW & Calvard S (1978) Picture thresholding using an iterative selection method. IEEE Trans. Syst. Man Cybern. 8, 630-632. Zhang B, Zerubia J, & Olivo-Marin JC (2007) Gaussian approximations of fluorescence microscope point-spread function models. Appl. Opt. 46, 1819-1829. 23 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14.
1902.04634	1	1902	2019-02-12T21:07:15	On the selectivity of KcsA potassium channel: asymptotic analysis and computation	[ "physics.bio-ph" ]	Potassium (K$^+$) channels regulate the flux of K$^+$ ions through cell membranes and plays significant roles in many physiological functions. This work studies the KcsA potassium channel, including the selectivity and current-voltage (IV) relations. A modified Poisson-Nernst-Planck system is employed, which include the size effect by Bikerman model and solvation energy by Born model. The selectivity of KcsA for various ions (K$^+$, Na$^+$, Cl$^-$, Ca$^{2+}$ and Ba$^{2+}$) is studied analytically, and the profiles of concentrations and electric potential are provided. The selectivity is mainly influenced by permanent negative charges in filter of channel and the ion sizes. K$^+$ is always selected compared with Na$^+$ (or Cl$^-$), as smaller ion size of Na$^+$ causes larger solvation energy. There is a transition for selectivity among K$^+$ and divalent ions (Ca$^{2+}$ and Ba$^{2+}$), when negative charge in filter exceeds a critical value determined by ion size. This explains why divalent ions can block the KcsA channel. The profiles and IV relations are studied by analytical, numerical and hybrid methods, and are cross-validated. The results show the selectivity of the channel and also the saturation of IV curve. A simple strategy is given to compute IV relations analytically, as first approximation. The numerical method deals with general structure or parameters, but the limitations and difficulties of pure numerical simulation are also pointed out. The hybrid method provides IV relations most effectively for comparison. The reason for saturation of IV relation is illustrated, and the IV curve shows agreement with the profile and scale of experimental results.	physics.bio-ph	physics	On the selectivity of KcsA potassium channel: asymptotic analysis and computation Zilong Songa,b, Xiulei Caoa,b, Tzyy-Leng Horngc,d, and Huaxiong Huanga,b February 14, 2019 aDepartment of Mathematics and Statistics, York University, Toronto, Ontario, Canada b Fields Institute for Research in Mathematical Sciences, Toronto, Ontario, Canada c Department of Applied Mathematics, Feng Chia University, Taichung 40724, Taiwan d National Center for Theoretical Sciences, Taipei Office, Taipei, Taiwan 10617 Abstract Potassium (K+) channels regulate the flux of K+ ions through cell membranes and plays significant roles in many physiological functions. This work studies the KcsA potassium channel, including the selectivity and current-voltage (IV) relations. A modified Poisson-Nernst-Planck system is employed, which include the size effect by Bikerman model and solvation energy by Born model. The selectivity of KcsA for various ions (K+, Na+, Cl−, Ca2+ and Ba2+) is studied analytically, and the profiles of concentrations and electric potential are provided. The selectivity is mainly influenced by permanent negative charges in filter of channel and the ion sizes. K+ is always selected compared with Na+ (or Cl−), as smaller ion size of Na+ causes larger solvation energy. There is a transition for selectivity among K+ and divalent ions (Ca2+ and Ba2+), when negative charge in filter exceeds a critical value determined by ion size. This explains why divalent ions can block the KcsA channel. The profiles and IV relations are studied by analytical, numerical and hybrid methods, and are cross-validated. The results show the selectivity of the channel and also the saturation of IV curve. A simple strategy is given to compute IV relations analytically, as first approximation. The numerical method deals with general structure or parameters, but the limitations and difficulties of pure numerical 1 simulation are also pointed out. The hybrid method provides IV relations most effectively for comparison. The reason for saturation of IV relation is illustrated, and the IV curve shows agreement with the profile and scale of experimental results. 1 Introduction Rapid communication in many organisms relies on fast propagation of electric signals, which in turn depend on a specialized class of protein molecules called ion channels. When the ion channels are opened on the cell membrane by either chemical ligands or membrane depolarization, they allow ionic flux across the cell membrane and lead to rapid changes of membrane potentials. Potassium (K+) channels regulate the flux of K+ ions through cell membranes and participate several physiological functions such as the maintenance of the resting membrane potential, the excitation of nerve and muscle cells, the secretion of hormones, and sensory transduction [13, 43]. When dysfunctional, ion channels would cause a number of diseases. Therefore, understanding how the molecular structure determines channel function is profoundly interesting, both for the biological and for the medical sciences [26, 1, 5]. The X-ray crystallographic structures of distinct potassium channels reveal a common architecture of the pore [25, 11]. Four subunits are symmetrically arranged around the channel axis, with each subunit having at least two transmembrane helixes separated by a re-entrant P-loop and selectivity filter (SF). K+ channels are the most extensively studied family of ion channels, both experimentally and computationally, and the KcsA structure [40, 42] has been the most popular one among K+ channels since it is the first K+ channel to be crystalized. Many computational and experimental data of KcsA is available for comparison. SF of K+ channels is the essential element to their permeation and selectivity mech- anisms [6]. Thousands of millions of K+ ions per second can diffuse in single file down their electrochemical gradient across the membrane at physiological conditions [2, 29]. Each subunit contributes to SF with a conserved signature peptide, namely TVGYG in most of the channels [26]. The carbonyl oxygens of the backbone of SF point toward the lumen and orchestrate the movements of ions in and out of the channel. These carbonyl oxygens together with the side-chain hydroxyl oxygen of a threonine residue define four 2 ion-binding sites in SF, designated S1-S4 starting at the extracellular side [45]. In addi- tion, K+ ion can bind in the central water-filled cavity of pore and two alternate positions at the extracellular side of pore [45]. SF is generally too narrow to accommodate a K+ ion with its hydration shell, and thus K+ ions must be dehydrated to enter SF, when attracted by the strong negative charges of carbonyl oxygens in SF. K+ ion must replace its solvation shell by the carbonyl oxygens in the backbone of SF. Each of these protein sites binds K+ ions with a tight-fitting cage of 8 carbonyl oxygen atoms that resembles the solvation shell of a hydrated K+ ion. Classical Poisson-Nernst-Planck (PNP) system has been widely applied to model ionic transport in biological setting as well as other areas [10, 28, 18]. Various analysis and computation [37, 38, 8] regarding this system have been attempted in the literature. Current-voltage (IV) relation is an important functional characteristic of ion channels and can be determined experimentally. PNP theory has been successfully applied to model wide ion channels, and has reproduced the experimental IV data quite successfully [44, 41]. However, when used in narrow ion channels, such as KcsA, the classical PNP system is not suitable anymore, due to the extremely narrow SF. This is because classical PNP neglects the size of ions and therefore overestimates the K ion occupancy of SF. Also, classical PNP does not consider solvation energy barrier that is significantly encountered by K+ ions when dehydrated to enter SF. Various modified PNP system have been proposed to include the steric or size effect of ions [16, 9, 21, 19, 23]. In this study, we employed Bikerman model with specific ion sizes [17], which is one of the widely accepted models in literature. In addition, solvation energy based on Born model is included in the present formulation based on dehydration of ions and its importance emphasized above [4]. The adoption of Bikerman model is because of its simplicity and availability of some analytical results, which we believe can provide more physical insights into the mechanism of ion channels. It is well known that potassium channels have high selectivity of potassium ion over sodium ion (K+ is 104 times more permeant than Na+) [13]. Though K+ and Na+ have the same valence and therefore they have the same electrostatic affinity to carbonyl oxygens in SF, K+ encounters less Born solvation energy barrier than Na+ when passing through SF due to its size slightly larger than Na+. However, few studies were found about selectivity between K+ and alkaline earth ions like Ca2+ and Ba2+. Alkaline earth ions generally 3 have stronger electrostatic affinity to SF than K+ due to their divalence, but at the same time also bear larger Born solvation energy barrier again due to their divalence. The blockage of KcsA by Ba2+ has demonstrated this strong competition of SF occupancy between electrostatic affinity and solvation energy [33]. Here we employed one-dimensional asymptotic analysis and numerical simulation of present model to study (i) the mechanism of channel selectivity among K+ and other ions; (ii) the mechanism causing IV curve to be saturated when voltage gets large as recorded in experiments [32]. More precisely, we give simple explanation and analytical formulas regrading the selectivity among K+, Na+, Cl−, Ca2+ and Ba2+. The selectivity is mainly influenced by permanent negative charges in SF and the ion sizes. The smaller ion size of Na+ compared with K+ gives a larger solvation energy barrier to enter filter, and hence its concentration is exponentially smaller (not selected). When negative charge in SF exceeds a critical value (given by ion size), SF starts to recruit divalent cations to coexist with K+ by squeezing some K+ out of SF, since divalent cations can do better in balancing the strong negative charge in narrow SF. Although Born solvation energy is increased by this recruitment of divalent cations into SF, electrostatic energy is decreased more for compensation and total energy is actually decreased then. We have studied the IV curves by analytical, numerical, and hybrid methods, and cross-validated the results. The results have revealed the reason for saturation of IV curve and pointed out the difficulties in numerical simulations for some cases. The IV curve also shows agreement with the profile and scale of experimental results. The manuscript is arranged as follows. Section 2 formulates the mathematical model, i.e., the modified PNP system with Bikerman model and Born model. Section 3 deals with the the equilibrium case with zero flux, implying the selectivity of channel. Section 4 provides analytical results for IV curve for non-equilibrium case. Direct numerical simulations are conducted in Section 5 and a hybrid computational-asymptotic analysis is done in Section 6. Finally, some concluding remarks are drawn. 2 Mathematical model We consider the Bikerman model with specific ion sizes [17], and include permanent charge and solvation energy into the Poisson-Nernst-Planck (PNP) formulation. The 4 original one-dimensional (1D) system for −L < x < L is (cid:32) n(cid:88) (cid:33) zkck − q(x) , − 1 A(x) ∂x(0r(x)A(x)∂xφ) = e0 ∂tci + 1 A(x) ∂xJi = 0, Ji = −A(x) k=1 Di kBT ci∂xµi, (1) where ci (i = 1, .., n) denote the concentrations of ions, φ is electric potential, A(x) is the cross section area, q(x) is the permanent charge (positive q means negative fixed charge), r(x) is the relative permittivity, and kB, T, 0, e0 are some constants (see Appendix A). The electro-chemical potentials are given by µi = kBT log(cia3 i ) − log cka3 k + zieφ + Wi, i = 1, .., n, (2) (cid:32) (cid:33)(cid:33) (cid:32) 1 − n(cid:88) k=1 (cid:18) 1 (cid:19) Wi(x) = z2 i e2 0 8π0ai − 1 r(x) where ai are the effective diameters of ions, and Wi is solvation energy . (3) Figure 1: Sketch of the potassium channel. Figure 1 schematically shows the setup, where the filter of channel lies between two chambers, linking extracellular and intracellular spaces, respectively. The length of filter is Lf and the length of each chamber is set as Lb (then L = Lb + 1 2Lf ), where some part of reservoir is included we consider a relative large Lb. The cross section area A(x) in filter is much smaller than that of the chamber region. The permanent negative charge q(x) due to carbonyl oxygens and threonine residues is confined in the small volume of the filter, so the effective q(x) in the model is extremely large compared with chamber concentrations [14, 15]. This further implies that the filter attracts counter-ions and thus the saturation of ions in filter means few water molecules in filter or ions are dehydrated. Therefore, the dielectric constant r(x) would be much smaller in filter, and this justifies the introduction 5 chamberfilterA(x)q(x)chamberof above solvation energy Wi(x), which gives the energy barrier from chamber to filter. In later analysis, the solvation energy also causes jumps in concentrations from chamber to filter while maintaining continuous electro-chemical potentials µi. In the chamber, the model is approximately the classic PNP system, where the size effect is negligible and q(x) = 0. With this in mind, we do a traditional nondimensional- ization with reference scales in the chamber. We set x = ai = , x L ai a0 ci = ci c0 , , A = A Ab , φ = φ φ0 Lf = Lf L , , Di = , Di D0 q c0 q = , t t = L2/D0 Wi kBT Wi = , r = r rb , , (4) where a0 is a reference diameter, D0 is a reference diffusion constant, Ab is reference (maximum) cross section area in chamber and rb is (maximum) relative permittivity at farther end of chamber (see (69) in Appendix A for their values). By removing the tilde, the dimensionless system in −1 < x < 1 is n(cid:88) k=1 − 2 1 A(x) ∂x(r(x)A(x)∂xφ) = zkck − q(x), ∂tci + 1 A(x) ∂xJi = 0, Ji = −A(x)Dici∂xµi, (5) (6) where i = 1, .., n and (cid:32) 1 − n(cid:88) (cid:19) k=1 W0, − 1 µi(x) = log[ci(x)] − log (cid:18) 1 Wi(x) = z2 i ai rbr(x) (cid:33) ck(x)a3 kδ + ziφ(x) + Wi(x), where the first term in µi is originally log(cia3 i δ) by dimensionalization but we removed the constant log(a3 i δ) from µi since this would not affect the system. The dimensionless parameters are = (cid:115) 0rbkBT 0c0L2 , e2 δ = a3 0c0, W0 = e2 8π0a0kBT . (7) Please refer to Appendix A for the estimates of parameters in this system. One easily see that with Wi =constant and as ckδ tends to 0, the above µi goes back to that in classical PNP system. This is the case in chamber region, whereas in filter region ck is quite large and ckδ terms can not be neglected. 6 3 Equilibrium case with zero flux In this section, we study the selectivity of the channel in equilibrium case, for sim- plicity. We will see the conclusions also hold for non-equilibrium case with finite fluxes. This is seen in analysis of this section that the boundary conditions (inducing finite flux when different) have negligible or exponential small impact on the results. This is also verified by analytical and numerical results in non-equilibrium case, as the selected ions are in equilibrium in filter (non-equlibrium outside), see Figures of µi in Sections 4 & 5 and analysis before (63). The same boundary conditions at two ends of chamber are used where i = 1, .., n and the electro-neutrality (EN) condition(cid:80) zicib = 0 is satisfied. There- ci(x) = cib, φ(x) = φb, (8) at x = ±1, fore, there's no flux across the filter. The aim is to study the relative concentrations of ions in filter under different situations, which would imply the selectivity. In general case, we notice that by definition of µi in (6) we can solve ci (i = 1, .., n) in terms of φ and µi (see Appendix B) ci = eµi−Wi−ziφ (1 + F δ) , F = n(cid:88) k=1 keµk−Wk−zkφ. a3 (9) For equilibrium case, by Ji = 0, we conclude that µi is constant throughout filter and (cid:32) 1 − n(cid:88) (cid:33) chamber µi(x) = Bi = log(cib) − log ciba3 kδ + ziφb + Wi(1) (10) k=1 where the constant Bi is determined by boundary conditions in (8). In this case, by substituting (10) into (9), ci is expressed explicitly in terms of φ. Since the filter region is quite small, it is natural to adopt some effective charge [15, 30]. We assume q(x) = q is a large constant in filter, and treat q as a crucial parameter. Depending on the relative magnitude of q, we have either electro-neutral (EN) case or non-EN case in filter. We also assume r(x) = r0 in filter, where r0 is constant (say 1/40, corresponding to original r = 2). Note by choice of scaling in (4), we have r(±1) = 1. 7 3.1 K+/Na+ selectivity In this subsection, we consider the case with three ions K+, Na+ and Cl− (respectively c1, c2 and c3), and study the selectivity between Na+ and K+. From the expression of c1 and c2 in (9,10), we get in the filter c1 c2 = eB1−W1(0) eB2−W2(0) . (11) Thus, the ratio c1/c2 is a constant independent of φ and x in filter. More precisely, we have Bi − Wi(0) ≈ −∆Wi + log cib + ziφb, i = 1, .., n, (12) where the O(δ) term in chamber has been dropped and ∆Wi is the barrier from chamber to filter due to solvation energy ∆Wi = Wi(0) − Wi(1) = z2 i ai ( 1 rbr0 − 1 rb )W0, i = 1, .., n. (13) Since the diameter of K+ is larger than that of Na+, the barrier of K+ from chamber to filter is smaller, i.e., a1 > a2 ⇒ ∆W1 < ∆W2. (14) From the data in (70) of Appendix A, we get ∆Wi ∼ W0 ∼ O(102), thus the term ∆Wi dominates the ratio (11), and hence c2 is always exponentially smaller than c1. As both c1 and c2 are at most at the order of O(q), the concentration c2 is exponentially small and negligible in the filter. This means that K+ is favored or selected in filter compared with Na+, and this fact is independent of q. Based on data in (69,70) of Appendix A, we get ∆W1 = 49.2, ∆W2 = 66.6. (15) This implies that, the term ∆Wi dominates in (11,12) unless the chamber contraction c2b is 107 times larger than c1b. Since boundary values cib, φb have negligible effect, one can imagine this conclusion holds for non-equilibrium case. One can rigorously prove this by noting µi is monotone in non-equilibrium case with finite flux. The high selectivity of SF for larger K+ over smaller Na+ has been also intensively studied by molecular dynamics (MD) [7, 24, 36, 32] and experiments [12, 20], just to name a few. Experiments show Na+ can block KcsA K+ current from intracellular side but not from extracellular side [12]. This observation is explained by MD studies that Na+ would 8 encounter a much larger energy barrier than K+ when approaching S2 binding site in multi-cation knock-on entering SF from extracellular side [7], and all binding sites are more selective to K+ than Na+ except the internal water cavity site lying at the entrance of SF from intracellular side [24]. 3.1.1 EN case From the data (70,71) of Appendix A, we get q ∼ O(1/δ). When q does not exceed the critical value we have the EN condition in filter q < 1 a3 1δ , c1 + c2 − c3 = q, (16) (17) which provides a nonlinear equation for φ, with the help of (9, 10). In fact, this is a quadratic equation for eφ. The analytic solution involves many exponential large and exponential small terms, and can easily lead to wrong or complex solutions by direct com- puation with softwares (like Mathematica). It's easy to prove that c3 is also exponentially small. As a leading order approximation, we get c∗ k = 0, c∗ 1 = q, φ∗ = −∆W1 + log c1b + φb + log(1 − a3 k = 2, 3, 1qδ) − log q. (18) which can also be obtained directly from analytical solution by keeping essential expo- nential small terms and dropping high-order exponentially small terms. Remark: In above analysis, by EN condition we mean that it is valid in most middle part of filter region. Actually, near the two edges of filter (or interface of filter and chamber), say x = ±s, there is a tiny boundary layer due to large q and small r in (5)1, where the variation of φ, ci is quite large. In the approximation (18), only some exponentially small terms are dropped, so the expressions are accurate enough. Figure 2 shows the dependence of above solution on q, with c1b = 1, φb = 0 and some data in (70) of Appendix A. The value of c2b (assumed as O(1)), the profiles of A(x) and r(x), as long as r(0) = 1/40, will not affect the above approximation in filter. This will be verified in direct numerical simulations. Based on selected values of parameters, the 1δ ≈ 790 in Figure 2. When q is near 0 or near this critical value, critical value is q = 1/a3 9 Figure 2: Dependence of φ∗, c∗ 1, c∗ 2 in filter on permanent charge q. the above solution is not valid, as indicated by some singularity in Figure 2a. When q exceeds the critical value, one should solve the full equation (5)1 in filter instead of EN condition, as we see in next subsection. For q > 1/a3 1δ, the subfigure in Figure 2a (the minimum of φ) and the curve in Figure 2b are based on next subsection. Figure 2b shows the selectivity of K+ and Na+ in filter. 3.1.2 Non-EN case For this case, we can not use EN condition and instead we should solve the full equation (5). Since length of filter is at the same scale of boundary layer in classical PNP of chamber region, we will introduce a new scale X = x/ to study the system. For the equilibrium case, the equation for φ is − 1 A(X) [r(X)A(X)φ(cid:48)(X)](cid:48) = n(cid:88) k=1 zkck − q(X), −∞ < X < ∞, (19) where prime denotes the derivative with respect to X. In above, we consider a relatively long chamber region, so the domain is set as ∞ as approximation (this causes essentially no difference). The position of interface between filter and chamber is X = S ≡ Lf /2, where S ∼ O(1). For simplicity, we consider a simple geometry (see Figure 3) that the cross section area A(X), the fixed charge q(X) and the relative permittivity r(X) are constants in either chamber or filter region. We denote A(X) = Af , r(X) = r0, q(X) = q, −S < X < S, A(X) = 1, r(X) = 1, q(X) = 0, X > S. (20) 10 2004006001/a13δ1000q-60-55-50ϕ850900950-100-300-500K,c1Na,c22004006001/a13δ1000q200400600800ciFigure 3: Sketch of the potassium channel with simplified geometry. Note that some typical values are Af = 1/30, q = 103, r0 = 1/40, (21) which will be used to show the results. Because of symmetry, we only consider the interval X ∈ [0,∞). For the chamber region, equation (19) reduces to classical Poisson-Boltzmann equation by neglecting the O(δ) term, − φ(cid:48)(cid:48)(X) = e−φ − eφ, S < X < ∞, where we have assumed the boundary conditions at ∞ φ(∞) = 0, c1(∞) + c2(∞) = c3(∞) = 1. It is easy to get √ φ(cid:48) = 2(e−φ/2 − eφ/2), S < X < ∞, and hence obtain the solution φ(X) in chamber region (see (75) in Appendix B). In filter, we have from equation (19) and symmetry condition that − r0φ(cid:48)(cid:48)(X) = c1 + c2 − c3 − q, φ(cid:48)(0) = 0, at X = 0, 0 < X < S, (22) (23) (24) (25) where ci (i = 1, 2, 3) are given by (9, 10). One can easily prove that the function φ(X) is monotonically increasing throughout the interval [0,∞), since c2 + c1 < q in filter. For filter region, since c2 is exponentially small (see the analysis below (11)), it can be neglected. In addition, by expression of c3 in (9,10) and some data (70) in Appendix A, we get c3 < eB3−W3(0)+φ, B3 − W3(0) ≈ −∆W3 ≈ −37.5. (26) 11 chamberfilterA=1qAfchamberFrom the fact that φ is increasing, we get that φ < 0, thus c3 is always exponentially small and can be neglected. Therefore, the filter equation (25)1 is simplified to By integration, we easily get √ − r0φ(cid:48)(cid:48)(X) = c1 − q, c1 = eB1−W1(0)−φ 1 + δa3 1eB1−W1(0)−φ . r0φ(cid:48) =(cid:112)2(G(φ) − G(φ0)), (cid:114)r0 (cid:112)G(φ) − G(φ0) (cid:90) φ φ0 1 2 X = dφ where φ0 ≡ φ(0) is to be determined, and the function G(φ) is given by G(φ) = q − c1dφ = qφ + 1 a3 1δ log(1 + a3 1δeB1−W1(0)−φ). (cid:90) φ At interface X = S, we have φ(S−) = φ(S+), r0Af φ(cid:48)(S−) = φ(cid:48)(S+). Denote φ(S±) = φs, then the two quantities φ0, φs are determined by Af (cid:112)r0(G(φs) − G(φ0)) = e−φs/2 − eφs/2, (cid:114)r0 (cid:90) φs (cid:112)G(φ) − G(φ0) dφ = S. 2 φ0 1 (27) (28) (29) (30) (31) Once they are found, we get the explicit solutions for filter and chamber. Figure 4: The profiles for case q = 1000 > 1/a3 1δ: (a) φ in both filter and chamber, (b) φ in chamber, (c) c1 in right-half interval of filter. Figure 4 shows the profiles of φ(X) and c1(X), with values in (21, 70) and c1(∞) = 1. In filter region, Figure 4(a) shows that the minimum value of φ is much smaller than the 12 0.51-0.5-1X-200-400-600ϕ123-1-2-3X-0.5-1.5-1ϕ0.10.20.3X0400600800c13(cid:88) 1 − δ EN case, and Figure 4(b) shows that φ ∼ O(1) in chamber region. Figure 4(c) shows the profile of c1 in right-half filter region, indicating that c1 = 1/a3 1δ in most middle part of filter and there is a inner transition point from exponential small to that value. This means that in most part of filter, it is fully packed a3 i ci = 0, (32) but it still needs the derivatives φ(cid:48)(cid:48)(X) to balance the large q. The solutions (e.g., mini- mum φ0 and interface value φs) are most influenced by dimensionless quantities Lf (po- i=1 sition S), Af and r in filter. Figure 5: The profiles for case q = 600 < 1/a3 1δ: (a) φ in both filter and chamber, (b) φ in chamber, (c) c1 in right-half interval of filter. Remark: In Section 3.1.1, we only considered the constant solution φ = φ∗ in middle part of filter. Actually the constant φ∗ is connected to the chamber by a standard boundary layer (BL) in filter and near two edges. The solution in filter can be easily constructed similar to above analysis, and is given (cid:114) r0 (cid:90) φ 2 φs (cid:112)G(φ) − G(φ∗) 1 X = S + dφ, 0 < X < S, (33) where φs determined by (31)1 with φ0 = φ∗ there. The results for φ and c1 are shown in Figure 5 for the case q = 600 < 1/a3 1δ with other parameters as before. One clearly see the typical BLs of φ near two edges in filter. 13 0.51-0.5-1X-20-40-60ϕ123-1-2-3X-0.2-0.5-0.8ϕ0.10.20.30.4X0100300400500600c13.2 K+/Ca2+ selectivity In this subsection, we consider the case with three ions K+, Ca2+ and Cl− (respectively c1, c2 and c3), and study the selectivity between K+ and Ca2+ (or Ba2+). In this case, one can not directly analyze the ratio c1/c2 anymore, since they have i in ∆Wi in (13), the barrier ∆W2 ≈ 274 for Ca2+ difference valences. Due to the factor z2 is much larger. Now, we consider the EN case in filter c1 + 2c2 − c3 = q. (34) With the help of (9, 10), this is a cubic equation for eφ and once φ is solved all ci can be recovered. The analytic solution for φ is quite complicated, and involves many exponentially large and small terms. One can not get right answer unless making proper approximations in different situations by keeping only leading exponential terms and neglecting high-order exponential terms. There are two situations. When q satisfies (16), we get the same approximation as in (18). When q is relatively large, 1 a3 1δ < q < 2 a3 2δ , 2 − a3 2qδ 1 − a3 (2a3 2)δ we get the leading-order approximation 1qδ − 1 a3 1 − a3 (2a3 2)δ c∗ c∗ 3 = 0, 1 = φ∗ = B2 − W2(0) − (B1 − W1(0)) − log c∗ 2 + log c∗ = ∆W1 − ∆W2 + log c2b − log c1b + φb − log(a3 , , c∗ 2 = 1 1qδ − 1) + log(2 − a3 2qδ). (35) (36) In above, φ∗ depends on the calculated c∗ 2, thus the size effect on φ∗ is through these two quantities. One can see that the boundary conditions affects φ∗, but have negligible 1 and c∗ influence on the selectivity. The conclusion on selectivity also applies to non-equilibrium case. Remark: In above approximation (36), c∗ 1 and c∗ 2 are determined by the constraints c∗ 1 + 2c∗ 2 = q, 1c∗ 2c∗ 1 + a3 δ(a3 2) = 1. (37) This implies that EN condition is satisfied, and at the same time SF is saturated with K+ and Ca2+. These two combined effects determine concentrations of K+ and Ca2+. It further implies, in the case of (35), the concentration of K+ itself can not balance q in 14 SF, and SF needs to recruit Ca2+ (by squeezing out some K+ at the same time) to help out the electrostatic balancing since Ca2+ has a larger valence in spite of its larger Born solvation energy as well. Figure 6: Dependence of φ and ci (i = 1, 2) in filter on charge q. Figure 6 shows the dependence of above solution (36) on q, with c1b = c2b = 1, φb = 0 and some data (70) in Appendix A. The first part of the curves is the same as Figure 2, 1δ ≈ 790. When q exceeds this critical value, the concentration of Ca2+ when q < 1/a3 increases while that of K+ decreases. When q crosses the critical value, the constant φ∗ in filter transits from previous state at about -55 to another state at about -225, see the embedded figure in Figure 6a. Based on the data (70) in Appendix A, the next critical value for saturation of c2 is q = 2/a3 2δ ≈ 4280. Figure 6 does not reach this value. The Barium Ba2+ has been used to block K+ channel for a long time [33]. The size of Ba2+ is larger than Ca2+, given in Appendix A. Since it also has +2 valence, the energy barrier (∼ 201.2) is still much larger than that of K+. The above analysis will not change, Baδ ∼ 1688. Figure 7 shows the results and in this case the critical value is q = 2/a3 and dependence on q, with same data as before. Figure 7b indicates that Ba2+ is more effective to block K+ due to larger size. Ba2+ specifically blocks K+ channels via electrostatic stabilization in the permeation pathway. At high concentrations of external K+, the block-time distribution of Ba2+ is double exponential, implies at least two Ba2+ binding sites in SF [33]. This coexistence of Ba2+ and K+ inside SF was also observed in MD computation [6] with Ba2+ at binding site S2 and K+ at binding site S0 forming a lock-in state impeding the translocation of Ba2+ [34]. 15 4001/a13δ120016002000q-50-100-150-200ϕ120016002000q-225.5-225.0-224.5-224.0-223.5ϕK,c1Ca,c24001/a13δ120016002000q500800ciFigure 7: Dependence of φ and ci (i = 1, 2) in filter on charge q. Remark: The above analysis is also valid for the case with four ions: K+, Na+, Ca2+ and Cl−. Based on the analysis in Section 3.1, the concentration of Na+ is always exponentially smaller than K+. Thus, K+ is favored compared with Na+, and adding Na+ will make no difference. The non-EN case will not be discussed here, since for relatively large q the two ions K+ and Ca2+ can coexist to maintain EN. For even larger q > 2/a3 2δ or near transition point q = 1/a3 1δ, we need to consider the non-EN case. The analysis is similar to Section 3.1.2, except that we have a more complicated G(φ) in (29) for this case. 4 Non-equilibrium case and flux-voltage relation In this section, we assume the same concentrations at two ends of chamber but with different electric potential. Then there is variation in electro-chemical potential µi across interval x ∈ [−1, 1], and we intend to study the flux-voltage relations at steady state for previous two cases. This section is restricted to relative long chamber region (length L), where some analytical flux-voltage relations are available. For general cases, numerical or semi-analytical solutions will be shown in the next section. 4.1 Fluxes of K+/Na+ case In this subsection, we consider the three-ion case with K+,Na+ and Cl−. At two ends x = ±1, we impose c1(±1) = 1, φ(−1) = V, φ(1) = 0. c2(±1) = c2b, c3(±1) = 1 + c2b, (38) 16 4001/a13δ120016002000q-50-100-150ϕ12001500q-153.5-153.0-152.5-152.0-151.5-151.0ϕK,c1Ba,c24001/a13δ12001600q500800ciIn this case, the results in Section 3.1 about selectivity of K+ and Na+ are still valid. Although Bi in (10) is not an exact constant anymore, the variation is small since µi is monotone. We have also pointed out in Section 3.1 that c2 is exponentially small unless c2 is 107 times larger than that of c1 near filter. Based on results on selectivity, now we study the relative variation ∆µi for each µi (i = 1, 2, 3) in chamber and filter. Since in chamber it is almost the classical PNP system, we get ci ∼ O(1), implying ∆µi = O(Ji), in chamber. (39) In filter, we have either ci ∼ O(q) or ci is exponentially small. Since the filter interval is small, as a first approximation, we have ∆µi ≈ Lf Ji Af c∗ i , in filter, (40) where Lf and Af are dimensionless quantities already. We know that the total variation (sum of above two, (39) and (40)) from left end to right end is O(1) with V ∼ O(1). From Section 3.1, we have c1 ∼ O(q) in filter, and then we get the estimate from some data (70,71) in Appendix A (41) This implies that J1 ∼ O(1), and the filter region can be neglected for variation of µ1. On the other hand, c∗ 3 are exponentially small in filter, thus J2 and J3 can only be 2 and c∗ ∼ 10−3 − 10−2. Lf Af c∗ 1 exponentially small, but this still gives finite variation ∆µ2, ∆µ3 in filter by (40). In this context, we can treat J2 = J3 = 0 when studying the chamber region, and therefore we only need to concentrate on J1-V relation. In the chamber, it is eligible to use the EN condition as first approximation for relative long chamber. We take constant cross section A(x) = 1 for illustration. By neglecting O(δ) term, we get the classic system 1(x) + c1φ(cid:48)(x) = −J1/D1 ≡ −J, c(cid:48) 2(x) + c2φ(cid:48)(x) = 0, c(cid:48) 3(x) − c3φ(cid:48)(x) = 0, c(cid:48) c1 + c2 = c3. (42) This can be solved explicitly for left half chamber −1 < x < 0 and right half chamber 0 < x < 1, given in Appendix B. Here x = 0 is treated as filter. By the continuity of µ1 17 at filter, we get (see Appendix B) (cid:18)(1 + c2b − J/2)2 1 + c2b log (cid:19) + V = log (cid:18)(1 + c2b + J/2)2 1 + c2b (cid:19) , − c2b (43) − c2b which provides the J-V relation. This can be obtained by solving a quadratic equation, and we select the reasonable root that satisfies J = 0 at V = 0, √ 2(1 + c2b)(1 + eV ) − 2 1 + c2b eV − 1 J = (cid:112)4eV + c2b(1 + eV )2 . (44) (45) For the special case c2b = 0, we have 2(eV /2 − 1) (eV /2 + 1) . J = The general case of A(x) causes no essential problem (see Appendix B), and finally we get (cid:90) 1 1 A(s) Lf /2 √ 2(1 + c2b)(1 + eV ) − 2 1 + c2b eV − 1 J (46) Since Lf ∼ O(), for special case A(x) = 1, this factor after J degenerates to 1−Lf /2 ∼ 1. ds = . (cid:112)4eV + c2b(1 + eV )2 Figure 8: Flux-voltage J-V relations with different boundary concentrations c2b. Remark: We have used EN condition in above system, which causes an O(J) error in estimate of the variation ∆µ1, due to classical BL near filter edge in chamber (see [39]). Also in (44), there is O() error by treating the filter as a point x = 0 as filter length is O(), but in (46) the exact point x = Lf /2 of filter edge is used. Later numerical calculations show that the above approximation is good for small V , and it slightly underestimates the flux for relatively large V . Figure 8 shows the J-V relations (note J1 = D1J) in (46) with A = 1 and different boundary concentrations c2b. It indicates that the flux J tends to saturate for relatively 18 c2b=0c2b=0.1c2b=0.5-10-5510V-2-112Jlarge V (the reason will be illustrated in later section), which agrees well with experimental measurements [32]. The presence of Na+ reduces the flux of K+ with the still tendency to saturate at large V. These generally agree well with experiment measurements in [32] except that there is a dip in experimental IV curves at moderate V corresponding to the blockage by Na+ and it becomes relieved at high V by a "punch-through" mechanism. The failure to predict the dip of IV curve caused by Na+ is due to the limitation of current analysis. Na+ is expected to bind at the water cavity site near the intracellular entrance of SF, and this peak of Na+ concentration at water cavity site is totally overlooked by current asymptotic analysis which assumes EN over there. Figure 9 shows the profiles of φ(x) and ci(x) (i = 1, 2, 3) with boundary values c2b = 0.1, V = 1 in (38) and parameter q < 1/a3 1δ is for illustration purpose as φ in most part of filter is approximated by φ∗. The exact values of q and φ∗ are not used in Figure 9 because they are so large, and the red dashed vertical lines mean a big jump to the two values. For larger q > 1/a3 1δ. The choice of q < 1/a3 1δ the results will not change much except that φ in filter has a profile like Figure 4(a). Figure 10 shows the profiles of µi(x) (i = 1, 2, 3) for each ion species. There is finite variation for µ1 in chamber, which causes the finite flux of c1. The µ2 and µ3 are constant in chamber, leading to 0 fluxes. Even though there's finite variation for µ2 and µ3 in filter, there's no flux since the concentrations c2 and c3 are essentially 0 in filter. Now we summarize the strategy for determining J-V relations, which also applies to other cases like next subsection. • from the equilibrium case, determine which ions (here K+, next subsection K+ and Ca2+) are prevalent in filter and which (here Na+ and Cl−) are 0 in filter. • set finite flux for only those ions prevalent in filter and set 0 flux for others, and then solve the chamber equations for left and right chamber regions • determine the J-V relations by using continuity of µi at filter for only those ions prevalent in filter (note that other µi are constant in chamber and have jumps at filter). It appears that we have only used chamber equations to approximation the J-V relations, but actually it is totally different to directly solve chamber equations without the filter 19 Figure 9: Profiles of φ(x) and ci(x) (i = 1, 2, 3) with c2b = 0.1, V = 1. Figure 10: Profiles of µi(x) (i = 1, 2, 3) with c2b = 0.1, V = 1. since all fluxes and variation of all µi would be finite and continuous in that case. It is also clear that, for the present case with filter, the chamber solutions of φ, ci (i = 1, 2, 3) in Figure 9 have jumps at filter, and the µ2 and µ3 in Figure 10 are constants in each chamber. 20 -1-0.50.51xϕ0.51ϕ-1-0.50.51x0.51qc1-1-0.50.51x0.10.05c2-1-0.50.51x0.51c3-1.0-0.50.51.0x-99.6-99.4-99.2-99.0-98.8μ1-1.0-0.50.00.51.0x-137.0-136.8-136.6-136.4-136.2μ2-1.0-0.50.00.51.0x-76.8-76.6-76.4-76.2-76.0μ34.2 Fluxes of K+/Ca2+ case In this subsection, we consider the three-ion case with K+,Ca2+ and Cl− (the case for Ba2+ is similar). At two ends x = ±1, we impose c2(±1) = c2b, c1(±1) = 1, φ(−1) = V, φ(1) = 0. c3(±1) = 1 + 2c2b, (47) The analysis on the variation of ∆µi (i = 1, 2, 3) are similar to the preceding sub- section, and we can follow the preceding strategy to determine the flux-voltage relations. Depending on the parameter q and results in Section 3.2 about selectivity of K+ and Ca2+, there are two cases. (1) When q < 1/a3 1δ, we get J2 = 0, J3 = 0 and finite J1. Then the results of J1-V relation will be similar to preceding subsection, and the profiles of ci and µi are similar. (2) When 1/a3 1δ < q < 2/a3 2δ is relatively large as in (35), we have J3 = 0 and finite J1 and J2, since both K+ and Ca2+ can exist in filter. Figure 11: Flux-voltage J-V relations with c2b = 1 and c2b = 10−3. Now we focus on the second case and take A(x) = 1 for illustration. We solve the following system in chamber ∂xc1 + c1∂xφ = −J1/D1 ≡ J1, ∂xc2 + 2c2∂xφ = −J2/D2 ≡ J2, ∂xc3 − c3∂xφ = 0, c1 + 2c2 − c3 = 0. (48) It is not easy to solve φ(x) and ci(x) (i = 1, 2, 3) directly, instead if we treat φ as the independent variable, we can solve x(φ) and ci(φ) (i = 1, 2, 3) explicitly. We denote 21 J1J2-6-4-2246V-3-2-1123J-6-4-2246V-0.003-0.002-0.0010.0010.0020.003J2solutions by xR(φ), ciR(φ) for the right half interval 0 < x < 1 and by xL(φ), ciL(φ) for −1 < x < 0, given in Appendix B. Then, by the continuity of µ1 and µ2 at x = 0, we get φ0L + ln c1L(φ0L) = φ0R + ln c1R(φ0R), 2φ0L + ln c2L(φ0L) = 2φ0R + ln c2R(φ0R), (49) where φ0L and φ0R are left and right limit values of φ at x = 0, which are defined by xL(φ0L) = 0, xR(φ0R) = 0. (50) All these four equations involve the fluxes J1, J2 and V , thus they determine J1, J2, φ0L, φ0R in terms of V . The general case of A(x) needs only slight modifications, see Appendix B. Figure 12: Profiles of φ(x) and ci(x) (i = 1, 2, 3) with c2b = 1, V = 1 and 1/a3 1 < q < 2/a3 2δ. Figure 11(a) shows flux-voltage J-V relations with c2b = 1, indicating that both fluxes J1 and J2 saturate for relatively large V . Figure 11(b) shows the flux J2 when c2b = 10−3 is set very small, indicating the flux almost proportionally gets smaller as In Figure 11(b), the flux J1 is omitted since it chamber concentration gets smaller. is almost the same as in 11(a) and in much larger scale. Figure 12 shows the profiles of φ(x) and ci(x) (i = 1, 2, 3) with boundary values c2b = 1, V = 1 and parameter 22 -1-0.50.51xϕ0.51ϕ-1-0.50.51x0.51c1c1-1-0.50.51x0.51c2c2-1-0.50.51x1234c3Figure 13: Profiles of µi(x) (i = 1, 2, 3) with c2b = 1, V = 1 and 1/a3 1 < q < 2/a3 2δ. 1/a3 1 < q < 2/a3 2δ. Figure 13 shows the profiles of µi(x) (i = 1, 2, 3) for each ion species. The finite variation of µ1 and µ2 in chamber causes the finite flux of c1 and c2, while µ3 is constant in chamber. 5 Computational analysis In this section, we solve the modified PNP system numerically. Our main objective is to verify our asymptotic analysis under simplifying conditions. Figure 14: Smooth functions A(x) and r(x) used in simulation. We use the dynamic process to simulate the steady state solutions for φ, ci and associ- ated fluxes. Some smooth dimensionless functions r(x) (connecting 1/40 and 1) and A(x) (connecting 1/30 and 1) will be used in the simulation, see Figure 14. Now we illustrate it by considering the 3-ions case with ci (i = 1, 2, 3) for K+, Na+, Cl−. This is to verify previous analytical results for both equilibrium and non-equilibrium cases. We adopt the initial conditions at t = 0, c1(x, 0) = 1, c2(x, 0) = 0.1, c3(x, 0) = 1.1. (51) 23 -1.0-0.50.00.51.0x-99.6-99.4-99.2-99.0-98.8μ1-1.0-0.50.00.51.0x-555.5-555.0-554.5-554.0μ2-75.8-75.6-75.4-75.2-75.0μ3-1-0.500.51x00.20.40.60.81A-1-0.500.51x00.20.40.60.81rThe boundary conditions are c1(±1, t) = 1, φ(−1) = V, φ(1) = 0. c2(±1, t) = 0.1, c3(±1, t) = 1.1, (52) Figure 15: Profiles of φ and ci (i = 1, 2, 3) near steady state for V = 0 and q = 600. Figure 16: Profiles of φ at steady state for V = 0 and different q, and comparison with analytical results in Figure 2. First, we set V = 0 and compare the numerical results with analytical results in (18) (or Figure 2). A series of cases with different q will be simulated. In the simulation, finite-volume method is used with non-uniform mesh points. More mesh points are used in filter, near filter edge, and in regions for large gradient of r, and there are totally 273 points. Very small time step (because of large q, small and small mesh size) is 24 -1-0.500.51x-60-50-40-30-20-100Numerical solutionAnalytical solution-1-0.500.51x0100200300400500600c1Numerical solutionAnalytical solution-1-0.500.51x00.511.522.533.5c2Numerical solutionAnalytical solution-1-0.500.51x00.20.40.60.811.2c3Numerical solutionAnalytical solution-1-0.500.51x-60-50-40-30-20-10010q=300q=400q=500q=600-0.0500.05-57-56-55-54AnalyticalNumerical2004006001/a13δq-60-55-50ϕK,c1Na,c2AnalyticalNumerical2004006001/a13δ1000q200400600800cichosen to ensure stability and accuracy of algorithm. After quite a long time, about 20 h on a computer (processor: 4 GHz, i76700K; memory: 32 GB), the solution tends to some steady state (i.e., all fluxes are almost 0). The profiles of φ and ci (i = 1, 2, 3) for q = 600 are shown as red curves in Figure 15, in comparison with the analytical results in blue curves from Section 3. The numerical and analytical solutions agree very well except a smoothing region near two edges of filter. For instance, the constant values of φ in filter show remarkable agreement, i.e., φ∗ = −57.0257,−57.0268 in numerical and analytical results. One can see that K+ is favored in the filter region, and all the other ions are essentially 0 in filter region. This agrees with results in (18). To see clearly the dependence on q, the profiles of φ for different q are shown in Figure 16a, showing that it is constant in filter region. In Figures 16b and 16c, the constant values of φ, c1, c2 in filter are compared with previous analytical results, where curves are from previous Figure 2 and dots are from numerical results. We have also tested different smoothing profiles of r(x) and A(x) and boundary conditions for concentrations, and as long as r is 1/40 (original value is 2 before scale) in part of filter, the minimum values of φ will not change. This also verifies the predictions in (18). Figure 17: The fluxes Ji (i = 1, 2, 3) near steady state for V = 1, q = 600. To test the analysis of non-equilibrium case, we set V = 1, q = 600 and others the same as above. After computation of about 20 h to t = 2, the system tends to some steady state. The fluxes are shown in Figure 17, indicating that only flux J1 is nonzero and goes to a constant 1.167 at steady state. This feature agrees with previous analysis. The previous predicted flux by formula (46) with A(x) = 1, c2b = 0.1 is J ≈ 0.51, and hence J1 = D1J ≈ 1. They differ by an O() with present ≈ 0.13, as it is natural for previous approximation. In addition the difference is partly due to the the smoothing of 25 -1-0.500.51x1.16711.16721.16731.16741.16751.16761.16771.1678J1-1-0.500.51x-1-0.500.51J210-3-1-0.500.51x-6-4-20246J310-4Figure 18: The profiles of ci (i = 1, 2, 3) and φ near steady state, for V = 1, q = 600. r(x) and A(x). The profiles of ci, φ (i = 1, 2, 3) are shown in Figure 18. Some features are similar to the equilibrium case, but the profiles are not symmetric anymore. The profiles of µi (i = 1, 2, 3) are shown in Figure 18. The numerical solutions in chamber are also compared with the previous analytical solutions (see Figure 9 and Appendix B) in dashed lines of embedded figures. All the profiles of ci, φ, µi (i = 1, 2, 3) except µ3 show agreement with previous analytical results. We have also tested different V , and compared with analytical flux-voltage curves in next section. Figure 19: The µi (i = 1, 2, 3) near steady state for V = 1, q = 600. Now we provide some insight and explanation for above wrong µ3, based on previous 26 -1-0.500.51x0100200300400500600c1Numerical solutionAnalytical solution-1-0.8-0.6-0.40.511.50.20.40.60.811.52-1-0.500.51x0123456c2Numerical solutionAnalytical solution-1-0.8-0.6-0.40.10.150.20.250.20.40.60.80.080.10.120.14-1-0.500.51x00.20.40.60.811.21.4c3Numerical solutionAnalytical solution-1-0.500.51x-60-50-40-30-20-10010Numerical solutionAnalytical solution-1-0.8-0.6-0.400.510.20.40.60.8-0.4-0.200.2-1-0.500.51x-99.8-99.6-99.4-99.2-99-98.8-98.61Numerical solutionAnalytical solution-1-0.500.51x-137.2-137-136.8-136.6-136.4-136.2-1362Numerical solutionAnalytical solution-1-0.500.51x-80-70-60-50-40-30-20-103Numerical solutionAnalytical solutionanalytical results. We can easily prove that µ3 is monotone in steady state by the positivity of c3. Thus, Figure 10c is correct and direct numerical result in Figure 19c is wrong. By definition of µ3 and matching with boundary conditions (two values of µ3 at boundaries do not differ much), in filter we approximately have (cid:32) 1 − 3(cid:88) log c3 − log (cid:33) ck(x)a3 kδ ∼ −37.5 + φ ∼ −90 (53) k=1 Since the second term is O(1) for present q = 600 not exceeding the critical value 790, we need c3 to be as accurate as e−90 ∼ 10−40. We know it is almost 0, but to compute correct µ3 in filter, it has to go to as small as 10−40. This is partially verified numerically, i.e., when we increase the accuracy of c3 in filter, the values of µ3 in filter as in Figure 19c will decrease further (in both cases V = 0 and V = 1). In addition, the accuracy of c3 would also affect other results in filter, to certain degree. For example, if we only keep accurate up to 10−10, the minimum values of φ are wrong (differ much from analytical results), and it works for φ when we keep accurate up to 10−15 (Figure 16(b) is based on this). The inaccuracy of µ3 is also one reason that the profile of c3 in Figure 15 has a relatively larger discrepancy with analytical resutls. When Ca2+ is present and with above q = 600, the results and features are very similar to above results (omitted here), and this agrees with previous analysis. We also tried for large q = 1000 in above 3-ion case and in a case with Ca2+, but the computation is very unstable and failed to capture the features in analysis. Now we give some explanation based previous analysis and provide some insight on the numerical difficulty. In such cases, the ions saturate in filter and thus the second term in µi of Eq. (6) is crucial and requires very high accuracy for ci in computation. Take the 3-ion case with Ca2+ in Section 3.2 for example, one can see that even for the simple case of algebraic equations from (34) and (9, 10), it is not straightforward to determine φ. Originally, the solution depends on identity (10), and from the solution in (36) we find that in this case ck(x)a3 kδ ∼ −170. (54) k=1 This causes the main difficulty of direct numerical simulation, as this term is essential to capture the behaviour in filter. One should be very cautious to calculate ci directly in simulation, since both Ca2+ and K+ are in the order O(q) but they need to be accurate to 27 (cid:32) 1 − 3(cid:88) log (cid:33) e−170 to capture this term. Other difficulty can also come from log ci terms, as some ion like Cl− is exponentially small (this is already illustrated in last paragraph for previous case). These difficulties can be avoided if the ci can be represented by φ, as φ behaves good in analysis and computation. This can be easily done for equilibrium case with help of formula (9,10), but not straightforward in dynamic case. We also briefly mention the 3-ion case of K+, Na+, Cl− with large large q, as in Section 3.1.2. It is similar for the difficulties from the two log terms in µi in (6). In addition, analytical solution or Figure 4c shows that there is an internal transition point for c1 in filter, where c1 changes from exponential small to O(q). In some part of filter, on the one hand c1 is exponential small, and on the other J1 ∼ c1∂xµ1 should be finite. It is not easy to capture the transition or to compute the form 0 ∗ ∞. 6 Hybrid computational-asymptotic analysis When q is large, direct numerical computation becomes challenging and inefficient. In addition, when is relatively large (i.e., short chamber length L), our analysis for the J-V relation in Section 4 fails since the EN assumption is no longer valid in the chamber. In this subsection, we provide an alternative hybrid method by combining asymptotic analysis in the filter with numerical computation in the chamber. We obtain an analytical solution in filter for non-equilibrium case by slightly modifying that from Section 3.1.2, and in the chamber we can simplify the system, which is generally easy to solve numerically (no such difficulties mentioned in last subsection) or relates to some special functions. We could also call the solutions in the subsection as semi-analytical solutions. We take the three ion case K+, Na+ and Cl− as illustration, and assume A = 1 and r = 1 in the chamber (the general case should not cause essential difficulty). The dimensional length can be either large or small (reflected in parameter ), say L = 10.5 nm in previous sections or L = 3 nm in more practical case. The system in the right 28 chamber by neglecting O(δ) term is 1(x) + c1φ(cid:48)(x) = −J, c(cid:48) 2(x) + c2φ(cid:48)(x) = 0, c(cid:48) 3(x) − c3φ(cid:48)(x) = 0, c(cid:48) − 2φ(cid:48)(cid:48)(x) = c1 + c2 − c3, s < x < 1. (55) with boundary conditions ci = cib, φ = 0 at x = 1. Here position s denote the edge of filter. We immediately get c2, c3 in terms of φ c2 = c2be−φ, c3 = c3beφ, so that − 2φ(cid:48)(cid:48)(x) = c1 + c2be−φ − c3beφ, s < x < 1. Multiplying φ(cid:48) on this equation and using (55)1, we obtain (56) (57) c1(x) = 2 1 2 [(φ(cid:48)(x))2 − (φ(cid:48)(1))2] − J(x − 1) + c1b − c2b(e−φ − 1) − c3b(eφ − 1). (58) Substituting into equation (57) and with c1b + c2b = c3b, we obtain 2φ(cid:48)(cid:48)(x) = −1 2 2[(φ(cid:48)(x))2 − (φ(cid:48)(1))2] + J(x − 1) + 2c3b(eφ − 1), s < x < 1. (59) Similarly for the left chamber with boundary conditions ci = cib and φ = V , we would have 2φ(cid:48)(cid:48)(x) = −1 2 2[(φ(cid:48)(x))2 − (φ(cid:48)(−1))2] + J(x + 1) + 2c3b(eφ−V − 1), −1 < x < s. (60) These two equations are to be solved with help of solution in filter or with some matching connection conditions. Remark: The final differential equation for φ seems complicated, but actually it relates to a special function, defined by Painlev´e II (PII) equation. Here we would like to bring attention to this connection, as Painlev´e transcendents have been studied intensively in last decades [3]. The reduction of steady state PNP system with ±1 ions to PII equation was mentioned in [35]. For the present 3-ion case, it is similar and we can adopt the transform y = eφ/2 √ 2(J)1/3 , z = Jx + C 2(J)2/3 , C = −J + 1 2 2(φ(cid:48)(1))2 − 2c3b, (61) 29 so that equation (58) becomes PII equation with parameter 0, y(cid:48)(cid:48)(z) = 2y3 + zy. (62) The typical solutions in present setting is that φ(x) either blows up to ∞ or to −∞ at x = x∗ as x decreases from 1, and this agrees with some features (like poles) of solutions of PII equations. But the reasonable solution in current case is connected to the filter solution at x = s before it reaches x∗. Next we would like to connect above solutions in chamber with filter solution. We take q > 1/a3 1δ for example. In general, for non-equilibrium case, one can not express ci in terms of φ and then directly construct the solution like Section 3.1.2. But we make use of the facts that Eq. (9) still holds in non-equilibrium case. In addition, for selected ions (K+ or K+ and Ca2+), µi are constants for filter region based on evidence from both analysis and simulation. Thus, the only modification of filter solution in (28, 29) is that the constant B1 is replaced by µ1(s), which relates to chamber solution. We can determine the solutions by using shooting method. Once we fix J and φ(cid:48)(1), we can compute the solution of φ and hence ci (i = 1, 2, 3) upto x = s. We treat the solution as a special function of arguments J, φ(cid:48)(1). With calculated B1 = µ1(s), the filter solution is known. Then, the connection conditions at x = s are (cid:112)2r0(G(φs) − G(φ0)) = φ(cid:48)(s), φs = φ(s), (cid:114) r0 (cid:90) φs Af dφ = (s − s0)/, (cid:112)G(φ) − G(φ0) 1 2 φ0 (63) (64) where s0 is position of minimum of φ or φ(cid:48) = 0 in filter. Similarly for the left chamber, with given V, J, φ(cid:48)(−1), the get the solutions and then the connection conditions at x = −s (cid:112)2r0(G(φ−s) − G(φ0)) = −φ(cid:48)(−s), φ−s = φ(−s), (cid:114) r0 (cid:90) φ−s Af 1 (cid:112)G(φ) − G(φ0) dφ = (s + s0)/. 2 φ0 Note that we have s0 = 0 for the equilibrium case V = 0, but in general the solution is not exactly symmetric. The final condition is µ1(s) = µ1(−s). (65) In brief, with given boundary value V , we have 7 nonlinear equations for 7 unknowns φ0, φ(±s), φ(cid:48)(±1),J and s0. The case q < 1/a3 1δ is simpler, and we do not need the two 30 integral conditions (64)3 and (65)3 anymore, which are replaced by φ0 = µ1(s) − W1(0) + log(1 − a3 1qδ) − log q. (66) Then, we have 6 nonlinear equations for 6 unknowns φ0, φ(±s), φ(cid:48)(±1),J. Figure 20: The profiles of ci and φ with V = 1 and q = 600. The above algorithm can be easily achieved in Mathematica (or Matlab) with only a few lines of code, the solutions for given V can be computed by finding roots of the 6 or 7 nonlinear equations. The computation is super quick, and the solution is found within seconds on a laptop. This is verified with V = 0, q = 1000 and data in (70), and it coincides with previous results in Section 3.1.2. For previous case q = 600, V = 1 in Section 4.1, the solutions are computed for comparison. The profiles of φ and ci (i = 1, 2, 3) are shown in Figure 20 with dashed lines from previous analytical solution, showing good agreement away from filter. The flux computed here is J ≈ 0.55 (or J1 ≈ 1.08), also indicating that the previous approximation J ≈ 0.51 in Section 4.1 slightly underestimates the flux. Remark: There is a fictitious singularity in the integrals in (63,64), i.e., the integrand is singular at φ = φ0, but the integral is like(cid:82) a 1√ xdx. We have used a little trick in pratical computation to ensure stability and accuracy, i.e., replace φ0 by φ0 + δ0, say δ0 = 10−10 (this would be helpful if one wants to repeat above computation). For quite small (long 0 31 -1.0-0.50.51.0x0.51.01.5c1-1.0-0.50.51.0x0.050.100.150.20c2-1.0-0.50.51.0x0.20.40.60.81.01.21.4c3-1.0-0.50.51.0x-0.20.20.40.60.81.0ϕFigure 21: The J-V relations with small (dimensional length L = 10.5 nm): (a) comparison of different methods (b) different q. dimensional L) and large c2b, the solution of φ is sensitive to boundary conditions φ(cid:48)(±1). It can easily blow up to ±∞, and only a narrow interval of φ(cid:48)(±1) with given J leads to solution of φ in whole interval [s, 1]. For different V , the flux-voltage (IV) relations by three different methods are com- pared in Figure 21a, where red curve is from current section, and dots and dashed lines are from previous numerical and analytical solutions. Although different approximations regarding boundary layer near filter or parameters A(x), r(x) are made, the three meth- ods provide similar results and trend for IV curve. The analytical solution underestimates the flux, due to the neglect of boundary layer near edge, while the slight difference be- tween numerical and hybrid methods are due to the smoothing of r(x) and A(x) used in numerical solutions. For different q, the flux-voltage J-V relations are computed by varying V , shown in Figure 21b, with reference curve from analytical result in Section 4.1. The flux in each curve saturates for large V , and as q increases the flux will increase. The saturation of flux is certainly a consequence of selectivity of filter, which is origi- nally due to parameters r and q. Without filter, the flux-voltage relations will be totally different, as indicated at the end of Section 4.1. With filter, the most important condition is continuity of µi for selected ions. To see the direct reason of saturation of flux for the K+/Na+ case, we analyze the profiles of c1 in chamber for different V , obtained by both analytical and hybrid methods. Figure 22 shows the profiles of c1 with parameters c2b = 0.1, q = 600 and three different V . The dashed lines from analytical results provide 32 NumericalHybridAnalytical12345V0.51.01.52.0Jq=1000q=900q=700q=600Analytical246810V0.51.01.52.0Jreasonable approximation in the region away from filter, but not as accurate as the solid lines near filter, which also capture the BL. Both indicate that c1 approaches 0 near the left edge of filter as V increases, and one can easily see this trend from the analytical expressions in Appendix B. The left edge of filter is important here since the flux is from left to right with positive V (otherwise we should analyze the right edge). As c1 can not be negative, this is the main restriction for the saturation of scaled flux J. Also note that the original flux J1 is controlled by diffusion constant D1, one may think the saturation is related to the diffusion limit [32, 31]. When c1 is near 0 at left edge, there are not enough ions available to go through the filter even with large V . As c2b increases, c1 will be more likely to reach this critical value, resulting in smaller saturation flux J. The reason of saturation of both fluxes for the case with Ca2+ in Section 4.2 is similar, except that the two fluxes are restricted by values of both c1 and c2 at the edge of filter (both approach 0 as V increases). Figure 22: The profiles of c1 in chamber for different V . The hybrid method in this section has the advantages of both efficiency and accuracy for the IV relation. The direct numerical computation is extremely time-consuming, even for one point in the IV curve of Figure 21(a), thus it can hardly be used to compare IV relations with experiments. The hybrid method can produce IV curves efficiently, say 20 min for one smooth curve in Figure 21(b). It also includes the boundary layer effect near filter of edge, and does not have the restriction for parameters (like or length L), in contrast to analytical approximations. Thus it can be readily used to compare with experiments or estimate parameters in the model. The data in (70) of Appendix A corresponds to relatively long dimensional length 33 V=1,HybridV=3,HybridV=10,HybridV=1,AnalyticalV=3,AnalyticalV=10,Analytical-1.0-0.50.51.0x0.51.01.52.02.53.0c1Figure 23: The J-V relations with different q, c2b and relatively large (small dimensional length L = 3 nm). L = 10.5 nm. But in more realistic case, L is much shorter based on molecular structure of KcsA channel. In order to compare with experiments, we adopt the dimensional length L = 3 nm (i.e., Lb = 2.5 nm), which leads to ≈ 0.46. We also compute the J-V relations for c1b = 1 and different c2b and q, shown in Figure 23(a). As c2b increases, the flux will decrease, while the flux will increase as q increases. From the present formulation, the dimensional flux and current are scaled by AbD0c0 L = 6.02 ∗ 106/s, e0AbD0c0 L = 0.96 pA (67) where L = 3 nm is used. Note also J1 = D1J where D1 = 1.96. Figure 23b shows the I-V relations with physical units for q = 1100, which are in similar order to figure 2B of experiment paper [32]. One could also make it more comparable by adjusting other parameters, say the cross section area A(x). The idea in this subsection can be applied to more general cases, say general A(x), slowly varying r(x) in chamber or with ion Ca2+. The formulation and solving process are quite similar, except that we might solve more than one equation in chamber region. We will not repeat this here. 7 Concluding remarks We have studied the selectivity of KcsA potassium channel and the current-voltage (IV) relation. With a 1D modified PNP system by keeping essential elements, many fea- tures of the channel have been demonstrated by both analytical formulas and numerical 34 c2b=0,q=1200c2b=0,q=1000c2b=0.1,q=1200c2b=0.1,q=1000246810V2468Jc2b=0,q=1100c2b=10mM,q=110050100200150250300V,mV2468101214I,pAsimulations. The selectivity among K+ and other ions are clearly illustrated with analyt- ical formulas. Saturation of IV curve is captured by various methods, and explanation is provided. We hope these methods in current work can be applied to other types of ion channels, and provide insights into the selectivity and IV relations. More work is needed to make comparison with experiments or calibrate some parameters in the model for different channels. Some feature in detailed 3D simulations such as pile-up of ions near filter may be missed in current 1D framework. This could be due to the boundary charge distribution (instead of local source charge) in filter and complex geometry of the channel. More work under 3D framework is ongoing as an extension of current work. Acknowledgment This work was initiated when Dr. Tzyy-Leng Horng was a Fields Research Fellow at Fields Institute. A Parameter values The data in this Appendix are mainly from [22, 13, 39, 27]. For dimensional system, the vacuum permittivity 0, elementary charge e0, Boltzmann constant kB and absolute temperature T are 0 = 8.854 × 10−12 C/(V · m), kB = 1.38 × 10−23 J/K, T = 300 K. e0 = 1.602 × 10−19 C, Some typical values are adopted as φ0 = kBT e0 ≈ 24 mV, c0 = 100 mM = 6.022 × 1025 m−3, D0 = 10−9 m2/s, a0 = 3 A, Lb = 10 nm, Lf = 1 nm, L = 10.5 nm, rb = 80, 2 rf = 2, Ab = 30 A , (68) (69) aK = 2.76 A, aNa = 2.04 A, aCa = 1.98 A, aCl = 3.62 A, aBa = 2.70 A. If we think of exact sphere instead of cube, the factor (π/6)1/3 ≈ 0.8 should be multiplied to above effective diameters of ions ai. 35 For dimensionless system, we have the estimates of dimensionless parameters δ = a3 0c0 ≈ 1.6 × 10−3, W0 = ≤ r ≤ 1, Af ≤ A ≤ 1, Lf = 0.095, ≈ 0.13, 1 40 e2 8π0a0kBT ≈ 93, (70) DK = 1.96, DNa = 1.33, DCa = 0.79, DCl = 2.03, aK = 0.92, aNa = 0.68, aCa = 0.66, aCl = 1.21, aBa = 0.9. The permanent charge and cross section area are estimated from a 3D Poisson-Boltzmann computation based on realistic molecular structure of KcsA. The corresponding dimen- sionless quantities for q and Af are e.g., [1000, 2000] q ∼ 103, Af ∼ 1 A 30 A 2 2 = 1 30 . B Some solutions and expressions From definition (6), we get then by multiplication of ai and summation, we obtain 1 −(cid:80)n n(cid:88) = i=1 C 1 − Cδ ci k=1 δcka3 k = eµi−Wi−ziφ, i = 1, .., n, i eµi−Wi−ziφ ≡ F, C = a3 n(cid:88) i=1 cia3 i , (71) (72) (73) (74) which implies C = , ci = eµi−Wi−ziφ (1 + F δ) . 1 + F δ The solution of (24) in chamber region is given by F (cid:33) (cid:32) φ(X) = 2 log √ 2X + m e √ 2X − m e , m = √ e 2S(eφs/2 − 1) eφs/2 + 1 , S < X < ∞. (75) For the system (42), we get for the left-half chamber −1 < x < 0 c3(x) = 1 + c2b − J 2 c2b(1 + c2b) c2(x) = c3(x) (x + 1), φ(x) = log c3(x) 1 + c2b + V, c1(x) = c3(x) − c2(x), , (76) 36 and for the right-half chamber 0 < x < 1 c3(x) = 1 + c2b − J 2 c2b(1 + c2b) c2(x) = c3(x) (x − 1), φ(x) = log c3(x) 1 + c2b , c1(x) = c3(x) − c2(x). , Based on the solutions, we get the µ1(x) for left chamber µ1(x) = log c1 + φ + W1 (cid:19) (cid:18) (cid:18) c2 (cid:32) = log = log = log c3(x) − c2b(1 + c2b) + log c3(x) 1 + c2b + V + W1 (cid:19) c3(x) + V + W1 (cid:33) − c2b 3(x) 1 + c2b [1 + c2b − J 1 + c2b 2 (x + 1)]2 − c2b + V + W1, (77) (78) substituting x = 0 give the left-hand side of (43) except the W1 term. For general A(x), the linear terms x + 1, x − 1 in c3(x) in (76,77) should be replaced by (cid:90) x (cid:90) x same except that J is multiplied by a factor(cid:82) 1 A(s) ds, −1 1 1 1 A(s) ds. Lf /2 and all the other expressions are the same. The final result for J-V relation is almost the 1 A(s) ds, (79) The system (48) is equivalent to a system for functions of φ c2 + 2c2 = − J2 x, c1 + c1 = − J1 x, c3 − c3 = 0, c1 + 2c2 − c3 = 0, (80) where dot represents derivative with respect to φ. Then the solutions xR(φ) and ciR(φ) (i = 1, 2, 3) for right-half interval 0 < x < 1 (i.e., φ0R < φ < 0 or 0 < φ < φ0R) are c3R(φ) = (2c2b + 1)eφ, (3c2b J1 − 2 J2)eλφ 3 J1 + 4 J2 c2R(φ) = c1R(φ) = c3R(φ) − 2c2R(φ), + 2(2c2b + 1) J2eφ 3 J1 + 4 J2 , λ = −2( J1 + J2) J1 + 2 J2 , (81) xR(φ) = 1 + 3c2b + 2 J1 + J2 − 6(2c2b + 1)eφ 3 J1 + 4 J2 + (3c2b J1 − 2 J2)eλφ ( J1 + J2)(3 J1 + 4 J2) . The solutions ciL(φ) and xL(φ) for left-half interval −1 < x < 0 (i.e., V < φ < φ0L or φ0L < φ < V ) are ciL(φ) = ciR(φ − V ), xL(φ) = xR(φ − V ) − 2. i = 1, 2, 3, 37 (82) For the general case of A(x), one only needs to make a transformation y =(cid:82) x ±1 right and left chamber equations. The only modifications of above solutions are yL(φ) = yR(φ − V ). yR(φ) = + (3c2b J1 − 2 J2)eλφ ( J1 + J2)(3 J1 + 4 J2) , 3c2b + 2 J1 + J2 − 6(2c2b + 1)eφ 3 J1 + 4 J2 1 A(s)ds for (83) For flux voltage relations, the equations in (49) will not change and the equations in (50) (cid:90) Lf /2 1 change to yR(φ0L) = References (cid:90) Lf /2 1 A(s) ds, yL(φ0L) = 1 −1 A(s) ds. (84) [1] Michael J Ackerman and David E Clapham. Ion channelsbasic science and clinical disease. New England Journal of Medicine, 336(22):1575 -- 1586, 1997. [2] Johan Aqvist and Victor Luzhkov. Ion permeation mechanism of the potassium channel. Nature, 404(6780):881, 2000. [3] Richard Beals and Roderick Wong. Special functions and orthogonal polynomials, volume 153. Cambridge University Press, 2016. [4] Max Born. Volumen und hydratationswarme der ionen. Zeitschrift fur Physik, 1(1):45 -- 48, 1920. [5] William A Catterall. Ion channel voltage sensors: structure, function, and patho- physiology. Neuron, 67(6):915 -- 928, 2010. [6] Declan A Doyle, Joao Morais Cabral, Richard A Pfuetzner, Anling Kuo, Jacqueline M Gulbis, Steven L Cohen, Brian T Chait, and Roderick MacKinnon. The structure of the potassium channel: molecular basis of k+ conduction and selectivity. science, 280(5360):69 -- 77, 1998. [7] Bernhard Egwolf and Benoıt Roux. Ion selectivity of the kcsa channel: a perspective from multi-ion free energy landscapes. Journal of molecular biology, 401(5):831 -- 842, 2010. [8] Allen Flavell, Michael Machen, Bob Eisenberg, Julienne Kabre, Chun Liu, and Xiao- fan Li. A conservative finite difference scheme for poisson -- nernst -- planck equations. Journal of Computational Electronics, 13(1):235 -- 249, 2014. 38 [9] Nir Gavish. Poisson -- nernst -- planck equations with steric effectsnon-convexity and multiple stationary solutions. Physica D: Nonlinear Phenomena, 368:50 -- 65, 2018. [10] Dirk Gillespie and Robert S Eisenberg. Modified donnan potentials for ion transport through biological ion channels. Physical Review E, 63(6):061902, 2001. [11] Eric Gouaux and Roderick MacKinnon. Principles of selective ion transport in chan- nels and pumps. science, 310(5753):1461 -- 1465, 2005. [12] Lise Heginbotham, Meredith LeMasurier, Ludmilla Kolmakova-Partensky, and Christopher Miller. Single streptomyces lividans k+ channels: functional asym- metries and sidedness of proton activation. The Journal of general physiology, 114(4):551 -- 560, 1999. [13] Bertil Hille et al. Ion channels of excitable membranes, volume 507. Sinauer Sunder- land, MA, 2001. [14] U Hollerbach, D Chen, W Nonner, and B Eisenberg. Three dimensional poisson- nernst-planck theory of open channels. In Biophysical Journal, volume 76, pages A205 -- A205. BIOPHYSICAL SOCIETY 9650 ROCKVILLE PIKE, BETHESDA, MD 20814-3998 USA, 1999. [15] Uwe Hollerbach, Duan P Chen, David D Busath, and Bob Eisenberg. Predicting function from structure using the poisson- nernst- planck equations: Sodium current in the gramicidin a channel. Langmuir, 16(13):5509 -- 5514, 2000. [16] Tzyy-Leng Horng, Tai-Chia Lin, Chun Liu, and Bob Eisenberg. Pnp equations with steric effects: a model of ion flow through channels. The Journal of Physical Chemistry B, 116(37):11422 -- 11441, 2012. [17] Tzyy-Leng Horng, Ping-Hsuan Tsai, and Tai-Chia Lin. Modification of bikerman model with specific ion sizes. Computational and Mathematical Biophysics, 5(1):142 -- 149, 2017. [18] Jerzy J Jasielec, Grzegorz Lisak, Michal Wagner, Tomasz Sokalski, and Andrzej Lewenstam. Nernst-planck-poisson model for the description of behaviour of 39 solid-contact ion-selective electrodes at low analyte concentration. Electroanalysis, 25(1):133 -- 140, 2013. [19] Mustafa Sabri Kilic, Martin Z Bazant, and Armand Ajdari. Steric effects in the dynamics of electrolytes at large applied voltages. ii. modified poisson-nernst-planck equations. Physical review E, 75(2):021503, 2007. [20] Meredith LeMasurier, Lise Heginbotham, and Christopher Miller. Kcsa: it's a potas- sium channel. The Journal of general physiology, 118(3):303 -- 314, 2001. [21] Tai-Chia Lin and Bob Eisenberg. A new approach to the lennard-jones potential and a new model: Pnp-steric equations. Communications in Mathematical Sciences, 12(1):149 -- 173, 2014. [22] Jinn-Liang Liu and Bob Eisenberg. Poisson-nernst-planck-fermi theory for modeling biological ion channels. The Journal of chemical physics, 141(22):12B640 -- 1, 2014. [23] Benzhuo Lu and YC Zhou. Poisson-nernst-planck equations for simulating biomolec- ular diffusion-reaction processes ii: Size effects on ionic distributions and diffusion- reaction rates. Biophysical journal, 100(10):2475 -- 2485, 2011. [24] Victor B Luzhkov and Johan Aqvist. K+/na+ selectivity of the kcsa potassium chan- nel from microscopic free energy perturbation calculations. Biochimica et Biophysica Acta (BBA)-Protein Structure and Molecular Enzymology, 1548(2):194 -- 202, 2001. [25] Roderick MacKinnon. Potassium channels and the atomic basis of selective ion conduction (nobel lecture). Angewandte Chemie International Edition, 43(33):4265 -- 4277, 2004. [26] Roderick MacKinnon, Steven L Cohen, Anling Kuo, Alice Lee, and Brian T Chait. Structural conservation in prokaryotic and eukaryotic potassium channels. Science, 280(5360):106 -- 109, 1998. [27] Jaakko Malmivuo, Robert Plonsey, et al. Bioelectromagnetism: principles and appli- cations of bioelectric and biomagnetic fields. Oxford University Press, USA, 1995. [28] Peter A Markowich. The stationary semiconductor device equations. Springer Science & Business Media, 2013. 40 [29] Christopher Miller. See potassium run. Nature, 414(6859):23, 2001. [30] Boaz Nadler, Uwe Hollerbach, and RS Eisenberg. Dielectric boundary force and its crucial role in gramicidin. Physical Review E, 68(2):021905, 2003. [31] Crina M Nimigean and Toby W Allen. Origins of ion selectivity in potassium channels from the perspective of channel block. The Journal of general physiology, 137(5):405 -- 413, 2011. [32] Crina M Nimigean and Christopher Miller. Na+ block and permeation in a k+ channel of known structure. The Journal of general physiology, 120(3):323 -- 335, 2002. [33] Kene N Piasta, Douglas L Theobald, and Christopher Miller. Potassium-selective block of barium permeation through single kcsa channels. The Journal of general physiology, 138(4):421 -- 436, 2011. [34] Christopher N Rowley and Benoıt Roux. A computational study of barium blockades in the kcsa potassium channel based on multi-ion potential of mean force calculations and free energy perturbation. The Journal of general physiology, 142(4):451 -- 463, 2013. [35] Isaak Rubinstein. Electro-diffusion of ions. SIAM, 1990. [36] Indira H Shrivastava, D Peter Tieleman, Philip C Biggin, and Mark SP Sansom. K+ versus na+ ions in a k channel selectivity filter: a simulation study. Biophysical journal, 83(2):633 -- 645, 2002. [37] Amit Singer and John Norbury. A poisson -- nernst -- planck model for biological ion channelsan asymptotic analysis in a three-dimensional narrow funnel. SIAM Journal on Applied Mathematics, 70(3):949 -- 968, 2009. [38] Zilong Song, Xiulei Cao, and Huaxiong Huang. Electroneutral models for a multidi- mensional dynamic poisson-nernst-planck system. Physical Review E, 98(3):032404, 2018. [39] Zilong Song, Xiulei Cao, and Huaxiong Huang. Electroneutral models for dynamic poisson-nernst-planck systems. Physical Review E, 97(1):012411, 2018. 41 [40] Ameer N Thompson, David J Posson, Pirooz V Parsa, and Crina M Nimigean. Molecular mechanism of ph sensing in kcsa potassium channels. Proceedings of the National Academy of Sciences, 105(19):6900 -- 6905, 2008. [41] Trudy A van der Straaten, John M Tang, Umberto Ravaioli, Robert S Eisenberg, and NR Aluru. Simulating ion permeation through the ompf porin ion channel using three-dimensional drift-diffusion theory. Journal of Computational Electronics, 2(1):29 -- 47, 2003. [42] Di Wu. Dynamic water patterns change the stability of the collapsed filter confor- mation of the kcsa k+ channel. PloS one, 12(10):e0186789, 2017. [43] Jie Zheng and Matthew C Trudeau. Handbook of ion channels. CRC Press, 2015. [44] Qiong Zheng, Duan Chen, and Guo-Wei Wei. Second-order poisson -- nernst -- planck solver for ion transport. Journal of computational physics, 230(13):5239 -- 5262, 2011. [45] Yufeng Zhou, Joao H Morais-Cabral, Amelia Kaufman, and Roderick MacKinnon. Chemistry of ion coordination and hydration revealed by a k+ channel -- fab complex at 2.0 a resolution. Nature, 414(6859):43 -- 48, 2001. 42
1506.08862	1	1506	2015-06-29T21:06:48	Rapid Evolution of the Photosystem II Electronic Structure during Water Splitting	[ "physics.bio-ph", "physics.chem-ph" ]	Photosynthetic water oxidation is a fundamental process that sustains the biosphere. A Mn$_{4}$Ca cluster embedded in the photosystem II protein environment is responsible for the production of atmospheric oxygen. Here, time-resolved x-ray emission spectroscopy (XES) was used to observe the process of oxygen formation in real time. These experiments reveal that the oxygen evolution step, initiated by three sequential laser flashes, is accompanied by rapid (within 50 $\mu$s) changes to the Mn K$\beta$ XES spectrum. However, no oxidation of the Mn$_{4}$Ca core above the all Mn$^{\text{IV}}$ state was detected to precede O-O bond formation. A new mechanism featuring Mn$^{\text{IV}}$=O formation in the S$_{3}$ state is proposed to explain the spectroscopic results. This chemical formulation is consistent with the unique reactivity of the S$_{3}$ state and explains facilitation of the following S$_{3}$ to S$_{0}$ transition, resolving in part the kinetic limitations associated with O-O bond formation. In the proposed mechanism, O-O bond formation precedes transfer of the final (4$^{\text{th}}$) electron from the Mn$_{4}$Ca cluster, in agreement with experiment.	physics.bio-ph	physics	Title: Rapid Evolution of the Photosystem II Electronic Structure during Water Splitting Authors: Katherine M. Davis†, Brendan T. Sullivan, Mark Palenik ‡, Lifen Yan, Vatsal Purohit§, Gregory Robison¥, Irina Kosheleva1, Robert W. Henning1, Gerald T. Seidler2, Yulia Pushkar* Affiliations: Department of Physics and Astronomy, Purdue University, West Lafayette, IN 47907, USA. 1Center for Advanced Radiation Sources, The University of Chicago, Chicago, IL 60637, USA. 2Department of Physics, University of Washington, Seattle, WA 98195, USA. Correspondence to: [email protected] Current addresses: † Department of Chemistry, Princeton University, Princeton, NJ 08544, USA. ‡ Naval research laboratory, Washington, DC 20375, USA § Department of Biology, Purdue University, West Lafayette, IN 47907, USA. ¥ Department of Physics and Astronomy, Hanover College, Hanover, IN 47243, USA Abstract: Photosynthetic water oxidation is a fundamental process that sustains the biosphere. A Mn4Ca cluster embedded in the photosystem II protein environment is responsible for the production of atmospheric oxygen. Here, time-resolved x-ray emission spectroscopy (XES) was used to observe the process of oxygen formation in real time. These experiments reveal that the oxygen evolution step, initiated by three sequential laser flashes, is accompanied by rapid (within 50 µs) changes to the Mn Kβ XES spectrum. However, no oxidation of the Mn4Ca core above the all MnIV state was detected to precede O−O bond formation. A new mechanism featuring MnIV=O formation in the S3 state is proposed to explain the spectroscopic results. This chemical formulation is consistent with the unique reactivity of the S3 state and explains facilitation of the following S3 to S0 transition, resolving in part the kinetic limitations associated with O-O bond formation. In the proposed mechanism, O-O bond formation precedes transfer of the final (4th) electron from the Mn4Ca cluster, in agreement with experiment. One Sentence Summary: Time-resolved XES indicates that oxygen formation in Photosystem II proceeds via the reduction of the Mn4 molecular model. IV centers, a result that has be rationalized with a Main Text: Plants, green algae and cyanobacteria rely on the photoactive metalloprotein complex photosystem II (PS II) for its role in photosynthetic energy conversion. Using the sun’s energy, PS II splits water into molecular oxygen (O2), protons and electrons (1-3). Its catalytic activity and quantum efficiency remain unmatched by synthetic systems developed for artificial photosynthesis (4, 5). PS II utilizes a Mn4Ca cluster and its associated protein environment, collectively referred to as the Oxygen Evolving Complex (OEC), as the catalytic center for water splitting. In 1970, Bessel Kok et al. described a potential water splitting mechanism in which the OEC cycles through five states (S0-S4), corresponding to the oxidation states of manganese, following sequential visible light absorptions, Figure 1 (6). Antenna pigments from the surrounding protein matrix absorb light and funnel energy towards P680, the chlorophyll a special pair responsible for charge separation. Within nanoseconds, a tyrosine residue (TyrZ) located • is subsequently reduced by the between P680 and the OEC is oxidized by the special pair. TyrZ OEC on a microsecond timescale. This process drives the water splitting reaction (1). The past forty years have yielded new insights into the structure of PS II (3, 7-12), as well as the nature and timescales of the S-state transitions, Figure 1 (13-15). IV state, also presented as Mn4 However, the critical step of O−O bond formation remains poorly characterized and thus cannot be implemented in artificial systems. The O−O bond is likely formed within microseconds during the S3 to S0 transient step of the Kok cycle culminating in O2 evolution, but • details remain unknown. A pre-eminent report by Babcock et al. yielded a rate of TyrZ reduction, t1/2 ~1 ms, following three flashes and associated this rate constant with the formation of the S4 state, subsequently capable of fast O2 evolution (16). It was pointed out later that such hypotheses face a serious kinetic challenge in that the timescale for molecular oxygen release, following the formation of the S4 state, is very short for the associated redox chemistry and bond formation dynamics (2). As an alternative hypothesis, S3 state peroxide formation was proposed (2). However, this has not been confirmed experimentally. Given the inherent experimental complications, many computational simulations have been performed to model the O−O bond IV via the formation path (17-20). The majority of these imply oxidation of the OEC past Mn4 formation of a MnVMn3 IV−O• (oxyl radical). This oxidized configuration is currently associated with S4 and would precede O-O bond formation (17, 21). Experimental proof of such an intermediate state is currently lacking and our data rule out its formation. Here, we examine the earliest dynamic in the S3 to S0 transition by time-resolved x- ray emission spectroscopy (TR-XES) utilizing dispersive detection to aid our understanding of this critical biological process (22, 23). From these results, we formulate a mechanistic hypothesis consistent with experimental data that sheds light on some of the key steps in O−O bond formation. XES probes occupied orbitals; in particular, the Mn Kβ spectral emission lines reflect the number of unpaired 3d electrons and, thus, provide information about the oxidation and/or spin states (24), Figure 2 (inset). The exchange interaction between the 3p hole and 3d valence electrons in the final state causes multiplet splitting that results in separate Kβ1,3 and Kβ# peaks, Figure 2. This coupling is directly linked to the electronic state of Mn such that an increase in the oxidation state results in decreased splitting between the Kβ lines. This effect is apparent in the Kβ1,3 peak position shift to lower energies with increasing oxidation. XES also allows for dispersive detection, in which the full emission spectrum is recorded during a single, intense, polychromatic x-ray pulse, Figure 2 (22). The temporal resolution is only limited by the time structure of the x-ray source. In our PS II experiments, we utilized multi-bunch x-ray exposures of 44 µs duration to match the microsecond kinetics of the OEC, Table 1. We previously determined that 66 µs of exposure at these conditions is undamaging to PS II (22, 25). Data collection was performed using a von Hamos style miniature x-ray emission spectrometer (miniXES – Figure S1) (22, 26), and a non-jet open-air sample delivery system (see SI and Figure S2) was used to supply fresh, unrecycled PS II for each measurement. Samples were excited given a defined number (0-3) of laser flashes (F) and probed at a time (Δt) after the final laser flash by a single x-ray pulse, Figures 2 & S3, Tables S1 & S2. A total of six beamtimes were accomplished, with two devoted to methodology development and four to data collection, Tables 2 and S2 & S8. Time-resolved measurements following zero, one and two laser flashes were recorded at Δt=500 µs, a delay time allowing for • with limited decay of the formed S-state, Tables 2, S3, S4, S6, S8 (13, the full reduction of TyrZ 27). These flash data correspond to the majority S-states S1, S2, and S3 respectively (Table S1). The obtained S1 state spectral shape and peak position are in good agreement with previous RT PS II measurements (22). A comparison of 0F and 1F spectra shows a reproducible shift of the Kβ1,3 peak to lower energies quantified by analysis of the first moment. Given the multiplet character of the spectra and the potential noise inherent for such a dilute sample, , surrounding the Kβ1,3 previous studies recommend the use of the statistical first moment, jEj∑ ⋅ jI jIj∑ peak (6485-6495 eV), to determine any changes to the electronic structure, Figure S8. The observed reproducible shift of the Kβ1,3 peak to lower energies following a single laser flash is consistent with previous cryogenic XANES (11, 28) and XES (29) results for the S1 to S2 state transition indicating Mn-centered oxidation. In contrast, the x-ray free electron laser (xFEL)- based TR-XES analysis of the same transition in PS II did not detect this shift, likely due to poor statistics (23). While the absolute shift between the first moments of 0F and 1F varies in magnitude between data sets likely due to counting statistics, Table S6 demonstrates that the values are consistently negative. Analysis of three combined data sets yields a first moment shift of -0.12 eV for the 0F → 1F transition, Table S3. Cryogenic measurements previously reported a -0.059 eV shift following spectral smoothing and background subtraction procedures (29). It should be emphasized that background subtraction (Figure S6) procedures can affect the first moment magnitudes, Tables S3, S6 & S7. For studies reporting small changes in the first moment values, statistical methods should be used to determine confidence in the detected changes. One-way ANOVA confirms that the first moment shift between 0F and 1F spectra detected in this study is statistically significant, Table 2. Overall, we consider our results for the 0F → 1F transition robust and in good agreement with previous characterization of the S1 to S2 transition in which one Mn center is oxidized from MnIII to MnIV. For the ensuing S-state transition, S2 → S3, previous cryogenic measurements determined that changes in the Mn Kβ emission spectrum are minimal (29). A small (-0.02 eV) shift, on the order of our systematic error (0.02 eV, see SI for more details), was reported (29). Statistical analysis shows that the first moments of 1F and 2F x-ray emission spectra are indistinguishable, Table 2. Emission data representative of the S1 (0F) to S3 (2F) transition are presented in Figure IV in the S3 state (30-33) Figure 1, implies S7. The proposed Mn oxidation to form Mn4 comparable XES shifts for the S1 → S2 and S2 → S3 transitions. Lack of such an effect was previously attributed to ligand-centered oxidation (29). However, we know that the OEC undergoes a major structural change during the S2 → S3 transition from both EXAFS (28) and femtosecond (fs) x-ray crystallography (12). If such a structural change significantly alters the hybridization of the Mn 3d and ligand orbitals, some neutralization of the Mn-centered oxidative shift could be proposed to explain these results. To probe samples enriched with S0, XES spectra were collected at a 40 ms delay following a third laser flash, (3F40ms). The first moment of 3F40ms is consistently shifted to higher energies (Tables S3-S6) and statistically distinct from 1F and 2F first moments, Table 2, supporting the expected reduction of Mn. Overall, the RT TR-XES results are in good agreement with Mn Kβ emission data for previously cryogenically trapped states (29). Having validated the experimental technique, we investigated the elusive transient S3 to S0 process also initiated by three laser flashes (3F). Figure 3 depicts the earliest evolution of the OEC electronic structure following three laser flashes and the associated first moment shifts of Mn Kβ1,3 measured at Δt ~50 µs (3F!"!!) and Δt ~200 µs (3F!""!!). The trend of increasing first moment is robust and statistically Note that changes between 2F and 3F!""!! are greater than between 2F and 3F!""!!. We significant, Table 2. Furthermore, the observed increase in the first moment cannot be explained by mixing of states only, see the SI for details regarding laser excitation. This result suggests that during the S3 to S0 transition the OEC undergoes a significant transformation at short timescales. currently attribute this to limited statistics or to transient oxidation of the OEC by the electron •. Transient oxidation would likely reverse the reductive trend in first moments arriving from TyrZ to one of increasing value. Contrary to the density functional theory (DFT)-based modeling predictions of O−O bond formation (17-20), we were unable to observe oxidation of the OEC, which would be associated with lower values of the first moment, at any observation time point following the third flash and measured over multiple beamtimes. This lack of evidence for oxidation past 2F (majority state S3) was consistently observed (Figure 3 & Tables S3–S7) thereby excluding formation of a MnVMn3 IV state kinetic intermediate. Current DFT models (17-20) also do not resolve the kinetic challenge identified by Prof. Renger (2). UV-Vis difference spectra show that • quite slowly, ~1 ms after the third flash. Given that this time constant is the electron leaves TyrZ comparable to the rate of O2 evolution, only a very short ~50 µs time window remains for all bond formation dynamics and product/substrate exchange to occur. While our results cannot be explained by existing DFT mechanistic models, they are in agreement with the only other published TR study probing the electronic structure evolution of the Mn centers via x-ray absorption spectroscopy (XAS) (13, 34). TR-XAS detected no oxidation and instead, suggests the gradual (milliseconds) reduction of the OEC initiated 250 µs into the S3 to S0 transition. In contrast to the XAS study, where only two energy points along the Mn K-edge were analyzed, we collect full spectra representing the complete electronic structure of the OEC at Δt ~50 µs and Δt ~200 µs. In addition, XAS and XES represent different electronic transitions. It is therefore possible that some early spectral changes may have previously escaped detection. A more recent XES study performed at the Linac Coherent Light Source indicates no changes to the Mn Kβ spectra 250 µs after the third laser excitation (35). As with the unobserved S1 to S2 shift at the xFEL, we attribute this discrepancy to the differences in experimental conditions that will likely be clarified in the future. Figure 4 introduces a mechanism based on the fs-XRD structure of PS II (36) and verification of structural models via EXAFS (37) that is consistent with recent experiments. DFT calculations in Gaussian09 (38) were used to perform geometry optimizations. We hypothesize that the anomalous O5 in the XRD structure corresponds to an OH bridge between Mn4A and Mn3B (also proposed by Prof Shen (36)). However, we assign it as a structural element of the S0 state similar to the S0 structure presented by Krewald et al. (39). While not critical for the main conclusions of this manuscript, this hypothesis implies that the crystals studied might have been produced partially or completely in the S0 state. Subsequent oxidation of S0 and deprotonation of the OH bridge give rise to an S1 state structure that agrees well with recent room temperature EXAFS (37). To ensure equivalency between energy inputs for each S-state transition in our model, we describe S-state transitions as proton-coupled electron transfer (PCET) events, Figure 4. A summary of the structural changes inherent to our mechanistic model is in Table S9. In agreement with experiment, the S1 to S2 transition results in minimal changes. The fundamental hypothesis in our proposed mechanism is the presence of a MnIV=O fragment in the S3 state, located on the last Mn center to be oxidized, Mn1D. Our model suggests that this center cannot oxidize in lower S-state transitions as it lacks a ligand capable of PCET. The high activation energy measured for the S2 to S3 transition (40) is in good agreement with the need for substrate binding to Mn1D and substantial structural changes to create a reactive MnIV=O fragment. The model also predicts significant shortening of one Mn-Ca vector previously observed experimentally (28), Table S9. The presence of MnIV=O agrees well with the distinct chemical reactivity of the S3 state in comparison to other S-states (41), provides rationale for the suppression of oxidative shifts in XES by changes in orbital hybridization, and decreases the “kinetic challenge” (2) associated with the S3 to S0 transition. The presence of a substrate oxygen as a MnIV=O−Ca bridge is also in agreement with the order of magnitude increase in substrate exchange rate upon Ca to Sr substitution (42). Single point energies were obtained for a broken symmetry calculation of S3. Diagonalizing the Heisenberg-Dirac-Van Vleck Hamiltonian yielded an S3 ground state spin of 3, in agreement with the most recent EPR data (33), as well as indicating radicaloid character of the oxygen in the MnIV=O−Ca fragment due to its high spin density of ρ=0.4. The potential for O−O bond formation was subsequently analyzed for IV=O, as these correspond to shortest O−O Mn1D distances in the S3 state model, Figure 4. Both were found to be larger than the driving force available from the Tyr−OH=Tyr−O• + e- + H+ couple. (Table S10?) However, in the case of O−O IV=O and H2O−Ca, the increase in energy can be compensated by bond formation between Mn1D IV=O] + [H2O−Ca] + binding of a new water molecule to Ca according to the equation: [Mn1D H2O = Mn1D−OOH + [H2O−Ca]. While this scheme is formally equivalent to the use of H2O from the protein environment and preserves Ca−H2O coordination, it seems less likely due to the IV=O]. Interestingly, the energy rise was similar absence of free water in close proximity to [Mn1D with and without the abstraction of an electron, Table S10. IV=O + H2O−Ca and Mn4A−O−Mn3B + Mn1D IV=O] + [H2O−Ca] + H2O = Mn1D−OOH + [H2O−Ca]+1e-+1H+ IV=O] + [H2O−Ca] + H2O = Mn1D−OOH + [H2O−Ca]+1H+ [Mn1D [Mn1D DFT calculations show an increase in required energy input for each following S-state transition, even when they are conducted similarly (as PCET), Figure 4. In manmade catalysts, the consecutive oxidations of the catalytic center required for O−O bond formation also necessitate the application of progressively increasing oxidation potentials. For instance, ruthenium-based water oxidation catalysts undergo several redox transitions such as RuII/RuIII, RuIII/RuIV, and RuIV/the catalytic state, occurring at progressively increasing redox potentials To overcome the kinetic challenge, we propose that the O-O formation step takes place in •. Rapid evolution of the OEC during the S3 the S3 to S0 transition prior to the reduction of TyrYZ to S0 transition has long been a primary target of PS II research. Based on UV-Vis difference spectroscopy (15, 45) and TR infrared spectroscopy (14) a deprotonation event has been proposed to occur early (0-300 µs) during this transition. Our results do not explicitly exclude a deprotonation event, but necessitate significant changes to the electronic structure of the 3d Mn frontier orbitals to explain the observed spectroscopic effect. The results presented here can be •)S3 state, occurring on the order of 100ns (14), better rationalized if the formation of the (TyrYZ triggers a sequence of events resulting in significant redox or structural changes to the OEC, such as the formation of the O−O bond. The most recent isotope exchange studies show that substrate exchange stops early (0-300 µs) in the transition (42) hinting at such a possibility. The paramount question is whether the O−O bond can be formed in the Mn4Ca cluster •. This possibility would require significant energy input such prior to electron transfer to TyrYZ as from a strong base or proton-pump to drive deprotonation, mechanical relaxation of the protein resulting in O−O bond formation, and/or stabilization of the peroxide product. Stored energy from S0 to S3 conversion can be as high as ~2.5 eV if no energy dissipation occurs, which is sufficient for O−O bond formation by proton abstraction, Figure 4, S9. In this case, the final •. A step of O2 evolution would correspond to the oxidation of the Mn4Ca(OOH) cluster by TyrZ •)S3OOH state structure is the hydrogen particularly interesting feature of the proposed (TyrYZ bond between the peroxide and the Mn4A−O−Mn3B bridge that is protonated in the S0 state. Thus, deprotonation of the -OOH fragment can take place under oxidation to result in (TyrYZ)S4(O2)•-, from which expulsion of O2 follows directly resulting in S0 and providing a logical conclusion of the cycle. it. The OEC works under a fixed potential provided by (43, 44). PCET can decrease the differences in consecutive oxidation steps but cannot completely eliminate the Tyr−OH=Tyr−O• + e- + H+ couple. Thus, any variation in redox potential between the S-state transitions will result in energy losses unless excess energy from low oxidation state (such as S0/S1 and S1/S2) transitions can be efficiently stored in the form of an altered protein conformation (mechanical energy) or as a re-distribution of charges (electrical energy). Our simplistic analysis of reaction energetics can be refined in the future by including the larger protein environment using molecular dynamics simulations in conjunction with TR-fs-XRD of S-state transitions. In summary, to analyze the evolution of the photosystem II electronic structure, we observed intermediate states of photosynthetic O2 production via microsecond resolution time- resolved X-ray emission spectroscopy at a synchrotron source. Consistent with the obtained full spectra is a mechanism involving the presence of MnIV=O in the S3 state, and O−O bond • reduction. This mechanism resolves the formation in the S3 to S0 transition prior to TyrYZ previously highlighted kinetic problems and opens to investigation the possibility of energy storage in the surrounding protein matrix. MnIV=O−Ca activated fragments mimicking the OEC have already been demonstrated in synthetic systems (46) and the use of such fragments to drive O−O bond formation will likely be discovered in the future. References and Notes: 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16. 17. 18. 19. 20. 21. 22. 23. 24. T. Wydrzynski, S. Satoh, Photosystem II: The Light-Driven Water:Plastoquinone Oxidoreductase. Govindjee, Ed., Advances in Photosynthesis and Respiration (Springer, Dordrecht, 2005). J. Kern, G. Renger, Photosystem II: Structure and Mechanism of the Water : Plastoquinone Oxidoreductase. Photosynth. Res. 94, 183-202 (2007). Y. Umena, K. Kawakami, J. R. Shen, N. Kamiya, Crystal Structure of Oxygen-Evolving Photosystem II at a Resolution of 1.9 Angstrom. Nature 473, 55-U65 (2011). F. E. Osterloh, Inorganic Nanostructures for Photoelectrochemical and Photocatalytic Water Splitting. Chem. Soc. Rev. 42, 2294-2320 (2013). J. R. Swierk, T. E. Mallouk, Design and Development of Photoanodes for Water-Splitting Dye-Sensitized Photoelectrochemical Cells. Chem. Soc. Rev. 42, 2357-2387 (2013). B. Kok, B. Forbush, M. McGloin, Cooperation of Charges in Photosynthetic Oxygen Evolution. I. A Linear Four Step Mechanism. Photochem. Photobiol. 11, 457-475 (1970). K. N. Ferreira, T. M. Iverson, K. Maghlaoui, J. Barber, S. Iwata, Architecture of the Photosynthetic Oxygen-Evolving Center. Science 303, 1831-1838 (2004). A. Zouni et al., Crystal Structure of Photosystem II from Synechococcus Elongatus at 3.8 Angstrom Resolution. Nature 409, 739-743 (2001). A. Guskov et al., Cyanobacterial Photosystem II at 2.9-Angstrom Resolution and the Role of Quinones, Lipids, Channels and Chloride. Nat. Struct. Mol. Biol. 16, 334-342 (2009). Y. Pushkar et al., Structure and Orientation of the Mn4Ca Cluster in Plant Photosystem II Membranes Studied by Polarized Range-Extended X-Ray Absorption Spectroscopy. J. Biol. Chem. 282, 7198-7208 (2007). C. Glockner et al., Structural Changes of the Oxygen-Evolving Complex in Photosystem II During the Catalytic Cycle. J. Biol. Chem. 288, 22607-22620 (2013). C. Kupitz et al., Serial Time-Resolved Crystallography of Photosystem II Using a Femtosecond X-Ray Laser. Nature 513, 261-265 (2014). M. Haumann et al., Photosynthetic O2 Formation Tracked by Time-Resolved X-Ray Experiments. Science 310, 1019-1021 (2005). T. Noguchi, H. Suzuki, M. Tsuno, M. Sugiura, C. Kato, Time-Resolved Infrared Detection of the Proton and Protein Dynamics During Photosynthetic Oxygen Evolution. Biochemistry-US 51, 3205-3214 (2012). F. Rappaport, M. Blancharddesce, J. Lavergne, Kinetics of Electron-Transfer and Electrochromic Change During the Redox Transitions of the Photosynthetic Oxygen-Evolving Complex. BBA-Bioenergetics 1184, 178-192 (1994). G. T. Babcock, R. E. Blankenship, K. Sauer, Reaction-Kinetics for Positive Charge Accumulation on Water Side of Chloroplast Photosystem II FEBS Lett. 61, 286-289 (1976). P. E. M. Siegbahn, Water Oxidation Mechanism in Photosystem II, Including Oxidations, Proton Release Pathways, O-O Bond Formation and O2 Release. Biochim. Biophys. Acta 1827, 1003-1019 (2013). T. Saito et al., Possible Mechanisms of Water Splitting Reaction Based on Proton and Electron Release Pathways Revealed for CaMn4O5 Cluster of PSII Refined to 1.9 Angstrom X-Ray Resolution. Int. J. Quantum Chem. 112, 253-276 (2012). K. Yamaguchi et al., Full Geometry Optimizations of the Mixed-Valence CaMn4O4X(H2O)(4) (X=OH or O) Cluster in OEC of PS II: Degree of Symmetry Breaking of the Labile Mn-X-Mn Bond Revealed by Several Hybrid DFT Calculations. Int. J. Quantum Chem. 113, 525-541 (2013). E. M. Sproviero, J. A. Gascon, J. P. McEvoy, G. W. Brudvig, V. S. Batista, Quantum Mechanics/Molecular Mechanics Study of the Catalytic Cycle of Water Splitting in Photosystem II. J. Am. Chem. Soc. 130, 3428- 3442 (2008). P. E. M. Siegbahn, O-O Bond Formation in the S4 State of the Oxygen-Evolving Complex in Photosystem II. Chemistry – A European Journal 12, 9217-9227 (2006). K. M. Davis et al., Fast Detection Allowing Analysis of Metalloprotein Electronic Structure by X-Ray Emission Spectroscopy at Room Temperature. J. Phys. Chem. Lett. 3, 1858-1864 (2012). J. Kern et al., Simultaneous Femtosecond X-Ray Spectroscopy and Diffraction of Photosystem II at Room Temperature. Science 340, 491-495 (2013). P. Glatzel, U. Bergmann, High Resolution 1s Core Hole X-Ray Spectroscopy in 3d Transition Metal Complexes - Electronic and Structural Information. Coordin. Chem. Rev. 249, 65-95 (2005). 25. 26. 27. 28. 29. 30. 31. 32. 33. 34. 35. 36. 37. 38. 39. 40. 41. 42. 43. 44. 45. 46. 47. K. M. Davis, I. Kosheleva, R. W. Henning, G. T. Seidler, Y. Pushkar, Kinetic Modeling of the X-Ray- Induced Damage to a Metalloprotein. J. Phys. Chem. B 117, 9161-9169 (2013). B. A. Mattern et al., A Plastic Miniature X-Ray Emission Spectrometer (MiniXES) Based on the Cylindrical Von Hamos Geometry. Rev. Sci. Instrum 83, 023901:023901-023909 (2012). J. Messinger, W. P. Schroder, G. Renger, Structure-Function Relations in Photosystem-II - Effects of Temperature and Chaotropic Agents on the Period 4 Oscillation of Flash-Induced Oxygen Evolution. Biochemistry-US 32, 7658-7668 (1993). Y. Pushkar, J. Yano, K. Sauer, A. Boussac, V. K. Yachandra, Structural Changes in the Mn4Ca Cluster and the Mechanism of Photosynthetic Water Splitting. P. Natl. Acad. Sci. USA 105, 1879-1884 (2008). J. Messinger et al., Absence of Mn-Centered Oxidation in the S2 to S3 Transition:Implications for the Mechanism of Photosynthetic Water Oxidation. J. Am. Chem. Soc. 123, 7804-7820 (2001). T.-a. Ono et al., X-Ray Detection of the Period-Four Cycling of the Manganese Cluster in Photosynthetic Water Oxidizing Enzyme. Science 258, 1335-1337 (1992). L. Iuzzolino, J. Dittmer, W. Dörner, W. Meyer-Klaucke, H. Dau, X-Ray Absorption Spectroscopy on Layered Photosystem II Membrane Particles Suggests Manganese-Centered Oxidation of the Oxygen- Evolving Complex for the S0-S1, S1-S2, and S2-S3 Transitions of the Water Oxidation Cycle. Biochemistry- US 37, 17112-17119 (1998). M. Haumann et al., Structural and Oxidation State Changes of the Photosystem II Manganese Complex in Four Transitions of the Water Oxidation Cycle (S0 → S1, S1 → S2, S2 → S3, and S3,S4 → S0) Characterized by X-Ray Absorption Spectroscopy at 20 K and Room Temperature. Biochemistry-US 44, 1894-1908 (2005). N. Cox et al., Electronic Structure of the Oxygen Evolving Complex in Photosystem II Prior to O-O Bond Formation. Science 345, 804-808 (2014). M. Haumann, A. Grundmeier, I. Zaharieva, H. Dau, Photosynthetic Water Oxidation at Elevated Dioxygen Partial Pressure Monitored by Time-Resolved X-Ray Absorption Measurements. P. Natl. Acad. Sci. USA 105, 17384-17389 (2008). J. Kern et al., Taking Snapshots of Photosynthetic Water Oxidation Using Femtosecond X-Ray Diffraction and Spectroscopy. Nature communications 5, 4371-4371 (2014). M. Suga et al., Native Structure of Photosystem II at 1.95 Å Resolution Viewed by Femtosecond X-Ray Pulses. Nature 517, 99-103 (2015). K. M. Davis, Y. N. Pushkar, Structure of the Oxygen Evolving Complex of Photosystem II at Room Temperature. The Journal of Physical Chemistry B 119, 3492-3498 (2015). M. J. Frisch et al. (Gaussian, Inc., Wallingford, CT, USA, 2009). V. Krewald et al., Metal Oxidation States in Biological Water Splitting. Chemical Science 6, 1676-1695 (2015). M. Karge, K. D. Irrgang, G. Renger, Analysis of the Reaction Coordinate of Photosynthetic Water Oxidation by Kinetic Measurements of 355 nm Absorption Changes at Different Temperatures in Photosystem II Preparations Suspended in Either H2O or D2O. Biochemistry-US 36, 8904-8913 (1997). N. Ioannidis, G. Schansker, V. V. Barynin, V. Petrouleas, Interaction of Nitric Oxide with the Oxygen Evolving Complex of Photosystem II and Manganese Catalase: A Comparative Study. JBIC, J. Biol. Inorg. Chem. 5, 354-363 (2000). H. Nilsson, F. Rappaport, A. Boussac, J. Messinger, Substrate-Water Exchange in Photosystem II Is Arrested before Dioxygen Formation. Nature communications 5, (2014). Y. Pushkar, D. Moonshiram, V. Purohit, L. Yan, I. Alperovich, Spectroscopic Analysis of Catalytic Water Oxidation by [RuII(bpy)(tpy)H2O]2+ Suggests That RuV═O Is Not a Rate-Limiting Intermediate. J. Am. Chem. Soc. 136, 11938-11945 (2014). M. H. V. Huynh, T. J. Meyer, Proton-Coupled Electron Transfer. Chem. Rev. 107, 5004-5064 (2007). M. Haumann, O. Bogershausen, D. Cherepanov, R. Ahlbrink, W. Junge, Photosynthetic Oxygen Evolution: H/D Isotope Effects and the Coupling between Electron and Proton Transfer During the Redox Reactions at the Oxidizing Side of Photosystem II. Photosynth. Res. 51, 193-208 (1997). J. Chen et al., A Mononuclear Non-Heme Manganese(IV)–Oxo Complex Binding Redox-Inactive Metal Ions. J. Am. Chem. Soc., 6388-6391 (2013). A. W. Rutherford, Orientation of Electron-Paramagnetic-Res Signals Arising from Components in Photosystem-II Membranes. Biochim. Biophys. Acta 807, 189-201 (1985). 48. 49. 50. 51. 52. 53. 54. 55. D. A. Berthold, G. T. Babcock, C. F. Yocum, A Highly Resolved, Oxygen-Evolving Photosystem II Preparation from Spinach Thylakoid Membranes. EPR and Electron-Transport Properties. FEBS Lett. 134, 231-234 (1981). G. Y. Han, F. Mamedov, S. Styring, Misses During Water Oxidation in Photosystem II Are S State- Dependent. J. Biol. Chem. 287, 13422-13429 (2012). T. Graber et al., Biocars: A Synchrotron Resource for Time-Resolved X-Ray Science. J. Synchrotron Radiat. 18, 658-670 (2011). J. I. Pacold et al., A Miniature X-Ray Emission Spectrometer (MiniXES) for High-Pressure Studies in a Diamond Anvil Cell. J. Synchrotron Radiat. 19, 245-251 (2012). B. Dickinson et al., A Short Working Distance Multiple Crystal X-Ray Spectrometer. Rev. Sci. Instrum. 79, 123112:123111-123118 (2008). V. K. Henner, P. G. Frick, T. S. Belozerova, V. G. Solovyev, Application of Wavelet Analysis to the Spectrum of Omega ' States and Ratio Re+E. European Physical Journal C 26, 3-7 (2002). A. D. Becke, Density-Functional Exchange-Energy Approximation with Correct Asymptotic-Behavior. Phys. Rev. A 38, 3098-3100 (1988). F. Weigend, R. Ahlrichs, Balanced Basis Sets of Split Valence, Triple Zeta Valence and Quadruple Zeta Valence Quality for H to Rn: Design and Assessment of Accuracy. Physical Chemistry Chemical Physics 7, 3297-3305 (2005). Acknowledgments: The spectrometer and TR-XES methodology development was supported by the DOE, Office of Basic Energy Sciences DE-FG02-12ER16340 (Y.P.). Measurements of Photosystem II were supported by NSF, CHE-1350909 (Y.P.) and the NSF Graduate Research Fellowship under Grant No. DGE0833366 (K.D.). Research at the University of Washington is supported by the DOE, Office of Basic Energy Sciences DE-SC0002194. PNC/XSD facilities at the Advanced Photon Source and research at these facilities are supported by the U.S. Department of Energy, Basic Energy Sciences, a Major Resources Support grant from NSERC, the University of Washington, Simon Fraser University, and the Advanced Photon Source. Use of the Advanced Photon Source, an Office of Science User Facility operated for the U.S. Department of Energy (DOE) Office of Science by Argonne National Laboratory, was supported by the U.S. DOE under Contract No. DE-AC02-06CH11357. Use of the BioCARS Sector 14 was also supported by grants from the National Center for Research Resources (5P41RR007707) and the National Institute of General Medical Sciences (8P41GM103543) from the National Institutes of Health. The time-resolved setup at BioCARS was funded in part through a collaboration with Philip Anfinrud (NIH/ NIDDK). We thank Prof. L. Slipchenko from Purdue University for providing computational resources and helpful discussion. Fig. 1. A current model of the Kok cycle depicting incident visible light photons and electron/proton release (13). The dashed region is based on previous analysis of the S3 to S0 transition in which the S4 state was proposed (13, 34). Fig. 2. Schematic of the experimental setup. Nanosecond laser pulses (1, 2 or 3) are used to advance the Kok cycle in the protein, Table S2. The pump/probe delay time, Δt, measured from the final laser flash to the center of the x-ray pulse, is set dependent on the desired S-state, Figure S3 and Table S2. X-ray fluorescence from the sample is reflected by 10 flat analyzer crystals onto a 2D-position sensitive detector. Kβ emission spectra are extracted to form snapshots of the electronic structure in time. Smoothed emission spectra are presented for the S3 to S0 transition. Fig. 3. Fully processed (background subtracted and smoothed) 2F, 3F!"!!, and 3F!""!! Kβ emission spectra for data sets (A) 3 (including the additional statistics, Table S7), (B) 4 and (C) their sum respectively. The region (6.485-6.495 keV) over which the first moment was calculated is highlighted and a magnified inset presented. For each set of spectra the trend in first moments is shown. Note that variations between data sets are due to counting statistics that cause noise on the edges of the selected region. Those moments statistically different from 2F data are indicated with an asterisk. See Tables S6 and S7 for the first moment magnitudes. Fig. 4. A mechanism of water oxidation has been developed to explain the spectroscopic results. Left: The orientation of molecular models are aligned for clarity. Purple, red and green atoms are Mn, O, and Ca respectively. Water molecules are shown in dark blue, hydroxides are light blue. Amino acids shown are those used in the DFT calculations. Bonds are shown for distances <2.5 Å. Magnifications of S3 and S4 show conversion of the MnIV=O-Ca reactive fragment to the peroxo species hydrogen bonding to the Mn4A−O−Mn3B bridge. Note that the S3 inset is rotated slightly from the common alignment to allow for easier visualization of the MnIV=O-Ca IV=O fragment. Arrows indicate the shortest distances most likely for O-O bond formation: Mn1D + H2O−Ca and Mn4A−O−Mn3B + Mn1D Right: On the energy diagram red ticks represent the energy change (ΔTyr) due to TyrYZ⋅ reduction in the PCET process: (Tyr−OH=Tyr−O⋅ + e- + H+). Green ticks show the changes in energy due to S-state transitions. The S3 to S0 transition is shown in greater detail including the energy levels of proposed intermediates. IV=O. Table 1. Experimental characteristics of pulsed x-ray source Characteristics BioCARS Peak energy 7.85 keV, FWHM ~500 eV Excitation Energy ~45 x 100 µm2 X-ray Spot Size Pulse Length 44 µs 3 x 1011 photons/pulse Photon Flux Dose Delivered per Pulse ~7 x 107 photons/µm2 Repetition Rate ~20 Hz Majority S-State S0 3F40ms 1 S1 0F 0.25 1 Flash 3F40ms 0F 1F 2F !!!"!" !!!""!" !!!""!" S4b S4a S'4 Table 2. ANOVA values for each S-state calculated from raw (unprocessed) data. S2 1F 0.05 <0.01 1 S3 2F 0.03 <0.01 0.98 1 !!!"!" !!!""!" !!!""!" S0 (117) S1 (78) S2 (82) S3 (97) S4a (76) S4b (100) S'4 (84) p-values are based on the first moments over the range 6.485 – 6.495 keV, and are calculated for all data, i.e. including every beamtime. The number of “samples” (i.e. threads) is shown in parentheses for each state. See Table S8 for additional comparisons between the states based on the number of X-ray pulses per state per beamtime and Figure S8 for a dotplot of all first moments used for p-value calculations. 0.41 0.78 0.01 <0.01 0.68 1 0.10 0.02 0.98 0.96 0.08 0.04 1 0.69 0.46 0.03 0.01 1 15 Supplementary Materials: Materials and Methods, optimized DFT coordinates Figures S1-S9 Tables S1-S10 References (47-55) 16 Supplementary Materials Materials and Methods Photosystem II Preparation: Beamline Description: PSII-enriched thylakoid-membrane particles were prepared from supermarket spinach (47, 48). Prior to use, samples were stored at -80°C in a buffer using sucrose as a cryo- protectant: 0.4 M sucrose, 5 mM CaCl2, 5 mM MgCl2, and 15 mM NaCl, 50 mM MES, pH 6.0. The oxygen evolution activity of PSII was measured by a Clark-type electrode in a Hansatech oxygraph. The activity of the preparation was 300 µmol O2/(mg Chl • hr) or greater under constant saturating illumination at 25°C utilizing 0.3 mM 2,6-dichloro-1,4-benzoquinone (DCBQ) as an artificial electron acceptor. The Chl a:b ratio was derived from the optical absorbance of chlorophyll extracted with 80% acetone: 20% water solution. This was measured with a Cary300 Bio UV-visible spectrophotometer. For all samples, this ratio was ~2.5:1 which indicates high enrichment of membrane particles with PSII. To ensure the high quality of the samples, low temperature X-band EPR spectra were recorded for the S1 and S2 states. S2 state samples were obtained by illuminating S1 state samples with 120 W Halogen lamp for 30 minutes while in a cold bath of ethanol and dry ice to maintain a sample temperature of 195 K, after which they were immediately frozen in liquid nitrogen. For TR-XES, the PSII samples were prepared on-site immediately prior to the measurements as follows. After thawing the stock pellet (~30mg Chl/mL) stored on dry ice for travel, the samples were diluted 1:1 with a 30% v/v glycerol in a buffer of 5 mM CaCl2, 5 mM MgCl2, and 15 mM NaCl, 50 mM MES, pH 6.5 and homogenized gently with a paintbrush. A solution of 50 mM PPBQ in DMSO was added to yield a final PPBQ concentration of 500 µM (28, 49). Clean mixed fiber thread of radius ~175 µm was tested to have no Mn and remain intact under laser and x-ray illuminations. The sample was painted onto these threads with a soft brush. Painting ensures best sample absorption onto the threads. Threads were wound onto small plastic spools and servo motors were used to slowly wind and unwind the threads between the plastic spools at a constant speed during painting. During measurements, the spool with freshly painted thread was placed on an axis above a reservoir of ice and covered with aluminum foil. Measurements were performed as soon as possible (<30 min from application of sample to data collection). All sample preparation, handling and storage environments were completely dark, save for dim green LEDs when unavoidable, to prevent state transitions prior to measurement. In addition, samples were maintained at a constant temperature of approximately 4°C during preparation. TR emission spectra were collected at sector 14-ID B, BioCARS (50), of the Advanced Photon Source, Argonne National Lab with an electron energy of 7.0 GeV and an average current of 100 mA. To avoid second order reflections and gallium fluorescence inherent to the spectrometer crystals, the undulator gap of U27 was set to 11.7 mm during the emission experiments yielding a ~500 eV FWHM undulator spectrum centered around 7.85 keV. In addition, the water-cooled KB mirrors for focusing were translated to their Si stripes and set to an angle of 3.8 mrad, thereby cutting the incident energy range sharply above the energy of the first harmonic. The vertical mirror was defocused to yield a final projection size of ~45 x 100 µm2 (V x H) given the sample surface angled 45° to the incident beam, and the incident flux was monitored downstream of both mirrors via a photodiode. X-ray pulses of ~44 µs were produced using the high heat load chopper (50) and delivered at frequency 10.3 or 20.6 Hz, see Table S2 for experimental parameters. Monochromatic beam was produced using a Si(111) channel-cut monochromator inserted into the beam path and calibrated via the KMnO4 pre-edge at 6.5433 keV. Monochromatized beam is necessary for the energy calibration of the position sensitive detector. The smallest possible undulator gap (10.5 mm), corresponding to a peak energy of 6.85 keV, was used such that the tails of the undulator spectrum reach 6.445 keV, the lowest collection energy of the GaP spectrometer. To improve the energy resolution of x-rays reflected by the monochromator, the beam incident on the channel-cut was collimated by decreasing the radius of curvature of the vertical mirror. The scattering signal used for calibration was collected with 1 kHz pulse frequency to increase the integrated flux. XES Spectrometer: For this study, we utilized a short working distance (SWD) miniature x-ray emission spectrometer (miniXES) in which multiple flat dispersive GaP 110 analyzers reflect x-ray fluorescence onto a Pilatus 100k (Dectris) 2-dimensional position sensitive detector (2D-PSD) Figure S1 (22, 25). To decrease attenuation, the internal spectrometer space is purged with He during use. The spectrometer crystals are arranged in a von Hamos geometry and designed for a collection range of ~50 eV about the Kβ main lines, with each crystal reflecting the complete Mn Kβ spectrum (6445-6510 eV) onto the PSD, Figure S1. The width of monochromatic elastic scatter measured by the spectrometer is a convolution of the monochromator spectral resolution (ΔE/E = 1.4×10-4) and the broadening introduced by the spectrometer itself, determined to be ~0.3 eV. Note that this resolution is distinct from the minimum energy shifts resolvable by the setup; our maximum obtained resolution is discussed in Data Processing / Averaging. Prior to measurements, the spectrometer was aligned horizontally and vertically via eyelets designed for this purpose. The thread position with respect to the spectrometer in the beam direction (see more in Sample Positioning) was optimized prior to PS II data collection using a sample with intense Mn Kβ fluorescent signal, typically MnCl2 solution painted on the threads. Multiple small (~100 µm) sample translations along the beam direction were implemented until all 10 crystals were fully visible on the Pilatus face. The thread position in the beam direction was fixed using the focus of a stationary, high magnification camera. A lens extension with a focal depth of ~50µm was used to ensure consistent spatial positioning. Sample Positioning: To provide the least amount of stray scatter into the spectrometer aperture, the samples were tilted vertically 45° towards the incident beam, Figure S2(A). Kinematic mounts were used to attach the sample delivery system, Figure S2(A), to a motorized X,Y,Z stage. These mounts provide a simple, highly reproducible method for exchanging samples with minimal shifts of thread position relative to the spectrometer. Immediately following the sample mounting procedure, the thread was aligned with the x-ray beam via a computerized transmission scan perpendicular to the beam (inboard/outboard –(cid:1) (cid:1)in Figure S1) using the downstream diode (Thorlabs PDA36A). To prevent the destruction of the diode, a reduction in beam intensity was necessary in addition to a protective layer of attenuation foil on the diode. Once the center of the thread was confirmed horizontally, the thread was brought into the focal point of the camera (cid:1)respectively, in Figure S1), with the using the vertical and along-the-beam motors ( y and z x inboard/outboard ( required. x (cid:2) position held constant. Typically, a few small steps of 50-100 µm were The sample delivery system, Figure S2(A), includes plastic goal posts protruding from an aluminum plate and angled 45° to the beam. Stainless steel syringe needles of radius 203 µm are securely fitted through holes in these posts to deliver sample reproducibly at the beam position. The needles are positioned far outside of the spectrometer’s acceptance volume to prevent contamination from ancillary Mn. A stepper motor is mounted on the upstream side of the sample assembly, and an aluminum spool of starting radius 9.55 mm or 11.2 mm (data set #1 only, see Table S2 for an experimental summary) is attached to the motor to pull and collect thread as it passes through the interaction region. See Figure S2(A). In general, sample positions were not changed except upon mounting and aligning in small increments as determined by transmission scan and by the camera system, see more in Establishing the Interaction Region and Timing Schemes. Note that the sample coated spool was covered completely with aluminum foil during the positioning time to prevent any unintentional x-ray or light exposure that could alter the protein state prematurely. Once the sample position was finalized, this covering was partially removed to allow free translation of the thread, and the collection spool was rotated repeatedly until the previously enclosed fresh sample reached the beam position. Spectrometer Calibration: An in situ calibration of the detector pixels is achieved by measuring the positions of the elastic scattering peaks while scanning the monochromator through the x-ray emission energy range (6445-6510 eV). The calibration methodology has been described in detail elsewhere (26, 51, 52). Note that the line width of the elastic peaks is the convolution of the monochromator and spectrometer energy resolutions. It is important to recognize that changes in sample position relative to the spectrometer alter the detector calibration. We thus developed a system to reproducibly position the sample (tolerance <50 µm), and calibrated the spectrometer daily following the completion of measurements on a given sample (typical sample consisted of 6-8 m of painted thread). In addition to this, MnCl2 solution emission spectra were collected after each sample on the same thread without any sample repositioning. These spectra were used as an internal reference. Establishing the Interaction Region and Timing Schemes: Pump/probe Alignment and Camera Setup: In addition to the criticality of reproducible sample position in the miniXES system, exact alignment between laser and x-ray is crucial for accurate TR data. Light excitation was provided by a Spectra-Physics Empower laser (527 nm, pulse length ~200 ns) placed at an angle θ~5° to the incident x-ray beam. Additional optics allowed for a FWHM spot size of ~220 µm, at a laser fluence of ~11 mJ/mm2. The laser spot size was selected to be about twice larger than x-ray spot size to ensure best alignment. The x-ray beam was visualized on a phosphor (SPI P47 coated scintillator) positioned at the previously established sample position relative to the spectrometer and the laser was moved to an overlapping position. To facilitate fine adjustments to the sample alignment and monitor the interaction region during data collection, a Prosilica camera directed at the sample was mounted vertically on a small X,Y,Z stage. To improve our precision, a lens extension with a focal depth of ~50 µm was used. Once in place, the focal region of the camera was aligned to a test thread known to be in alignment with the spectrometer setup. The thread was then removed and replaced by the phosphor mounted at the same position, see Figure S2. It is critical that the phosphor surface appears in focus on the camera. X-ray/laser alignment was checked periodically using the phosphor. No deviations were found during 4 days of beamtime. Once these alignments are complete, fine adjustments of the sample/spectrometer alignment for optimal emission signal are made by small spectrometer translations only. From that point the sample position is ‘fixed’ at the focal region of the camera, the spectrometer position is finalized, and data collection starts. X-ray/Laser Synchronization: For verification of timing, the phosphor was replaced by a small Si photodiode aligned at the beam position (in the focus of the camera). Synchronization between the pump (laser) and probe (x-ray) pulses was performed by recording the respective diode signals on an oscilloscope. A field programmable gate array (FPGA) was used to maintain the timing between the laser and x-ray pulses. The programmable delay between laser and x-rays pulses was adjusted to overlap in time the peaks of each pulse that was subsequently defined as time zero. Given the broad 44 µs FWHM X-ray and ~200 ns FWHM laser pulses, it is more exact to synchronize via peak position instead of rising edges. It follows that spectra recorded at a set delay include fluorescence from ±20 µs, see Figure S3. Photosystem II advancement in the Kok cycle: In this work we do not deconvolute the XES spectra but instead analyze changes in the first moments as a function of the number of laser flashes and time delay to the probe pulse. For a prediction of S-state composition, we refer to Han et al. who recently established the maximal experimentally achievable S-state composition after each laser flash at room temperature for PS II membrane preparations from spinach, Table S1 (49). To ensure the most efficient S-state , assuming the threads are fully saturated with sample, the ratio of incident photons to PS II conversion, we emulate previous experimental laser settings. At a laser fluence of ∼11 mJ/mm2 centers is ∼560. This matches the only other XES experiment by Messinger et al. in which advancement to ∼65% of the S0 state was determined by EPR (29). Given we are able to probe oxidative effect more than ∼30%. the created states long before any decay occurs, unlike the manual freezing in Messinger et al., we set this value as a lower bound on our 3F40ms data. This convolution of states will be similar for each of the S4 time delays, given they are pumped with the same number of laser flashes. It is clear, therefore, that any reductive effects we observe are in reality more significant than they appear. It should also be emphasized that convolution with S2,3 data is not likely to lessen any Data Collection: Data collection scans were synchronized in the following manner – for parameters α, β and γ see Table S2: (i) the stepper motor translates the sample α microns through the beam position transporting fresh material into the beam; (ii) the specified number of laser flashes are delivered to the fresh sample; (iii) subsequently, the ~44 µs x-ray pulse arrives at a delay, β; (iv) immediately following the closure of the shutter, the cycle repeats. Note that both laser and x-ray pulse at the same frequency γ. During the repetition time, the Pilatus continuously collects a single image. Sample threads were mounted in ~5 to ~10 m lengths and monitored every ~0.5 m resulting in multiple Pilatus images per generated state. These images were directly summed during data processing for improved statistics. To evaluate the effects of adding up data obtained on different (and thus re-positioned relative to spectrometer) threads, we first analyzed the strong MnCl2 spectra. Highly concentrated MnCl2 solution (~1 M) was painted onto each thread at the conclusion of PS II measurements prior to sample exchange, and a single PSD image was collected. First moments were calculated around the range 6490-6495 eV (as the MnCl2 peak is higher in energy than PS II) and a standard deviation of 0.02 eV was found. To verify that summing threads would not affect the resolution, we also checked with data that had been split randomly into two groups and observed shifts between 0.005 – 0.02 eV. Consistent with both these methods, we consider 0.02 eV to be our maximal resolution for PSII XES collection. Note also that averaging of multiple threads broadens the data slightly due to minor variations in thread position. These variations are random and therefore can be assumed to be even about the Kβ1,3 peak with limited effect on the first moments. Given the sensitivity to sample positioning, each thread was used to collect all S-states/intermediates and only one calibration file was used to process the data reducing the error due to the calibration files themselves. However, calibrations were still taken daily to ensure large shifts in the monochromator calibration or other major effects did not occur. To completely reduce any effects of position or monochromator calibration drift on our S-state shifts, we collected all spectra in approximately equally intensities (equal number of detector exposures) for every thread. It follows that any changes in the absolute position of the sample do not affect the relative shifts between the states. Processing PS II spectra A total of six beamtimes were accomplished, with two devoted to methodology development and four to data collection. For every beamtime (data sets 1 – 4, Table S2), each TR XES spectrum is a direct summation of the Pilatus exposures for all individual threads using a single calibration. Differing data sets are also summed for greater statistics. Results from different beamtimes were merged using the strong MnCl2 signal as an internal energy-calibration reference. Similar to PS II, all MnCl2 spectra in the data set are summed and calibrated with one calibration file. Data processing / averaging Averaging and a Systematic Error Estimation Statistical first moments were analyzed for PS II data processed in three different ways: completely unprocessed, with background subtraction only, and after smoothing and background removal, Tables S6 & S7. No significant differences were found in the first moment trends between these methods; errors introduced by processing do not alter unprocessed data trends that are significant to within a 95% confidence interval, see Table 2. Examples of data for the S3 to S0 transition are shown in Figure S4 and first moments of the raw data are in Table S3-S7. Due to high levels of background, and relatively low statistics, consistent background removal was necessary to reliably compute first moment values for comparison with other published works both past and future. A quartic polynomial was fit to the spectral region excluding the Kβ peaks, Figure S6. The computed polynomial was then subtracted from the spectrum. Note that removal of a linear background was found to create significant variability in the first moments and is thus insufficient. Data were then normalized to the Kβ1,3 peak maximum for the purpose of comparison. For both raw and background subtracted data some variability in the first moment trends between beamtimes persists due to noise in the spectral range. To combat this problem, smoothing techniques were tested. Wavelet Transforms The wavelet transform method was chosen to smooth non-physical rapid fluctuations in the data while maintaining the integrity of the spectral shape. The wavelet transformation of our spectrum f(E) is (53): w(a,E) = C−1/2 ψ a−1/2 ψ* "E )d "E "E − E f ( ∞∫ −∞ # % $ & ( ' a (1) E −b a % ' & " ψa,b(E) = ψ $ # where Cψ is a normalization constant. The base wavelet function, , is dilated a b ; and translated to allow for decomposition of the spectrum. Dilation is described by characterizes translation. For our purposes, we use the Mexican Hat wavelet function, see equation 2 (53). ψ(E) = (1− E2)e−E2/2 Wavelet functions are calculated for the range of 0.01 to 30, over the interval 6453.7 eV to 6513 eV. The energy-frequency plane of w(a,E) allows for visualization of the most appropriate cutoff for the dilation parameter used in the spectral reconstruction. Figure S5 provides an example spectrogram. Note that the intensity in z is determined by the Our chosen parameters leave the Kβ1,3 peak adequately smoothed but without significant broadening while the Kβ’ peak maintains some noisy character. It is possible to completely eradicate noise in Kβ’ peak, but the process results in significant broadening and symmetrizing of the Kβ1,3 peak, an effect we want to avoid as it can skew the first moment calculations. Figure S6 compares the wavelet reconstruction of 3F!""!! state data to the ‘raw’ spectrum. value (53). w(a,E) (2) a a Error Determination: ! For some experiments, it is common to present the standard error of the mean (SEM) from the second moment, essentially an evaluation of the error on the first moment (our measure of spectral position, mathematically the weighted arithmetic mean). The SEM can be written as follows: SEM= !!!!and!!!!= (!!!!)!!! !!! (3) µ2 is the weighted second moment calculated from the intensities, Ii, and emission energies, Ei, at every point, as well as the weighted first moment, !. N is the number of observations. Given that the SEM is highly dependent on the sampling/binning of data into N groups, we prefer to avoid this error analysis method. Additionally, SEM results can be misleading in that overlapping error bars do not necessarily exclude statistically significant differences. Analysis of Variance ANalysis Of VAriance (ANOVA) between experimental groups was used to determine the confidence level for the observed first moment shifts between states. One-way ANOVA, calculated from the first moments of raw unprocessed normalized data, yields the p-values in Table 2. These values represent the probability of the null hypothesis, i.e. that a difference in the first moment between the two states listed is a random statistical variation. In this study we take the 95% confidence interval (1.96σ) to be significant. The first moment trends for raw data match well with those for smoothed, background- subtracted spectra (i.e. ‘processed’), Table S3 within our confidence levels, Table 2. In other words, for those states in which we likely reject the null hypothesis, smoothing and background subtraction do not alter the trends in the changes to the first moments. We present first moment shifts for data collected on the same threads, Tables S3, S5, & S7. However, ANOVA analysis between the same states (excluding 2F) from different beamtimes shows no statistical variation, p≥0.19, Tables 2 & S4. The only outlier 2F spectrum from data set 3 has a dip on the low energy side of the Kβ1,3 peak between 6.485 – 6.488 keV. This noise causes a significant variation in first moments between data sets and skews the mean of summed data to lower energies. Unfortunately, this spectral noise is not easy to remove without completely excluding the 2F data from that beamtime, after which statistical invariance is re-established. However, in the full data set, Table S4, inclusion versus exclusion of beamtime #3 2F data only changes the first moment by 0.01eV, again, below our systematic error. To preserve statistics we include these data. DFT Calculations: Undamaged coordinates from the fs-XRD structure were used as the initial geometry (36). Geometry optimizations were performed in Gaussian09 (38) on a 94 atom model of S3 including the Mn core and a portion of the surrounding backbone using the bp86 (54) functional and def2 tzvp basis set (55). Eight single point calculations representing the possible relative spin orientations of the Mn centers with the spin lying along a single axis were used to build the Heisenberg-Dirac-Van Vleck (HDVV) Hamiltonian, where the total energy is given as Here, !! is a constant background energy, !! and !! are the spin operators of the individual Mn centers with indices i and j, and !!" is the coupling constant between them. The values of !!" and !! were determined by minimizing the means squared error in energy between the HDVV !=!!− !!"!!⋅!! Hamiltonian and the energies obtained through DFT. The S3 ground state spin and energy were obtained from the lowest eigenvalue of the fitted HDVV Hamiltonian. !" (4) Fig. S1. A) 3D model of the miniXES spectrometer in the von Hamos configuration using the GaP 440 Bragg reflections. Multiple crystals are used for increased solid angle collection and are positioned for optimal use of the detector. The yellow square represents the sample plane. (B) Schematic explanation of the dispersive direction on the detector face due to the Bragg reflections. (A) (B) Fig. S2. Sample delivery and sample positioning system. (A) Plastic (green) goalposts are mounted at 45° to the incident beam. Sample coated thread (red) passes through two narrow steel needles embedded as guides in the plastic piece (green). A computer controlled step motor rotates a spool in defined intervals to translate the sample and collect the thread after it passes the interaction region between the goalposts. The kinematic mount is below the white plate to allow for reproducible sample exchange. (B) The plastic piece (green) in (A) can be exchanged to allow for alignment equipment to be placed at the exact sample position. Here is a visualization of both the small timing photodiode (gold) and the alignment phosphor and how they attach to the apparatus. Fig. S3. Timing schematic showing the example delay time 50 µs. Laser (green)/X-ray (purple) synchronization was done via the pulse centers. The number of x-ray bunches (dark purple) per 44 µs pulse is variable depending on storage ring parameters during each beamtime. Data sets 1- 3 had approximately 4000, data set 4 only ~300. 1 0.95 0.9 0.85 0.8 0.75 ) . u . a ( y t i s n e t n I d e z i l a m r o N 6450 6460 500(cid:43)s 200(cid:43)s 50(cid:43)s 6490 6500 6470 6480 Emission Energy (eV) Fig. S4. Mn Kβ XES for OEC during the S3 ! S0 transition at delay times of 500 µs, 200 µs, and 50 µs from the third laser flash. Data from different beamtimes are aligned using the MnCl2 standard. S/N reflects the amount of data collected per state and improves with the merging of several data sets. The 500 µs spectrum is a sum of data sets 1 – 3; the 200 µs spectrum includes data from data sets 2 and 3; and the 50 µs data set is from data set 3 only. See Tables S2 & S8 for details. Background levels from stray scatter entering the detector are significant, making up approximately three quarters of the total signal. It is clear, however, that the signal to background remains essentially static with little beamtime-to-beamtime variation. Emission Energy (eV) Fig. S5. Example spectrogram for the 500 µs raw data from data set 2. The red line is our chosen frequency cutoff, below which we begin to lose spectral features or experience spectral w(a,E) broadening due to the wavelet reconstruction. The grey-scale color is based on the value at each point. Original Data Reconstructed Signal Quartic Background . u . a / y t i s n e t n I 6460 6470 Fig. S6. Example of unprocessed 3F!""!! data from data set 2 overlaid with its wavelet- Emission Energy / eV reconstructed spectrum. The quartic background (black line) shown was calculated for the reconstructed signal. 6510 6480 6490 6500 ) . u . a ( y t i s n e t n I d e z i l a m r o N 1 0.8 0.6 0.4 0.2 0 S1 S3 6486 6488 6490 6492 6494 Emission Energy (eV) Fig. S7. Fully processed (background subtracted and smoothed) 0F (majority S1) versus 2F (majority S3) Kβ emission spectra for data sets 2 and 3 collected on the same threads. Fig S8. Dot plot of all calculated first moments from raw data. Each dot represents a calculated first moment from a thread collected during the beamtime corresponding to its shape in the legend. The bars represent the 95% (solid) and 99% (dashed) confidence intervals. Fig S9. Visualization of the key energetic steps in the water oxidation mechanism from Table S10. Energies in parentheses represent the energies obtained from broken symmetry DFT calculations. This figure is an addition to Figure 4 showing more details. Table S1. Maximal experimentally achievable S-State composition at room temperature. Flash No. 0 1 2 3 As determined by Han et al. (49). S1 100 10 S2 90 5 23 ± 1 S3 76 22 ± 1 S0 74 Table S2. S-state dependent scan parameters State No. of Laser β* (µs) α* (µm) γ*** (Hz) Data Sets 600 / 450 300 / 300 600 / 450 / 450 /450 / 450 600 / 450 S0 S1 S2 S3 S4a S4b S4' * Translation in µm (spool radius: 11.2 / 9.55 mm, for data sets 1 / 2,3,4) Delay Time * Laser and x-ray Frequency 40,000 n/a 500 500 50 200 500 1 – 4 1 – 3 1 – 3 2 – 4 3 & 4 2 – 4 1 – 3 10 20 20 20 20 20 20 Pulses 3 0 1 2 3 3 3 Table S3. First moments of unprocessed versus processed data. Unprocessed Data Processed Data 1st Moments – 6490 eV for beamtimes: 1 – 3 0.085 0.093 0.069 n/a n/a 0.078 2 & 3 0.087 0.095 0.066 0.068 0.090 0.081 Majority S-state Trends 1 – 3 0.497 0.541 0.418 n/a n/a 0.485 2 & 3 0.474 0.518 0.359 0.401 0.526 0.448 S2 →S′4 →S0 →S1 S2 →S′4 →S0 →S1 Majority S-State Flash 3F40ms 0F 1F 2F 3F!""!! 3F!""!! S0 S1 S2 S3 S4b S'4 Beamtimes 1 – 3 2 & 3 Max data S2 →S3 →S′4 →S0 →S4b →S1 S2 →S3 →S′4 →S0 →S1 →S4b S3 →S2 →S′4 →S0 →S4b →S1 S3 →S2 →S′4 →S0 →S4b →S1 First moments are calculated over the range 6.485 – 6.495 keV. Absolute magnitude variations between processed and unprocessed data occur primarily due to background removal. Only states taken on the same threads are compared, excluding the row labeled ‘max data’ which assumes the statistical invariance between beamtimes to compare the maximum collected statistics for each state. Table S4. First moments of unprocessed data from all beamtimes. 1st Moments – 6490 eV 0.085 0.088 0.066 0.070 0.086 0.088 0.071 0.088 0.088 0.0750 Flash Majority S-State 3F40ms 0F 1F 2F 3F!"!! 3F!""!! 3F!""!! S0 S1 S2 S3 S4a S4b S'4 First moments are calculated over the range 6.485 – 6.495 keV. All beamtimes are included for comparison due to the statistical invariance between the states. Additional statistics were collected for S3, S4n and S4 for data set 3, see Table S8. The columns are split to reflect the adjusted first moments including these statistics. Table S5. First moments of unprocessed 50µs delay data. Flash 3F40ms 2F 3F!"!! 3F!""!! Majority S-State S0 S3 S4n S4 1st Moments – 6490 eV for beamtimes: 2, 3 & 4 0.086 0.069 n/a 0.088 3 & 4 0.084 0.067 0.085 0.092 First moments are calculated over the range 6.485 – 6.495 keV. Only states taken on the same threads are compared in each column. Table S6. First Moments – 6490 eV for individual data sets. Raw 3 1 Flash 3F40ms 0 1 2 1 0.083 0.088 0.075 n/a Majority S-State 4 S0 0.57 S1 n/a S2 n/a S3 0.51 S4a 0.51 S4b 0.57 S!4 n/a First moments are calculated over the range 6.485 – 6.495 keV. Only states taken on the same threads are compared in each column. Additional statistics were collected for S3, S4n and S4 for data set 3. The columns are split to reflect the adjusted first moments. 4 2 0.093 0.55 0.092 0.072 0.57 n/a 0.104 0.074 0.067 0.058 n/a 0.51 0.076 0.051 0.060 0.082 n/a n/a 0.080 0.088 0.090 n/a 0.079 0.090 0.086 0.093 n/a 0.53 0.084 0.070 2 0.53 0.44 0.53 0.53 0.35 0.38 0.42 0.28 0.39 n/a 0.57 0.58 0.44 0.66 0.54 0.46 0.41 !!!"!" n/a !!!""!" n/a !!!""!" 0.073 1 0.53 0.55 0.49 n/a n/a n/a 0.50 4 0.63 n/a n/a 0.59 0.58 0.64 n/a n/a Processed 3 Raw – background Data Sets 2 0.49 0.49 0.34 0.43 n/a 0.43 0.44 3 0.42 0.58 0.34 0.34 0.57 0.68 0.41 0.41 0.61 0.58 Raw Flash 2 0.069 Raw-background Processed !!!"!" 0.088 !!!""!" 0.091 Table S7. First moments -6490 eV for averaged data sets 3 and 4. Majority S-State S3 S4a S4b First moments are calculated over the range 6.485 – 6.495 keV. Only states taken on the same threads are compared in each column. These are calculated including the additional statistics collected for data set 3. 0.50 0.58 0.59 0.51 0.60 0.62 Data Sets 3 Table S8. Approximate number of x-ray pulses per state per beamtime. Majority S-State S0 S1 S2 S3 S4a S4b S!4 Additional statistics were collected for S3, S4n and S4 for data set 3. The columns are split to reflect the additional x-ray pulses for these states. Unless specified, figures and calculations did not reflect these data to preserve the comparison between threads. 55,333 60,233 55,333 57,178 48,889 107,222 57,178 48,889 111,667 57,178 48,889 110,556 57,400 265,805 195,300 158,972 226,400 168,845 228,623 162,039 275,289 217,734 277,512 107,222 n/a n/a 2 1 Flash 3F40ms 0 1 2 43,250 60,000 62,267 72,800 42,083 61,556 62,000 n/a n/a 60,889 !!!"!" n/a !!!""!" n/a !!!""!" 43,750 60,889 4 Total n/a S3! S4OOH!+H2O! Table S9. Interatomic distances (Å) based on the optimized coordinates for each S-state of the proposed model. ! Mn1-Mn2! Mn1-Mn3! Mn1-Mn4! Mn2-Mn3! Mn2-Mn4! Mn3-Mn4! Mn1-Ca! Mn2-Ca! Mn3-Ca! Mn4-Ca! 2.81% 3.57% 5.34% 2.80% 5.14% 2.71% 3.14% 3.37% 3.45% 3.87% 2.79% 3.68% 5.56% 2.86% 5.19% 2.74% 3.58% 3.32% 3.34% 3.67% S1! 2.74% 3.31% 4.87% 2.79% 5.10% 2.72% 3.50% 3.31% 3.40% 3.61% S0! 2.84% 3.50% 5.43% 2.73% 5.24% 2.93% 3.58% 3.27% 3.45% 3.88% S2! 2.76% 3.39% 5.16% 2.80% 5.25% 2.78% 3.48% 3.32% 3.35% 3.80% Table S10. DFT energies and spin states E (Hartree) ΔE (eV) S0 S1 S2 S3 S3OOH -8023.4311 -8022.8449 -8022.2197 -8021.5906 -8097.4059 S3+ H2O = S3OOH/H2O + H+ S3 = S4OOH+H++e- S4(Mn3B-OO-Mn1D) S3 = S4(Mn3B-OO-Mn1D)+H++e- S4OOH + H2O -8097.4031 -8020.9200 n/a 15.95 17.01 17.12 17.62 18.76 18.25 Spin State 1/2 1 1/2 3 1 1/2 1/2 Broken Symmetry E (Hartree) -8021.5878 -8097.4351 -8020.9514 -8097.4206 Broken symmetry ΔE (eV) Broken symmetry spin 3 1 1/2 1/2 16.75 17.31 0.39 17.70 0.08 S3 + H2O = S4OOH/H2O+H++e- S3OOH/H2O= S4OOH/H2O+e- * Broken symmetry calculations only available for S3 and S4 states. Unless it is specified, ΔE is the total change in energy from the previous state according to equation Sn+1=Sn+H++e-, see also Fig. 4 for scheme of Kok cycle. Whenever a proton and electron were removed, ΔE includes their energies. Such a presentation avoids estimating separate proton energies, which might have large errors. Optimized Cartesian Coordinates S0 C -6.63321 1.87587 1.01510 C -5.86435 0.56014 0.90321 C -4.49165 0.62932 0.22296 O -3.84687 -0.43384 0.10380 O -4.11284 1.79073 -0.18391 C -0.84672 -4.82847 -4.55234 C 0.32852 -3.90080 -4.25068 C 0.13495 -2.97397 -3.04387 O -0.93416 -3.02102 -2.38902 O 1.12588 -2.18747 -2.80521 C 4.13044 1.14442 -3.62258 N 4.92460 1.05067 -2.49280 C 3.04035 0.34667 -3.37052 C 4.30444 0.23243 -1.60546 N 3.15783 -0.20941 -2.10918 C 5.97422 1.41024 1.56162 N 5.56022 2.52946 0.86325 C 4.89235 0.60538 1.86969 C 4.23878 2.39067 0.76184 N 3.78931 1.25234 1.35992 C 4.18780 -3.68410 1.81550 C 3.15024 -2.72654 1.24167 O 2.25630 -2.31764 2.05254 O 3.25416 -2.42910 0.01175 C -1.96284 -4.22591 3.43695 C -1.50124 -3.22785 2.36778 O -2.15796 -3.09582 1.30882 C -3.13451 -5.11268 3.02099 O -0.43568 -2.58321 2.70127 C 1.22205 1.75281 4.65986 C 0.88255 1.08003 3.34078 O 1.00444 -0.18092 3.29127 O 0.51860 1.85354 2.38930 O 0.32023 -2.21285 0.08310 Ca -1.86527 -1.52324 -0.59413 Mn 1.68933 -1.36058 -0.98925 O -0.92898 -0.25156 1.37679 Mn 0.54659 -1.28763 1.60484 O 1.45995 0.03324 0.55275 Mn -0.09785 1.17814 0.63292 O -1.55789 2.32222 0.74678 Mn -2.30877 2.31233 -0.92165 O -0.81702 0.58051 -1.10075 O -2.06168 4.61179 1.86295 O -3.29780 -3.62029 -1.17834 O -2.92807 -0.70277 -2.77683 O -2.97370 1.98652 -2.67907 O -3.41675 4.43536 -0.51320 H -5.68826 0.11960 1.89841 H -6.44881 -0.19404 0.35077 H -6.82841 2.30943 0.02375 H -1.04236 -5.50798 -3.71026 H -1.76681 -4.25444 -4.73037 H 1.25365 -4.47329 -4.07014 H 0.55144 -3.25307 -5.11515 H 2.19179 0.11495 -4.00526 H 4.71211 -0.01402 -0.62928 H 4.80357 -0.33576 2.40190 H 2.83474 0.83000 1.21552 H 3.55425 3.07477 0.26803 H 5.17295 -3.49244 1.36990 H 3.89296 -4.71073 1.54706 H -4.00160 -4.50730 2.72276 H -3.43853 -5.77028 3.85057 H -2.86839 -5.74930 2.16437 H -1.08540 -4.82498 3.72816 H 2.29604 1.62516 4.86056 H 0.97903 2.82088 4.62824 H -0.25242 0.81081 -1.85890 H -1.73511 3.69740 1.53870 H -2.55609 4.40847 2.67413 H -2.52931 -3.85239 -1.74859 H -3.06492 -3.92396 -0.27438 H -3.85740 -0.98218 -2.71932 H -2.96164 0.31028 -2.79420 H -2.41164 2.44020 -3.33163 H -4.20971 3.90353 -0.30279 H -3.01818 4.64679 0.38144 C 1.16072 5.81891 -0.68194 C 1.29736 4.62647 -1.64531 C 0.42360 3.46429 -1.18703 O 0.93215 2.66163 -0.33895 O -0.75284 3.40689 -1.67678 H 1.46392 5.53655 0.33588 H 0.11979 6.16853 -0.63941 H 2.34343 4.28771 -1.68385 H 0.98335 4.91948 -2.65769 H 1.79489 6.65528 -1.01295 H 4.40801 1.74931 -4.47807 H 7.02045 1.25342 1.81469 H 4.23205 -3.61024 2.90891 C -6.18112 3.05918 1.13271 C -5.73667 1.59731 1.12514 C -4.44238 1.28914 0.35486 O -4.04232 0.09500 0.38168 O -3.87397 2.24735 -0.26709 C -1.64341 -4.29609 -4.70566 C -0.31704 -3.67090 -4.27952 C -0.40289 -2.76692 -3.04789 O -1.48168 -2.60361 -2.45009 O 0.72805 -2.20713 -2.73575 C 3.56746 1.39556 -3.31530 N 4.55016 0.64441 -2.69256 C 2.37486 0.82643 -2.94354 C 3.94682 -0.32606 -1.96345 N 2.62878 -0.24062 -2.10350 C 6.29660 0.16348 1.14900 N 6.04124 1.42121 0.63408 C 5.11444 -0.48019 1.46225 C 4.71443 1.52595 0.64521 N 4.11288 0.40793 1.14012 C 3.61563 -4.37243 1.15671 C 2.69501 -3.20717 0.81588 O 1.91813 -2.81406 1.74395 O 2.76794 -2.73976 -0.36515 C -2.48985 -4.16555 3.22114 C -1.90797 -3.16321 2.21907 O -2.56090 -2.83465 1.20357 C -3.79571 -4.82307 2.77969 O -0.74138 -2.72425 2.55504 C 1.65113 1.08355 4.75853 C 1.13496 0.61674 3.40764 O 1.03561 -0.64728 3.25104 O 0.88008 1.51301 2.54652 H -2.21929 -3.62904 4.32858 H -0.64294 -5.44168 -5.44455 H -6.06610 2.61854 1.59439 H -7.60158 1.72043 1.51696 H 5.76513 1.58248 -2.28857 H 0.67169 1.25855 5.47264 S1 O -0.05697 -2.19128 -0.08713 Ca -2.20790 -1.04792 -0.51628 Mn 1.23817 -1.37443 -1.04678 O -0.93599 -0.22118 1.45582 Mn 0.36142 -1.47012 1.54633 O 1.43916 -0.29347 0.51587 Mn 0.05344 1.17321 0.70221 O -1.14432 2.50313 0.88641 Mn -1.69844 2.50889 -0.89583 O -0.72532 0.87438 -0.92411 O -1.47505 4.89417 1.84838 O -3.91527 -2.92407 -1.20210 O -3.17882 -0.10292 -2.60135 O -2.30267 2.37312 -2.69603 O -2.59457 4.48110 -0.62293 H -5.58333 1.22449 2.15116 H -6.52026 0.95062 0.69564 H -6.34878 3.43037 0.11143 H -2.05074 -4.94329 -3.91580 H -2.39503 -3.52247 -4.91549 H 0.43932 -4.44300 -4.05990 H 0.11509 -3.06275 -5.09135 H 1.35929 1.11885 -3.18669 H 4.46914 -1.04805 -1.34390 H 4.89870 -1.45426 1.88737 H 3.09335 0.19951 1.08702 H 4.12824 2.37760 0.31226 H 4.57796 -4.26886 0.63836 H 3.14039 -5.29987 0.80042 H -4.56453 -4.06808 2.56535 H -4.17736 -5.49490 3.56438 H -3.65396 -5.41682 1.86468 H -1.70893 -4.91373 3.43083 H 2.75188 1.07513 4.73452 H 1.31652 2.10786 4.96149 H -1.20174 3.93898 1.62878 H -2.08931 4.79494 2.59486 H -3.22029 -3.19896 -1.83977 H -3.66856 -3.33691 -0.34637 H -4.12758 -0.01470 -2.40740 H -2.85099 0.85766 -2.72446 H -1.55077 2.52513 -3.29510 H -3.46379 4.05746 -0.42133 H -2.25287 4.80247 0.26293 C 2.59128 5.28784 0.19203 C 2.36717 4.40673 -1.04738 C 1.20831 3.42857 -0.83770 O 1.43916 2.47781 -0.00366 C 6.49804 2.35684 -0.54105 C 5.80126 1.04873 -0.91450 C 4.39099 0.85324 -0.33568 O 3.91887 -0.30571 -0.33551 O 3.83436 1.92977 0.08364 C 1.03953 -4.51143 4.70169 C -0.18246 -3.68892 4.29942 C 0.03087 -2.77981 3.08713 O 1.13715 -2.71955 2.52086 O -1.02633 -2.09954 2.75490 C -3.60550 1.53667 3.48411 N -4.56813 1.13142 2.57530 C -2.47908 0.81621 3.17388 C -4.01508 0.21039 1.74958 N -2.74841 -0.00067 2.09139 C -6.13695 0.89318 -1.36326 N -5.76301 2.12226 -0.85192 C -5.02121 0.13275 -1.65981 C -4.43222 2.09477 -0.84831 N -3.93993 0.92005 -1.33236 C -4.05241 -3.98091 -1.19488 C -3.02146 -2.91775 -0.83933 O -2.18592 -2.61675 -1.75099 O -3.07015 -2.43392 0.33565 C 2.08203 -4.45907 -3.08272 O 0.13631 3.63925 -1.46321 H 2.80969 4.67064 1.07452 H 1.69459 5.88431 0.41447 H 3.27893 3.82685 -1.26552 H 2.13475 5.02566 -1.92573 H 3.43419 5.97797 0.03318 H 3.79876 2.24760 -3.94314 H 7.31312 -0.20114 1.27875 H 3.76161 -4.45332 2.24076 H -2.62824 -3.61312 4.16594 H -1.51133 -4.90704 -5.61226 H -5.41857 3.70354 1.59439 H -7.11737 3.18304 1.70060 H 5.54090 0.86020 -2.65358 H 1.32188 0.39884 5.55020 S2 C 1.59042 -3.37619 -2.11746 O 2.24934 -3.09562 -1.09200 C 3.32618 -5.21217 -2.61677 O 0.48385 -2.82434 -2.49282 C -1.40767 1.16077 -4.84472 C -1.06109 0.67417 -3.44895 O -1.09844 -0.59229 -3.26471 O -0.75965 1.55594 -2.59022 O -0.19506 -2.17175 0.11742 Ca 2.02207 -1.24259 0.59034 Mn -1.42918 -1.23839 1.04597 O 0.91362 -0.33540 -1.50157 Mn -0.50782 -1.44134 -1.54456 O -1.43268 -0.13972 -0.51624 Mn 0.12735 1.16488 -0.75962 O 1.35178 2.43808 -1.02633 Mn 2.05990 2.55180 0.68577 O 0.97445 0.88916 0.80636 O 2.30877 4.43713 -2.55585 O 3.56457 -3.24571 1.34235 O 2.99637 -0.29126 2.65942 O 2.66631 2.30651 2.47144 O 2.69826 4.24955 0.44324 H 5.68477 0.97002 -2.01017 H 6.39386 0.17011 -0.61576 H 6.66495 2.42145 0.54398 H 1.34632 -5.18792 3.89117 H 1.89598 -3.86165 4.92891 H -1.04575 -4.33513 4.07006 H -0.51534 -3.04136 5.12780 H -1.50100 0.82894 3.64134 H -4.54323 -0.27293 0.93321 H -4.89719 -0.85951 -2.07974 H -2.95473 0.59941 -1.23931 H -3.76874 2.88713 -0.51437 H -5.00516 -3.77938 -0.68819 H -3.67911 -4.95290 -0.83595 H 4.16273 -4.52212 -2.44028 H 3.63876 -5.94986 -3.37223 H 3.13854 -5.74907 -1.67542 H 1.23832 -5.14411 -3.26433 H -1.97958 2.09551 -4.78608 H -0.46753 1.37560 -5.37590 H 1.80112 3.66976 -2.16818 H 3.18797 4.05189 -2.71112 H 2.82358 -3.46849 1.94777 H 3.31914 -3.63424 0.47459 H 3.95623 -0.44117 2.66947 C 6.50874 1.94899 -1.01160 C 5.82206 0.62622 -0.67490 C 4.43811 0.70950 -0.01710 O 3.88341 -0.34735 0.31983 O 3.96645 1.90969 0.12772 C 0.08434 -5.64532 3.79005 C -1.05538 -4.73093 3.34490 C -0.62023 -3.55734 2.45439 O 0.59298 -3.32457 2.27043 O -1.62183 -2.89565 1.97168 C -4.00857 1.11126 3.28003 N -4.93054 0.60033 2.37880 C -2.80864 0.52858 2.95314 C -4.27815 -0.24349 1.53946 N -2.99492 -0.30633 1.87023 C -5.84557 1.44501 -1.71649 N -5.56613 2.44191 -0.80052 S3 H 2.87871 0.72676 2.63026 H 1.94473 2.62824 3.04236 H 2.56123 4.51153 -0.49752 C -1.51408 5.75567 0.78904 C -1.74211 4.39435 1.46836 C -0.73999 3.36734 0.95032 O -1.12934 2.56953 0.03975 O 0.42363 3.44305 1.45688 H -1.66179 5.68185 -0.29799 H -0.48852 6.10644 0.96947 H -2.76184 4.02968 1.28004 H -1.60017 4.49174 2.55534 H -2.21678 6.50675 1.18084 H -3.80248 2.28257 4.24479 H -7.18314 0.63123 -1.50376 H -4.19166 -4.04499 -2.28093 H 2.26370 -3.95653 -4.04757 H 0.82182 -5.12203 5.59170 H 5.88456 3.22386 -0.81932 H 7.47509 2.43588 -1.04441 H -5.49441 1.52738 2.45135 H -1.96303 0.39597 -5.40025 C -4.70899 0.70627 -1.98858 C -4.26934 2.29734 -0.53475 N -3.70914 1.26834 -1.22839 C -3.95208 -3.38952 -2.08432 C -2.88197 -2.48017 -1.49516 O -1.93126 -2.14852 -2.27840 O -3.00876 -2.13830 -0.28429 C 2.51084 -3.89426 -3.24725 C 1.87645 -2.92990 -2.23569 O 2.38844 -2.76545 -1.11015 C 3.67489 -4.71656 -2.69848 O 0.81776 -2.34362 -2.70138 C -1.14895 1.91287 -4.69539 C -0.75320 1.23839 -3.39324 O -0.71892 -0.02891 -3.37510 O -0.50119 2.01710 -2.41491 O -0.17112 -2.11084 -0.19689 Ca 1.71793 -1.22931 1.00243 Mn -1.42308 -1.28521 0.87449 O 1.13071 -0.01993 -1.37571 Mn -0.30023 -1.10394 -1.68974 O -1.23717 0.12915 -0.55538 Mn 0.29706 1.43201 -0.66793 O 1.56864 2.72857 -0.88139 Mn 2.25194 2.59419 0.81073 O 1.22297 1.16701 0.91701 O 3.18882 -3.29047 1.48292 O 2.85132 -0.27134 3.07869 O 2.99458 2.46920 2.55281 O 3.30650 4.45212 0.14258 H 5.68264 0.01430 -1.58111 H 6.44604 0.01599 -0.00205 H 6.66439 2.56003 -0.11044 H 0.57492 -6.11811 2.92671 H 0.85393 -5.07967 4.33321 H -1.83227 -5.29068 2.80017 H -1.56637 -4.28916 4.21738 H -1.81550 0.63053 3.37922 H -4.73767 -0.77046 0.71079 H -4.52036 -0.13652 -2.64505 H -2.74358 0.88599 -1.07975 H -3.68147 2.90454 0.14761 H -4.94463 -3.10215 -1.71225 H -3.75361 -4.41817 -1.74543 H 4.46824 -4.06546 -2.30588 H 4.10652 -5.35481 -3.48570 H 3.34900 -5.36757 -1.87410 C -7.15766 -1.34529 0.19618 C -5.87834 -2.10165 -0.16308 C -4.68135 -1.23953 -0.57828 O -3.56491 -1.79677 -0.68634 O -4.90990 0.00840 -0.81510 C 5.02593 -4.77601 -2.91368 C 4.86039 -3.31694 -2.49008 C 3.48051 -2.98121 -1.87650 O 2.58668 -3.85045 -1.90860 O 3.43298 -1.79055 -1.39341 H 1.70696 -4.54050 -3.63445 H -2.09950 2.44642 -4.54817 H -0.38912 2.65913 -4.96648 H 2.37658 -3.71594 1.84243 H 3.15576 -3.45475 0.51289 H 3.76567 -0.56627 2.92269 H 2.88853 0.72079 2.92267 H 2.36115 2.88965 3.16386 H 2.76420 4.37208 -0.67711 C -1.76371 5.76068 0.73024 C -1.35181 4.68955 1.75285 C -0.39720 3.67071 1.13963 O -0.87460 2.85822 0.30045 O 0.82648 3.75528 1.52665 H -2.23348 5.30042 -0.14982 H -0.88944 6.33497 0.38964 H -2.24094 4.14209 2.10302 H -0.86296 5.14995 2.62266 H -2.48137 6.46519 1.17672 H -4.28833 1.81834 4.05239 H -6.84162 1.32207 -2.13615 H -3.92794 -3.36946 -3.18053 H 2.83275 -3.28016 -4.10540 H -0.29072 -6.44416 4.44943 H 5.90942 2.53932 -1.71969 H 7.49301 1.76897 -1.47224 H -5.88651 0.91529 2.25154 H -1.26047 1.17360 -5.49665 O -0.33001 -0.58965 1.96240 H 4.10933 3.92318 -0.06182 S3OOH C 3.50208 2.78702 -2.93271 N 4.31668 2.80403 -1.80987 C 2.70174 1.67971 -2.79499 C 3.98072 1.74504 -1.02966 N 3.00696 1.04820 -1.60232 C 3.98731 4.26237 2.03389 N 3.27361 4.96404 1.07878 C 3.43827 3.00720 2.22264 C 2.30469 4.12609 0.70702 N 2.36280 2.93897 1.36954 C 4.70530 -1.09047 3.00926 C 3.50585 -0.72423 2.13027 O 2.36104 -0.92661 2.65821 O 3.73371 -0.28174 0.97485 C -1.08405 -4.32827 3.45211 C -0.71533 -3.31228 2.35542 O -1.13034 -3.49760 1.18658 C -1.63723 -5.65252 2.92702 O -0.00516 -2.33405 2.79179 C -0.67757 2.31338 4.51569 C -0.44031 1.53633 3.22785 O 0.15830 0.42184 3.32702 O -0.87082 2.07614 2.16124 O 1.13217 -1.80795 0.32799 Ca -1.12826 -1.94692 -0.85219 Mn 2.00587 -0.57065 -0.65533 O -1.14795 -0.54845 1.14014 Mn 0.56019 -0.76185 1.71387 O 0.97434 0.67032 0.53384 Mn -1.05426 1.12885 0.37315 O -2.83984 1.41484 0.41757 Mn -3.16453 0.94574 -1.32653 O -1.32429 0.45341 -1.28949 O -0.04984 -4.11530 -1.32599 O 0.30873 -0.05547 -3.16627 O -3.45999 0.47593 -3.15146 O -5.39021 1.80267 1.54254 H -5.53652 -2.73014 0.67412 H -6.05734 -2.79569 -1.00272 H -7.48967 -0.71339 -0.63971 H 4.92537 -5.45078 -2.05057 H 4.24650 -5.06299 -3.63344 H 5.63017 -3.01971 -1.75946 H 4.99351 -2.64209 -3.35317 H 1.90890 1.27822 -3.42301 H 4.43359 1.50439 -0.07365 H 3.70554 2.17651 2.86640 H 1.78637 2.08315 1.14888 H 1.52749 4.31968 -0.02734 H 5.55025 -0.41776 2.80688 H 5.01902 -2.11370 2.74871 H -2.50789 -5.48058 2.27951 H -1.93910 -6.31192 3.75797 H -0.88696 -6.18528 2.32358 H -0.19388 -4.47979 4.08260 H -0.38586 3.36324 4.37434 H -1.75432 2.29427 4.74277 H 0.91136 -3.92363 -1.54370 H -0.04817 -4.30784 -0.36648 H -0.36489 0.38988 -2.54437 H -2.91768 1.08105 -3.68960 H -4.40789 1.78949 1.42357 C -1.69436 5.91840 -0.57252 C -1.24971 4.96620 -1.69391 C -1.55223 3.50691 -1.32984 O -0.75899 2.98653 -0.47274 O -2.55892 2.97324 -1.88366 H -1.18384 5.66995 0.36843 H -2.77819 5.83747 -0.40021 H -0.16541 5.06706 -1.86101 H -1.77096 5.20442 -2.63286 H -1.46318 6.96537 -0.82722 H 3.56325 3.54780 -3.70241 H 4.84492 4.70234 2.53976 H 4.43775 -1.07011 4.07383 H -1.82641 -3.82671 4.09713 H 6.01383 -4.94799 -3.37490 H -6.99822 -0.68883 1.06424 H -7.97064 -2.04785 0.44537 H 4.91842 3.56435 -1.51312 H -0.11927 1.86829 5.34813 O 0.79347 -1.06682 -2.19890 H -5.64688 1.24729 0.77968 O -1.91210 -1.84125 -3.26395 H -2.57066 -1.08431 -3.26521 H -1.08221 -1.39805 -3.55953 S4OOH + H2O C 7.28709 0.65419 0.42507 C 6.13942 1.61051 0.10388 C 4.81611 0.95752 -0.31250 O 3.82021 1.71144 -0.44536 O 4.82054 -0.30713 -0.50997 C -4.00800 5.03726 -3.58358 C -4.13603 3.66723 -2.92090 C -2.86088 3.19336 -2.20196 O -1.83178 3.88407 -2.25374 O -3.03890 2.06140 -1.59172 C -4.02700 -2.16265 -3.04563 N -4.67240 -2.24540 -1.82319 C -3.05620 -1.20581 -2.88698 C -4.08401 -1.36908 -0.97391 N -3.09796 -0.72470 -1.58956 C -4.59802 -3.46451 2.23830 N -4.11637 -4.31463 1.26054 C -3.73105 -2.40457 2.42784 C -2.96846 -3.76469 0.87181 N -2.69132 -2.61696 1.55065 C -4.67105 1.58246 2.61532 C -3.38377 1.19859 1.90459 O -2.30939 1.27015 2.56839 O -3.49177 0.84863 0.68397 C 1.37656 4.36449 3.38453 C 0.99996 3.32074 2.32791 O 1.47164 3.39288 1.17415 C 2.21969 5.52218 2.85450 O 0.18995 2.41357 2.77436 C 0.18750 -2.22569 4.64920 C 0.11464 -1.48346 3.32708 O -0.34246 -0.29372 3.36031 O 0.51307 -2.10704 2.29777 O -0.94057 1.95787 0.31602 Ca 1.52820 1.78319 -0.75898 Mn -1.90662 0.78353 -0.64117 O 1.19457 0.42581 1.26685 Mn -0.46609 0.90025 1.74371 O -1.03884 -0.48121 0.53834 Mn 0.94493 -1.24964 0.50337 O 2.63572 -1.82260 0.63036 Mn 3.13786 -1.49554 -1.11351 O 1.44252 -0.61501 -1.12478 O 0.73434 3.94544 -1.48902 O -0.34347 -0.21995 -2.80687 O 3.53826 -1.08967 -2.91841 O 4.87516 -2.93697 -0.51927 H 5.91194 2.26419 0.96107 H 6.41655 2.29155 -0.71865 H 7.52910 0.02006 -0.43952 H -3.77049 5.81464 -2.84348 H -3.19757 5.04113 -4.32538 H -4.95536 3.65014 -2.18414 H -4.39179 2.89097 -3.66174 H -2.31163 -0.84692 -3.58610 H -4.39264 -1.22218 0.05430 H -3.74465 -1.55252 3.09908 H -1.95370 -1.92965 1.29893 H -2.29463 -4.14991 0.11215 H -5.49488 0.92913 2.29957 H -4.92949 2.61189 2.32283 H 3.13922 5.15337 2.38043 H 2.49609 6.20720 3.67130 H 1.67232 6.09948 2.09542 H 0.44181 4.71985 3.84721 H -0.19739 -3.24635 4.52295 H 1.24436 -2.30214 4.94648 H -0.22844 3.87461 -1.76201 H 0.71227 4.32317 -0.58834 H 0.42019 -0.60608 -2.21514 H 3.07412 -1.73445 -3.48329 H 4.37929 -3.08946 0.31306 C 0.75728 -6.09721 -0.62586 C 0.33245 -4.98068 -1.59407 C 0.96103 -3.64508 -1.19495 O 0.30257 -2.96624 -0.32194 O 2.06687 -3.34139 -1.71840 H 0.44238 -5.86454 0.40146 H 1.85040 -6.21859 -0.63039 H -0.76296 -4.87646 -1.59432 H 0.66029 -5.21987 -2.61614 H 0.30453 -7.05794 -0.91492 H -4.30832 -2.77510 -3.89420 H -5.52925 -3.66994 2.76123 H -4.54030 1.54721 3.70333 H 1.90981 3.82026 4.18243 H -4.94647 5.31522 -4.08927 H 7.02647 -0.01127 1.26118 H 8.19458 1.21233 0.70576 H -5.37190 -2.92656 -1.54740 H -0.37125 -1.69526 5.42861 O -0.62689 0.94113 -2.00868 H 5.38707 -2.11357 -0.33783 O 2.23789 1.34387 -3.14401 H 2.78148 0.49562 -3.12125 H 1.39197 1.05601 -3.53977
1804.10603	1	1804	2018-04-27T17:39:28	Cell membrane disruption by vertical nanopillars: the role of membrane bending and traction forces	[ "physics.bio-ph", "cond-mat.soft", "q-bio.CB" ]	Gaining access to the cell interior is fundamental for many applications, such as electrical recording, drug and biomolecular delivery. A very promising technique consists of culturing cells on nano/micro pillars. The tight adhesion and high local deformation of cells in contact with nanostructures can promote the permeabilization of lipids at the plasma membrane, providing access to the internal compartment. However, there is still much experimental controversy regarding when and how the intracellular environment is targeted and the role of the geometry and interactions with surfaces. Consequently, we investigated, by coarse-grained molecular dynamics simulations of the cell membrane, the mechanical properties of the lipid bilayer under high strain and bending conditions. We found out that a high curvature of the lipid bilayer dramatically lowers the traction force necessary to achieve membrane rupture. Afterwards, we experimentally studied the permeabilization rate of cell membrane by pillars with comparable aspect ratios but different sharpness values at the edges. The experimental data support the simulation results: even pillars with diameters in the micron range may cause local membrane disruption when their edges are sufficiently sharp. Therefore, the permeabilization likelihood is connected to the local geometric features of the pillars rather than diameter or aspect ratio. The present study can also provide significant contributions to the design of 3D biointerfaces for tissue engineering and cellular growth.	physics.bio-ph	physics	Cell membrane disruption by vertical nanopillars: the role of membrane bending and traction forces. Rosario Capozzaa, Valeria Caprettinia,b, Carlo A. Gonanoa, Alessandro Boscaa, Fabio Moiaa, Francesca Santoroc, Francesco De Angelisa* aIstituto Italiano di Tecnologia, via Morego 30, 16163 Genova, Italy bUniversità degli studi di Genova, Genova, Italy cCenter for Advanced Biomaterials for Healthcare, Istituto Italiano di Tecnologia, 80125 Napoli, Italy corresponding author's email: [email protected] KEYWORDS: nanopillars, lipid bilayer, nanopore, membrane disruption, molecular dynamics, intracellular delivery ABSTRACT Gaining access to the cell interior is fundamental for many applications, such as electrical recording, drug and biomolecular delivery. A very promising technique consists of culturing cells on nano/micro pillars. The tight adhesion and high local deformation of cells in contact with nanostructures can promote the permeabilization of lipids at the plasma membrane, providing access to the internal compartment. However, there is still much experimental controversy regarding when and how the intracellular environment is targeted and the role of the geometry and interactions with surfaces. Consequently, we investigated, by coarse-grained molecular dynamics simulations of the cell membrane, the mechanical properties of the lipid bilayer under high strain and bending conditions. We found out that a high curvature of the lipid bilayer dramatically lowers the traction force necessary to achieve membrane rupture. Afterwards, we experimentally studied the permeabilization rate of cell membrane by pillars with comparable aspect ratios but different 1 sharpness values at the edges. The experimental data support the simulation results: even pillars with diameters in the micron range may cause local membrane disruption when their edges are sufficiently sharp. Therefore, the permeabilization likelihood is connected to the local geometric features of the pillars rather than diameter or aspect ratio. The present study can also provide significant contributions to the design of 3D biointerfaces for tissue engineering and cellular growth. INTRODUCTION Direct access to the intracellular compartment is an open challenge [1] with many potential applications, such as gene transfection [2, 3], biomolecule delivery [4], and electrical recording in electroactive cells [5]. The main difficulties are related to the impermeability of the plasma membrane, which, after a billion years of evolutionary defenses, strictly controls the trafficking in and out of the cell. The most popular methods for intracellular delivery are electroporation [6], chemical transfection, and virus-mediated transduction, although novel methods have been considered recently [7, 8, 9, 10]. Among them, local membrane permeabilization through vertical nanopillars or other 3D nanostructures is emerging as a robust approach [11, 12, 13, 14, 15, 16, 17]. Essentially, the concept relies on arrays of vertical standing nanostructures in a fakir bed-like configuration. When cells are cultured on these arrays, the plasma membrane exhibits tight adhesion to the pillars or even engulfment-like events [18]. These processes often lead to a spontaneous increase in membrane permeability [19] that can be used to deliver molecules into the cytosol by bypassing the conventional biochemical pathways or to record intracellular electrical activity via the enhanced electrical coupling between the conductive pillar and the cell [5]. In general, there is a wide interest in designing novel 3D bio-interfaces for tissue engineering and cellular growth [20] to investigate how cells interact and proliferate onto these types of geometries [21, 22] and which structures improve the cell viability [23]. However, the exact mechanism enabling local permeabilization is still not fully understood, and many controversies still exist. Recent studies suggested that the internalization of molecules is not due to the temporary disruption of the plasma membrane in contact with the 3D nanostructures but is instead driven by membrane deformation [24] or enhancement of the clathrin-mediated endocytosis process of the cell at the interface with sharp edges [25]. On the contrary, a previous study, developing a simple 2 but effective mechanical continuum model of elastic cell membranes [26], ascribes such complex behaviors to actual penetration of the plasma membrane. In that work, two main mechanisms at the interface are considered, namely "impaling", where cells land on a bed of nanowires, and "adhesion-mediated" permeabilization, which occurs as cells spread on the substrate and generate adhesion force. In the former, the force leading to the membrane disruption is gravity, whereas in the latter mechanism, this force is the adhesion-force provided by membrane proteins. In both cases, membrane permeabilization occurs when the nanopillar generates sufficient tension to overcome a critical membrane tension value. However, the pillar is modeled as a cylindrical probe with a hemispherical tip, and the effect of local geometry (e.g., the sharpness of pillar edge) is not taken into account. Hence, only the diameter, the height and the spacing between the pillars determine the penetration forces for a cell line of a given stiffness. This approach may explain the experimental reports of spontaneous poration observed in vertical nanostructures of small diameters on the order of 50-200 nm [27, 28, 29] or below. Indeed, as we will subsequently show, we found experimentally that spontaneous permeabilization may occur even for much larger pillar diameters of approximately 2 µm. Such a finding is also confirmed by another study that recently reported spontaneous membrane disruption by pillars of 1 µm in diameter, even in absence of adhesion with the substrate [30]; hence, the models mentioned above cannot fully explain the membrane permeabilization. In other words, the reasons why these large nanopillars can effectively permeabilize the membrane are still unclear, thus demanding a more complex scenario that includes not just gravity and adhesion but also traction forces, membrane deformations (bending) and local geometric features (sharp edges on the pillars). Obviously, biological mechanisms and surface properties also participate in the increase in cell internalization processes, but the present mechanical model will not take them into account. Molecular dynamics simulation is a powerful tool that may elucidate the behavior of these systems. However, to the best of our knowledge, there are still few molecular dynamics studies investigating the mechanical properties of a membrane when in contact with a nanostructure [31]. Furthermore, most molecular dynamics studies tackle the problem only in the regime of small bending deformations [32, 33, 34], which is not applicable for this case of study. In this work, we first undertake molecular dynamics simulations in synergic combination with a mechanical model of the cell membrane that goes beyond the linear response approximation. We 3 found that the bending of the membrane is characterized by an elastic regime at low bending angles, followed by a plastic one at higher angles. We also included the local geometry of the pillar by investigating the role of curvature or edges at the pillar tip (different from the pillar diameter), and we found that a high curvature favors the rupture even at very low tensile strength. Afterwards, we show that simulation results are supported by experimental data suggesting that when cells are cultured on pillars with diameters in the micrometric range, their permeabilization likelihood is strongly increased in the presence of sharp edges. We conclude that, under the given conditions, the local curvature may dramatically affect the lipid bilayer permeabilization to a greater extent than the effect of the pillar diameter. a) c) 1 µm b) d) F F 1 µm F Figure 1. SEM images showing a FIB cross sections of a cell cultured on an array of 2µm diameter nanopillars. The samples were fixed and stained following a recently developed ROTO protocol (see S2 for more details) and embedded in a thin film of epoxy resin. The cell membrane (indicated by red arrows) can assume a "tent-like" configuration (!) or be tightly wrapped around the pillars ("). c) Adhesion-induced 4 mechanism of permeabilization. d) Traction-induced mechanism of permeabilization. Inset: the membrane's bending is dictated by the local radius of curvature rather than by the nanopillar diameter. RESULTS AND DISCUSSION Figure 1 shows a SEM/FIB cross-sectional image of cell cultured on nanopillars with a diameter of 2 µm and height of 1.5 µm, obtained by using a technique explained elsewhere [35]. This image displays possible configurations of the cell/nanostructure interface, and highlights the proximity of the cell membrane to the pillars: the cell can tightly wrap around the 3D pillar (b) or assume a "tent-like" configuration (a). In both cases, the cell membrane conforms to the pillar's head geometry, where we assume most of tensile strength is concentrated, via adhesion with the substrate and forces exerted through the cell body. Therefore, the membrane curvature is dictated by the pillar's edge. Furthermore, the forces acting on the membrane can be oriented in any direction depending on the local configuration of the system, as sketched in panels c) and d) of Figure 1. Traction and bending cannot be considered separately, and they may cooperate to lower the threshold for local permeabilization or nanopore creation. For instance, when the traction force is mainly directed laterally (parallel to the substrate), the pillar may behave like a knife edge, meaning that only the size of the edge matters and the diameter plays a minor role. Simulation model. We used a 2D coarse-grained model of the cell membrane in which the lipid molecule is made of a hydrophilic Lennard-Jones (L-J) particle (the head) and a hydrophobic tail made of five L-J beads (Figure 2b). We included an additional harmonic interaction between the L-J beads of the tail. The water is simulated as a single L-J bead [36]. First, we tested the validity of our system by simulating free lipid molecules randomly dispersed in water and checking that the lipids undergo a self-assembly process that leads to the formation of the lipid membrane. As expected, under the conditions we used, a bilayer is formed spontaneously with the hydrophilic heads pointing toward water and the tails clustering together, minimizing their interaction with water (see S3). The average distance between hydrophilic heads is on the order of experimental values [37] (~0.8 nm). Clearly, with a 2D model, we cannot reproduce all of the 3D phenomena, for example, the lipid diffusion through the membrane's surface. In addition, the molecular dynamics model presented here considers just the lipid bilayer and not the cytoskeleton and membrane proteins that certainly affect the mechanical properties of the cell. However, we carefully checked that our numerical simulations can qualitatively reproduce many real processes (self-assembly, bilayer, micelles, vesicles, etc.) that depend on the geometrical 5 features of molecules [37]. This method can be regarded as a good compromise between the computational cost and the accuracy of the mechanical analysis and allows a statistical analysis of results. After the bilayer formation, we replicate the structure to create a system size at will and set up the initial configuration (Figure 2a). The typical size of our system is on the order of 100 nm. Modeling of the plasma membrane mechanical response. The mechanical properties of a material are often described by means of a mathematical relation linking the strain % (ratio between a displacement and a rest length) to the applied stress & (force per unit area). For an ideal linear material, the applied stress is given by &=(∙%, where E is the Young's modulus (see S4). The Young's modulus E is a characteristic of the material, so it is an intensive parameter in the sense that it does not depend on extensive properties of the considered object (e.g., length and mass); however, membrane deformation is often described in terms of stiffness and applied forces [26, 33, 37]. For example, the area stretch modulus KA (characterizing the rigidity of a membrane to traction force) and the bending stiffness Kbend (linking the curvature to the applied moment) are extensive parameters because they depend on the membrane's length and thickness (see S5). We decided to calculate both intensive and extensive quantities. Since the cell membrane may undergo large deformation when in contact with nanostructures, the approximation of a linear material is no longer valid, and the membrane could also exhibit plastic behavior (hysteresis). In this case, the Young's modulus can be interpreted as the derivative of the stress & with respect to the strain %, formally: ∆&=((&,%)∙∆% More simply, extracting the strain-stress relation is a standard strategy for characterizing the elastic properties of a material. For that purpose, we simulated a cell membrane loaded by cylinders measuring the displacements and the applied forces, as shown in Figure 2a. 6 F c) rupture a) y z x b) hydrophilic head water hydrophobic tail Figure 2. a) Snapshot of a simulated segment of membrane subject to a horizontal traction force F. Here, water is indicated in blue, hydrophilic heads in gray and hydrophobic tails in red. Inset: blow-up of the simulated system showing the molecular arrangement in detail. b) Coarse-grained model of water and the lipid molecule with a single bead for the hydrophilic head and five beads for the hydrophobic hydrocarbon tail. c) Strain- stress curve showing three loading regimes: a thermally dominated regime (thermal noise larger than the applied stress), an elastic regime, and failure for F > 28 pN (corresponding to s > 9.2 MPa). connecting the bending moment /0 to the deflection of the beam (the membrane) through the Afterwards, we calculated the strain-stress curve through the classical Euler-Bernoulli beam model Young's modulus E and a local linearization. In the simple case of pure traction, we calculate the stress-strain curve by applying a traction force to the edges of the membrane (Figure 2a) and estimating the ultimate tensile strength as the force 7 FTS we need to apply to break the membrane. After embedding two cylinders into the membrane segment, we keep one fixed, while the other one applies a force F along the positive direction of the x axis. The force is increased in steps ∆1=3.5 56 and kept constant for a time interval ∆7= 1.2 :;. During the step-wise increase in force, we monitored the strain % of the membrane, and we found that the membrane ruptures at 1=28 56 (corresponding to an ultimate tensile stress &= 9.2 />?). Therefore, the value 1@A=28 56 represents the tensile strength of the membrane in the straight configuration. The longitudinal strain e xx is the ratio between the displacement Dsx and the rest length lx = 92 nm. The applied stress sxx here is simply the force on the membrane's section: =e xx sD l x x F x A t = F x ll zy =s xx The plot of the strain-stress curve (Figure 2c) clearly shows three loading regimes: fluctuates, and the thermal noise overwhelms the effects of the applied force. • For 0≤&≤4.5 />?, the stress does not significantly affect the strain. The membrane • For 4.5 />?<&<9.2 />?, the membrane responds elastically, and we estimate a Young's modulus (=50 />? by linear regression (blue dashed curve in Figure 2c). This • For &>9.2 />?, membrane rupture occurs. From Young's modulus E, we calculate the area-stretch modulus GH=I∙HJKL ≈1.5 NO/NQ. As for typical free biological cell membranes, which range from 100−250 NO/NQ [26, 37]. In fact, expected, KA and FTS are approximately two orders of magnitude smaller than the ones calculated value is in the range of experimental measurements [37]. our system is representative of a longitudinal section of a three-dimensional membrane; thus, the domain's depth must be taken into consideration. 8 a) Fx Fy F2y sy z x c) y 30 20 10 ) ) m m n ( n R ( R b) ) N p ( F Fx 10 5 0 10 5 0 10 5 0 0 F2y Fy F2y Fx Fy 10 20 30 d(nm) 40 50 sy 300 d) ) a P M ( x a m σ 200 100 elastic plastic 0 0 40 50 0 0 Rmin Rmin 20 Sy(nm) Sy(nm) 30 10 0.8 Figure 3 Configuration of the membrane in response to a small (a) and high (b) deformation sy. c) Behavior of the estimated radius of curvature R as a function of the deformation sy. For sy > 35 nm, the radius of curvature R reaches the minimum Rmin, corresponding to the membrane wrapping around the pillar's tip. d) Stress-strain curve showing the elastic and plastic regimes. 0.4 εyx 0.2 0.6 In the following, we describe the calculation of the membrane bending stiffness Kbend. Previous approaches estimated Kbend by analyzing the height (spatial) fluctuation spectrum of a membrane [38, 39, 40, 41]. However, these methods present serious drawbacks due to the slow convergence of long wavelengths [32, 33] and assume a regime of small deformations. In contrast, a membrane in contact with nanopillars could be strongly deformed. For this reason, we studied the bending without invoking neither the hypothesis of small deformations nor the hypothesis of a linear material. Hence, the elastic parameters such as E and Kbend could depend on the strain. 9 We employed a simulation setup similar to the one already used in the literature [32]. A segment of membrane with free ends is pushed in the center through a nanopillar (with a hemispherical tip of radius rp~2 nm) moving upward and two immobile cylinders (Figure 3a, b) interacting just with the hydrophobic tails. The segment is free to flow under the lateral cylinders to avoid stretching and to study the mechanical response under pure bending. During the course of simulations, we measured the x and y components of the force on the cylinders and the pillar and the vertical displacement sy, and we estimate the local radius of curvature R as the best circle approximating the average membrane profile (Figure 3c). For small values of vertical displacement sy, R is independent on the pillar size, and it rapidly decreases reaching the plateau Rmin~4 nm at sy~35 nm (Figure 3c). The value Rmin represents the smallest possible radius of curvature considering the membrane thickness is ST≈4.5 :N and corresponds to the membrane wrapping around the pillar's tip. As explained in detail in S6 and S7, for a linear material, the stress & and the local radius of curvature R are both proportional to the bending moment /0. It is then possible to show that the stress-strain relation reads as: &U=(∙ where ST≈4.5 nm is the average membrane thickness. Since both &U and the curvature strain %UT estimated as (= ∆WL∆XLY. clearly identified. In fact, for values of curvature strain %UT<0.3, the membrane responds elastically, while for %UT>0.3, it is plastically deformed. This latter regime, characterized by a ST/2V =(∙%UT can be obtained from the simulations, even for a non-linear material, the Young's modulus can be The stress-strain curve is reported in Figure 3d. Two distinct regimes, elastic and plastic, can be smaller Young's modulus, suggests a hysteretic behavior. In fact, upon retraction of the nanopillar, the membrane does not return to its initial straight configuration, at least in the time interval available to our simulations. Each point in Figure 3d is obtained from a statistical average over ten different simulations. We fitted the points of the stress-strain curve (Figure 3d) with two lines, and we estimated the Young's modulus for the elastic regime (ZK[\= 600±150 />? and plastic regime (_K[\= 75±25 />?. Not surprisingly, the value of (_K[\ is (considering the error) equal to the E calculated for the pure stretching, indicating that for %UT>0.3, the bending corresponds to an effective stretching of the outer membrane monolayer. 10 As shown in S5, the relation GaZbc= IdQ∙STe allows us to estimate the bending stiffness in the elastic GaZbcZK[\≈(4.5±2.6)∙10fdgO and plastic regime GaZbc_K[\≈(0.57±0.37)∙10fdgO. The value of GaZbc_K[\h is of the same order of magnitude of the ones reported in literature [37], while GaZbcZK[\h is one order of magnitude larger. This large value of GaZbcZK[\h can be attributed to the limits We stress that the estimation of GaZbcZK[\h is affected by a large uncertainty related to the dependency of our 2D modeling, which cannot correctly reproduce the interlayer diffusivity of lipid molecules. on the cube of membrane thickness ly (see S5) and the one related to Young's modulus E. Membrane mechanical response to local curvature. In the following, we investigate how curvature and bending affect the ultimate tensile strength FTS. In fact, the edges of 3D nanostructures in contact with the cell membrane could be very sharp, and their sharpness is usually very difficult to control experimentally. The lipid bilayer can adhere to the nanopillar edge and, then, follow its curvature. Traction forces are on the order of 1 nN [18] and in principle too small to promote spontaneous rupture of membrane. What is the relation between the ultimate tensile strength FTS and the curvature? To clarify this point, we performed simulations by stretching and bending the membrane simultaneously. The three-cylinder configuration is shown in Figure 4a. Two cylinders (1 and 3) are embedded into the membrane, while cylinder 2 is kept fixed. Cylinder 3 applies a constant force F to the membrane along the positive direction of the x axis and lower than the membrane's ultimate tensile strength, FTS = 28 pN. Cylinder 1 rotates the membrane around an axis passing through cylinder 2 and directed along z, imposing a curvature on the membrane. We found that the smaller F is, the higher is the curvature (the smaller R) needed to bring the membrane to the rupture, as reported in the snapshots of the system in Figure 4 a, b. In Figure 4c, the local radius R is plotted as a function of FTS; each value of R was obtained from a statistical average over ten different simulations. The behavior of the local radius of curvature R as a function of FTS clearly shows that even a tiny force can promote a rupture provided that R is small, i.e., the edge is sharp. 11 F 3 b) FTS=25pN 2 R F 3 1 d) 1.1 ) ) m m n n ( ( h h - - h h d d 1 0.9 0.8 0 25 20 40 deq(R+w)/R deq(R+w)/R w dh-h deq deq 60 80 100 R(nm) R(nm) a) FTS=7pN 2 R 1 c) 30 25 20 15 10 5 ) m ) n m ( R n ( R 0 10 15 20 FTS(pN) FTS(pN) Figure 4 The three-cylinder simulation geometry used to calculate the effect of the local radius curvature R on the ultimate tensile strength FTS. Cylinder 3 applies a constant force F, and cylinder 1 rotates around cylinder 2 (at rest) (a, b). A smaller value of F corresponds to a smaller R needed to bring the membrane to the rupture. The yellow shaded area suggests a possible configuration of the membrane in contact with a nanopillar. c) Behavior of the critical radius of curvature R as a function of FTS. d) Average distance between hydrophilic heads, dh-h, as a function of the bending radius R, as calculated in the segment where the membrane is bent. The blue dashed line is the equilibrium distance deq in the absence of bending, and the red dashed line shows the behavior of dh-h from an analytical model. Here, deq=0.83 nm, w=ly/2=2.25 nm. What is the reason for the curvature-induced membrane weakening? Figure 4d shows the variation in the average distance between hydrophilic heads, dh-h, as a function of the bending radius R, calculated only in the segment where the membrane is bent. In the absence of bending, dh-h=deq=0.83 nm (blue dashed line in Figure 4d), which is the equilibrium distance after the self-assembly process of the membrane (see S3). After the membrane is bent, dh-h deviates from the equilibrium and rapidly increases as the radius of curvature R decreases. 12 Let us consider an initially straight bar composed of n elements at distance deq and finite thickness ST (that is, the average thickness of the membrane). When the bar is bent, the distance dh-h between the elements of its upper surface (see inset in Figure 4d) increases following the relation djfj= dmn (the red dashed line shown in Figure 4d). The values of djfj versus V, calculated from klKY/Qk the numerical simulations, follow approximately the theoretical red dashed line, suggesting that the membrane is effectively deformed like a bar composed of discrete elements. When dh-h increases, the membrane is brought out of its equilibrium distance deq, and the probability of nanopore formation increases as well. In this situation, traction forces further increase the probability of defect formation (lowering the activation energy for nanopore formation) and favor breakdown. Experimental investigation of permeabilization on sharp pillars. To confirm the effect of the edge on the cell membrane permeabilization, we fabricated two sets of pillars, denoted as "sharp" and "smooth"; the pillars had an identical pitch of 5 µm, comparable diameter of approximately 2 µm, height of 2.5 µm and 1 µm, respectively, and their most important difference was the edge sharpness, as reported in Figure 5a and 5b (see S1 for more details on the fabrication process). We estimated their radius of curvature as Rsharp~20 ± 5 nm for the sharp case and Rsmooth~250 ± 20 nm for the smooth one from the cross sections of the pillars, as shown in the inset of Figure 2a. Afterwards, we cultured NIH-3T3 cells on these arrays of pillars and administered the impermeable dye propidium iodide in solution together with the permeable calcein AM dye to verify the healthiness of the cells. Surprisingly and contrary to some previous literature results [26], we found that in the case of large but sharp pillars, the dye entered the cell body with a probability of approximately 70%, staining the cells red, as demonstrated by fluorescence images reported in Figure 5c (see SI for more details) maintaining their viability, confirmed by the green stained. These results have been acquired on 400 cells in 6 different cell cultures. In contrast, cells cultured on the smooth pillars showed no sign of dye internalization (no red color) and a green color indicating cell viability but no permeabilization (see Figure 5d). In this last case, the presence of shorter pillars ensures stronger adhesion with the substrate that, according to previous studies [26], is assumed to be one of the factors increasing the permeabilization probability. However, our fluorescence analysis of smooth pillars did not show such an increase, displaying permeabilization in very few cases that we have estimated to represent less than 10% (30 cells over 320 in 5 different cell cultures). 13 The radius of curvature of sharp pillars (Rsharp~20 ± 5 nm) falls in the range of values where, according to the simulation results reported in Figure 4c, we expect a significant decrease in tensile strength for rupture. This can explain the increase of permeabilization, as shown in Figure 5c by fluorescence images. For sake of clarity we remark that the novelty here is represented by the role of the sharp edge and not by a fine tuning of the aspect ratio. Similar findings were also recently observed by another group that studied plasmid transfection driven by cells interfacing with 3D nanostructures [30]. We consider these results of great importance in the field for two reasons: i) Cell membrane permeabilization is reported at pillar sizes where permeabilization is not theoretically expected [26]. ii) The comparison between sharp and smooth pillars highlights the importance of membrane curvature and its role as key player in the permeabilization processes. a) cross section cut 50 nm b) c) 1 µm d) 1 µm 20 µm 14 Figure 5 a) SEM image of a "sharp" pillar with diameter of 2 qr, height of 2.5 qr. The inset highlights a cross circle with the scale bar). b) SEM image of a "smooth" pillar with diameter of 2 qr, height of 1 qr. The section of the sharp edge, with an estimated radius of curvature of Rsharp~20 ± 5 nm (compare the approximating estimated radius of curvature is Rsmooth~250 ± 20 nm. c) Fluorescence image of cells cultured on sharp pillars fabricated with spacing of 5 µm (green spots) and treated with both permeable calcein AM (green) and impermeable dye propidium iodide (red) administrated in solution. Most of the cells present green and red staining, with a permeability likelihood very close to 70%. d) Corresponding fluorescence image of cells cultured on smooth pillars and treated with the same dyes. In this case, only a limited number of cells contain propidium iodide and are hence stained in red, meaning that the cell membrane is successfully permeabilized in a few cases (see S2 for more details). A specific red staining is probably due to fraction of death cells and DNA dispersed in the cell culture. CONCLUSIONS In summary, we studied, through a 2D coarse-grained molecular dynamics model, the mechanical behavior of a cell membrane interacting with micro- and nanopillars. To identify the stress-strain relations and the rupture conditions, we performed simulations of a lipid bilayer subjected to different loading conditions: longitudinal traction, pure bending and combined traction and bending. We have also proposed a simple mechanical model for estimating the main intensive elastic parameters, such as the Young's modulus and the ultimate tensile stress, in a regime of large stress and deformation. Notably, we did not introduce any restrictive hypothesis on the constitutive relation for the membrane's elasticity: actually, the stress-strain relation could be non- linear. Moreover, we extracted the ultimate tensile loads from the simulations without postulating any value a priori. We found that the bending of a membrane is characterized by an elastic regime at low bending angles, followed by a plastic one at higher angles, namely, spatial configurations that are irreversible within the time window of the simulations. Importantly, the simulation shows that bending of the membrane (e.g., at the nanopillar's edge) dramatically lowers the rupture force. To support these theoretical findings, we investigated experimentally the permeabilization of cells cultured on micropillars. Surprisingly, we found that spontaneous permeabilization events may occur on pillars of diameters in the micrometer range, which is much larger than previously observed. A comparison between cells cultured on pillars with a comparable aspect ratio but different sharpness at the edges suggests that the local curvature of the membrane could be 15 responsible for the strong increase in permeabilization, with a probability that increases from 10% in the smooth case to the 70% in the sharp case. Therefore, the simulation results suggest a direct interpretation of the experimental data. In principle, even step edges can cause local membrane disruption provided that they are sharp enough. The results obtained from our simulations could also be applied to the understanding of the internalization of substrate-free nanowires [42, 43], since cellular traction forces exist in that situation as well. In fact, the internalization process often starts at the sharp tip of nanowires, where the curvature is higher. In conclusion, local geometric features at the cell-substrate interface can dramatically affect cell permeabilization to a greater extent than the effect of the pillar aspect ratio and spacing. These results may provide important information for the field of biointerfaces and tissue engineering, offering valuable insight into designing devices for gene transfection, intracellular delivery and electrical recording. METHODS Simulation details. We performed Molecular Dynamics simulations of a lipid bilayer in water by using Langevin dynamics that allows to calculate not only the forces due to bending, but also to the thermal fluctuations. Periodic boundary conditions were imposed in all directions and the temperature was kept at 301.7 K by means of a Langevin thermostat applied to all particles. We assume the existence of three types of particles: "water-like" particles (w), hydrophilic "head" particles (h), and hydrophobic "tail" particles (t). A water molecule is approximated as a single "w" particle, while a lipid molecule is composed of one "h" particle and five "t" particles joined together by harmonic interactions. The three types of particles interact by means of Lennard-Jones potentials with the depth of the potential well stu=6.94∙10fQdO, except for shh=11.8∙10fQdO and vtu=v=0.4456 :N, with v distance at which the inter-particle potential is 0 (see S3 for more details). All interactions are truncated at Vtuw=2.5v, except for w- t and h-t interactions where Vtuw=2d/xv=0.5:N, which makes the latter interactions repulsive. 16 When randomly dispersed in water, the lipids undergo a self-assembly process because of the assigned interactions, eventually forming (after t=10 ns) a lipid bilayer with thickness of ST≈ 4.5:N and average distance between the hydrophilic heads yZz≈0.83 :N. After the bilayer "Shape factor" [37], {=}w/(?~∙Sh), where }w is the area of hydrocarbon chains, ?~ is the head thickness, and Sh is the tail length. In our case {≈1 and we have indeed a bilayer (see S3 for formation, we replicate the structure to create a system size of about 100 nm corresponding to a straight and stable membrane. The equilibrium curvature of the membrane can be predicted estimating the two-dimensional more details). All simulations have been executed using the LAMMPS package for large scale molecular dynamics simulations. Configurations were stored every 3 ps for further structural and dynamic analyses. Fabrication of pillars. For the fabrication of the silicon pillars arrays, we used first photolithography to pattern a chrome hard mask on a silicon wafer substrate and then dry etching to define the pillars in the silicon bulk using a customized multi-step process to control the shape of the pillars' top edges (see S1 for technical details). The fabrication of the silicon pillar arrays has been done using a photolithographic chrome hard mask on a silicon wafer substrate. Silicon pillars have been fabricated by dry etching in an ICP- RIE reactor (Sentech SI 500) using a customized multi-step silicon etching process based on SF6 and C4F8 gases. ICP source has been set at 1200 W, RF power at 20 Wand reactor pressure at 1.33 Pa. After dry etching, any residual of chrome has been removed using Chrome Etch 18. More details on the fabrication process can be found in the Supplementary Information (S1). Cell culture. The cells that appear in Figures 1 and 5 are NIH-3T3 plated with a density of 1,5x104 cell/cm2 in DMEM complemented with 1% penicillin streptomycin antibiotics and 10% inactivated fetal bovine serum (FBS); cells are incubated at controlled temperature, humidity and CO2 concentration for 48h. A viability test has been performed and both calcein AM and propidium 17 iodide (2µM) have been administer to the cultures (see S2 for more details). After 20 minutes of incubation, the samples have been washed three times with PBS and then analyzed through a fluorescence microscope. Cell imaging. Figures 1a and 1b are SEM images of a NIH-3T3 cell cross section. The cells have been fixed and stained using a recently implemented RO – T – O procedure [35] and samples have been infiltrated increasing concentration of resin in ethanol. The excess of resin has been removed prior to polymerization in order to allow the detection of the cells on the 3D nanostructures using SEM [35]. FIB/SEM cross section images have been made with a FEI Helios Nanolab 650 dual beam microscope. A 1µm layer of Pt has been deposited on the selected region by a Gas Injection System (GIS) integrated to the microscope before the cutting. An initial cut was made with a high current (9.3nA) of the ionic beam, and then a cleaning cross section was performed using a 0.79 nA ion current with a 30.0kV accelerating voltage. The imaging was made using the immersion mode of the microscope and backscattered electrons have been detected with TLD detector. Corresponding Author Francesco De Angelis, E-mail: [email protected], Via Morego 30, 16163 Genoa, Italy. Author Contributions R. C. conceived and performed the MD simulations, the data analysis and wrote the manuscript. V. C. carried out the experiments and associated data analysis. C. A. G. performed the mechanical and mathematical analysis of simulation data. A. B. and F. M. fabricated the nanopillars to be used for cell culture. F. S. assisted in the experimental analysis and imaging. F. D. A. conceived the experiment, supervised the work, analyzed data. All authors contributed to the preparation of the manuscript and have given approval to the final version. Notes The authors declare no competing financial interest. ACKNOWLEDGMENTS 18 This work was supported by the European Research Council under the European Union's Seventh Framework Programme (FP/2007-2013) / ERC Grant Agreement no. (616213), CoG: Neuro-Plasmonics. ASSOCIATED CONTENT Supporting Information S1: Fabrication method S2: Cell culture and imaging S3. MD simulation of Membrane self-assembly S4. Strain-stress curve S5. Bending stiffness S6. Maximum stress on the membrane's section S7. Stress-strain relation for bent membrane References [34] [34] [1] M. P. Stewart, A. Sharei, X. Ding, G. Sahay, R. Langer and K. F. Jensen, "Break and enter: in vitro and ex vivo strategies for intracellular delivery," Nature, vol. 538, p. 183–192, 2016. [2] P. Washbourne and A. K. McAllister, "Techniques for gene transfer into neurons," Curr. Opin. Neurobiol., vol. 12, p. 566–573, 2002. [3] G. He, H.-J. Chen, D. Liu, Y. Feng, C. Yang, T. Hang, J. Wu, Y. Cao and X. Xie, "Fabrication of Various Structures of Nanostraw Arrays and Their Applications in Gene Delivery," Adv. Mater. Interfaces, p. 1701535, 2018. [4] Y. Ma, R. J. M. Nolte and J. J. L. M. Cornelissen, "Virus-based nanocarriers for drug delivery," Adv. Drug Deliv. Rev., vol. 64, p. 811–825, 2012. [5] C. Xie, Z. Lin, L. Hanson, Y. Cui and B. Cui, "Intracellular recording of action potentials by nanopillar electroporation," Nat. Nanotechnol., vol. 7, p. 185–190, 2012. [6] J. Teissié, N. Eynard, B. Gabriel and M. Rols, "Electropermeabilization of cell membranes," Adv. Drug Deliv. Rev., vol. 35, p. 3–19, 1999. [7] S. Yoon, M. G. Kim, C. T. Chiu, J. Y. Hwang, H. H. Kim, Y. Wang and K. K. Shung, "Direct and sustained intracellular delivery of exogenous molecules using acoustic-transfection with high frequency ultrasound," Sci. Rep., vol. 6, p. 20477, 2016. 19 [8] Q. Fan, W. Hu and A. T. Ohta, "Efficient single-cell poration by microsecond laser pulses," Lab Chip, vol. 15, p. 581–588, 2015. [9] R. Xiong, S. K. Samal, J. Demeester, A. G. Skirtach, S. C. De Smedt and K. Braeckmans, "Laser-assisted photoporation: fundamentals, technological advances and applications," Adv. Phys. X, vol. 1, no. 4, p. 596–620, 2016. [10] N. Saklayen, M. Huber, M. Madrid, V. Nuzzo, D. I. Vulis, W. Shen, J. Nelson, A. A. McClelland, A. Heisterkamp and E. Mazur, "Intracellular Delivery Using Nanosecond-Laser Excitation of Large-Area Plasmonic Substrate," ACS Nano, vol. 11, p. 3671–3680, 2017. [11] A. R. Durney, L. C. Frenette, E. C. Hodvedt, T. D. Krauss and H. Mukaibo, " Fabrication of Tapered Microtube Arrays and Their Application as a Microalgal Injection Platform," ACS Appl. Mater. Interfaces, vol. 8, p. 34198−34208, 2016. [12] C. Chiappini, J. O. Martinez, E. De Rosa, C. S. Almeida, E. Tasciotti and M. M. Stevens, "Biodegradable Nanoneedles for Localized Delivery of Nanoparticles in Vivo: Exploring the Biointerface," ACS nano, vol. 9, p. 5500, 2015. [13] C. Chiappini, E. De Rosa, J. O. Martinez, X. Liu, J. Steele, M. M. Stevens and E. Tasciotti, "Biodegradable silicon nanoneedles delivering nucleic acids intracellularly induce localized in vivo neovascularization," Nature Mat., vol. 14, p. 532, 2015. [14] A. Hai, J. Shappir and M. E. Spira, "In-cell recordings by extracellular microelectrodes," Nature Methods, vol. 7, p. 200, 2010. [15] A. Hai and M. E. Spira, "On-chip electroporation, membrane repair dynamics and transient in-cell recordings by arrays of gold mushroom-shaped microelectrodes," Lab Chip, vol. 12, p. 2865, 2012. [16] V. Caprettini, A. Cerea, G. Melle, L. Lovato, R. Capozza, J.-A. Huang, F. Tantussi, M. Dipalo and F. De Angelis, "Soft electroporation for delivering molecules into tightly adherent mammalian cells through 3D hollow nanoelectrodes," Sci. Rep., vol. 7, p. 8524, 2017. [17] A. K. Shalek, J. T. Robinson, E. S. Karp, J. S. Lee, D.-R. Ahn, M.-H. Yoon, A. Sutton, M. Jorgolli, R. S. Gertner, T. S. Gujral, G. MacBeath, E. G. Yang and H. Park, "Vertical silicon nanowires as a universal platform for delivering biomolecules into living cells," PNAS, vol. 107, p. 1870, 2010. [18] X. Xie, A. Aalipour, S. Gupta and N. Melosh, "Determining the Time Window for Dynamic Nanowire Cell Penetration Processes," ACS Nano, vol. 9, p. 11667−11677, 2015. [19] Z. C. Lin, C. Xie, Y. Osakada, Y. Cui and B. Cui, Iridium oxide nanotube electrodes for sensitive and prolonged intracellular measurement of action potentials, vol. 5, Nat. Comm., 2014, p. 3206. 20 [20] M. M. Stevens and J. H. George, "Exploring and Engineering the Cell Surface Interface," Science, vol. 310, p. 1135, 2005. [21] J. Huang, S. Grater, F. Corbellini, S. Rinck, E. Bock, R. Kemkemer, H. Kessler, J. Ding and J. Spatz, "Impact of Order and Disorder in RGD Nanopatterns on Cell Adhesion," Nano Lett., vol. 9, p. 1111, 2009. [22] H. Persson, Z. Li, J. O. Tegenfeldt, S. Oredsson and C. N. Prinz, "From immobilized cells to motile cells on a bed-of-nails: effects of vertical nanowire array density on cell behaviour," Sci. Rep., vol. 5, p. 18535, 2015. [23] M. Toma, A. Belu, D. Mayer and A. Offenhäusser, "Flexible Gold Nanocone Array Surfaces as a Tool for Regulating Neuronal Behavior," small, p. 1700629, 2017. [24] A. Sharei, J. Zoldan, A. Adamo, W. Y. Sim, N. Cho, E. Jackson, S. Mao, S. Schneider, M.- J. Han, A. Lytton-Jean, P. A. Basto, S. Jhunjhunwal, J. Lee, D. A. Heller, J. W. Kang, G. C. Hartoularos, K.-S. Kim, D. G. Anderson, R. Langer and K. F. Jensen, "A vector-free microfluidic platform for intracellular delivery," PNAS , vol. 110, p. 2082, 2013. [25] W. Zhao, L. Hanson, H.-Y. Lou, M. Akamatsu, P. D. Chowdary, F. Santoro, J. R. Marks, A. Grassart, D. G. Drubin, Y. Cui and B. Cui, "Nanoscale manipulation of membrane curvature for probing endocytosis in live cells," Nat. Nanotech., vol. 12, p. 750, 2017. [26] X. Xie, A. M. Xu, M. R. Angle, N. Tayebi, Verma, P. and N. A. Melosh, "Mechanical Model of Vertical Nanowire Cell Penetration," Nano Lett., vol. 13, p. 6002−6008, 2013. [27] W. Kim, J. K. Ng, M. E. Kunitake, B. R. Conklin and P. Yang, "Interfacing Silicon Nanowires with Mammalian Cells," J. Am. Chem. Soc., vol. 129, p. 7228–7229, 2007. [28] R. Yan, J.-H. Park, Y. Choi, C.-J. Heo, S.-M. Yang, L. P. Lee and A. P. Yang, "Nanowire- based single-cell endoscopy," Nat. Nanotechnol., vol. 7, p. 191–196, 2012. [29] R. Singhal, Z. Orynbayeva, R. V. K. Sundaram, J. J. Niu, S. Bhattacharyya, E. A. Vitol, M. G. Schrlau, E. S. Papazoglou, G. Friedman and Y. Gogotsi, "Multifunctional carbon- nanotube cellular endoscopes," Nat. Nanotechnol., vol. 6, p. 57–64, 2011. [30] J. Harding F., S. Surdo, B. Delalat, C. Cozzi, R. Elnathan, S. Gronthos, H. Voelcker N. and G. Barillaro, "Ordered Silicon Pillar Arrays Prepared by Electrochemical Micromachining: Substrates for High-Efficiency Cell Transfection," ACS Appl. Mater. Interfaces, vol. 8, p. 29197−29202, 2016. [31] C. H. Huang, P. Hsiao, F. Tseng, S. Fan, C. Fu and R. Pan, "Pore-Spanning Lipid Membrane under Indentation by a Probe Tip: a Molecular Dynamics Simulation Study," Langmuir, vol. 27, pp. 11930-42, 2011. 21 [32] S. Kawamoto, T. Nakamura, S. O. Nielsen and W. Shinoda, "A guiding potential method for evaluating the bending rigidity of tensionless lipid membranes from molecular simulation," J. Chem. Phys., vol. 139, p. 034108, 2013. [33] M. Hu, P. Diggins and M. Deserno, "Determining the bending modulus of a lipid membrane by simulating buckling," J. Chem. Phys., vol. 138, p. 214110, 2013. [34] H. Noguchi, "Anisotropic surface tension of buckled fluid membranes," Phys. Rev E, vol. 83, p. 061919, 2011. [35] F. Santoro, W. Zhao, L.-M. Joubert, L. Duan, J. Schnitker, Y. van de Burgt, H.-Y. Lou, B. Liu, A. Salleo, L. Cui, Y. Cui and B. Cui, "Revealing the Cell−Material Interface with Nanometer Resolution by Focused Ion Beam/ Scanning Electron Microscopy," ACS Nano, vol. 11, p. 8320, 2017. [36] M. Venturoli, M. M. Sperotto, M. Kranenburg and B. Smit, "Mesoscopic models of biological membranes," Physics Reports, vol. 437, p. 1 – 54, 2006. [37] J. N. Israelachvili, Intermolecular and Surface Forces, London: Academic Press, 2011. [38] R. Goetz, G. Gompper and R. Lipowsky, "Mobility and Elasticity of Self-Assembled Membranes," Phys. Rev. Lett. , vol. 82, p. 221, 1999. [39] E. Lindahl and O. Edholm, "Mesoscopic Undulations and Thickness Fluctuations in Lipid Bilayers from Molecular Dynamics Simulations," Biophys. J. , vol. 79, p. 426, 2000. [40] G. Brannigan, L. Lin and F. Brown, "Implicit solvent simulation models for biomembranes," Eur. Biophys. J., vol. 35, p. 104, 2006. [41] S. J. Marrink and A. E. Mark, "Effect of Undulations on Surface Tension in Simulated Bilayers," J. Phys. Chem. B , vol. 105, p. 6122, 2001. [42] J. F. Zimmerman, R. Parameswaran, G. Murray, Y. Wang, M. Burke and B. zimm, "Cellular uptake and dynamics of unlabeled freestanding silicon nanowires," Sci. Adv., vol. 2, p. 1601039, 2016. [43] J.-H. Lee, A. Zhang, S. S. You and C. M. Lieber, "Spontaneous Internalization of Cell Penetrating Peptide-Modified Nanowires into Primary Neurons," Nano Lett., vol. 16, p. 1509, 2016. GRAPHICAL ABSTRACT 22 1 µm cut 50 nm Curvature-induced permeabilization F cross section R Cell membrane 23
1512.01292	1	1512	2015-12-04T00:41:30	On Improving the Performance of Nonphotochemical Quenching in CP29 Light-Harvesting Antenna Complex	[ "physics.bio-ph", "physics.chem-ph", "quant-ph" ]	We model and simulate the performance of charge-transfer in nonphotochemical quenching (NPQ) in the CP29 light-harvesting antenna-complex associated with photosystem II (PSII). The model consists of five discrete excitonic energy states and two sinks, responsible for the potentially damaging processes and charge-transfer channels, respectively. We demonstrate that by varying (i) the parameters of the chlorophyll-based dimer, (ii) the resonant properties of the protein-solvent environment interaction, and (iii) the energy transfer rates to the sinks, one can significantly improve the performance of the NPQ. Our analysis suggests strategies for improving the performance of the NPQ in response to environmental changes, and may stimulate experimental verification.	physics.bio-ph	physics	On Improving the Performance of Nonphotochemical Quenching in CP29 Light-Harvesting Antenna Complex Gennady P. Bermana, Alexander I. Nesterovb, Richard T. Sayrec, Susanne Stilld aTheoretical Division, T-4, Los Alamos National Laboratory, and the New Mexico Consortium, Los Alamos, NM 87544, USA bDepartamento de F´ısica, CUCEI, Universidad de Guadalajara, Av. Revoluci´on 1500, Guadalajara, CP 44420, Jalisco, M´exico cBiological Division, B-11, Los Alamos National Laboratory, and the New Mexico Consortium, Los Alamos, NM 87544, USA dDepartment of Information and Computer Sciences, and Department of Physics and Astronomy, University of Hawaii at M¯anoa, 1860 East-West Road, Honolulu, HI 96822, USA Abstract We model and simulate the performance of charge-transfer in nonphotochemical quenching (NPQ) in the CP29 light- harvesting antenna-complex associated with photosystem II (PSII). The model consists of five discrete excitonic energy states and two sinks, responsible for the potentially damaging processes and charge-transfer channels, respec- tively. We demonstrate that by varying (i) the parameters of the chlorophyll-based dimer, (ii) the resonant properties of the protein-solvent environment interaction, and (iii) the energy transfer rates to the sinks, one can significantly improve the performance of the NPQ. Our analysis suggests strategies for improving the performance of the NPQ in response to environmental changes, and may stimulate experimental verification. Keywords: Electron transfer, photosynthesis, noise, correlations, nonphotochemical quenching PACS: 03.65.Yz, 05.60.Gg,05.40.Ca,87.15.ht,87.18.Tt 1. Introduction Photosynthesis in plants and algae is powered by rapid transfer of excitation energy to the reaction cen- ter (RC). At full sunlight intensities the rate of photon capture exceeds the rate of downstream electron transfer catalyzed by the cytochrome b6f complex by a factor of 10 [1]. In this case, the PSII RC becomes over-reduced blocking further photochemistry. As a result, damag- ing processes occur. In particular, the energy of excess chlorophyll excited states can be used for production of singlet oxygen species which can destroy the pho- tosynthetic organism. To survive intense sunlight fluc- tuations, photosynthetic organisms have evolved many protective strategies, including NPQ [2] (and references therein). NPQ is a strategy for partial suppression of the dam- aging channels. Excessive sunlight energy is transferred into quenching channels, including energy dissipation by the xanthophyll cycle carotenoids. The initiation of NPQ includes four main stages: (1) protonation pro- cesses (on timescales up to a few milliseconds) which are accompanied by decreasing pH in particular regions of the LHC; (2) geometrical reorganizations due to con- formational changes of the protein-solvent environment (on timescales up to a few minutes and more); and (3) very rapid exciton transfer (ET) to charge transfer or damaging channels, on timescales up to a few picosec- onds. Here, we consider only stage (3) of the NPQ process: energy transfer inside the LHC and to the sinks. The success of NPQ depends on the efficiency with which energy gets transferred away from damaging processes. In this paper, we ask how this efficiency depends, on one hand, on parameters that can be changed by the plant via adaptation (mutation), and, on the other hand, on noise characteristics of the protein-solvent environ- ment. Specifically, we analyze stage (3) of the NPQ mechanism in CP29 LHCs of plants and green algae by a charge-transfer state (CTS), discussed in [3] (see also [4, 5, 6])1. We build on the model developed in [7, 8], and demonstrate numerically that the performance of the NPQ mechanism can be significantly improved by varying (i) characteristics of the chlorophyll dimer, (ii) 1Note, that there are many NPQ related mechanisms discussed in the literature, but a consensus does not yet exist, see e.g., [2], and references therein. Preprint submitted to Physics Letters A August 13, 2018 resonant properties of protein-solvent interaction, and (iii) the ET rates to sinks. Our analysis suggests strate- gies for restoring NPQ efficiency in response to envi- ronmental changes. 2. Model and main equations Energy transfer processes of the NPQ mechanism can be modeled by describing the sites associated with light-sensitive chlorophyll or carotenoid molecules by discrete excitonic energy states, ni (where n enumer- ates the sites of the LHC). Both the damaging and the quenching channels can be characterized by their cor- responding energy sinks, S ni, that provide independent continuum electron energy spectra [7, 8]. These sinks can have very complex structures. They can be respon- sible for many quasi-reversible chemical reactions, such as primary charge separation processes in the photosyn- thetic RCs [9, 10], creation of CTS and singlet oxy- gen production [7, 8], and coherent quantum effects [11, 12, 13, 14] even at ambient temperature. Generally, each sink, S ni, is connected to a particular site, ni. The energy transfer from this state to its sink is characterized by the corresponding ET rate, Γn. For those sites which are not connected to sinks, the corresponding ET rates vanish. In our model, five discrete excitonic energy states and two energy sinks are embedded in a protein-solvent en- vironment. The environment is modeled by a random telegraph process (RTP). (See Fig. 1.) For simplicity, we include only the first excited states of the chlorophyll and carotenoid molecules. More general consideration should also include their long-lived excited states [2]. This generalization is straightforward within our frame- work. Under reasonable assumptions, the quantum dynam- ics of the ET can be described by an effective non- Hermitian Hamiltonian [7, 8], H = H − iW, where, H = Xn εnnihn +Xm,n Vmnmihn, m, n = 0, 1, . . . , 4, is the dressed Hamiltonian, and W = 1 2 Xn={0,4} Γnnihn. (1) (2) In (1), εn is the renormalized energy of the discrete state, ni, and the parameter, Γn, in (2) is the tunneling rate to the n-th sink (only Γ0 and Γ4 are included). In our model, the dimer, which consists of two chlorophylls, Chla5 and Chlb5, each in an excited elec- tron state, participates in the NPQ process, as described Figure 1: Schematic of the NPQ model consisting of five discrete excitonic states, ni, (n = 0, ..., 4), and two independent sinks, S 0i (connected to the damaging state), and S 4i (connected to the CTS that helps dissipate energy as part of the NPQ strategy). The blue rectangular indicates the dimer based on the excited states of a pair of chlorophyll molecules. The red dashed lines indicate non-zero matrix elements used in numerical simulations. in [3]. We use the following notation (see Fig. 1 and [3]): The discrete electron state, 1i ≡ Chla∗ 5i, is the ex- cited electron state of Chla5. The discrete electron state, 5i, is the excited electron state of Chlb5. The 2i ≡ Chlb∗ discrete state, 3i ≡ (Chla5 − Zea)∗i, is the heterodimer excited state, where "Zea" denotes the carotenoid Zeax- anthin. The discrete state, 4i ≡ (Chla− 5 − Zea+)∗i, is the CTS of the heterodimer. The sink, S 4i, is the con- tinuum part of the CTS (channel), associated with dissi- pation of energy, and thereby suppression of the damag- ing channel. The discrete part of the damaging channel is the state, 0i, and the sink, S 0i, is the continuum part of the damaging channel. All matrix elements, Vmn, in the Hamiltonian (1) de- scribe interactions between the discrete excitonic states. Each sink is characterized by two parameters: the rate, Γ0 (Γ4), of ET into the sink, S 0i (S 4i), from the cor- responding attached discrete state, 0i (4i), and the ef- ficiency (cumulative time-dependent probability), η0(t) (η4(t)), for the exciton to be absorbed by the correspond- ing sink. Note, that both, Γ0 and Γ4, characterize only the rates of destruction of the exciton in the CP29 LHC. They do not describe any subsequent chemical reactions that take place in the damaging and the charge-transfer channels. In this sense, our model describes only the primary NPQ processes in the ET, and it does not de- scribe the processes which occur in both sinks. The latter occur on relatively large timescales, and require a detailed knowledge of the structures of the sinks, and additional methods for their analysis. The dynamics of the system can be described by the 2 The evolution of the average components of the den- sity matrix is described by the following system of or- dinary differential equations [7, 8]: (3) d dt hρi = i[hρi, H] − {W, hρi} − iBhρξi, (12) hρξi = i[hρξi, H] − {W, hρξi − iBhρi − 2γhρξi, (13) (dm − dn)mihn, B = Xm,n dm = λmσ, hρξi = hξρi/σ. (14) (15) (16) The average, h . . . i, is taken over the random process. We used the approach developed in [22, 23, 24] for the RTP to derive Eqs. (12) and (13). Employing Eqs. (12) -- (16), one can show that the fol- lowing normalization condition is satisfied, 4 Xn=0 hρnn(t)i + η0(t) + η4(t) = 1. (17) Eq. (17) requires that the total probability of finding the exciton among the five discrete levels and in two sinks is unity for all times. Resonant noise The ET rate of the "donor-acceptor" dimer depends on the noise characteristics of the RTP. In particular, there exist the conditions of "resonant noise" resulting in a maximal ET rate [18, 21]. According to [18], the ET rate, ΓDA, is given by: ΓDA = 8γVDA2d2 (d2 − ε2)2 + 4γ2ε2 , d = (λD − λA)σ, Liouville-von Neumann equation, ρ = i[ρ, H] − {W, ρ}, where {W, ρ} = Wρ + ρW. We define the ET efficiency of tunneling to all N sinks as, η(t) = 1 − Tr(ρ(t)) = Z t 0 Tr{W, ρ(τ)}dτ. (4) This can be expressed as the sum of time-integrated probabilities of trapping an electron into the n-th sink, [13, 14]: d dt with ηn(t), η(t) = Xn ηn(t) = ΓnZ t 0 ρnn(τ)dτ , (5) (6) where ηn(t) is the efficiency of the n-th sink. In partic- ular, for our model (Fig. 1), complete suppression of the damaging channel at all times occurs if η0(t) = 0. This would be the most desirable outcome, leading to maximal effectiveness of the NPQ mechanism. In the presence of the protein-solvent noisy envi- ronment, the evolution of the system can be described by the following effective non-Hermitian Hamiltonian [7, 8], Htot = H − iW + V(t), where the operator, V(t) = Xm,n λmn(t)mihn , (7) (8) with m, n = {0, 1, . . . , 4}, describes the influence of the protein-solvent noisy environment. The noise matrix el- ements, λmn(t), lead to both relaxation and decoherence processes. In what follows, we restrict ourselves to the diagonal noise effects. (See also [15, 16, 17], and refer- ences therein.) Then, one can write, λmn(t) = λnδmnξ(t) , (9) where where λn is the coupling constant at site, n, and ξ(t) is a random process. Generalization to local protein-solvent environments can be done following [18]. We describe the protein-solvent noisy environment by a random telegraph process (RTP), ξ(t), with the fol- lowing properties [19, 20], hξ(t)i = 0, hξ(t)ξ(t′)i = σ2e−2γt−t′, (10) (11) where σ is the amplitude of noise, and 2γ is the decay rate of the noise correlation function. 3 is the renormalized amplitude of the noise. As one can see from Eq. (18), the rate, ΓDA, has a maximum, Γ (max) DA = 4γVDA2 , pε4 + 4γ2ε2 − ε2 at the "resonant amplitude" d(res) DA = (ε4 + 4γ2ǫ2)1/4 ≈ ε (last expression is valid for 2γ ≪ ε, which is assumed in our numerical simulations). (18) (19) (20) Weakly and strongly coupled chlorophyll dimer In the numerical simulations we report on in the next section, we call the chlorophyll-based dimer, Chla∗ 5i − 5i ≡ 1i − 2i weakly coupled if V12/δ ≪ 1, where Chlb∗ δ = ε1 −ε2. In this case, two orbitals (eigenfunctions) of the dimer, ϕ−i and ϕ+i, are close to the unperturbed site states, ϕ−i ≈ 1i and ϕ+i ≈ 2i [7]. We call the dimer strongly coupled if V12/δ & 1. In this case, the two orbitals of the dimer, ϕ−i and ϕ+i, represent mixtures of the two site states [7]. The corresponding eigenener- gies are: E± = (ε1 + ε2)/2 ± pδ2 + 4V122/2 (E+ > E−). Usually, the chlorophyll-based dimer, 1i−2i, is weakly coupled in the CP29 LHC. However, according to [3], in the NPQ regime this dimer becomes strongly cou- pled. This transformation of the dimer from weakly to strongly coupled can occur, for example, due to an asymmetric modification (as a result of asymmetric pro- tonation) of local potentials at sites 1i and 2i. 3. Results of numerical simulations We analyzed the dependence of NPQ performance on the following parameters: (1) the resonant properties of the protein-solvent environmental noise (controlled by parameter d), (2) the coupling strength of chlorophyll- based dimer (controlled by V12/δ; we fix V12, and vary δ), and (3) the rates to the sinks, controlled by Γ0. We numerically computed the solutions of Eqs. (12) and (13) for the density matrix components, averaged over noise realizations. In all numerical simulations, we set = 1. Then, the values of parameters in energy units are measured in ps−1 (1ps−1 ≈ 0.66meV). Time is mea- sured in ps. Noise amplitudes, dn (n = 0, ..., 4), were chosen close to those discussed in the literature. In particular, "res- onant conditions" (dn − dm ≈ εn − εm), discussed in [18], were used for the dimer transitions, mi ⇄ ni (m, n = 1, 2, 3, 4). The noise correlation decay rate was 2γ = 20 ps−1. The rates to the sinks were: Γ0 = 1 and 2 ps−1, and Γ4 = 10 ps−1. Fig. 2 shows the results of our numerical simula- tions: the efficiency, η4(t), of the sink S 4i, which is associated with the beneficial charge-transfer channel. The efficiency, η4(t), saturates at relatively short times, tsat = (2 − 4) ps. Note, that in the saturation regime, all density matrix elements, ρnm(t) (for discrete states), ap- proach zero, and the total probability accumulates in the two sinks. Table 1 provides an overview of the resulting efficiency, η4. Figure 2: Time dependence of the NPQ efficiency, η4(t). Solid/dotted curves demonstrate weakly/strongly coupled chlorophyll dimer. Res- onant noise on the transition 1i ⇄ 0i (d0 = −90): Γ0 = 2 (red), Γ0 = 1 (blue). Non-resonant noise on the transition, 1i ⇄ 0i (d0 = 0): Γ0 = 2 (green). The insert demonstrates the dynamics of the NPQ efficiency for short times. Fixed parameters: V10 = V20 = 30, V12 = 15, V13 = V34 = 25, ε0 = −90, ε1 = 60, ε1 − ε2 = δ, ε3 = 45, ε4 = 30, Γ4 = 10, γ = 10, d1 = 60, d2 = 90, d3 = 45, d4 = 30. Initial conditions: ϕ(0)i = ϕ−i. Table 1: Efficiency η4. Improving the NPQ performance by transition from weakly to strongly coupled chlorophyll dimer The solid red curve in Fig. 2 corresponds to the weakly coupled chlorophyll dimer, 1i − 2i (V12 = 15, δ = ε1 − ε2 = −90, V12/δ ≈ 0.17) with a relatively high rate to the damaging sink, Γ0 = 2, and resonant environ- mental noise applied to the dimer 1i − 0i. In this case, the asymptotic efficiency of the NPQ is relatively small, η4 ≈ 80%. This means that 20% of the photosynthetic organisms die in this regime. Suppose that the weakly coupled chlorophyll dimer, 1i−2i, is transformed in the NPQ regime to be strongly coupled, as assumed in [3]. The result of this transfor- mation is demonstrated by the red-dotted curve (V12 = 15, δ = −15, V12/δ = 1). The saturated efficiency ap- proaches, η4 ≈ 85%. The outcome of the transformation from weakly to strongly coupled dimer thus improves the efficiency, η4, by approximately 5%. This improve- ment is the result of destructive interference effects for 4 the probability amplitude of state 0i. Similar effects of improvement of the NPQ efficiency occur when the ET rate, Γ0, to the damaging channel is set to Γ0 = 1. Weakly coupled chlorophyll dimer now results in η4 ≈ 87.5% efficiency (blue curve in Fig. 2, δ = −90, Γ0 = 1), while strongly coupled chloro- phyll dimer is performing ≈ 2.5% better, resulting in η4 ≈ 90% (blue-dotted curve, δ = −15, Γ0 = 1). Even though the absolute values of the NPQ efficiencies have increased (in comparison with the case Γ0 = 2), the rel- ative improvement is less. Improving the NPQ performance by transition from res- onant to non-resonant environmental noise NPQ efficiency improves by approximately 10% when non-resonant environmental noise is applied to the dimer 1i − 0i: weak coupling and Γ0 = 2 lead to η4 ≈ 80% under resonant noise (red curve), and to η4 ≈ 90% under non-resonant noise (green curve). The most favorable case of the ones we tried here is strong coupling and Γ0 = 2 under non-resonant noise (green-dotted curve), resulting in an efficiency of ≈ 97.5%. We would expect that if we do an optimiza- tion over parameters, than the efficiency should be even closer to 100%. Restoring NPQ efficiency in response to environmental change The results presented in Fig. 2 can be used to de- velop strategies for restoring NPQ efficiency in response to changes in the environment. To summarize, the effi- ciency, η4, is lowest (≈ 80%) with weak coupling, larger Γ0, and resonant noise. By switching to strong coupling, and also to lower Γ0, the efficiency increases to ≈ 90%. There is another way to change the efficiency from 80% to 90%, and that is simply by switching from resonant to non-resonant noise. This suggests that pH-dependent modifications of the choroiphyll dimer coupling in CP29 could restore NPQ efficiency in response to environment changes. Imagine that the CP29 dimer operates with weak coupling and large Γ0, under non-resonant noise, where the NPQ ef- ficiency is η4 ≈ 90% (corresponding to the green curve in Fig. 2). Now, suppose that the protein-solvent en- vironment experiences a modification which results in resonant noise on the dimer 1i − 0i. This modifica- tion increases the ET rate to the damaging channel, S 0i, and, correspondingly, decreases the NPQ efficiency to η4 ≈ 80% (red curve in Fig. 2). This effect could then be counteracted by switching to strong dimer coupling and lower Γ0, thereby restoring the efficiency to 90%. 5 As we discussed above, efficiency would be partially restored by a transition from weakly to strongly cou- pled chlorophyll-based dimer. This transformation in- creases the NPQ efficiency, η4, by ≈ 5% (red-dotted curve, η4 ≈ 85%). Another way is to keep the weakly coupled dimer, 1i − 2i, but to decrease the ET rate, Γ0, to the damaging channel, which improves the NPQ efficiency to ≈ 87.5%. 4. Conclusion 5 − Chlb∗ We demonstrated that the efficiency of the charge- transfer nonphotochemical quenching in CP29 can be improved by transformation from a weakly to a strongly coupled Chla∗ 5 dimer. This transformation could be related to the NPQ mechanism by the CTS discussed in [3]. We also demonstrated strategies for restoring the NPQ efficiency when the environment changes. We analyzed numerically only a limited number of the possible system-environment scenarios. Many additional considerations and generalizations can be studied using our approach. The mathematical advantages of our approach are that it (i) allowed us to derive an exact and closed system of ordinary differential equations, which is easy to analyze and to solve numerically and (ii) does not require the use of small parameters and uncontrolled perturbative techniques. We expect that the results discussed in this paper will be useful for a better understanding of the NPQ mechanisms in photosynthetic organisms, and will stimulate new experimental studies. In particular, the experimental verification of possible mechanisms of transformation from a weakly to a strongly coupled chlorophyll-based dimer, in the NPQ regime, would be of significant interest. One of the important theoretical problems for future research is to understand how the relationship between (in)efficient use of information and energy dissipation [25, 26, 27] comes to bear on the approach presented here. This work was carried out under the auspices of the National Nuclear Security Administration of the U.S. Department of Energy at Los Alamos National Lab- oratory under Contract No. DE-AC52-06NA25396. A.I.N. acknowledges the support from the CONACyT and the suppo345). rt during his visit of the B-Division at LANL. G.P.B. and R.T.S. acknowledge the support from the LDRD program at LANL. S.S. is grateful for support from the Foundational Questions Institute (Grant No. FQXi-RFP3-1 [20] A. I. Nesterov, G. P. Berman, J. M. S´anchez M´artinez, and R. Sayre, Noise-assisted quantum electron transfer in photosyn- thetic complexes, J. Math. Chem., 51, 1 (2013). [21] S. Gurvitz, A.I. Nesterov, and G.P. Berman, Noise-assisted quantum electron transfer in multi-level donor-acceptor sys- tem, arXiv:1404.7816 [physics.bio-ph]. [22] V. Klyatskin, Stochastic Equations through the Eye of the Physicist, (Elsevier, 2005). [23] V. Klyatskin, Dynamics of Stochastic Systems, (Elsevier, 2005). [24] V. Klyatskin, Lectures on Dynamics of Stochastic Systems, (El- sevier, 2011). [25] S. Still, D.A. Sivak, A.J. Bell, and G.E. Crooks, Thermody- namics of prediction, Phys. Rev. Lett., 109, 120604 (2012). [26] A. L. Grimsmo, Quantum correlations in predictive processes, Phys. Rev. A, 87 060302 (2013). [27] A. L. Grimsmo and S. Still, Quantum predictive filtering, arXiv:1510.01023 [quant-ph] (2015). References [1] Z. Perrine, S. Negi, R.T. Sayre, Optimization of photosynthetic light energy utilization by microalgae, Algal Research, 1, 134 (2012). [2] B. Demming-Adams, G. Garab, and W. Adam III Govindjee (Eds), Non-Photochemical Quenching and Energy Dissipation in Plants, Algae and Cyanobacteria, (Springer, 2014). [3] T.K. Ahn, T.J. Avenson, M. Ballottari, Y.C. Cheng, K.K. Niyogi, R. Bassi, and G.R. Fleming, Architecture of a charge- transfer state regulating light harvesting in a plant antenna pro- tein, Science, 320, 794 (2008). [4] A. Dreuw, G.R. Fleming, and M. Head-Gordon, Charge- transfer state as a possible signature of a zeaxanthin- chlorophyll dimer in the non-photochemical quenching pro- cess in green plants. J. Phys. Chem. B, 107, 6500 (2003). [5] A. Dreuw, G.R. Fleming, and M. Head-Gordon, Role of Electron transfer quenching of chlorophyll fluorescence by carotenoids in non-photochemical quenching of green plants, Biochem. Soc. Trans., 33, 858 (2005). [6] Y.C. Cheng, T.K. Ahn, T.J. Avenson, D. Zigmantas, K.K. Niyogi, M. Ballottari, R. Bassi, and G.R. Fleming, Kinetic modeling of charge-transfer quenching in the CP29 minor complex, J. Phys. Chem. B, 112, 13418 (2008). [7] G.P. Berman, A.I. Nesterov, S. Gurvitz, and R.T. Sayre, Pos- sible role of interference and sink effects in nonphotochemi- cal quenching in photosynthetic complexes, arXiv:1412.3499 [physics.bio-ph] [8] G.P. Berman, A.I. Nesterov, G.V. Lopez, and R.T. Sayre, Superradiance transition and nonphotochemical quenching in photosynthetic complexes, J. Phys. Chem. C., 119, 22289 (2015). [9] M. Pudlak and R. Pincak, Modeling charge transfer in the pho- tosynthetic reaction center, Phys. Rev. E, 68, 061901 (2003). [10] M. Pudlak, Primary charge separation in the bacterial reac- tion center: Validity of incoherent sequential model, J. Chem. Phys., 118, 1876 (2003). [11] 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). [12] D. Ferrari, G.L. Celardo, G.P. Berman, R.T. Sayre, and F. Borgonovi, Quantum biological switch based on superradiance transitions, J. Phys. Chem., 118, 20 (2013). [13] P. Rebentrost, M. Mohseni, I. Kassal, S. Lloyd, and A. Aspuru- Guzik, Environment-assisted quantum transport, New J. Phys., 11, 033003 (2009). [14] A.W. Chin, A. Datta, F. Caruso, S.F. Huelga, and M.B. Ple- nio, New J. Phys., Noise-assisted energy transfer in quantum networks and light-harvesting complexes, 12, 065002 (2010). [15] R. Marcus and N. Sutin, Electron transfers in chemistry and biology, Biochimica et Biophysica Acta, 811, 265 (1985). [16] D. Xu and 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, Chemical Physics, 182, 97 (1994). [17] M. Merkli, G.P. Berman, and S.T. Sayre, Electron transfer re- actions: Generalized spin-boson approach, J. Math. Chem., 51, 890 (2013). [18] A .I. Nesterov and G. P. Berman, Role of protein fluctuation correlations in electron transfer in photosynthetic complexes, Phys. Rev. E, 91, 042702 (2015). [19] 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). 6
1008.0953	1	1008	2010-08-05T11:46:15	Pearling in cells: A clue to understanding cell shape	[ "physics.bio-ph", "cond-mat.soft", "q-bio.CB" ]	Gradual disruption of the actin cytoskeleton induces a series of structural shape changes in cells leading to a transformation of cylindrical cell extensions into a periodic chain of "pearls". Quantitative measurements of the pearling instability give a square-root behavior for the wavelength as a function of drug concentration. We present a theory that explains these observations in terms of the interplay between rigidity of the submembranous actin shell and tension that is induced by boundary conditions set by adhesion points. The theory allows estimation of the rigidity and thickness of this supporting shell. The same theoretical considerations explain the shape of nonadherent edges in the general case of untreated cells.	physics.bio-ph	physics	Pearling in cells: A clue to understanding cell shape ROY BAR-ZIV, TSV I TLUSTY, ELISHA MOSES, SAMUEL A. SAFRAN , AND ALEXANDER BERSHADSKY Departments of Physics of Complex Systems, ‡Materials and Inter faces, and ¶Molecular Cell Biolog y, Weizmann Institute of Science, Rehovot 76100, Israel ABSTRACT Gradual disruption of the actin cy toskeleton induces a ser ies of structural shape changes in cells leading to a transformation of cylindr ical cell extensions into a per iodic chain of ‘‘pearls.’’ Quantitative measurements of the pearling instability g ive a square-root behavior for the wavelength as a function of drug concentration. We present a theor y that explains these obser vations in terms of the interplay between r ig idity of the submembranous actin shell and tension that is induced by boundar y conditions set by adhesion points. The theor y allows estimation of the r ig idity and thickness of this suppor ting shell. The same theoretical considerations explain the shape of nonadherent edges in the general case of un- treated cells. The parameters that determine cell shape are generally well known: adhesion to the substrate, membrane elasticity, and cy toskeleton mechan ics (1– 4). Yet, a quantitative and predic- tive model for cell shape is still lack ing. In particular, the interplay of the cy toskeleton rigidity w ith the tension produced by constraints subjected on the cell by adhesion points has not been accounted for in a quantitative framework. The importance of actin in the majority of processes of cell morphogenesis leads us to probe in detail the changes arising on gradual disruption of the actin cy toskeleton, by using the drug latrunculin A (LatA). LatA is known to bind monomeric actin in a 1:1 complex, sequestering it and thereby allow ing control of the level of polymerized actin by var ying the drug concentration (5–7). We show here that a gradual increase in LatA concentration leads first to arborization (the formation of numerous radial tubular protrusions), a phenomenon that has prev iously been described for other drugs that disrupt the actin cy toskeleton. Further increase of the LatA concentration induces an instability of these tubes, converting them into a chain of pearls. This ‘‘pearling’’ is a general phenomenon of tubes under tension that can be found in a w ide variety of physical systems (8), including phospholipid bilayers (9). We quantify the dynamics of the instability in detail and present a theor y that explains these phenomena in terms of the competition between the tension in the membrane and the rigidity of the actin cy toskeleton that opposes it. It is inter- esting that the same theor y can quantitatively explain the shape of adherent cells during the arborization process. More- over, the theor y describes the shape of both cells w ith dis- rupted actin cy toskeleton, as well as untreated, normally adhering cells. Thus, a simple description in terms of rigidity, sustained by a h ’ 1mm thick actin shell underlying a tense lipid bilayer (10, 11) can explain major features of cell morphogen- esis. Comparison of theor y and experiment allows estimation of the actin-shell thickness and elastic moduli in both arborized and pearled states and prov ides a quantitative measure of the rigidity as it is reduced by LatA. Obser vations SVT2 cells (12) were plated on coverslips and treated w ith increasing concentration of LatA (0.08 – 40 mM). Untreated cells are polygonal, w ith lamellipodia and protrusions concen- trated in one or a few locations on the cell peripher y (Fig. 1a). Their edges are cur ved, and the protrusions are at the vertices. This morpholog y is disrupted by the addition of LatA in two stages. First, the outer envelope shrinks to form a round cell body enclosing the nucleus and most of the cy toplasm, leav ing radial tubular protrusions still attached to adhesion points (a ‘‘sun’’ w ith ‘‘rays’’). In the second stage, some of the tubes destabilize, transforming into the pearled state characteristic of tense cylinders. Complete recover y of cell shape occurs on removal of LatA (data not shown). Careful examination of the cell morpholog y during arboriza- tion shows that the shape is produced by stretching the cell sur face between adhesion points creating cur ved edges (13– 15). As the concentration increases, the hanging sur face droops, and its cur vature increases (Fig. 1 b and c). The part that is near the cell body is well approx imated by a circular shape. With progressively higher concentrations of LatA, the cell is rounded, and the cur vature of the sur face between protrusions changes sign (Fig. 1d). At this stage, instability of the cylinders of ten occurs (Fig. 1e). Electron microscopy of the tubular extensions taken af ter treatment w ith the drug confirmed that these cylindrical regions contain microtubule bundles. Real-time obser vations of dynamic behav ior, such as buckling w ithin the tubular protrusions, may be attributed to these microtubules because they grow and push against the membrane enveloping the cell. However, the pearling instability cannot be inhibited by total disruption of the microtubules, which is achieved by pretreat- ment of the cells w ith 10mM nocodazole (data not shown). The Rayleigh-like pearling instability of these extensions is depicted in Fig. 2. Pearling appears as a fin ite-amplitude, peristaltic modulation of the cylinder, characterized by a well- defined wavelength. The shape in the highly nonlinear state of the instability is that of isolated spherical pearls of membrane connected by thin tubular sections. The instability occurs predominantly in thinner tubes. Obser v ing the dynamics of the instability, we found that as the modulation progresses in time, an in itially long wavelength can evolve into a final shorter one. The tension in itiating the instability of ten is induced by elongation, related either to motion of the cell as the adhesion sites remain fixed or to the motion of microtubule bundles inside the cylindrical extension. Some tubes are unstable to the pearling instability even w ith no drug added, but this is a relatively rare event that occurs in ex isting long tubular protrusions, for example, those that connect neighboring cells. To quantify the characteristics of the instability we treated the cells w ith LatA at various concentrations for 1 h and then fixed them. Cells were fixed by 3% paraformaldehyde in PBS; FIG. 1. Microscope images taken in dif ferential inter ference contrast (363, oil-immersion objective) of SVT2 cells in var ying degree of arborization af ter treatment w ith LatA. Concentrations F of the drug are (a) control F 50; (b ) F 50.62 mM; (c) F 51.25 mM; (d) F 52.5 mM; and (e) F 510 mM. a and b are most probably mononuclear, whereas in c-e we chose more spherical, multinuclear cells. This illustrates the arborization by enhancing both the symmetr y and the number of arbors. (Bar 5 20 mm.) this fixation did not change the morpholog y of the pearled tubular protrusions. Measuring the percentage of fixed cells that have undergone pearling shows that this occurs in a range of LatA concentration generally higher than that required for arborization. As the drug concentration, F, is increased, the probability for pearling also rises. Fig. 3 (Lower) shows a fit of the measured probability to a normal distribution (i.e., this cur ve is its integral—an error function) w ith a mean of Fc5 2.5 6 1 mM and a variance that is of the same order as Fc. The wavelength l also shows a clear trend w ith F. Fig. 3 (Upper) shows the measurement of the dimensionless wave- ' qR0 5 2pR0/l, that characterizes the inst abilit y number, k at dif ferent drug concentrations, averaged over many cells. Var ying F by over two decades, we obt ain a clear power law behav ior k 5 (0.42 6 0.03) 3 F0.51 6 0.04 over most of the range covered. We also obser ved the characteristic change of shape asso- ciated w ith actin disruption, as well as the pearling of tubular extensions on treatment w ith cy tochalasin D (results not shown). However, the quantitative relation of the wavelength to the concentration of cy tochalasin was not apparent in this case. The complicated mechansism by which cy tochalasin acts to disrupt the actin cy toskeleton creates a nonlinear relation- ship between drug concentration and the extent of actin polymerization (16 –18). The linear relation of drug concen- tration to F actin disruption seems to be a un ique property of LatA, which is also the only drug whose specificity for actin has been shown genetically (6, 7). Theor y Whereas the same forces of tension and elastic stress act on the cell independent of its shape, the changes in geometr y modify its stability properties. The driv ing force for the pearling instability is the tension in the cell. Typical measured values of this ef fective tension in untreated cells are s. 431022 erg/cm2 (19 –21). The precise origin of tension is an interesting question but is not crucial for our model. Most probably it lies in passive elements such as the membrane, coming from the boundar y constraints set by the adhesion points that link the cell to the substrate. A lternatively, active actomyosin contractility could contribute, but in this case the ef fective tension energ y Us would only be redefined. In multicellular tubes like blood vessels, the activation of actomyosin contractility by the drug angiotension II was shown to induce the pearling instability (22). LatA is highly specific and should not af fect the actin- myosin interaction directly. Its addition should therefore not af fect this active component until the actin filaments get too short as a result of LatA-induced depolymerization. We addressed the question of passive versus active tension product ion in a pre l im inar y ex per iment by us ing 2 ,3- butanedione monox ime (BDM, 10 –30 mM), a known inhibitor of interactions of actin w ith myosin (23). We obser ved LatA- induced pearling even at the highest concentration of BDM (data not shown), suggesting the primar y importance of the passive component. FIG. 2. Pearling in an SVT2 cell treated w ith 2.5 mM LatA and fixed in paraformaldehyde. A ll samples included in the statistical analysis were fixed and then v iewed in a dif ferential inter ference contrast microscope by using a computer-enhanced v ideo system. Dynamical obser vations were conducted in a temperature-stabilized dish held at 37°C. (Bar 5 10 mm.) The force that resists pearling is caused by the shear rigidity in the solid-like actin cortex, which resides beneath the mem- brane (10, 11). The shear strength of an actin shell ref lects its response to a deforming strain and is characterized by its Young’s modulus E. In an actin gel, E increases w ith the density of filaments and the number links between them [typically, E ’ 102–103 Pa for highly cross-linked actin gels (24 –27)]. Increasing the LatA concentration F decreases the amount of polymerized actin in a linear fashion (5) and consequently reduces its Young’s modulus E. Our measure- ments are per formed far enough from the percolation point (where the network is so disrupted that its macroscopic shear rigidity falls to zero )i that we can assume a linear dependence: E(F) 5 E0 2 aF. The Young’s modulus thus van ishes at F0 5 E0/a. Because of the order-of-magn itude dif ference be- tween the actin mesh size (#0.1 mm) and the size of the min imal shape that we obser ve ($1 mm), we can use a continuum description of the actin cy toskeleton. We also assume that the actin is homogenous and does not redistribute. Electron microscope pictures indicate no tendency of the actin to redistribute. Such a spatial inhomogeneity w ill not af fect the in itial instability, although it may contribute to the higher order, nonlinear ef fects. Pearling. To understand the origin of pearling in the tubular extensions, we consider a single tube under the combined inf luence of tension and rigidity. Both a thin elastic shell and a solid rod are considered; we w ill see that the experimental case corresponds to the solid rod. Using a linear approx ima- tion, we consider the stability of a tube subject to a peristaltic perturbation of its radius, r(z), of the form r(z) 5 R0 1 u sin(qz) 2 (u2y4R0) , where R0 is the tube radius, q is the iIn fact, cooperative ef fects cause the Young modulus to van ish at the percolation point, which is expected to occur at volume fractions somewhat below F0. Percolation ef fects are expected to be sign ificant only in a narrow v icin ity around the saturation point, where the network is totally disrupted. This cannot be detected w ithin our experimental precision. Saturation of the pearling wavelength may be related to the van ishing of the shear rigidity at a fin ite concentration of drug. (Upper) Nondimensional wave number k 5 2pR0/l of the FIG. 3. pearled state as function of the LatA concentration F. Samples were scanned by eye to identify cells w ith pearled tubular protrusions, and still v ideo pictures were transferred to the computer for analysis. The diameter of pearls 2 R, distances between them l, and the diameter of tube sections connecting pearls d were measured. In many of the cells for which we identified pearling, the instability was in the nonlinear state w ith a periodic array of isolated pearls rather than a small- amplitude sinusoidal modulation. In such cases, the wavelength was taken as the distance between pearls. To determine the dimensionless wavenumber of the instability 2pR0/l, we needed to know R0, the in itial unperturbed tube radius, which was unknown for fixed cells. Hence, we used volume conser vation along the tube: k 5 2pR0/l 5 2p(4R3/3l3 1 d2/l2 2 2 Rd2/l3)1/2. For each drug concentration, we averaged up to 20 tubes and repeated the measurements at least once for almost all drug concentrations, w ith good reproducibility in each case. At low concentrations (F # 0.5 mM) the wavelength measure- ments are noisy and therefore unreliable. At high concentrations, a trend to saturation of the wavenumber begins to be obser ved. Ac cordingly, we do not present the ver y low F points and did not consider them or the highest F point in fitting the power law. (Lower) Percent age of cells in which at least one c ylinder has undergone pearling as a function of L atA concentration F. About 100 cells were counted per drug concentration. The line is a fit to an error function centered around Fc 5 2.5 mM and w ith a w idth s 5 3 mM. This is an integ ral of the Gaussian distribution erf(F) 5 F e xp[(F9 2 F c)2/2s 2]dF9. A f raction of about (1 2 C(2ps2)21/22‘ C) 5 25% of the cells do not pearl even at high F. wavenumber of the modulation, and u ,, R0 is its amplitude. Volume conser vation is ensured by the term u2y4R0. (This assumes that the cell does not exchange volume through its membrane on the time scales involved, and that osmotic ef fects are negligible. Note that the depolymerization of actin can only produce a small perturbation on the overall osmotic pressure. The addition of monomers would in any case tend to increase the volume of the cell, contrar y to our obser vations.) The area of the tube S is reduced by a factor of S u D 2 R 0 and so the sur face-energ y gain, per un it area, is Us 5 s 3 (dS)y(S) . 3 ~k 2 2 1 ! , dS S 5 1 4 k 2 2 U T 5 A f luid cylinder under sur face tension s is thus unstable to long modes w ith k # 1 (8), and it is up to the elastic forces to oppose the instability. In the case of artificial lipids, the bending energ y of the f luid bilayer overcomes the instability when the dimensionless ratio of energies is large enough ky(sR0 2) $ 2y3 (9) , where k is the bending modulus of the lipid (28). In cells, the elastic restoring force is mainly caused by the underlying actin cortex and is much stronger. For the simple case of a thin shell w ith thickness h ,, R0, elastic deformations of a thin shell can be decomposed into two decoupled components. We w ill call the in-plane deformations stretching, while regarding out-of-plane deformations as bend- ing (in the case of a full, rigid tube, they remain strongly coupled) (29). The stretching is usually much larger than the bending, giv ing as a criterion for instability that the dimen- sionless energ y ratio 2y(1 2 n2) 3 Ehys should be larger than 1, where n is the Poisson ratio. Note that this stability criterion does not depend on the tube radius R0. As mentioned before, thinner tubes are unstable whereas thicker ones are not, so that stability does depend on R0. We conclude that the thickness of the actin layer is close to the tube radius. For a solid rod (h 5 R0), stretching and bending couple to give the total elastic energ y per un it area k 4D , D 2S 1 1 4 E S u 1 1 1 144 6 R 0 where E 5 3ER0y1 1 n is an ef fective cylindrical stretching modulus (30). The exact cylindrical stress function adapted to our boundar y conditions is x(r , z) 5 (Euy3k2I1(qR0)) 3 J0( iqr) sin( qz) (31). Expanding in k yields the series UT. Compared to the full numerical solution, we find that keeping terms to second order gives an accuracy that is better than 0.1% for k # 1, while tak ing only the constant term gives an error of 15% at most. For all relevant wavelengths (k # 1), the k-dependent terms may be neglected (as can easily be verified numerically). The new dimensionless criterion for stability is 2 UTy Us(k 5 0) 5 Eys $ 1 , which does depend on R0. It therefore is easier to destabilize thin tubes, requiring less drug and less disruption of the actin cy toskeleton. Because tubes are typi- cally unstable when R0 # 1–2 mm, and using R0 ’ h, we get an estimate for the thickness of the actin cortex h # 1 mm, in agreement w ith published values (10, 11). This also gives an estimate for the average Young’s modulus at which pearling occurs, Ep 5 (1 1 n)y3 3 syR0 ’ 20 Pa. As we shall see below, this rigidity is sign ificantly lower than that of an untreated cell. It is tempting to conjecture that a similar mechan ism determines the min imal thickness of the adherent cell w ithout drugs. The threshold for pearling occurs when the rigidity is below a critical value corresponding to a critical concentration Fc. Typically, Fc is on the order of 3 mM but depends on the statistical distribution of tensions (and possibly actin rigidity) in our sample of cells. Within a linear theor y, we can find the grow th rate of the instability by considering a small harmon ic perturbation to a full rod u(z,t) 5 uevt 3 sin(qz). The time rate of change of the sur face and elastic energies per un it length is U 5 pv(u2yR0) 3 [s(k2 2 1) 1 E] . The first term represents the tension energ y, whereas the second term is the positive shear rigidity contribution that tends to stabilize the tube. The grow th rate of the instability is determined by equating U w ith the v iscous dissipation of the Poiseuille f low of water in the deforming tube (32), W 5 h(16p3y5) 3 v2k22 3 u2, where h is the cy toplasm v iscosity. The resulting dispersion relation takes the form v~k ! 5 5 16p2 3 k 2 hR 0 @s~k 2 2 1 ! 1 E# . . The typical time scale is determined by the grow th of the fastest mode (k 5 1/2), t 5 (64p2y5 3 hR0ys) . We can use the typical experimental time scale for development of the instability t ’ 3 3 102 sec to obtain an estimate for the cy toplasmatic v iscosity, h’ 103 poise. Experimental measure- ments of the v iscosity of the cy toplasm are still controversial, and our estimate tends to support those of ref. 21 rather than those of ref. 24. Moreover, we find that the typical wavenumber of pearling at criticality is k c 5 s 2 E 5 F 2 F c F 0 2 F c s Comparing to the experiment gives not only the correct square-root behav ior, but the measured prefactor also gives an estimate for F0 on the order of 10 mM. This simple theor y seems to reasonably estimate the value at which the actin has lost almost all ef ficacy. However, the simple linear-dispersion relation cannot predict the full nonlinear dynamic relation that is measured experimentally. The linear theor y does not explain why short wavelengths k . 1 are obser ved, but this is known to be characteristic of the nonlinear regime of pearling. Furthermore, the assumption of a linear dependence of the shear rigidity on the concentration may be unrealistic at some limit. The experimental power law does not show the ex istence of a (radius-dependent) critical concentration Fc. This can be attributed to the distribution of tensions and rigidities, which can be high enough to cause pearling at F , Fc and even in untreated cells. Surprisingly, the wavenumber goes to 0 exactly at the limit of untreated cells, hinting at a mechan ism of ‘‘marginal’’ stability in untreated cells, which keeps the rigidity in thin tubular protrusions regulated on the verge of instability. Cell Shape. The same physical considerations used to un- derstand pearling can also explain specific features of normal cell shape and their changes on gradual disruption of the actin cy toskeleton. Our simplified model regards the cell as an envelope that is attached to the substrate at several discrete locations and can mediate tension, along w ith a thin actin cortex shell supplying rigidity beneath it. The balance of forces in the cell w ill then depend drastically on the geometr y. In its in itial natural state, the adhering cell has a f lattened shape, which is a stable configuration. Most of the elastic energ y is located around the edge, giv ing an ef fective line tension g. Approx imating the cross section of the elastic edge as ph2 gives g . pEh2. Fig. 4 depicts the calculated equilibrium shapes of the cell for decreasing g. Min imizing analy tically the total energ y gdl 1 sdS (elastic line tension on the edge together w ith the sur face tension on the whole cell) gives the force – balance equation s 5 g/Rp. Volume constraints are supposedly taken care of by adjustment of the cell’s vertical profile. The equi- librium shape describes a positive-cur vature circular arc be- tween the adhesion points, w ith radius Rp 5 g/s ’ pEh2/s. Approx imating for the untreated cell Rp(F 50) 5 10mm and h ’ 1 mm, whiles tak ing for the tension values typically measured in cells s 5 4 3 1022 erg/cm2 (19 –21), we obtain an estimate of E0 ’ 100 Pa, consistent w ith values reported in refs. 24 –27 (compared w ith Ep ’ 20 Pa for pearling). As seen in both experiment and calculation Rp ’ E0 2 aF indeed decreases w ith F. At high values of F, Rp decreases to a value smaller than the typical distance between adhesion points, and the tubular shape begins to become apparent (Fig. 1 d and h and the theoretical shapes, Fig. 4 c and g). The shape of the membrane near the cell center can be shown analy tically to be close to a semicircle, i.e., the distance between rays at the contact to the cell center is L ’ 2 Rp. This prediction is well borne out experimentally (L/Rp 5 2 w ithin 10% on average). FIG. 4. Shapes of cells calculated from the theor y as a function of the changing ratio between the balanced sur face tension and ef fective elastic line tension at their edges. The calculation begins by placing a symmetric polygonal cell w ith n adhesion points lying on the un it circle (n 5 3 for a–d and n 5 7 for e–h). The parameter that determines the shape is then the radius Rp 5 g/s, measured in un its of the distance between adhesion points. The values for a–d are Rp 5 2.31, 1.15, 0.80, and 0.58 and for e–h are Rp 5 1.15, 0.58, 0.40, and 0.23. a and e represent untreated cells, each w ith dif ferent symmetr y, determined by the number of adhesion points. In a–d, we depict a more polar cell w ith three adhesion points, whereas in e–h, a more spherical cell like those in Fig. 1 c–e is produced. As the concentration of LatA F increases, the radius of cur vature Rp that defines the arcs of cell membrane that hang between the adhesion points decreases (b and f ). Eventually, when Rp is too small for the membrane to bridge the distance between adhesion points, the cell develops cylindrical protrusions (c and g). When the actin cortex rigidity of the collapsing cell is decreased below a critical value, these grow ing protrusions become unstable w ith respect to the pearling instability (d and h). Perspective and the cell height were added for illustration purposes, along w ith pearls on one cylinder in the final stages (d and h). At higher concentrations, the low sur face-to-volume ratio favors a change of sign in the cur vature of the outer envelope, the arborization process is complete, and we obtain a ‘‘sun’’ (Fig. 4 d and h). This shape has a circular (min imal sur face) core contain ing the nucleus along w ith most of the cell cy toplasm and rays attached to the original adhesion points. Summar y In conclusion, we have shown that the phenomenon of pearling can be quantitatively explained by a simple theor y that takes into account the elastic energetics of shape. The energ y includes the rigidity of the actin cy toskeleton, which is dis- rupted by the drug LatA and which competes w ith a tension produced by external constraints such as adhesion points. The obser vation of pearling in other systems such as stretched neuronal axons (33–37) can be attributed to the same physical mechan isms. Similarly, because the termin i of peripheral axons are ver y thin, they of ten demonstrate a pearling mor- pholog y (terminal varicosities) (38). The appearance of pearl- ing in untreated cells and the behav ior of the wavenumber at low values of concentration F may indicate that the untreated cell regulates its rigidity marginally, just suf ficiently to coun- teract the instability in ex isting tubular protrusions. The theor y generally can account for cell shape in all of those regions that don’t adhere, both in untreated cells and in the various shapes that appear as the cy toskeleton is disrupted. Irrespective of other factors that are involved in determin ing cell shape, the elastic energ y considerations that we have raised must always be taken into account when discussing cell morphogenesis. S.A.S. acknowledges a grant to the Research Center for Self Assembly from the Israel Science Foundation. E.M. acknowledges support from the US-Israel Binational Foundation Grant 94-00190. A.B. acknowledges the support of La Fondation Rapael et Regina Levy. S.A.S., E.M., and A.B. acknowledge support from the Miner va Foundation, Mun ich, Germany. 1. Janmey, P. A. (1991) Cur r. Opin. Cell Biol . 3, 4 –11. 2. Wang, N., Butler, J. P. & Ingber, D. E. (1993) Science 260, 1124 –1127. 3. Chicurel, M. E., Chen, C. S. & Ingber, D. E. (1998) Cur r. Opin. Cell Biol . 10, 232–239. 4. Sackmann, E. (1994) FEBS Lett . 346, 3–16. 5. Coue, M., Brenner, S. L., Spector, I. & Korn, E. D. (1987) FEBS Lett . 213, 316 –318. 6. Ayscough, K. R., Str yker, J., Pokala, N., Sanders, M., Crews, P. & Drubin, D. G. (1997) J. Cell Biol . 137, 399 – 416. 7. Ayscough, K. R. (1998) Methods Enz ymol . 298, 18 –25. 8. Lord Rayleigh (1892) Philos. Mag. 34, 145–154. 9. Bar-Ziv, R. & Moses, E. (1994) Phys. Rev. Lett . 73, 1392–1395. 10. Hartw ig, J. H. & Shevlin, P. (1986) J. Cell Biol . 103, 1007–1020. 11. Keller, H. & Eggli, P. (1998) Cell Motil . Cytoskeleton 40, 342–353. 12. Aaronson, S. A. & Todaro, G. J. (1968) J. Cell Physiol . 42, 141–148. 13. A lbrecht-Buehler, G. (1987) Cell Motil . Cytoskeleton 7, 54 – 67. 14. Zand, M. S. & A lbrecht-Buehler, G. (1992) Cell Motil . Cytoskel- eton 13, 195–211. 15. Zand, M. S. & A lbrecht-Buehler, G. (1992) Cell Motil . Cytoskel- eton 21, 15–24. 16. Goddette, D. W. & Frieden, C. (1986) J. Biol . Chem. 261, 15974 –15980. 17. Bershadsky, A. D., Gluck, U., Den isenko, O. N., Sklyarova, T. V., Spector, I. & Ben Ze’ev, A. (1995) J. Cell Sci . 108, 1183–1193. 18. Morris, A. & Tannenbaum, J. (1980) Nature (London) 287 637– 639. 19. Sheetz, M. P. & Dai, J. (1996) Trends Cell . Biol . 6, 85– 89. 20. Sheetz, M. P. & Dai, J. (1995) Biophys. J. 68, 988 –996. 21. Evans, E. & Yeung, A. (1989) Biophys. J. 56, 151–160. 22. A lstrfm P., Eguı´luz V. M., Colding-Jfrgensen M., Gustafsson, F. & Holstein-Rathlou, N.-H. (1999) Phys. Rev. Lett . 82, 1995–1998. 23. Cramer, L. P. & Mitchison, T. J. (1995) J. Cell Biol . 131, 179 –189. 24. Thoumine, O. & Ott, A. (1997) J. Cell Sci . 110, 2109 –2116. 25. Xu, J., Schwartz, W. H., Kas, J. A., Stossel, T. P, Janmey, P. A & Pollard, T. D. (1998) Biophys. J. 74, 2731–2740. Isambert, H. & Maggs, A. (1996) Macromolecules 29, 1036 –1040. 26. 27. F. Gittes, F., Schnurr, B., Olmsted, P. D., MacK intosh, F. C. & Schmidt, C. F. (1997) Phys. Rev. Lett . 79, 3286 –3289. 28. Helfrich, W. (1973) Z. Naturforsch. C 28, 693–703. 29. Landau, L. D. & Lifshitz, E. M. (1981) Theor y of Elasticit y (Pergamon, Oxford), Ch. 2. 30. Love, A. E. H. (1944) A Treatise on the Mathematical Theor y of Elasticit y (Dover, New York), 4th Ed., Ch. 11. 31. Purser, F. (1902) R. Ir ish Acad . Trans. 32, 31– 60. 32. Bar-Ziv, R., Moses, E. & Nelson, P. (1998) Biophys. J. 75, 294 –320. 33. Young, J. Z. (1944) Nature (London) 153, 333–335. 34. Young, J. Z. (1944) Nature (London) 154, 521–522;. 35. Ochs, S., Jersild, R. A., Pourmand, R. & Potter, C. G. (1994) Neuroscience 61, 361–372. 36. Ochs, S., Pourmand, R. & Jersild, R. A. (1996) Neuroscience 70, 1081–1096. 37. Pourmand, R., Ochs, S. & Jersild, R. A. (1994) Neuroscience 61, 373–380. 38. Hubbard, J. I. (1974) The Per ipheral Ner vous System (Plenum, New York).
1507.03621	2	1507	2015-11-13T21:56:08	Reactive Boundary Conditions as Limits of Interaction Potentials for Brownian and Langevin Dynamics	[ "physics.bio-ph" ]	A popular approach to modeling bimolecular reactions between diffusing molecules is through the use of reactive boundary conditions. One common model is the Smoluchowski partial absorption condition, which uses a Robin boundary condition in the separation coordinate between two possible reactants. This boundary condition can be interpreted as an idealization of a reactive interaction potential model, in which a potential barrier must be surmounted before reactions can occur. In this work we show how the reactive boundary condition arises as the limit of an interaction potential encoding a steep barrier within a shrinking region in the particle separation, where molecules react instantly upon reaching the peak of the barrier. The limiting boundary condition is derived by the method of matched asymptotic expansions, and shown to depend critically on the relative rate of increase of the barrier height as the width of the potential is decreased. Limiting boundary conditions for the same interaction potential in both the overdamped Fokker-Planck equation (Brownian Dynamics), and the Kramers equation (Langevin Dynamics) are investigated. It is shown that different scalings are required in the two models to recover reactive boundary conditions that are consistent in the high friction limit where the Kramers equation solution converges to the solution of the Fokker-Planck equation.	physics.bio-ph	physics	REACTIVE BOUNDARY CONDITIONS AS LIMITS OF INTERACTION POTENTIALS FOR BROWNIAN AND LANGEVIN DYNAMICS∗ S. JONATHAN CHAPMAN†, RADEK ERBAN‡, AND SAMUEL A. ISAACSON§ Abstract. A popular approach to modeling bimolecular reactions between diffusing molecules is through the use of reactive boundary conditions. One common model is the Smoluchowski partial adsorption condition, which uses a Robin boundary condition in the separation coordinate between two possible reactants. This boundary condition can be interpreted as an idealization of a reactive interaction potential model, in which a potential barrier must be surmounted before reactions can occur. In this work we show how the reactive boundary condition arises as the limit of an interac- tion potential encoding a steep barrier within a shrinking region in the particle separation, where molecules react instantly upon reaching the peak of the barrier. The limiting boundary condition is derived by the method of matched asymptotic expansions, and shown to depend critically on the relative rate of increase of the barrier height as the width of the potential is decreased. Limiting boundary conditions for the same interaction potential in both the overdamped Fokker-Planck equa- tion (Brownian dynamics), and the Kramers equation (Langevin dynamics) are investigated. It is shown that different scalings are required in the two models to recover reactive boundary conditions that are consistent in the high friction limit (where the Kramers equation solution converges to the solution of the Fokker-Planck equation). Key words. reactive boundary conditions, Brownian dynamics, Langevin dynamics AMS subject classifications. 1. Introduction. Let X(t) ∈ Ω ⊂ Rd and V (t) ∈ Rd denote the stochastic processes for position and velocity at time t of a molecule moving in the d-dimensional domain Ω (where d ∈ N) according to the Langevin dynamics (LD): dX(t) = V (t) dt, dV (t) = −(cid:0)β V (t) + D β ∇ϕ(X(t))(cid:1) dt + β √ 2D dW (t), (1.1) where β is the friction constant, D is the diffusion constant, ϕ : Ω → R is the potential (to be specified later) and W (t) denotes a d-dimensional Brownian motion. Here β is assumed to have units of 'per time', D units of 'distance squared per time', and ϕ is assumed to be non-dimensional. In physical units, the potential energy of a molecule at x is then kBT ϕ(x), where kB is the Boltzmann constant and T is the absolute temperature. If m denotes the mass of the molecule, the Einstein relation gives that mDβ = kBT , so that excepting the noise term, (1.1) follows from Newton's second law of motion. Passing to the overdamped limit β → ∞ in (1.1), we obtain the Brownian dy- Oxford, OX2 6GG, United Kingdom, e-mail: [email protected]. ∗The research leading to these results has received funding from the European Research Council under the European Community's Seventh Framework Programme (FP7/2007-2013)/ ERC grant agreement n◦ 239870, and by National Science Foundation award DMS-1255408. †Mathematical Institute, University of Oxford, Radcliffe Observatory Quarter, Woodstock Road, ‡Mathematical Institute, University of Oxford, Radcliffe Observatory Quarter, Woodstock Road, Oxford, OX2 6GG, United Kingdom, e-mail: [email protected]. Radek Erban would like to thank the Royal Society for a University Research Fellowship and the Leverhulme Trust for a Philip Leverhulme Prize. This prize money was used to support research visits of Samuel Isaacson in Oxford. §Department of Mathematics and Statistics, Boston University, 111 Cummington Mall, Boston, MA 02215, USA; e-mail: [email protected]. 1 2 S. J. CHAPMAN, R. ERBAN, S. A. ISAACSON namics (BD) model for X(t) as follows: dX(t) = D ∇ϕ(X(t)) dt + √ 2D dW (t). In the special case that ϕ ≡ 0 the molecule simply moves by Brownian motion, √ dX(t) = 2D dW (t). (1.2) (1.3) Equation (1.3) is a popular description for the movement of molecules within cells. It has been used to model spatial transport in a number of computational pack- ages for simulating intracellular processes, including Smoldyn [2,3,30], GFRD [34,35] and FPKMC [27, 28]. A common approach for then coupling molecular interactions (diffusion-limited reactions) to (1.3) in these packages is to postulate that reactions can occur by one of several possible mechanisms if the corresponding reactants are sufficiently close [1, 16]. The LD model (1.1) with ϕ ≡ 0 provides a more microscopic description of diffu- sion than the BD model (1.3). It computes both position and velocity of a molecule by assuming that the molecule is subject to a normally distributed random force during each time increment. In particular, LD can be considered as an intermediate descrip- tion between detailed molecular dynamics (MD) simulations and BD simulators [12]. Typical full-atom MD simulations use time steps of the order of 1 fs = 10−15 s [24], while Smoldyn discretizes the equation (1.3) with times steps ranging from nanosec- onds to milliseconds, depending on a particular application [30]. Stochastic descrip- tions which compute both position and velocity of diffusing particles, including LD, are applicable on intermediate time scales [12, 13]. One advantage of Smoldyn or similar BD packages is that they can simulate whole- cell dynamics. BD models based on equation (1.3) have been applied to a number of biological systems including signal transduction in E. coli [26], actin dynamics in filopodia [18], the MAPK pathway [34] and intracellular calcium dynamics [10]. In these applications, the positions of all diffusing molecules are updated according to (1.3) and the distances between each pair of possible reactants (for bimolecular reactions) are calculated. Each reaction then occurs (with a given probability) when the computed distance is smaller than a specified reaction radius (as in Smoldyn), or alternatively occurs when the computed distance exactly equals a specified reaction- radius (as in GFRD and FPKMC). To our knowledge, there is no established spatio- temporal simulator of intracellular processes based on the LD model. In order to develop one, one has to first investigate how bimolecular reactions might be described in the LD context. One possible way to implement bimolecular reactions in LD is to adopt the same approach as in BD. That is, the positions and velocities of a diffusing molecule would evolve according to the LD model (1.1) (with ϕ ≡ 0) and each bimolecular reac- tion would occur (with a given probability) if the distance between two reactants is smaller than the reaction radius. However, as normally formulated this description of bimolecular interactions would not make use of a molecule's velocity (as is available in LD). That is, the LD bimolecular reaction model would not provide any more physical detail than a BD model. In this work we step back from the normal BD bimolecular reaction model to the more microscopic reaction mechanism of a molecular interaction potential. The general potential forms we consider represent an irreversible bimolecular reaction as a molecule surmounting a steep potential barrier in the separation coordinate from a stationary target molecule, after which it enters an infinitely deep well (the "bound" REACTIVE BOUNDARY CONDITIONS AS LIMITS OF INTERACTION POTENTIALS 3 state). We show that the popular Smoluchowski partial-adsorption BD reaction model [22] can be derived in the simultaneous limit that the width of the barrier ap- proaches zero and the height of the barrier becomes infinite. Using the same potential model, we then examine what limiting reactive mechanism arises in the corresponding LD model, obtaining a specular reflection boundary condition. We conclude by show- ing that in the high friction limit, β → ∞, the LD model with the specular reflection bimolecular reaction model converges back to the Smoluchowski partial-adsorption BD reaction model, consistent with the kinetic boundary layer studies of [7, 8, 23]. Our results demonstrate how an interaction potential model for a bimolecular reac- tion can be parametrized in either LD or BD models to be consistent with a BD model based on a partial-adsorption reaction mechanism. 2. Problem Setup. We consider the movement of a molecule that can undergo an irreversible bimolecular reaction with a second stationary molecule, hereafter called the reactive target, which is modelled as a sphere of radius rb. Let Br(0) ⊂ Rd denote the d-dimensional ball of radius r centered at the origin. Then the reactive target is given as ∂Brb (0). We assume that the diffusing molecule moves within the d- dimensional domain Ω which satisfies BL1 (0) \ Brb(0) for rb < L1 < ∞. (2.1) (cid:16) (cid:17) ⊂ Ω, Equation (2.1) defines Ω as the domain which is the exterior to the reactive target (sphere ∂Brb(0)) and which includes all points which have a distance less than L1 − rb > 0 from the reactive target. A simple example of Ω satisfying (2.1), for which we can find some explicit solutions in Section 3.1, is given as Ω = Rd \ Brb (0). (2.2) This is a standard ansatz for deriving formulae for the reaction radius in BD descrip- tions, e.g. in the Smoluchowski or Doi models [1,16,32]. In most cases, we will further assume that Ω is bounded with smooth boundary, i.e. equation (2.1) is altered by a bound from above (cid:16) BL1(0) \ Brb (0) (cid:17) ⊂ Ω ⊂ BL2(0), for rb < L1 < L2 < ∞. (2.3) A spherically symmetric example of domain Ω satisfying condition (2.3) is given as Ω = BL(0) \ Brb (0), for rb < L < ∞. (2.4) If dimension d = 1, then by the "surface of the ball" we mean the origin, i.e. ∂Brb(0) = {0} and rb = 0. Then equation (2.4) reduces to the finite interval Ω = [0, L], which we use in our numerical examples. We assume that the molecule is adsorbed instantly upon reaching the surface of the reactive target, so that trajectory X(t) is terminated, if X(t) ∈ ∂Brb(0). (2.5) In order to reach the surface, the molecule has to overcome a potential barrier, which, denoting r = x, is given by (cid:18) r − rb (cid:19) ε ϕ ψ 0, ϕ(x) ≡ ϕ(r) = , for rb ≤ r ≤ rb + ε, for r > rb + ε, (2.6) S. J. CHAPMAN, R. ERBAN, S. A. ISAACSON 4 where ε > 0 and ϕ > 0 are positive constants, and ψ : [0, 1] → [0, 1] is a smooth function with a (unique) global maximum at ψ(0) = 1, and ψ(1) = 0. These imply ψ(z) < ψ(0), 0 < z ≤ 1, with dψ dz (0) < 0. We assume that ε in (2.6) is chosen sufficiently small so that ε (cid:28) (L1 − rb) for rb and L1 given in assumption (2.1). Instead of studying the Langevin equation (1.1) we shall work with the corre- sponding equation for the probability density that X(t) = x and V (t) = v, denoted p(x, v, t), which satisfies the Kramers equation (also called the phase-space Fokker- Planck equation) ∂p ∂t + v · ∇xp = β ∇v · [v p + D p∇xϕ + β D ∇vp] , (2.7) where ∇x (resp. ∇v) denotes the gradient in x (resp. v) variable. Considering the overdamped limit (1.2), the corresponding equation for the probability distribution p(x, t) is given as the Fokker-Planck equation for x ∈ Ω, v ∈ Rd, = D∇x ·(cid:2)∇xp + p∇xϕ(cid:3), ∂p ∂t for x ∈ Ω. (2.8) In this paper, we show that equations (2.7) and (2.8) with fully adsorbing boundary condition (2.5) on ∂Brb(0) are in suitable limits equivalent to a diffusion process with partially adsorbing (reactive, Robin) boundary condition on ∂Brb (0), namely to the problem with ∂p ∂t = D ∆xp, for x ∈ Ω, D ∇xp(x) · x rb = K p(x), for x ∈ ∂Brb (0), (2.9) (2.10) where K is a suitable Robin boundary constant (reactivity of the boundary) and x/rb is a unit normal vector to sphere ∂Brb(0) at point x. Considering spherically symmetric Ω given by (2.2) or (2.4), and a spherically symmetric initial condition, we can rewrite the Fokker-Planck equation (2.8) as (cid:18) (cid:20) ∂p ∂r ∂p ∂t = D rd−1 ∂ ∂r rd−1 + p dϕ dr (cid:21)(cid:19) , (2.11) (cid:18) (cid:19) where p(r, t) is the radial distribution function. Then the limiting Robin boundary problem (2.9) -- (2.10) is given as ∂p ∂t = D rd−1 ∂ ∂r rd−1 ∂p ∂r , with D ∂p ∂r (rb) = K p(rb). (2.12) The Robin boundary constant K in equations (2.10) and (2.12) is (together with D) an experimentally determinable (macroscopic) parameter. In this way, we are able to parametrize both BD and LD models using experimental data. REACTIVE BOUNDARY CONDITIONS AS LIMITS OF INTERACTION POTENTIALS 5 The rest of this paper is organized as follows. In Section 3, we investigate the BD description (1.2). To get some insights into this problem, we first consider the specific case of a linear interaction potential and spherically symmetric domain (2.2) for d = 3. In this case, we can explicitly solve the corresponding Fokker-Planck equation as shown in Section 3.1, and prove the convergence of this solution, as ε → 0 and ϕ → ∞, to the solution of a model involving a reactive Robin boundary condition. We then continue in Section 3.2 with an asymptotic analysis, in the same dual limit, of the full BD model (1.2) in general domain (2.3). In Section 4, we investigate the LD model (1.1) and derive a boundary condition in the limit ε → 0. This boundary condition is then used in Section 5 to connect the LD model to a BD model with Robin boundary condition in the dual limits of high friction, β → ∞, and large potential barrier, ϕ → ∞. Numerical examples supporting our analysis are provided in Sections 4.1 and 5.1. We conclude with discussion in Section 6. 3. Brownian Dynamics. In this section we consider the overdamped problem in which the particle moves by BD, i.e. its position X(t) evolves according to equa- tion (1.2) with the interaction potential given by (2.6). 3.1. Simple three-dimensional example with explicit solution. Before investigating the general problem in Rd, as a warm-up we first consider a simplified special case to gain insight into how to recover the Robin boundary condition (2.12) from an interaction potential. We consider the spherically symmetric domain (2.2) for d = 3. The steady-state spherically-symmetric Fokker-Planck equation (2.11) is then given by (cid:18) (cid:20) ∂p ∂r (cid:21)(cid:19) 0 = D r2 ∂ ∂r r2 + p dϕ dr , for rb < r < ∞, (3.1) where the fully adsorbing boundary condition (2.5) implies a Dirichlet boundary con- dition at r = rb, i.e. p(rb) = 0. We consider constant concentration p∞ > 0 far away from the reactive surface, i.e.1 r→∞ p(r) = p∞. lim (3.2) We also assume that the potential ϕ(r) is given by (2.6) where ψ : [0, 1] → [0, 1] is the linear function ψ(z) = 1 − z. (3.3) Then (r) ≡ dϕ dr − ϕ 0, ε , rb < r < rb + ε, rb + ε < r. Substituting into (3.1), and making use of the boundary conditions at r = rb and r = ∞, we can solve piecewise to obtain a solution on (rb, rb + ε) and a solution on 1Of course, p as defined here is not a probability distribution, since it does not integrate to unity on the infinite domain r > rb. The problem as stated arises from a limiting process in which p is re-scaled appropriately as the domain size tends to infinity. S. J. CHAPMAN, R. ERBAN, S. A. ISAACSON 6 (rb + ε,∞). On each interval the solution is defined up to an unknown constant. By enforcing continuity of the flux at r = rb we can eliminate one constant to obtain (cid:20) ϕ(r − s) (cid:21) ε 1 s2 exp (cid:90) r A rb p∞ − A r p(r) = ds, for rb ≤ r ≤ rb + ε, (3.4) , for r > rb + ε, where the last constant, A, is determined by the continuity of p(r, t) at r = rb + ε. This gives with A = (rb + ε) p∞ 1 + (rb + ε)β(ε, ϕ) , (cid:90) 1 0 exp[ ϕ(1 − s) ] (rb + εs)2 ds. β(ε, ϕ) = ε To recover the Robin condition (2.12) as ε → 0 and ϕ → ∞, we need (cid:20) (cid:21) lim ε→0 ϕ→∞ lim r→(rb+ε)+ D (r) − Kp(r) = 0, ∂p ∂r which is equivalent to or simplifying, (rb + ε)(cid:0)1 + (rb + ε)β(ε, ϕ)(cid:1) = K, D + K(rb + ε) lim ε→0 ϕ→∞ β(ε, ϕ) = lim ε→0 ϕ→∞ D Kr2 b . (3.5) (3.6) We note that (cid:90) 1 0 0 ≤ β(ε, ϕ) ≤ ε r2 b exp[ ϕ(1 − s) ] ds = ε (exp[ ϕ ] − 1) r2 b ϕ . (3.7) Therefore, β(ε, ϕ) will have a finite limit if ε exp[ ϕ ]/ϕ has a finite limit. This moti- vates the choice D ϕ exp[−ϕ ] . ε = Using (3.5), we have (cid:12)(cid:12)(cid:12)(cid:12)β(ε, ϕ) − ε r2 b (cid:90) 1 0 exp[ ϕ(1 − s) ] ds K (cid:90) 1 0 (cid:12)(cid:12)(cid:12)(cid:12)ε (cid:12)(cid:12)(cid:12)(cid:12) = (cid:18) exp[ ϕ(1 − s) ] 1 (rb + εs)2 − 1 r2 b ≤ D ε (2rb + ε) K r4 b , (3.8) (cid:19) (cid:12)(cid:12)(cid:12)(cid:12) ds (3.9) REACTIVE BOUNDARY CONDITIONS AS LIMITS OF INTERACTION POTENTIALS 7 which converges to zero as ε → 0. We therefore conclude that β(ε, ϕ) = lim ε→0, ϕ→∞, where ε and ϕ are related by (3.8) lim ε→0, ϕ→∞, where ε and ϕ are related by (3.8) exp[ ϕ(1 − s) ] ds = lim ϕ→∞ D ϕ exp[−ϕ ] K r2 b 0 exp[ ϕ(1 − s) ] ds = D Kr2 b , so that we get (3.6). In particular, we recover the Robin boundary condition with Robin constant K. The steady-state solution to the limiting Robin boundary condition problem (2.12) with (3.2) is given by (cid:90) 1 (cid:90) 1 0 ε r2 b  , p(r) = p∞ rb r 1 D Krb 1 + for rb < r < ∞. 1 − (cid:12)(cid:12)(cid:12) ε (cid:16) (cid:18) (cid:16) (cid:16) ε (cid:17) r We now examine the error between this limit and the solution (3.4) for two cases: rb ≤ r ≤ rb + ε and r > rb + ε. For the latter we have p(r) − p(r) = p∞ 1 + D Krb − rb β(ε, ϕ) − β(ε, ϕ) r 1 + (rb + ε)β(ε, ϕ) We can estimate that the denominator is greater than r. Consequently, p(r) − p(r) < p∞ ε r 1 + D Krb + rb β(ε, ϕ) + p∞ r2 b − β(ε, ϕ) Using (3.7) and the definition of ε (equation (3.8)), we have β(ε, ϕ) ∼ O(1) as ε → 0. Using (3.9), we conclude that the second term on the right hand side is O(ε/r) as ε → 0. Thus, we conclude (cid:17) (cid:17)(cid:16) + r2 b Kr2 b 1 + D Krb (cid:16) D (cid:17) (cid:12)(cid:12)(cid:12)(cid:12) D Kr2 b r (cid:19) (cid:17)(cid:12)(cid:12)(cid:12) . (cid:12)(cid:12)(cid:12)(cid:12) . = O(ε), for r > rb + ε. (3.10) p(r) − p(r) = O On the other hand, we have sup r∈(rb,rb+ε) p(r) − p(r) ≥ p(rb) − p(rb) = p(rb) = p∞D Krb + D > 0. The maximum error between the two models is therefore O(1) as ε → 0 (where ϕ satisfies (3.8)), illustrating the non-uniformity of the error within the region where the potential interaction is non-zero. Note, while the maximum norm (i.e. L∞ norm) of the difference between solutions does not converge on (rb, rb + ε), both p(r) and p(r) are uniformly bounded on (rb, rb + ε) since p(r) ≤ A β(ε, ϕ) ≤ 2 D p∞ K rb p(r) ≤ p∞. , sup r∈(rb,rb+ε) sup r∈(rb,rb+ε) S. J. CHAPMAN, R. ERBAN, S. A. ISAACSON 8 As such, using that the maximum norm error is O(ε/r) on (rb + ε,∞) by (3.10), we find that the q-norm converges for any 3 < q < ∞, (cid:18)(cid:90) ∞ (cid:19) 1 q (cid:16) (cid:17) (cid:107)p(r) − p(r)(cid:107)q = p(r) − p(r)q r2 dr = O 1 q ε . rb Here, the lower bound q > 3 is an artifact of our working on an unbounded domain, which requires integrability of p(r) − p(r)q r2 on (rb,∞). If our domain was bounded, i.e. given by equation (2.4), and we used the Dirichlet boundary condition that p(L) = p∞, we would expect a similar estimate to hold for all 1 ≤ q < ∞. In summary, we have found that in the special case of a linear potential barrier, by taking the height of the barrier ϕ → ∞ and then choosing the width of the barrier, ε, to decrease exponentially in the height (i.e. according to equation (3.8)), we can recover the solution to the diffusion equation with Robin boundary condition. We may therefore interpret the Robin boundary condition model for bimolecular reactions between two molecules as approximating an underlying interaction potential, in which the two molecules must surmount a steep potential barrier before entering a bound state represented by a deep well. Remark. In [31] it was shown how the Robin constant can be chosen to give the same diffusion-limited reaction rate in two steady-state spherically-symmetric models both including the same fixed interaction potential, with one using a zero Dirichlet boundary condition at rb (as in (3.1)), and the other using a Robin condition at rb + ε for any fixed ε > 0. The argument in [31] can be modified to match diffusion-limited reaction rates in the two models considered in this section (i.e. (3.1) with p(rb) = 0, and the steady-state diffusion equation with Robin boundary condition satisfied by ¯p(r)). It is satisfying to note that choosing K to match the diffusion limited rates in this manner also gives the relationship (3.8) between K, ε and ϕ. We therefore conclude that choosing K to match the diffusion-limited reaction rates of the two models of this section is sufficient to give convergence of p(r) to ¯p(r) as ε → 0 and ϕ → ∞. 3.2. Asymptotics for a general BD model with a general interaction potential. To analyze equation (2.8) for arbitrary d ∈ N with general potential ϕ(r) in the form (2.6), we follow the approach of Pego [29] and Li [25]. We use the method of matched asymptotic expansions in a general bounded domain Ω which satisfies (2.3) and has a smooth boundary. First, we consider an inner solution in a boundary layer near ∂Brb (0). Let r − rb ε z = , (3.11) denote the stretched distance of a point x from the reactive boundary. We define an inner solution p(z, x, t). The second argument should actually be x = x/x, but we follow the method of [29] and instead assume that for all λ > 0 (i.e. the length of x is accounted for by z). We note the identities that p(z, λx, t) = p(z, x, t), ∇x · x = ∇xz = d − 1 r = d − 1 rb + εz , x ε . (3.12) REACTIVE BOUNDARY CONDITIONS AS LIMITS OF INTERACTION POTENTIALS 9 In the new coordinate system ∇xp → ∇x p + ∆xp → ∆x p + 1 ε 1 ε ∂ p x, ∂z d − 1 rb + εz where we have used (3.12) and that x · ∇y ∂ p ∂z (z, y, t) ∂ p ∂z (cid:12)(cid:12)(cid:12)y=x + 1 ε2 ∂2 p ∂z2 , = 0, as p is assumed constant when the second argument is varied along the x direction. From (2.8) we find the inner problem (for 0 < z < 1) ∂ p ∂t = D ∆x p + d − 1 rb + εz 1 ε ∂ p ∂z + 1 ε2 ∂2 p ∂z2 + ϕ ε2 ∂ p ∂z dψ dz + d − 1 rb + εz ϕ ε dψ dz p + ϕ ε2 p d2ψ dz2 (cid:21) . (cid:20) We now expand the inner and outer solutions in ε and denote the leading-order terms by p0 and p0 respectively. The leading-order behavior of the inner solution then satisfies ∂2 p0 ∂z2 + ϕ ∂ p0 ∂z dψ dz + ϕ p0 d2ψ dz2 = 0. Note that we will find below that in the distinguished limit ϕ depends on ε in a logarithmic manner (see equation (3.13)) so that retaining ϕ here is consistent with neglecting terms of O(ε). Solving this equation and using the Dirichlet boundary condition at z = 0, we find (cid:90) z exp(cid:2) ϕ (ψ(z (cid:48)) − ψ(z))(cid:3) dz (cid:48) , p0(z, x, t) = A(x, t) where A(x, t) is an unknown constant, also satisfying the dilation property 0 A(λx, t) = A(x, t), for λ > 0. The leading order term of the outer solution, p0, satisfies the diffusion equation (2.9) away from the reactive boundary. Our matching conditions are lim x→∂Brb (0) p0(x, t) = lim z→1 p0(z, x, t) = A(x, t) lim x→∂Brb (0) ∇p0(x, t) · x = lim x→∂Brb (0) ∂p0 ∂r = lim z→1 1 ε ∂z Combining these two equations, for x ∈ ∂Brb(0) we find D ∇xp(x) · x rb 0 exp(cid:2) ϕ ψ(z(cid:48))(cid:3) dz(cid:48) ∼ D ϕ ε(cid:82) 1 D p0(x, t) ∂p0 ∂r (x, t) = = D ε exp(ϕ) (cid:90) 1 exp(cid:2) ϕ ψ(z (cid:20) ∂ p0 0 + ϕ p0 dψ dz (cid:48))(cid:3) dz (cid:21) (cid:12)(cid:12)(cid:12)(cid:12)dψ dz = (0) (cid:48) , A(x, t) . ε (cid:12)(cid:12)(cid:12)(cid:12) p0(x, t), where the last approximation is obtained using Laplace's method in the limit ϕ → ∞. Thus we recover the desired Robin boundary condition (2.10) if ϕ → ∞ and ε → 0 such that (cid:12)(cid:12)(cid:12)(cid:12)dψ dz (cid:12)(cid:12)(cid:12)(cid:12) . (0) ε = D ϕ K exp(ϕ) (3.13) 10 S. J. CHAPMAN, R. ERBAN, S. A. ISAACSON In summary, we have used the method of matched asymptotic expansions to ex- amine scaling limits for a general set of potential interactions between two particles. The potentials were assumed to be short-range, forming a high barrier as the separa- tion between the molecules approaches a fixed "reaction-radius", and then a deep well once this barrier is surmounted. In the limit that the height of the barrier, ϕ → ∞, and the width of the barrier, ε, decreases to zero exponentially in the height, we re- cover the solution to the diffusion equation with Robin boundary condition. We may therefore interpret bimolecular reactions between two molecules modeled with a Robin boundary condition as an approximation to one of many possible underlying poten- tial interactions. These interactions are characterized by the two molecules needing to surmount a steep potential barrier before entering a bound state represented by a deep well. Remark. If ε is chosen to approach zero slower than (3.13), then we recover a zero Neumann boundary condition at the reactive surface, ∂p0 ∂r (x, t) = 0, for x ∈ ∂Brb(0). (3.14) Likewise, if ε is chosen to approach zero faster than (3.13), then we recover the zero Dirichlet boundary condition that p(x, t) = 0, for x ∈ ∂Brb(0). 4. Langevin Dynamics. We now consider the LD model (1.1) in a bounded d-dimensional domain satisfying (2.3). The reactive target is again taken to be the surface of the d-dimensional sphere of radius rb about the origin. It is again assumed that the molecule is adsorbed instantly upon reaching the surface of the sphere, i.e. we consider the boundary condition (2.5) together with spherically symmetric interaction potential ϕ given by (2.6). We leave unspecified any boundary condition on ∂Ω \ ∂Brb(0), as it is not needed in the following analysis. Instead of studying the Langevin equation (1.1) we work with the corresponding Kramers equation (2.7). Since x/rb is the normal to ∂Brb(0) at x, the adsorbing boundary condition (2.5) means that the Kramers equation (2.7) is coupled with the Dirichlet boundary condition p(x, v, t) = 0 for x ∈ ∂Brb(0) and v · x > 0, (4.1) and the unspecified boundary condition on ∂Ω\ ∂Brb(0). We are interested in various limits of (2.7) as ε → 0, ϕ → ∞ and β → ∞ in which the interaction potential can be approximated by an appropriate reactive boundary condition. In the BD models of Section 3 our goal was to derive the widely-used Robin boundary condition. To our knowledge, in the LD model we now investigate there is no standard reactive boundary condition for bimolecular reactions. Therefore, we wish to see what, if any, reactive boundary condition arises when considering similar limits of ε and ϕ. As in Section 3.2, to study these limits we will match an inner solution within an O(ε) boundary layer about ∂Brb (0) to an outer solution when far from the reactive boundary. Using the same notation as in Section 3.2, we denote by z the stretched distance from the boundary, given by (3.11). We also introduce a re-scaled radial velocity, vr vr = v · x vr , ν = REACTIVE BOUNDARY CONDITIONS AS LIMITS OF INTERACTION POTENTIALS 11 where x = x/x is a unit vector in the x direction, vr = v · x is the radial velocity into Ω from ∂Brb(0), and vr is a scaling constant in the radial velocity that will be specified later. In the inner region we denote the density by p(z, x, ν, v, t), where we assume that p is constant whenever x is varied in the radial direction, and when the component of the velocity, v, in the radial direction is varied. That is p(z, x, ν, v, t) = p(z, λx, ν, v, t), p(z, x, ν, v, t) = p(z, x, ν, v + α x, t), for λ > 0, for α ∈ R. In addition to the identities (3.12), we note that ∇xν = v vr(rb + εz) − ν x rb + εz . (4.2) Using (3.12) and (4.2) we find the derivative operators in (2.7) transform in the new coordinates to (cid:18) v vr(rb + εz) − ν x rb + εz (cid:19) ∂ p , ∂ν ∇xp → ∇x p + ∇vp → ∇v p + ∆vp → ∆v p + x + 1 ε 1 vr 1 v2 r ∂ p ∂z ∂ p ∂ν ∂2 p ∂ν2 , x, where we have used that x · ∇v(cid:48) (z, x, ν, v(cid:48) , t) ∂ p ∂ν Using (2.6), equation (2.7) can be transformed to (cid:18) + v · ∇x p + ∂ p ∂t vr ν ε β v · v (cid:32) + ∂ p vr(rb + εz) ∂z dp + v · ∇v p + ν = 0. (cid:12)(cid:12)(cid:12)v(cid:48)=v (cid:19) ∂ p − vr ν2 rb + εz = ∂ν ∂ p ∂ν + ϕ D vr ε dψ dz ∂ p ∂ν + βD∆v p + βD v2 r ∂2 p ∂ν2 (cid:33) . We consider an asymptotic expansion of the inner solution, p (in the boundary layer about ∂Brb(0)) as ε → 0 with all other parameters held fixed, p ∼ p(0) + p(1)ε + p(2)ε2 + . . . . Similarly we also consider an expansion of the outer solution, valid far from ∂Brb(0), for which we will abuse notation and also denote it by p, p ∼ p(0) + p(1)ε + p(2)ε2 + . . . . At leading order O(ε−1) we find that the inner solution satisfies vr ν ∂ p(0) ∂z To simplify this equation, we let dψ dz ∂ p(0) ∂ν = 0. − ϕ β D vr vr =(cid:112) ϕ β D. (4.3) (4.4) 12 (a) S. J. CHAPMAN, R. ERBAN, S. A. ISAACSON (b) Fig. 4.1: (a) Characteristic curves of (4.5) for the linear potential (3.3) are shown as blue lines. The curve z = ν2/2 is drawn as a thicker black line, dividing the upper half plane into a region where the moving particle will escape over the barrier (ν < 0, with z < ν2/2), a region where all trajectories are reflected (z > ν2/2), and a region where p(0) = 0 with all trajectories originating on the z = 0 axis (ν > 0 and z < ν2/2) because of boundary condition (4.6). (b) Probability of escaping through the left boundary as a function of the incoming velocity for ε = 0.03 (blue line), ε = 10−2 (green line) and ε = 10−3 (red line). We compute the trajectories using (4.9) -- (4.10) for ∆t = 10−7, D = 1, L = 1, β = 103 and ϕ = 2.593. The vertical dashed line at −√ 2 ϕ β D separates the two cases of boundary condition (4.7). This choice emphasizes large velocities, for which we expect the particle to be most likely to escape over the (reactive) potential barrier. Substituting (4.4) into (4.3), we obtain ν ∂ p(0) ∂z − dψ dz ∂ p(0) ∂ν = 0. The boundary condition (4.1) implies p(0)(0, x, ν, v, t) = 0, for ν > 0. The solution to (4.5) is constant along the characteristic curves (4.5) (4.6) −ψ(z) = ν2 2 + C, for C an arbitrary constant. These curves are shown in Figure 4.1(a) for the linear potential (3.3). The curve 1 − ψ(z) = ν2 2 divides the half-plane into three regions. In the region where ν > 0 and 1 − ψ(z) < ν2/2, p(0) is zero due to the boundary condition (4.6). This region corresponds to particles originating at the boundary and moving into the domain. Since particles are only adsorbed at the boundary, and not emitted, we find that the solution is −3−2−1012300.20.40.60.81νz⌫z−150−100−50000.20.40.60.81velocityatx=εprobability of escape ε=0.03ε=0.01ε=0.001−√2ϕβDREACTIVE BOUNDARY CONDITIONS AS LIMITS OF INTERACTION POTENTIALS 13 zero throughout the region. Where ν < 0 and 1 − ψ(z) < ν2/2 the solution will be determined by matching to the outer solution. This region corresponds to particles entering from outside the boundary layer with sufficient velocity to escape over the potential barrier, and thereby exit through the reactive surface at z = 0. Finally, in the region 1 − ψ(z) > ν2/2, trajectories with ν < 0 are reflected in a symmetric manner. For points with ν < 0 the solution will again be given by matching to the outer solution, while for those with ν > 0 the solution is determined by reflection. This region corresponds to particles that enter from outside the boundary layer with insufficient velocity to escape over the potential barrier. These particles are then reflected, moving back out of the boundary layer. Note, in the original variables the curve separating these three regions is given by (cid:18) r − rb (cid:19) ε 1 − ψ (v · x)2 2 ϕ β D = v2 r = . 2 ϕ β D For x ∈ ∂Brb (0), we match the inner solution as z approaches the edge of the bound- ary layer to the outer solution as the particle's spatial position, y, approaches x. That is, we require lim z→1 p(0)(z, x, ν, v, t) = lim y→x p(0)(y, v, t). Using that ψ(1) = 0, this matching condition implies the outer solution satisfies the following reactive boundary condition for vr ≥ 0 and x ∈ ∂Brb(0) p(0)(x, v, t) = p(0)(x, v − 2 vr x, t), 0 ≤ vr < √ 0, vr > 2 ϕ β D. 2 ϕ β D, √ (4.7) (cid:40) √ When the moving particle reaches the reactive boundary with radial velocity in the outward direction greater than 2 ϕ β D it leaves the domain (i.e. undergoes reac- tion). In contrast, when the particle reaches the boundary with a slower radial velocity in the outward direction it is reflected back into the domain along the direction of the normal at the point where it hit the reactive boundary. This "specular reflection" boundary condition is also obtained in the corresponding deterministic Newtonian me- chanics model. We demonstrate this explicitly for a simple one-dimensional example in Appendix A. A version of this boundary condition, given in terms of an arbitrary threshold velocity, is assumed in the kinetic boundary layer investigations of the one-dimensional Kramer's equation in [7, 8], and the 3D spherically-symmetric steady-state Kramer's √ equation in [23]. It is also (briefly) mentioned in [23] that the specific threshold of 2ϕβD we derive is what one might impose across an interface where the potential is discontinuous with a jump of size ϕ (again, for the 3D spherically-symmetric steady- state Kramer's equation). Our asymptotic analysis shows hows this specific threshold velocity arises in the general Kramer's equation as the limit of a shrinking potential boundary layer bordering a Dirichlet boundary condition. We obtain an effective jump in potential at the reactive boundary, as opposed to an interface within the domain. Contrast this to the limit of the BD problem from Section 3.2, where in taking ε → 0 with ϕ fixed the influence of the potential is completely lost (e.g. a zero Dirichlet boundary condition is recovered). limit that ε → 0 equivalent to solving the (zero-potential) Kramers equation for x ∈ Ω, v ∈ Rd, We therefore conclude that the LD model with the interaction potential is in the + v · ∇xp = β ∇v · [v p + β D ∇vp ] , (4.8) ∂p ∂t 14 S. J. CHAPMAN, R. ERBAN, S. A. ISAACSON with the specular reflection reactive boundary condition (4.7) on ∂Brb (0) and what- ever boundary condition was imposed on ∂Ω. That is, in the limit that the width of the potential approaches zero, with the barrier height held fixed, we find that the potential can be approximated by a velocity threshold boundary condition. Here par- ticles moving sufficiently fast relative to the height of the potential barrier undergo bimolecular reactions when reaching the reactive boundary, while those moving too slow are reflected back into the domain. This result should be applicable for general short-range potential interactions that form a high barrier as the separation between two molecules approaches a fixed "reaction-radius", and then a deep well once this barrier is surmounted. In contrast to the diffusive case, taking the barrier height ϕ → ∞ (with β fixed) leads to a complete loss of reaction; all particles reaching the reactive boundary are simply reflected back into the domain. 4.1. A numerical example showing recovery of boundary condition (4.7) as ε → 0. We consider the LD model (1.1) for d = 1 and the linear potential (3.3). In the one-dimensional case, our computational domain is interval Ω = [0, L]. We choose a small time step ∆t and compute the position X(t + ∆t) and the velocity V (t + ∆t) from the position X(t) and the velocity V (t) by √ (4.9) X(t + ∆t) = X(t) + V (t) ∆t, V (t + ∆t) = V (t) − β V (t) ∆t + D β ϕ ε χ[0,ε](X(t)) ∆t + β (4.10) where χ[0,ε] : R → {0, 1} is the characteristic function of the interval [0, ε] and ξ is a normally distributed random variable with zero mean and unit variance. We implement adsorbing boundaries at both ends x = 0 and x = L of the simulation domain [0, L] by terminating the computed trajectory whenever X(t) < 0 or X(t) > L. In this paper, we are interested in understanding the dependence of the behavior of the LD model (1.1) on its parameters ε, ϕ and β. In particular, we choose the values of other parameters equal to 1, namely 2D∆t ξ, (4.11) In this section, we are interested in the limit ε → 0. We want to illustrate the boundary condition (4.7). Therefore we fix the values of ϕ and β and simulate the LD model (1.1) for different values of ε. D = L = 1. For each value of ε, we compute many trajectories according to (4.9) -- (4.10), starting from the middle of the domain, i.e. X(0) = L/2, with initial velocity V (0) sampled from the normal distribution with zero mean and variance β D. Whenever a particle enters the region [0, ε], we record its incoming velocity. Then we follow its trajectory in the region [0, ε] and record one of two possible outputs: (1) the particle leaves Ω through its left boundary (i.e. X(t) < 0); or (2) the particle returns back to the region (ε, L] of domain Ω (i.e. X(t) > ε). The fraction of particles which left the domain Ω through its left boundary as a function of the incoming velocity is plotted in Figure 4.1(b). This curve can be given its incoming velocity. We also plot the vertical dashed line at −√ interpreted as the probability that the particle escapes over the potential barrier, 2 ϕ β D in Figure 4.1(b). This threshold separates the two cases of boundary condition (4.7). We observe that the probability of escape converges to the step function as ε → 0, i.e. we have numerically confirmed boundary condition (4.7). We will return to this example in Section 5.1 when we study the convergence of the LD model to the diffusion process with a Robin boundary condition. In particular, we REACTIVE BOUNDARY CONDITIONS AS LIMITS OF INTERACTION POTENTIALS 15 use values β = 103 and ϕ = 2.593 in Figure 4.1(b). This choice of ϕ will be explained in the following section and then used again in one of our simulations presented in Section 5.1. 5. From Langevin Dynamics to Brownian Dynamics. We now study the overdamped limit of the LD model (1.1). We wish to show that in the overdamped limit where β → ∞, taking ϕ → ∞ in a β-dependent manner will recover a Robin boundary condition for the limiting diffusion equation. To study the β → ∞ limit, we extend the asymptotic analysis of the one-dimensional Kramers equation in [14] to the d-dimensional (zero-potential) Kramers equation (4.8). The kinetic boundary layer studies in [7, 8, 23] previously investigated the relationship between the velocity threshold in the specular reflection boundary condition and effective adsorption rate, K, in the Robin condition. Our approach here differs from those studies, which primarily used numerical solutions of truncated moment equations or basis function expansions to estimate empirically determined formulas for the Robin constant, K. We focus on deriving an explicit formula relating how the potential barrier height, ϕ, should be chosen as β → ∞ to recover a specified Robin constant. We begin by re-scaling velocity as v = β η and let f (x, η, t) = p(x, v, t). Sub- √ stituting into (4.8), we obtain +(cid:112) ∂f ∂t β η · ∇xf = β ∇η · [η p + D ∇ηp] . (5.1) We expand f in powers of β−1/2 as f (x, η, t) ∼ f0(x, η, t) + Substituting into (5.1), we find 1√ β f1(x, η, t) + 1 β f2(x, η, t) + . . . . ∇η · [η f0 + D ∇ηf0] = 0, ∇η · [η f1 + D ∇ηf1] = η · ∇xf0, ∇η · [η f2 + D ∇ηf2] = η · ∇xf1 + ∂f0 ∂t . (5.2) (5.3) (5.4) Implicit in these equations is the assumption that we are interested in timescales for which t (cid:29) 1 β . We therefore interpret (5.2) as implying that the velocity distribution component of f0 relaxes to equilibrium on a faster timescale than t. Denote by τ this faster timescale. As discussed in the introduction to [9], up to a normalization constant there is a unique solution to (5.2) that corresponds to the equilibrium solution of the fast-timescale, time-dependent equation = ∇η · [η f0 + D ∇ηf0] . ∂f0 ∂τ This equilibrium solution is then f0(x, η, t) = (x, t) exp (cid:35) (cid:34) −η2 2D , 16 S. J. CHAPMAN, R. ERBAN, S. A. ISAACSON where (x, t) is independent of η. We similarly find that the general solution to (5.3) is given by (cid:35)  exp (cid:34) where ξ(x, t) is also independent of η. Substituting these into (5.4) we see that −η2 (cid:16) − ∇x(x, t) · η + ξ(x, t) (cid:17)  ∂ − d(cid:88) ∇η · [η f2 + D ∇ηf2] = (cid:34) −η2 ηiηj + η · ∇xξ(x, t) f1(x, η, t) = (cid:35) ∂2 exp 2D . , 2D ∂t ∂xi∂xj i,j=1 Integrating both sides of this equation for all η ∈ Rd, and using that f2(x, η, t) → 0 and ∇ηf2(x, η, t) → 0 as η → ∞, we conclude that (x, t) satisfies the diffusion equation ∂ ∂t = D∆x. The probability density that the particle has position x at time t and has not reacted is given by (cid:90) Rd u(x, t) = p(x, v, t) dv, implying that to leading order u is proportional to . As such, we expect as β → ∞ that u satisfies the diffusion equation. We now show that , and hence u, satisfies a Robin boundary condition as β → ∞ when ϕ is chosen to approach infinity in a β-dependent manner. To leading order in β, the outward flux through a point, x ∈ ∂Brb(0), is J(x, t) : = − Rd ∼ − (cid:112) (v · x) (v · x) p(x, v, t) dv (cid:90) (cid:20) (cid:90) = (2 π D β)d/2(cid:16) (cid:110) v(cid:12)(cid:12) v · x < −(cid:112)2 ϕ β D Rd β, t) + f0(x, v/ D ∇x(x, t) · x (cid:112) , β, t) 1√ β f1(x, v/ (cid:17) (cid:110) v(cid:12)(cid:12) vr < −(cid:112)2 ϕ β D (cid:111) = (cid:21) (cid:111) where as in previous sections x = x/x. Let dv (5.5) R( x) = (cid:90) (cid:90) R( x) R( x) (cid:20) (cid:34)(cid:114) denote the set of velocities at which particles may escape through the reactive bound- ary. The specular reflection boundary condition (4.7) implies J(x, t) = − (v · x) p(x, v, t) dv (cid:21) (cid:112) β, t) + 1√ β exp[−ϕ] (x, t) + (cid:114) (cid:112) (cid:32)(cid:114) ∼ − (v · x) f0(x, v/ f1(x, v/ β, t) dv = (2 π D β)d/2 β D 2π D 2π exp[−ϕ] ξ(x, t) √ (cid:33)(cid:16) ϕ) (cid:17)(cid:35) + exp[−ϕ] + ϕ π erfc( 2 D ∇x(x, t) · x , REACTIVE BOUNDARY CONDITIONS AS LIMITS OF INTERACTION POTENTIALS 17 for x ∈ ∂Brb(0) and x = x/rb. Comparing with (5.5), we find that to leading order , and hence u, satisfy the reactive Robin boundary condition for β → ∞ for x ∈ ∂Brb(0), D ∇x(x, t) · x =(cid:112) β K(ϕ, D) (x, t) where K(ϕ, D) = 1 − exp[−ϕ] D √ 2 π exp[−ϕ] − erfc( 2 . ϕ) (cid:114) (cid:114) (cid:32) ϕ π (cid:114) (cid:33) ϕ = ln 1 K β D 2 π , (5.6) (5.7) As ϕ → ∞, the denominator of K approaches one. This suggests the scaling so that (cid:112) β K(ϕ, D) → K, as β → ∞. We then find (x, t) satisfies the desired Robin boundary condition, for x ∈ ∂Brb(0), √ in the limit β → ∞. More generally, we could impose the equality in equation (5.6). Solving for β, we obtain the following relation between ϕ and β: D ∇x(x, t) · x = K (x, t) β K(ϕ, D) = K (cid:18)√ 2 π exp[ϕ] −(cid:112)2 ϕ − (cid:114) π 2 erfc((cid:112) β = K 2 D (cid:19)2 ϕ) exp[ϕ] . (5.8) In Figure 5.1(a), we compare both formulae (5.7) and (5.8) for K = D = 1. As expected, they are equivalent in the limit β → ∞. In the following section, we will present illustrative simulations, confirming that (5.8) provides a more accurate ap- proximation of the limiting Robin boundary condition. Further improvements to these formulas could presumably be made by incorporating a more detailed representation of the kinetic boundary layer near ∂Brb(0). 5.1. A numerical example illustrating limit β → ∞. As in Section 4.1, we consider the LD model (1.1) for d = 1 in the interval Ω = [0, L] with the linear potential (3.3) near the left-hand boundary only. We choose a small time step ∆t and compute the position X(t + ∆t) and the velocity V (t+∆t) from the position X(t) and the velocity V (t) by (4.9) -- (4.10). We implement adsorbing boundaries at both ends x = 0 and x = L of the simulation domain [0, L] by terminating the computed trajectory whenever X(t) < 0 or X(t) > L. In this section, we want to show that both equations (5.7) and (5.8) correctly recover the Robin boundary condition in the limit β → ∞. In particular, we choose the values of other parameters equal to 1, namely (compare with (4.11)) K = D = L = 1. (5.9) We vary β and we use either equation (5.7) or equation (5.8) to calculate the corre- sponding value of ϕ (these values are plotted in Figure 5.1(a)). 18 (a) S. J. CHAPMAN, R. ERBAN, S. A. ISAACSON (b) Fig. 5.1: (a) Dependence of ϕ on β computed by equation (5.7) (blue solid line) and equation (5.8) (red dashed line) for K = D = 1. (b) Splitting probability of escaping through the left boundary as a function β. In each simulation, we estimate the splitting probability as an average over 105 realizations. We use ε = 10−3, D = L = K = 1 and ϕ computed according to equation (5.7) (blue squares) or equation (5.8) (red circles). The dashed line denotes the theoretical BD result (5.10). For each value of β, we compute 105 trajectories according to (4.9) -- (4.10), starting from the middle of the domain, i.e. X(0) = L/2, with initial velocity V (0) sampled from the normal distribution with zero mean and variance β D. Each trajectory is calculated until it leaves the domain [0, L] either through the left or right boundary point. In Figure 5.1(b), we plot the probability that a trajectory leaves the domain Ω through the left boundary (the so-called splitting probability), estimated as the fraction of all trajectories which are terminated because X(t) < 0. Our goal is to illustrate that equations (5.7) and (5.8) can be used to connect the LD model with the limiting Robin boundary problem (2.9) -- (2.10). Let Π(x) be the probability that the BD particle leaves the domain Ω = [0, L] through the left boundary. Since d2Π dx2 (x) = 0, we find for x ∈ Ω, with D dΠ dx (0) = K (Π(0)−1) and Π(L) = 0, Π(x) = K (L − x) D + KL . (cid:18) L (cid:19) Π 2 Since all trajectories start from the middle of the domain, X(0) = L/2, we have = K L 2(D + KL) = 0.25%, (5.10) for the parameter values given by (5.9). This value is plotted in Figure 5.1(b) as the black dashed line. We confirm that the results estimated from simulations ap- proach (5.10) as β → ∞. We also confirm that simulations based on the higher-order approximation (5.8) converge more quickly to the limiting Robin boundary problem than the simulations based on equation (5.7). 10210310410512345βϕ equation (5.7)equation (5.8)1021031041050.150.20.250.30.350.4βsplitting probability LD simulations, equation (5.7)LD simulations, equation (5.8)BD limit (5.10)REACTIVE BOUNDARY CONDITIONS AS LIMITS OF INTERACTION POTENTIALS 19 6. Discussion. We have considered three parameters, ε (potential width), ϕ (potential height) and β (friction constant), and studied several limits of these pa- rameters which lead to the Robin (reactive) boundary condition (2.10). Parameters ε and ϕ are shared by both the BD and LD models. In Section 3, we have shown that the BD model can recover the Robin boundary condition in the limit ε → 0 and ϕ → ∞ when these parameters are related by (3.13). For the case of the linear potential (3.3) this relation can be rewritten as ϕ − ln(ϕ) = ln . (6.1) (cid:18) D (cid:19) Kε The LD model has an additional parameter β. In Section 5, we have derived two formulae (5.7) and (5.8) which relate the LD model to the BD model with the Robin boundary condition (2.10). Both results, (5.7) and (5.8), are equivalent in the limit β → ∞. Equation (5.8) is more accurate for finite values of β, while (5.7) is simpler and easier to interpret. It is given as (cid:32) (cid:114) (cid:33) ϕ = ln 1 K β D 2 π . (6.2) Considering that (experimentally determinable) parameters K and D are given con- stants, we can compare our BD result (6.1) with our LD result (6.2). They can be both used to specify the height of the potential barrier ϕ, which is given as a function of ε in (6.1) and as a function of β in (6.2). On the face of it, the LD result (6.2) does not depend on the parameter ε. How- ever, the derivation of the LD result (6.2) is only valid for small ε. More precisely, our conclusion for the LD model (1.1) can be stated as follows: β (cid:28) 1, then ϕ is independent of ε and can be written in the (1) If ε (cid:28) (cid:113) D (2) If 1 (cid:29) ε (cid:29)(cid:113) D form (6.2); β , then ϕ is independent of β and is given by (6.1). In our illustrative computations in Figure 5.1(b) we have used ε = 10−3. In this case, the BD result (6.1) implies that ϕ = 9.118. We observe that this value is higher than the values of ϕ plotted in Figure 5.1(a) which are used for our simulations in Figure 5.1(b). We can also substitute this value ϕ = 9.118 into (6.2). We obtain β = 5.224 × 108. For these values of β and ε we are in case (2), so that the LD result (6.2) would no longer be applicable. We instead recover the limiting Robin boundary condition (in the limit β → ∞ with ε = 10−3) by using the BD result (6.1). The above results (6.1) -- (6.2) can be used to connect the experimentally deter- minable parameters K and D with parameters of computer simulations to design reaction-diffusion models based on LD, i.e. to simulate diffusion-limited bimolecular reactions between Kramers particles. There are also other situations where our anal- ysis will be applicable. One of them is adsorption to surfaces [15, 17]. Our set up includes interactions with a reactive surface of a sphere and can be used in modeling interactions of small molecules with large reactive spheres, for example, for adsorp- tion of polymers to a surface of a virus [17] or for coating of spherical particles by reactive polymers [33]. We also expect our results should be easy to extend to general non-spherical reaction surfaces, assuming sufficient regularity. Another possible ap- plication area is modeling excluded-volume effects. We have observed that the short 20 S. J. CHAPMAN, R. ERBAN, S. A. ISAACSON range repulsive interaction potential (2.6) leads to zero Neumann boundary condi- tion (3.14) if ε is chosen to approach zero slower than (3.13) (in the limit ϕ → ∞). A similar potential mechanism has been used to enforce Neumann boundary conditions on global domain boundaries in [5], and can be used to model excluded-volume effects in models of intracellular macro-molecular crowding [4, 6]. Since LD requires a smaller time step than the overdamped BD model, the numer- ical simulations are in general more computationally intensive for LD. However, for many biological applications the LD model is only required close to a reactive surface (where we have a non-zero potential). In particular, one could replace the compu- tationally intensive LD model with the BD model in the part of the computational domain which is far from the reactive surface [12]. In some applications, one could further substitute the BD simulation algorithms by even coarser and more efficient simulation techniques, including lattice-based simulations [11, 19] or even mean-field equations [15, 20]. In this way, one could design LD simulation methods which simu- late intracellular processes on comparable time scales as the BD simulation packages which are available in the literature [3, 30, 34]. In some applications, one could also design an efficient first-passage-time scheme by replacing discretization (4.9) -- (4.10) (which uses fixed time step ∆t) by estimating the next time when a Kramers particle becomes sufficiently close to the reactive target [21]. Similar approaches have been used to accelerate BD simulations in the literature [27, 28, 35]. Appendix A. A simple example illustrating how specular reflection arises in a classical mechanical system. In this section we illustrate with a simple example how the specular reflection boundary condition (4.7) arises in a classical mechanical system in the absence of noise. We consider the simple case of a particle moving in the infinite one-dimensional potential, (cid:40)− kBT ϕ kBT ϕ(x) = ε x, x ≤ 0, x > 0, 0, and experiencing friction, with friction constant β. Note, for the purposes of this section we only make use of the value of the potential for x ∈ (−∞, 0]. Newton's equation when the particle's position, X(t), satisfies X(t) ≤ 0 with negative velocity is then m dV dt = −mβV + kBT ϕ ε . Using the Einstein Relation this reduces to dV dt = −βV + Dβϕ ε , the one-dimensional analogue of the LD model (1.1) with the noise term neglected. We consider the initial conditions V (0) = v0 < 0, X(0) = 0, and ask for what initial velocity range the molecule successfully moves a distance ε to the left before changing direction and falling down the potential gradient. This REACTIVE BOUNDARY CONDITIONS AS LIMITS OF INTERACTION POTENTIALS 21 is consistent with a molecule reaching the reactive boundary in the Kramer's equa- tion model (2.7). Given the molecule's negative initial velocity, until the molecule's direction of motion reverses and the molecule moves back across the origin we find (cid:18) (cid:19) V (t) = v0 − Dϕ ε exp [−βt] + Dϕ ε . The velocity then becomes zero at time ∗ = t 1 β ln (1 − α) , where At this time, the molecule's position is v0 − Dϕ ε (α − ln(1 − α)) . ∗) = X(t = (cid:19) (cid:18) 1 β Dϕ εβ α = < 0. εv0 Dϕ (1 − exp [−βt ∗]) + Dϕt∗ ε , The molecule then successfully travels further than −ε, analogous to penetrating the "reactive boundary" at x = −ε, if ∗) < −ε, X(t (α − ln(1 − α)) < −ε. Dϕ εβ or equivalently that As ε → 0 We therefore find that the molecule successfully penetrates the reactive boundary in the limit that ε → 0 if and only if . X(t 0 2ϕβD ∗) ∼ −εv2 v0 < −(cid:112)2ϕβD, consistent with the specular reflection boundary condition (4.7). REFERENCES [1] I. Agbanusi and S. Isaacson, A comparison of bimolecular reaction models for stochastic reaction-diffusion systems, Bulletin of Mathematical Biology, 76 (2014), pp. 922 -- 946. [2] S. Andrews, N. Addy, R. Brent, and A. Arkin, Detailed simulations of cell biology with smoldyn 2.1, PLOS Computational Biology, 6 (2010), p. e1000705. [3] S. Andrews and D. Bray, Stochastic simulation of chemical reactions with spatial resolution and single molecule detail, Physical Biology, 1 (2004), pp. 137 -- 151. [4] S. Arjunan and M. Tomita, A new multicompartmental reaction-diffusion modeling method links transient membrane attachment of E. coli MinE to E-ring formation, Systems and Synthetic Biology, 4 (2010), pp. 35 -- 53. 22 S. J. CHAPMAN, R. ERBAN, S. A. ISAACSON [5] P. J. Atzberger, S. A. Isaacson, and C. S. Peskin, A microfluidic pumping mechanism driven by non-equilibrium osmotic effects, Physica D, 238 (2009), pp. 1168 -- 1179. [6] M. Bruna and S. Chapman, Excluded-volume effects in the diffusion of hard spheres, Physical Review E, 85 (2012), p. 011103. [7] M. A. Burschka and U. M. Titulaer, The kinetic boundary layer for the Fokker-Planck equation: Selectively absorbing boundaries, Journal of Statistical Physics, 26 (1981), pp. 59 -- 71. [8] M. A. Burschka and U. M. Titulaer, The kinetic boundary layer for the Fokker-Planck equation: A Brownian particle in a uniform field, Physica A, 112 (1982), pp. 315 -- 330. [9] L. Desvillettes and C. Villani, On the trend to global equilibrium in spatially inhomoge- neous entropy-dissipating systems: the linear Fokker-Planck equation, Communications on Pure and Applied Mathematics, 54 (2001), pp. 1 -- 42. [10] U. Dobramysl, S. Rüdiger, and R. Erban, Particle-based multiscale modeling of intracel- lular calcium dynamics. submitted to Multiscale Modelling and Simulation, available as http://arxiv.org/abs/1504.00146, 2015. [11] S. Engblom, L. Ferm, A. Hellander, and P. Lötstedt, Simulation of stochastic reaction- diffusion processes on unstructured meshes, SIAM Journal on Scientific Computing, 31 (2009), pp. 1774 -- 1797. [12] R. Erban, From molecular dynamics to Brownian dynamics, Proceedings of the Royal Society A, 470 (2014), p. 20140036. [13] R. Erban, Coupling all-atom molecular dynamics simulations of ions in water with Brownian dynamics, submitted, preprint available as: http://arxiv.org/abs/1508.02805 (2015) [14] R. Erban and S. J. Chapman, Reactive boundary conditions for stochastic simulations of reaction-diffusion processes, Physical Biology, 4 (2007), pp. 16 -- 28. [15] R. Erban and S. J. Chapman, Time scale of random sequential adsorption, Physical Review E, 75 (2007), p. 041116. [16] R. Erban and S. J. Chapman, Stochastic modelling of reaction-diffusion processes: algo- rithms for bimolecular reactions, Physical Biology, 6 (2009), p. 046001. [17] R. Erban, S. J. Chapman, K. Fisher, I. Kevrekidis, and L. Seymour, Dynamics of polydisperse irreversible adsorption: a pharmacological example, Mathematical Models and Methods in Applied Sciences (M3AS), 17 (2007), pp. 759 -- 781. [18] R. Erban, M. Flegg, and G. Papoian, Multiscale stochastic reaction-diffusion modelling: application to actin dynamics in filopodia, Bulletin of Mathematical Biology, 76 (2014), pp. 799 -- 818. [19] M. Flegg, J. Chapman, and R. Erban, The two-regime method for optimizing stochastic reaction-diffusion simulations, Journal of the Royal Society Interface, 9 (2012), pp. 859 -- 868. [20] B. Franz, M. Flegg, J. Chapman, and R. Erban, Multiscale reaction-diffusion algorithms: PDE-assisted Brownian dynamics, SIAM Journal on Applied Mathematics, 73 (2013), pp. 1224 -- 1247. [21] P. Hagan, C. Doering, and C. Levermore, Mean exit times for particles driven by weakly colored noise, SIAM Journal on Applied Mathematics, 49 (1989), pp. 1480 -- 1513. [22] J. Keizer, Nonequilibrium statistical thermodynamics and the effect of diffusion on chemical reaction rates, J. Phys. Chem., 86 (1982), pp. 5052 -- 5067. [23] G. R. Kneller and U. M. Titulaer, Boundary layer effects on the rate of diffusion con- trolled reactions, Physica A, 129A (1985), pp. 514 -- 534. [24] B. Leimkuhler and C. Matthews, Molecular Dynamics: with deterministic and stochastic numerical methods, Springer, 2015. [25] X. Li, J. Lowengrub, A. Raetz, and A. Voigt, Solving PDEs in Complex Geometries: a Diffuse Domain Approach, Comm.Math. Sci., 7 (2009), pp. 81 -- 107. [26] K. Lipkow, S. Andrews, and D. Bray, Simulated diffusion of phosphorylated CheY through the cytoplasm of Escherichia coli, Journal of Bacteriology, 187 (2005), pp. 45 -- 53. [27] A. J. Mauro, J. K. Sigurdsson, J. Shrake, P. J. Atzberger, and S. A. Isaacson, A first-passage kinetic Monte Carlo method for reaction-drift-diffusion processes, Journal of Computational Physics, 259 (2014), pp. 536 -- 567. [28] T. Opplestrup, V. Bulatov, A. Donev, M. Kalos, G. Gilmer, and B. Sadigh, First- passage kinetic Monte Carlo method, Physical Review E, 80 (2009), p. 066701. [29] R. L. Pego, Front migration in the nonlinear Cahn-Hilliard equation, Proc. R. Soc. Lond. Ser. A Math. Phys. Eng. Sci., 422 (1989), pp. 261 -- 278. [30] M. Robinson, S. Andrews, and R. Erban, Multiscale reaction-diffusion simulations with Smoldyn, Bioinformatics, (2015), p. doi: 10.1093/bioinformatics/btv149. [31] D. Shoup and A. Szabo, Role of diffusion in ligand binding to macromolecules and cell-bound REACTIVE BOUNDARY CONDITIONS AS LIMITS OF INTERACTION POTENTIALS 23 receptors, Biophys. J., 40 (1982), pp. 33 -- 39. [32] M. Smoluchowski, Versuch einer mathematischen Theorie der Koagulationskinetik kolloider Lösungen, Zeitschrift für physikalische Chemie, 92 (1917), pp. 129 -- 168. [33] V. Šubr, Č. Koňák, R. Laga, and K. Ulbrich, Coating of DNA/Poly(L-lysine) complexes by covalent attachment of poly[N-(2-hydroxypropyl)methacrylamide], Biomacromolecules, 7 (2006), pp. 122 -- 130. [34] K. Takahashi, S. Tanase-Nicola, and P. ten Wolde, Spatio-temporal correlations can drastically change the response of a mapk pathway, PNAS, 107 (2010), pp. 19820 -- 19825. [35] J. van Zon and P. ten Wolde, Green's-function reaction dynamics: a particle-based ap- proach for simulating biochemical networks in time and space, Journal of Chemical Physics, 123 (2005), p. 234910.
1808.07956	2	1808	2018-08-27T12:19:18	Principles for sensitive and robust biomolecular interaction analysis - The limits of detection and resolution of diffraction-limited focal molography	[ "physics.bio-ph", "physics.app-ph" ]	Label-free biosensors enable the monitoring of biomolecular interactions in real-time, which is key to the analysis of the binding characteristics of biomolecules. While refractometric optical biosensors are sensitive and well-established, they are susceptible to any change of the refractive index in the sensing volume caused by minute variations in composition of the sample buffer, temperature drifts and nonspecific binding to the sensor surface. Refractometric biosensors require reference channels as well as temperature stabilisation and their applicability in complex fluids such as blood is limited by nonspecific bindings. Focal molography does not measure the refractive index of the entire sensing volume but detects the diffracted light from a coherent assembly of analyte molecules. Thus, it does not suffer from the limitations of refractometric sensors since they stem from non-coherent processes and therefore do not add to the coherent molographic signal. The coherent assembly is generated by selective binding of the analyte molecules to a synthetic binding pattern - the mologram. Focal Molography has been introduced theoretically and verified experimentally in previous papers. However, further understanding of the underlying physics and a diffraction-limited readout is needed to unveil its full potential. This paper introduces refined theoretical models which can accurately quantify the amount of matter bound to the mologram from the diffracted intensity. In addition, it presents measurements of diffraction-limited molographic foci. These improvements enabled us to demonstrate a resolution in real-time binding experiments comparable to the best SPR sensors, without the need of temperature stabilisation or drift correction and to detect small molecules label-free in an endpoint format. The presented experiments exemplify the robustness and sensitivity of diffractometric sensors.	physics.bio-ph	physics	Principles for sensitive and robust biomolecular interaction analysis - The limits of detection and resolution of diffraction-limited focal molography Andreas Frutiger,1, ∗ Yves Blickenstorfer,1, ∗ Silvio Bischof,1 Csaba Forró,1 Matthias Lauer,2 Volker Gatterdam,1 Christof Fattinger,2, † and János Vörös1, ‡ 1Laboratory of Biosensors and Bioelectronics, Institute of Biomedical Engineering, ETH Zürich, 8092 Zürich, Switzerland 2 Roche Pharma Research and Early Development, Roche Innovation Center Basel, 4070 Basel, Switzerland (Dated: August 28, 2018) Label-free biosensors enable the monitoring of biomolecular interactions in real-time, which is key to the analysis of the binding characteristics of biomolecules. While refractometric optical biosensors such as SPR [Surface Plasmon Resonance] are sensitive and well-established, they are susceptible to any change of the refractive index in the sensing volume caused by minute variations in composition of the sample buffer, temperature drifts and most importantly nonspecific binding to the sensor surface. Refractometric biosensors require reference channels as well as temperature stabilisation and their applicability in complex fluids such as blood is limited by nonspecific bindings. Focal molography does not measure the refractive index of the entire sensing volume but detects the diffracted light from a coherent assembly of analyte molecules. Thus, it does not suffer from the limitations of refractometric sensors since they stem from non-coherent processes and therefore do not add to the coherent molographic signal. The coherent assembly is generated by selective binding of the analyte molecules to a synthetic binding pattern -- the mologram. Focal Molography has been introduced theoretically [C. Fattinger, Phys. Rev. X 4, 031024 (2014)] and verified experimentally [V. Gatterdam, A. Frutiger, K.-P. Stengele, D. Heindl, T. Lübbers, J. Vörös, and C. Fattinger, Nat. Nanotechnol. 12, 1089 (2017)] in previous papers. However, further understanding of the underlying physics and a diffraction-limited readout is needed to unveil its full potential. This paper introduces refined theoretical models which can accurately quantify the amount of biological matter bound to the mologram from the diffracted intensity. In addition, it presents measurements of diffraction-limited molographic foci i.e. Airy discs. These improvements enabled us to demonstrate a resolution in real-time binding experiments comparable to the best SPR sensors, without the need of temperature stabilisation or drift correction and to detect low molecular weight compounds label- free in an endpoint format. The presented experiments exemplify the robustness and sensitivity of the diffractometric sensor principle. PACS numbers: I. INTRODUCTION Diffractive lenses or focusing holograms proposed by Augustin-Jean Fresnel are known to humanity since two hundred years and have ever since experienced appli- cability in various fields such as photography [1], tele- scopes [2], spectroscopy [3], optical tweezers [4], and X- ray lenses [5]. Yet, nature has discovered this principle much earlier. In certain organisms, biomolecules are as- sembled to create a focusing hologram for image forma- tion in the eye [6]. Recently, thanks to advances in photo- lithography [7] and non-fouling photoactivatable surface chemistries in particular [8], it became possible to apply the holographic principle to highly sensitive molecular ∗A. Frutiger and Y. Blickenstorfer contributed equally to this work. †Electronic address: [email protected] ‡Electronic address: [email protected] detection. These molecular holograms can be used for real-time label-free detection of molecules by molecular recognitions in complex samples [7]. This enables the monitoring of biomolecular interactions, which is key to the analysis of binding characteristics of biomolecules in a broad range of applications [9]. To date, the biosens- ing field is dominated by refractometric optical sensors and most prominently, thanks to their high surface sen- sitivity, by techniques based on evanescent waves, such as surface plasmons or dieletric waveguide modes [10, 11]. These analytical tools are well-established to perform label-free binding assays with high sensitivity and low limits of detection [0.1-1 pg/mm2] [12, 13]. Refracto- metric biosensors [e.g. surface plasmon resonance, SPR] measure the refractive index change upon receptor-ligand binding in the vicinity of the sensor surface. However, they are susceptible to any change in refractive index within the evanescent field caused by fluctuations in tem- perature, buffer composition and most importantly non- specific binding to the sensor surface. This inherent fea- 2 FIG. 1: Focal molography incorporates the four essentials of a highly sensitive diffractometric biosensor (a) A sub-micrometer affinity modulation of specific binders is exposed to a biological sample [i.e. blood]. The mode of a high refractive index waveguide provides perfect darkfield illumination of the molecules in the vicinity of the sensor surface and enhances the light intensity. The shape of the pattern acts as a diffractive lens, which concentrates the diffracted signal into a focal spot, whereas the background intensity is diluted over the entire solid angle. (b) The molographic pattern with the actual focal spot superimposed [bottom view] and enlarged in (c). The Airy disk of the mologram [red dot] monitors the binding activity of billions of recognition sites on an area that is nearly five orders of magnitude larger than the tiny focal spot. ture of refractometric sensors often manifest as drift and causes jumps in the sensor signal - e.g. during sample exchange [14]. Therefore, these sensors typically operate continuously and in well-defined buffers because measure- ments in serum or plasma exhibit artifacts and stability problems. The mentioned limitations arise from the fundamen- tal inability of a refractometric sensor to distinguish be- tween the molecules of the target analyte and all other influences that affect the refractive index of the sensing volume. Even for evanescent field sensors, the sensing volume is still enormous compared to the small volume of the target molecules. This makes it virtually impos- sible to compensate for these influences - even with a differential measurement [14]. There is, however, a phys- ical phenomenon that only measures the refractive index difference between the target molecules and the refractive index of their displacement volume, namely the scatter- ing of light. This rejects most of the influences from temperature and buffer changes by measuring only the refractive index contrast in the nanoenvironment of the binding events. It is a common belief in the biosensing community that the most sensitive detection methods for label-free biomolecular interaction analysis in real-time are based on the refractometric sensing principle. In this context it is sometimes not esteemed that biomolecular interac- tions can be detected with high sensitivity by the scat- tering of light [7, 15, 16]. The single molecule detection method iSCAT [15] is based on interferometric detection of scattering. It demonstrates exquisite sensitivity for the analysis of biomolecular interactions through scat- tering. The single molecular sensitivity of iSCAT allows analyzing the heterogeneity in an ensemble of a molec- ular species. Yet, in other applications it is sufficient to determine an averaged quantity of the ensemble. The accurate quantification of a biomarker concentration falls in this category. In such a measurement, single molec- ular sensitivity can be beneficial but is neither required nor should its importance be overestimated. In the case of iSCAT, the single molecular sensitivity comes at the cost of a relatively complicated setup, since the noise has to be sufficiently low to detect every single protein in- dependently. In addition, iSCAT cannot distinguish be- tween different types of similarly sized proteins. There- fore, for measuring in complex fluids, iSCAT is currently limited by non-specific binding similarly to refractomet- ric sensors. In both cases, the specificity is mostly deter- mined by the choice of surface chemistry [15]. A protein- repellent [non-fouling] surface chemistry is not enough to measure in complex samples since there is always a sig- nificant amount of defects in the ad-layer and therefore of nonspecific binding to the sensor [17, 18]. Conversely, molecular holograms are diffractometric sensors, which offer an additional mechanism to reduce the effect of nonspecific binding. This is achieved by constraining the specific binding to a coherent scattering system - i.e. a molecular hologram. The blueprint of this hologram is encoded into the surface ad-layer. Namely, the recognition sites compose a coherent binding pat- tern - i.e. a mologram. The constructive interference of the scattered fields relates all bound analyte molecules and yields a quadratic scaling of the measured inten- sity with respect to the analyte number. On the other hand, the scattered field of randomly bound background molecules interferes with equal probability constructively or destructively. Therefore, only its variance affects the coherent signal and thus it scales linearly with particle number. This implies that the nonspecific binding is sup- pressed efficiently for a sufficiently large ensemble of an- alyte molecules. In addition, other random scattering [noise] sources experience the same repression with re- spect to the signal. Therefore, it is considerably simpler to detect an ensemble of molecules coherently than to count them individually. In other words, a diffractomet- ric sensor is inherently self-referencing on the sub-micron length scale of regions of constructive and destructive in- terference. A pure diffractometric biosensor consists of coherently arranged binding sites without any diffractive power, or in other words a massless affinity modulation [16]. A sen- sor with these properties is extremely robust and only produces a signal in the presence of the analyte [7]. The conception of an affinity modulation that has no opti- cal modulation is reasonable to physicists. However, ad- vanced molecular engineering capabilities are required to achieve this experimentally, since most nanolithographic techniques such as imprinting or lift-off techniques pro- duce an inherent optical modulation due to coherent de- fects in the affinity keys [19, 20]. Even worse, these co- herent defects will give rise to an affinity modulation for background molecules, severely compromising the rejec- tion of nonspecific binding. This fact makes the imple- mentation of sensitive diffractometric biosensors interdis- ciplinary and demanding. Focal molography is the first diffractometric sensor that may exhibit resolutions in direct binding assays com- parable to the best refractometric sensors [7, 16]. Briefly, in focal molography the mologram is situated on a high refractive index slab waveguide and illuminated by the fundamental TE mode [Fig. 1]. When the affinity modu- lation is exposed to a biological sample the analyte binds to the mologram. This induces an optical grating that diffracts light from the guided TE mode into a diffraction- limited focal spot. From the diffraction efficiency, the presence of molecules at the interaction sites is quanti- fied. We will now outline the four pillars that need to be fulfilled for diffractometric biosensors to be highly sensi- tive and robust. Previously reported diffractometric con- cepts for biomolecular interaction analysis [21 -- 24][61] do 3 not incorporate all four pillars. This is the reason why they are limited in sensitivity or robustness. The concept - focal molography - was introduced with all the essentials necessary for highly sensitive and robust diffractometric biomolecular interaction analysis [16] [Fig. 1]. [I] The first essential is a sub-micrometer affinity modulation on a non-fouling monolithic surface-layer for efficient rejec- tion of nonspecific binding. In the first demonstration of focal molography, this has been achieved by the reactive immersion lithography [RIL] process, which produces an affinity modulation consisting of active regions [ridges] and passive regions [grooves] on a non-fouling brushed copolymer ad-layer [7]. Ideally, this affinity modulation should be massless. [II] In focal molography, the molo- gram is situated on an asymmetric high refractive in- dex slab waveguide, which provides the second essential - a proper dark-field illumination of the coherent affin- ity modulation. The two dimensional light sheet of the guided TE mode only illuminates the first 100 nm of the sample solution close to the surface. This avoids any background scattering from particles in the sample solu- tion that are further away. [III] The high refractive in- dex waveguide also provides the third essential, namely an increase in the field intensity at the scatterer loca- tion. In other words, guided to free-space mode coupling is more efficient than free-space to free-space coupling for a given amount of coherent biological matter [25]. [IV] The fourth and last essential is the observation of the diffracted signal in the far-field of the mologram defined in terms of Fraunhofer distance. This near to far-field transformation increases the signal to noise ratio due to the directed character of the diffracted signal compared to the distributed background from random scatterers. However, the Fraunhofer distance of a linear diffraction grating with a length of 400 µm is roughly 50 cm for visible wavelengths. By using a lens, the far-field can be observed much closer to the sensor surface. In our case, the mologram itself performs the near to far-field transformation by focusing the intensity holographically onto an Airy disk only a few hundred microns away from the sensor surface. The binding information of billions of recognition sites on an area five orders of magnitude larger is therefore contained in the tiny Airy disk [Fig. 1(b),(c)]. This enables compact technical realizations of diffractometric biosensors. From another viewpoint, focal molography can also be seen as a "chemical radio", at least in the eyes of a physical chemist [26]. The transmission of radio signals is based on the modulation of an RF carrier signal and the subse- quent demodulation at the receiver. Molography applies this principle at optical frequencies to the transmission of chemical signals. Molecules recognize the affinity modu- lation in the mologram and interact with it. The molec- ular interaction renders a coherent molecular pattern in the form of a diffractive lens. This diffractive lens modu- lates the momentum of the guided mode with the spatial frequency of the mologram. The demodulation in k-space is performed by Fourier optics and the molographic sig- nal is separated from the carrier wave in the focal plane of the lens. Recently, the first experimental measurements with focal molography have been performed using the non- diffraction-limited ZeptoReader [Zeptosens AG], sub- stantially compromising the fourth pillar [7]. The em- phasis of that publication was the demonstration of the robust operation of focal molography and its insensitiv- ity to nonspecific binding in complex samples rather than achieving high sensitivity. Nevertheless, a real-time de- tection limit of 5 pg/mm2 was achieved and we made the projection that noise levels can be reduced by at least two orders of magnitude by observing the molographic signal in a reader capable of resolving the diffraction-limited focus. The aim of this contribution is therefore to charac- terize the molographic signal in the proper far-field and thus to explore the resolution limits of diffraction-limited focal molography for massless affinity modulations ex- perimentally as well as to refine some of the theoretical concepts. First, we introduce a measurement setup that allows static and real-time measurement of diffraction- limited molographic focal spots. Second, we present a semi-analytical framework with which the field distribu- tion in the focal spot can be accurately computed by summation of the scattered fields of individual molecules [dipole scatterers] on the waveguide surface. Third, we demonstrate that the simulated and experimental field distributions are in excellent agreement with each other and that the Airy disk dimensions of the mologram are consistent with the Airy disk of a diffraction-limited lens. Fourth, we show that our synthetic holograms produce diffraction-limited focal spots at least up to mologram diameters of 400 µm on high refractive index slab waveg- uides. Furthermore, it was verified that the analytical predictions for the intensity of the focal spot through cou- pled mode theory made by Fattinger [16] coincide with Rayleigh scattering and can accurately describe the ex- perimentally measured intensities for a given amount of coherent biological matter. Fifth, we address the rele- vance of different background sources that can scatter intensity into the focal plane and produce an inhomoge- neous speckle pattern that limits the resolution and ac- curacy of the molographic measurement. Based on this discussion, a figure of merit for molography is formulated that allows direct comparison of different molographic arrangements with different waveguides and mologram sizes. Next, we apply our theoretical insights to calcu- late limits of detection for molography on Ta2O5 slab waveguides for endpoint and real-time detection. These predictions are then verified experimentally. In partic- ular, the low molecular weight [< 300 Dalton] molecule vitamin B7, commonly known as biotin, is detected label- free by molography in an endpoint measurement without any calibration of the sensor. These biotin molograms are most likely the faintest man made holograms that have 4 ever been measured. In addition, we demonstrate that it is possible to fabricate an affinity modulation without a detectable optical modulation and use it to acquire real- time binding curves with 500 pM Streptavidin [SAv] in buffer that exhibit baseline noise levels below 100 fg/mm2 over 20 mins which are comparable to the best commer- cially available label-free detection method [27]. How- ever, while the commercial system is temperature stabi- lized to 0.01 ◦C, we achieved this stability without any temperature control demonstrating the potential of focal molography for extremely sensitive and robust, real-time, label-free molecular interaction analysis. II. DIFFRACTION-LIMITED MOLOGRAPHY A. Measurement of foci formed by diffraction-limited molograms The realization and quantification of diffraction- limited molographic experiments incorporates the design of a microscope, waveguide coupler, fluidics as well as the development of appropriate algorithms for evalua- tion of the acquired images. The setup [MoloReader] which we developed for this purpose is displayed in Fig. 2(d), as well as in Fig. 15 and a functional schematic is shown in Fig. 2(a). The setup allows to couple a TE polarized He-Ne laser beam [632.8 nm wavelength] via a grating coupler [coupling angle -10.6◦, period 318 nm, length 500 µm] into a thin-film optical waveguide [145 nm thick Ta2O5] on a glass substrate [D263 Schott, 700 µm]. Molecules located on the waveguide are illuminated by the evanescent field of the fundamental guided TE mode [N = 1.814, penetration depth 82 nm] in a dark-field manner [Fig. 2(b)]. For most of the experiments pre- sented in this paper, the molograms composed of bind- ing sites for the protein molecule Streptavidin [SAv]. The molograms on the waveguide consist of alternating ridges, where SAv binds to immobilized biotin [MW: 227 g/mol]; and grooves, which are backfilled with an inert PEG molecule [MeO-dPEG12, MW: 570 g/mol]. The PEG backfilling is performed to obtain a massless affinity mod- ulation, as well as for blocking of free amine groups. As a side note, despite the higher molecular mass of the PEG, these molograms exhibit a non-detectable mass modula- tion [Movie 3]. Most likely because PEG is the more flexi- ble molecule and has a smaller refractive index increment than biotin [0.12 compared to 0.16 ml/g]. The SAv bound mologram is denoted as [NH-biotin/SAvNH-PEG12] and was fabricated by reactive immersion lithography as in- troduced previously [Fig. 2(c)] [7]. We developed a new version of the illumination setup that achieved higher peak-to-peak mass modulations of 540 pg/mm2 [27 %] compared to the previously published 283 pg/mm2 [14 %] [7] thanks to the higher spectral and spatial coherence of the laser source used [405 nm] [Figs. 13 and 14]. This value was determined from a STED [stimulated emis- sion depletion microscopy] measurement on a Leica SP8 5 FIG. 2: Detection of diffraction-limited molographic spots (a) Schematic representation of the setup for the experimental demonstration of diffraction-limited molography: Light of a He-Ne laser is coupled into the fundamental TE mode of a high refractive index slab waveguide via a grating coupler [GC]. The light propagates along the waveguide and is scattered at the molecules of interest [Streptavidin] that form a focusing hologram. These molecules are captured from solution by binding to a coherent affinity modulation on top of a non-fouling polymer layer that is fabricated by reactive immersion lithography [7]. The molographic signal [the intensity of the focal point] of one of ten molograms in a row is collected by a microscope objective and captured by a camera. The total power in the waveguide is monitored with a photodiode, which measures the light diffracted by a physical out-coupling grating which is etched into the waveguide. (b) Enlarged view of the waveguide with the field profile of the TE mode. (c) In active regions [ridges] immobilized receptors [biotin] capture the protein of interest [SAv] and form a coherent assembly, whereas inert regions backfilled with polyethylene glycol [grooves] do not recognize the protein. (d) The experimental setup [MoloReader] in operation. The microscopy objective is focused on the focal plane of the mologram. STED as described in our previous publication [7]. When impinging on the mologram, a small portion of the light is coupled out into two converging beams that form two diffraction-limited foci above and below the waveguide. The diffracted light of the lower beam is collected by a 20 x, 0.4 NA microscopy objective and visualized by a CMOS camera. The molographic pattern has a diameter of 400 µm, a numerical aperture of 0.33, a focal length of 900 µm in glass [ns = 1.521] and a sickle shaped central recess area [Bragg recess area] of 50 µm width to avoid second order Bragg reflections [16]. The recess area is formed by two concentric circles with 1040 µm and 1140 µm diameter, respectively. B. From protein molecules to molographic signals - Simulations of molographic foci The qualitative intensity distribution in and around the diffraction-limited molographic spot can be described by the coherent superposition of individual Rayleigh scat- terers or by a mean-field approach through Fourier optics both yielding the same results [30]. Here we chose the first method to investigate the expected intensity dis- tribution in the focal plane by summation of the scat- tered electric field of individual dipoles [molecules] lo- cated on the mologram [Fig. 3(a)]. We wrote a GPU- based Python framework that can simulate the intensity distribution of a large amount [few 100 millions] of scat- terers on a plane with typically 150 x 150 pixels reso- lution within roughly an hour. This semi-analytical ap- proach has the great advantage of calculating the field 6 at the optical interfaces, we used the dipole potential approach outlined by Novotny and Hecht [29]. Further- more, multi-body interactions were disregarded because the scattering cross-section of a typical protein is only in the order of 10−24 m2 and the total diffracted power is typically less than 1 %. C. Comparison between analytical, numerical and experimental results 1. Shape of the molographic Airy disk Fig. 4 illustrates the diffraction-limited focus obtained by simulations [Figs. 4(a),(b)] and experiments [Figs. 4(c),(d)] as an axial and radial slice through the focal point of the mologram. The experimental molographic spot was acquired from a SAv555 [Alexa FluorTM 555 la- beled, Thermo Fisher Scientific] mologram in water. The fluorophore was only used for quality control and its scat- tering cross section is negligible compared to the one of SAv. Therefore, we will only refer to SAv for the rest of the paper. The chip was fabricated by reactive im- mersion lithography in DMSO [7] followed by a 15 min incubation of 1 µM SAv in PBS-T buffer [pH 7.4; 0.05 % Tween20]. This yields a 540 pg/mm2 peak-to-peak sinu- soidal surface mass modulation. The experimental focal spot was acquired under water immersion by pipetting 10 µl of DI water on the chip and performing a z-stack with the MoloReader [vertical resolution 1.3 µm]. The com- puted focal spot was obtained from a simulation of 4.75 million SAv molecules [7.9 pg/mm2 peak-to-peak modu- lation] sinusoidally distributed on the ridges of the molo- gram with water as the cover medium. This amount of SAv molecules was sufficient to demonstrate the excellent agreement of the numerical results with the measured ex- perimental intensities. The simulated and experimental images only differ by the speckle pattern caused by scat- tering of the guided wave at non-coherent dipoles, which were not taken into account in the simulations. The Airy disk radius for a diffraction-limited lens is determined by the wavelength and the numerical aper- ture of the mologram ∆x = 0.61 λ NA which leads to 1.17 µm for our molograms. The Airy disk radii found in the measured [solid blue] and the simulated curves [dashed green] in Figs. 4(e) and 4(f), are 1.09 µm and 1.07 µm in x and 1.34 µm and 1.22 µm in y-direction, respectively. The Airy disk is slightly elongated in y-direction in both experiment and simulation due to symmetry breaking of the central Bragg recess area of the mologram [Fig. 3(a)]. Without considering the central recess area the Airy disk is perfectly symmetrical and has the size of the focal spot of a diffraction-limited lens [Fig. 17]. In the ex- periment, there is additional broadening by scattering of the guided mode at waveguide imperfections into other guided modes with a small y-component in the propa- gation vector. In the extreme case of a contaminated [strongly scattering] waveguide, the molographic spot at- FIG. 3: Simulation of molographic foci (a) The molographic signal emerges from the superposition of the scattered elec- tric fields of many individual protein molecules on the sur- face of the waveguide [proteins are not drawn to scale but their number density corresponds to the 2.6 pg/mm2 at the detection limit [Fig. 8]]. This field is computed for every pixel on a specified screen in the focal plane of the mologram. (b) The scattered field is calculated by modeling the proteins as Rayleigh scatterers excited by the evanescent field of the waveguide mode, which is obtained by solving the eigenvalue problem of the slab waveguide [28]. nc, nf and ns are the refractive indices of cover, film and substrate. The dielectric interfaces can be accounted for by computing the dipole po- tentials of the two layer interface as described by Novotny and Hecht [29]. (c) The optical properties necessary to determine the polarizability of the protein dipole i.e. refractive index and radius can be calculated from its molecular mass and the refractive index increment for proteins in water [16]. only where it is to be determined [compared to FDTD or FEM approaches]. The exact procedure is outlined in the Supplemental Material [SM] Section II and shall only be summarized here. First, the eigenmode equation of the dielectric waveguide was solved according to Marcuse [28] for the fundamental TE mode in order to calculate the excitation field at the position of the scatterers, which were placed on the molographic pattern [Fig. 3(b)]. Pro- teins can be modeled as Rayleigh scatterers due to their small size of only a few nm [31]. The dipole strength of a protein molecule depends only on its molecular mass and the immersion medium [Fig. 3(c)]. This is because the radius and the refractive index of the resulting sphere are related to the molecular mass and can be calculated as described in the SM Section II C. For most practical pur- poses the exact composition of the protein is negligible for its scattering properties. To account for reflections 7 mation of Rayleigh scatterers [RS] without considering multiple reflections of the scattered light at the interfaces of the waveguide. Here, we briefly show that the analyti- cal expressions for the two approaches are equivalent and verify them by numerical simulations and experiments. Fattinger [16] used coupled mode theory to obtain an expression for the ratio of power diffracted from the molecular assembly to the power guided by the waveg- uide [both powers are expressed in power per unit length] [33]. Here, we adapt this expression to yield the intuitive transfer function between the intensity on the waveguide surface and the average intensity in the Airy disk [deriva- tion in the SM Section III] which reads (cid:18) dn (cid:19)2 D2 Iavg,CMT = 5.59 · NA2 dc λ4 ηmod[A] 2∆Γ 2I0 (1) The subscript CMT stands for coupled mode theory. NA is the numerical aperture of the mologram, dn dc the refrac- tive index increment for proteins in water [34], D the di- ameter of the mologram, λ the wavelength. Here, we have adapted and generalized the canonical surface mass mod- ulation ∆Γcan introduced by Fattinger [16] with the con- cept of the analyte efficiency of the modulation ηmod[A]. The analyte efficiency of the modulation is analogous to the diffraction efficiency of gratings with different grating functions [35]. The surface mass density modulation can , where mmod is the mass be computed from ∆Γ = mmod A+ of the modulation and A+ the area of the ridges [see SM Section I]. For a sinusoidal surface mass density modula- tion [obtained to a first approximation from phase mask lithography] this is equal to the peak-to-peak value. Here we note that Fattinger defined the canonical surface mass density modulation differently. His definition would cor- respond to the molographic surface mass density [see SM Section I] and is therefore a factor two smaller. The pref- actor in Eq. (1) arises from various considerations such as taking into account the relative power incident on the Airy disk and its size. The intensity on the waveguide surface is given by f − N 2(cid:1) (cid:0)n2 nc N teff (n2 f − n2 c) I0 = 2 Pwg (2) where Pwg is the power per unit line [W/m] in the waveg- uide, teff the effective thickness of the waveguide, N the effective refractive index of the fundamental TE mode, nf, nc the refractive indices of the waveguide film and the cover medium. The expression for Rayleigh scattering [neglecting the optical interfaces] is stated by Fattinger [16]. It can be written in the following form [see SM Section III] (cid:1)2 (cid:0)n2 (n2 P − n2 c c)2 P + 2n2 D2 λ4 2∆Γ ηmod[A] ρP2 2 I0 (3) Iavg,RS = 1.268·π2NA2n2 c FIG. 4: The normalized intensity distribution in the vicinity of the focal spot. The contour plots show the vertical and the horizontal focal plane of the normalized intensity signal obtained by simulations (a),(b) and experiments (c),(d). The line plots (e)-(g) show the cross sections evaluated through the focal point along each axis. Both the simulated [dashed green] and the experimentally obtained [solid blue] curves in the focal plane exhibit the shape of a Bessel function as ex- pected from ideal lenses. The Airy disk is slightly enlarged in y-direction due to the sickle shaped recess area in the middle of the mologram [16]. λ tains a sickle shape [so called m-line [32]]. The depth of field of the mologram also follows the equation for a diffraction-limited lens: ∆z = 2ns NA2 [Chapter 4 in Ref. [29]]. The depth of field depends on the refractive in- dex of the medium in which it is observed. Since the thickness of our chip is smaller than the focal length of the mologram, we observe the molographic focus in air [Fig. 16]. Yet, the simulation was carried out in an in- finitely thick glass slide, therefore the depth of field had to be compressed by a factor 1/1.521 to match the ex- periment. After this adjustment, the experimental and the simulated depth of field amounted to 11.23 µm and 12.61 µm, respectively [Fig. 4(g)]. These are in close agreement with the expected value of 11.62 µm for a diffraction-limited lens. 2. Quantitative intensity in the molographic focus by analytical predictions, simulations and experiments The quantitative intensity within the Airy disk is amenable through coupled mode theory [CMT] or sum- 8 sphere. For SAv, we computed the following values: ρP = 1.412 g/cm3 and nP = 1.598, which yield an equivalent dc = 0.36 ml/g [air] that has to be used in the CMT dn model for comparison purposes. To compare the analytical expressions [Eqs. (1) and (3)] with experiments, molograms of different diameters and numerical apertures were fabricated on a chip. 10 molograms of decreasing diameters [400, 343, 296, 255, 221, 193, 168, 148, 131, 117 µm] and numerical apertures of [0.33, 0.29, 0.25, 0.21, 0.19, 0.16, 0.14, 0.13, 0.11, 0.1] at a constant focal length of 900 µm were designed in the first row of the phase mask. The same 10 molograms were arranged in the opposite order in the second row of the phase mask. The diameters were chosen such that the area differs by a factor 1.4 from one mologram to the next [compensated for the Bragg recess area]. To use the analytical expression for molograms with the Bragg re- cess area the area has to be corrected by a factor Amologram where A+ is the area of the ridges and Amologram is the area of the molographic footprint [ridges + grooves + Bragg recess area]. The experimental design of two rows was chosen because each mologram alters the mode shape slightly and therefore the foci of the last molograms in a row are increasingly distorted in y-direction. Thus, the first five molograms of either row were used on three dif- ferent measurement fields on the same chip [see Fig. 5(c) and Ref. [7] for chip geometry]. The investigated molo- grams were the same SAv-molograms as described in the last section. 2A+ The simulation was performed by placing exactly the amount of SAv molecules sinusoidally on the ridges of the mologram that exhibits the same diffraction efficiency as a peak-to-peak surface mass modulation of 540 pg/mm2 [for the 400 µm mologram these are 231 million individ- ual dipole scatterers, see SM Section II for the necessary conversions]. The proteins were placed directly on the waveguide [the field at z = 0 was used to calculate the dipole moment]. The scattered intensity was computed on a 150 x 150 grid around the focus whereas individual grid points were spaced 110 nm apart, which is equivalent to the pixel size of the camera used in the experiment. Fig. 5a shows the intensity distribution in the focal spot of the 10 molograms with varying diameters in air ob- tained from simulations and experiments, the last row shows the underlying mologram. It can be seen that the intensity distributions of simulation and experiment are in perfect agreement over the entire range of diameters investigated. It has to be noted, that it is nontrivial to achieve diffraction-limited focusing for molograms up to a diameter of 400 µm on high refractive index waveguides, since already small gradients in the thickness [and there- fore also in the effective refractive index] can cause an accumulated phase shift between the guided mode and the synthetic hologram [designed for constant effective refractive index]. Therefore, after a certain propagation distance, which we call the dephasing length, light scat- tered at the first and the last line of the mologram can interfere destructively [see SM Section IV and Fig. 12]. FIG. 5: Comparison between analytical models [CMT and RS], numerical simulations and experiments for SAv molo- gram in air. (a) First row: Simulation of the intensity dis- tribution in the focal plane for molograms with constant fo- cal length [900 µm], constant sinusoidal surface mass density modulation [540 pg/mm2, peak-to-peak] and varying diame- ters with air as cover medium. Second row: Corresponding experimental measurement of the intensity distribution in the focal plane. Third row: Schematic of the underlying molo- grams. (b) Absolute average intensity values over the Airy disk for different molograms derived from measurements [me- dian is in black other measurements in grey], simulations and the two analytical models [Eq. (1) and 3]. The diameters of the molograms in the analytical formula were adjusted in order to account for the missing binding sites in the central Bragg recess area. The small differences between measured and calculated intensity for some molograms can be explained by alterations of the wave front of the guided mode due to preceding molograms (c) Chip geometry with the molograms positions taken for the analysis. If one inserts the definition of the refractive index incre- dc = 0.182 ml/g ment for proteins in dilute solutions [ dn [water]] [34, 36] dn dc = 3 2 1 ρP nc P − n2 n2 c n2 P + 2n2 c (4) one can easily verify that they yield the same result. ρP is the dry mass density of the protein calculated according to [37] and nP the refractive index of the dry protein III. BACKGROUND AND NOISE ANALYSIS FOR MOLOGRAPHY WITH MASSLESS AFFINITY MODULATION 9 Fig. 5(b) compares the average intensity in the focal point analytically [RS and CMT], numerically and exper- imentally. The numerical intensity values were obtained by averaging the intensity on the screen over the Airy disk of the diffraction-limited lens. Experimentally, the mean intensity was calculated by subtracting the average background and then averaging on a circle of the size of the Airy disk centered at the maximum intensity. As can be readily seen, the CMT model and the RS model show nearly perfect agreement with the experimental results. However, there are a few effects which are not accounted for in these simple analytical models. These include free space attenuation [all the dipoles are assumed to be at the center of the mologram], the angle dependence of Rayleigh scattering [31], the symmetry breaking of the central Bragg recess area, the observation in a half space with a denser medium and reflections at the interfaces [38]. The numerical simulations incorporate them [see SM Section II], which result in a slightly lower intensity than the analytical models [factor 1.33 in air]. The fact that the experiment agrees closer with the simple ana- lytical models can be explained by uncertainties in the measurement. These arise for example from the deter- mination of the surface mass modulation on the molo- gram. Due to the nature of the quantification procedure [quantitative fluorescence on the sub-micron scale using STED], we expect to have an uncertainty of roughly 10 % in this measurement. Other possible sources of error, and most likely the prominent ones, are the estimation of the guided power Pwg at the mologram location [SM Section VI C]. One can also see in Fig. 5 that the molo- grams closer to the in-coupling grating have higher in- tensities and their median values show less deviation to the curves expected from the analytical and numerical models. This can readily be explained by the alteration of the wave front [guided-guided mode coupling] at every preceding mologram and at waveguide imperfections. In summary, we have shown that the RS and CMT models are equivalent and show excellent agreement with numerical simulations and experiments for molograms with different diameters and numerical apertures in air. The analytical models are therefore a valid tool to make predictions of the limit of detection and to determine the surface mass modulation from the measured intensity in the molographic focal spot. Furthermore, we demon- strated the manufacturing of diffraction-limited molo- grams with diameters up to 400 µm on a high refractive index waveguide. Using much larger molograms [>1 mm diameter] is not reasonable, since expressed proteins are valuable and often limited in biological experiments [39]. Besides the intensity that originates from coherently arranged molecules on the waveguide, various back- ground sources scatter intensity into the area of the fo- cal spot. This can either obscure the coherent signal or limit its accuracy due to the stochastic variation of the background. We analyzed the limit of resolution for the important case of molograms with massless affinity mod- ulations. For such molograms the signal of the empty mologram is hidden in the speckle background. Fattinger [16] provided a first estimation of the limit of detection by comparing the power diffracted by the mologram to the background power incident on the Airy disk. The back- ground power was estimated by distributing the waveg- uide radiation loss uniformly over the solid angle [4π]. While this serves as good first approximation, we now refine the approach. First, we distinguish between back- ground and noise. Whereas we describe the background as the mean intensity in the focal plane, we consider its spatial and/or temporal fluctuations as noise. Although the noise determines the limit of detection, it is worth- while to investigate the background because the ampli- tude of the noise is in a fixed ratio to the mean back- ground intensity. This is due to the nature of the speckle pattern [40], which will be explained in more detail be- low. The propagation loss [or attenuation] of a dielectric op- tical waveguide is an important quantity for its charac- terization. We evaluate the background with the help of the radiation loss as it has been performed in Ref. [16] for a first estimation of the limit of detection. However, two issues arise when approximating the background from the propagation loss. First, the attenuation constant is a sum of absorption and scattering loss α = αabs + αsca, where we define the propagation loss as Pwg (x) = Pwg (0) e−αx. The scattering loss provides additional background pho- tons to the area of the focal spot, whereas the absorption loss does not contribute any additional light. Therefore, determining the background intensity with the propaga- tion loss is only possible when the absorption is small compared to the scattering. The second issue when esti- mating the background from the attenuation arises from the anisotropy of the scattering. The out-coupled power is not distributed isotropically over the solid angle. In or- der to determine the intensity in the focal plane, an addi- tional parameter is needed to account for the anisotropy. This anisotropy parameter aani is explained in more de- tail in Fig. 6. The average intensity of the background in the focal plane Ibg can be conveniently written in terms of the scattering loss and the anisotropy parameter [de- tailed derivation in the SM Section VII A]. Ibg = NA2 4 aaniαscaPwg (5) 10 be explained. Any scattering is caused by an underlying stochastic refractive index distribution within the angles of the numerical aperture of the optical system. This stochastic process is transformed by the coherent illumi- nation to spatial intensity fluctuations in the focal plane - a speckle pattern. It is of utmost importance to distin- guish between dynamic and static scattering processes. The speckle pattern of a dynamic scattering process ex- hibits a timescale much shorter than the required band- width of the sensor. It will be averaged to a homoge- neous background. This background can be subtracted, which renders all dynamic scattering processes negligi- ble. Conversely, static scattering backgrounds lead to speckle patterns that are relatively stable over the time course of the measurement and generally unknown a pri- ori to the measurement. This generates an uncertainty when we determine the mass density on the mologram be- cause the relative contribution of the static background to the intensity of the molographic focus is unknown. As a side note, if the numerical aperture of the objective and the mologram match, background speckles and the molo- gram focus have the same size. Fortunately, the statistics of speckle patterns are well-known and speckles exhibit a negative exponential distribution of the intensity [40]. The 99.7 quantile [Definition of the LOD, generally stated as µ + 3σ] of the exponential distribution is always in a fixed ratio to the mean and therefore knowing the mean background allows estimating the noise and the limit of detection. Fig. 7(a) illustrates the scattering at molecules in so- lution in the evanescent field above the waveguide. If the distance between two molecules changes by roughly half a wavelength, the interference for a given speckle in the focal plane can switch from completely constructive to completely destructive. When comparing this short distance to the diffusivity of proteins [42] it is apparent that the measurement time [approximately 1 s] is sev- eral orders of magnitude longer than the diffusion time over these length scales. Therefore, the scattering of any molecule in solution is a dynamic process and does not contribute to the noise in the background. The scattering of randomly adsorbed proteins [Fig. 7(b)] on the waveguide surface has static and dynamic components. Most of these molecules adsorb reversibly to the low-energy non-fouling surface and therefore have affinities in the mM range. Interactions with such affini- ties exhibit short lifetimes [µs-ms] [43]. Therefore, the rate at which new proteins adsorb and desorb on the sur- face is much faster than the acquisition of a single data point. On the other hand, a minority [below 10 pg/mm2] [17] binds quasi irreversibly to incoherent surface defects present on any monolithic surface ad-layer [18]. These are the static components of nonspecific binding. How- ever, the contribution is extremely weak compared to the coherent signal. This can easily be understood if one re- calls that the coherent signal scales with the number of adsorbed particles squared whereas the nonspecific back- ground scales linearly with this number [16]. To give an Illustration of the anisotropy parameter aani. (a) The FIG. 6: scattered power is distributed isotropically in all directions. Only angles that can be collected by the numerical aperture [dashed lines] of the objective contribute to the background. This results in the expression for Ibg,iso. The intensity is then multiplied with aani to match the average measured intensity in the focal plane [see Fig. 19] (b) In reality, scattering is an anisotropic process. There are three effects that contribute to anisotropy. (i) The asymmetry of the waveguide leads to a stronger scattering into the substrate due to its higher optical density [29]. (ii) Forward scattering is usually dominant over backward scattering [41]. (iii) Scattering into guided modes of the waveguide is more efficient than into freely propagating modes [29]. We call the product aaniαsca the scattering leakage, since it refers to the light leaking into the focal plane of the mologram. Next, we aim to show which scattering mechanisms exist, how they contribute to the background and what needs to be considered when designing a waveguide for molographic sensing. We assume these mechanism to be non-correlated, which allows us to add the individual contributions such that (cid:88) aaniαsca = aani,iαsca,i (6) i where the total scattering leakage aaniαsca corresponds to the experimentally measured value [see SM Section VIII D]. Fig. 7 shows six possible scattering sources for background photons: (a)-(c) are scattering processes in- herent to the biosensing experiment and (d)-(e) depend on the waveguide manufacturing. We will now inves- tigate the importance of each mechanism qualitatively. If possible, we will treat the relevant sources quantita- tively. Before analyzing each scattering process individ- ually, the relation between background and noise shall 11 of the two waveguide sidewalls [Fig. 7(d)] has been stud- ied extensively and is not negligible [28, 45 -- 49]. In order to quantify the scattering we have adapted the analyti- cal formulas derived by Lacey and Payne [46, 50] for the more general case of an asymmetric waveguide [see SM Section VII B]. This model has the advantage of provid- ing an analytical solution for the attenuation constant [Eq. (75)]. However, more relevant for our analysis is the expression for the scattering leakage caused by the roughness of the waveguide. aani,rαsca,r = πns NA π(cid:0)n2 f − N 2(cid:1) ×(cid:16)(cid:0)n2 (cid:1)2 f − n2 teff (n2 c) f − n2 ns(cid:90) 2 + NA c π ×Lcβ π 2 − NA ns σ2 λ2N nf + J 2(cid:0)n2 f − n2 s (cid:1)2(cid:17) (7) 1 1 + ((β − kns cos θ) Lc)2 dθ FIG. 7: Possible sources of background intensity due to scat- tering (a) molecules in the solution (b) non-specifically bound molecules on the waveguide surface (c) large particles such as dust or cell bodies on the waveguide surface (d) waveguide surface roughness (e) refractive index inhomogeneities inside the waveguide (f) refractive index inhomogeneities inside the substrate example, the 10 pg/mm2 of irreversibly bound molecules upon serum exposure correspond to roughly 100 million molecules per mm2. The same signal intensity can be achieved with a coherent arrangement of the square root of this number, which is about 10,000 molecules per mm2. This corresponds to only 1 fg/mm2 of coherent matter or four orders of magnitude less than the molecular mass from irreversible nonspecific binding. This consideration exemplifies again the insensitivity of focal molography to nonspecific binding [7]. The value is so low that in most applications other background sources will be limiting. The scattering originating from large particles [see Fig. 7c] will be static or dynamic depending of the flow condi- tions. The analytical treatment of this mechanism is dif- ficult because such particles exceed the Rayleigh regime and the evanescent field of the waveguide. However, one can at least state that the large size of these scatter- ers causes strongly anisotropic forward scattering [Mie scattering regime [44]]. Besides the strong anisotropy, the influence of this scattering process can be reduced by controlling the number of adsorbed particles through careful handling of the chips and filtering of the samples prior to analysis. We will therefore not cover this scat- tering process in our theory since for most experiments it can be minimized to a negligible level [see Fig. 24]. The static scattering process at the surface roughness where J accounts for the waveguide asymmetry. J = cos tfβ (cid:32) (cid:0)n2 f − N 2(cid:1)1/2 (cid:32) (cid:1)1/2 (cid:0)N 2 − n2 f − N 2)1/2 (n2 sin N c (cid:33) tfβ + (8) (cid:0)n2 f − N 2(cid:1)1/2 (cid:33) N σ is the the RMS roughness and Lc the correlation length of the roughness. β = 2πN corresponds to the momen- λ tum of the mode in propagation direction and tf is the thickness of the waveguide. Eq. (7) can be used to esti- mate the background for different parameters. Yet, since it is difficult to determine a single correlation length from an AFM measurement the model should be applied care- fully. It can be improved by taking into account the full information of the power spectral density instead of as- suming an exponential decay of the autocorrelation. Yet, this model is less intuitive. In addition, the assumption of no correlation between the two sidewall roughnesses and the two-dimensionality of the model can result in some inaccuracy. Despite these simplifications, the de- scribed model is a valuable tool for the estimation of the background intensity from the roughness properties of the waveguide surface [see SM Section VIII E]. Further- more, it is important to notice that the scattering leakage caused by roughness aani,rαsca,r is fairly constant with re- spect to NA. Once characterized, this enables a straight- forward comparison between experiments of molographic system with different numerical apertures. The volume scattering due to refractive index inho- mogeneities in the waveguiding film [Fig. 7(e)] is a static process and amenable by similar theoretical investigation as the surface scattering [48]. However, the applicability of these models are limited since the characterization of the detailed distribution of those inhomogeneities by an orthogonal method is not trivial. For our waveguides the volume scattering is much smaller than the scattering from surface roughness of the waveguide sidewalls [SM section VIII B] and can therefore be neglected. Yet, this holds only true for thin waveguides with a high refractive index. Finally, the scattering from substrate inhomogeneities can be neglected because it is extremely low for a well- chosen material [Fig. 7(f)]. Based on the qualitative analysis we conclude that for a general waveguide one has to consider the volume and surface scattering of the waveguide for the background analysis aaniαsca = aani,rαsca,r + aani,vαsca,v. Which of these processes is prevailing is determined by the con- figuration of the waveguide. In general, the relative im- portance of volume scattering increases with waveguide thickness [see SM Section VIII F] On the other hand, the higher refractive index of the film, the larger is the in- dex contrast and the light intensity at the two waveg- uide sidewalls, which increases the contribution of sur- face scattering. Therefore, surface scattering is likely to be dominant for thinner waveguides with a high re- fractive index contrast [which is the case for the Ta2O5 waveguide discussed in this publication, see SM Section VIII F], whereas volume scattering will be limiting for thick waveguides with a low refractive index [high frac- tion of power in the waveguide and hardly any index contrast at the waveguide sidewalls]. Yet, also the signal scales with the waveguide prop- erties. Therefore, instead of minimizing the background one needs to maximize the signal to background ratio. We define a figure of merit for a waveguide in order to easily identify the relevant parameters for this optimiza- tion. More generally, a figure of merit of a molographic biosensor can be formulated. The figure of merit for focal molography is the ratio of signal to background intensity, which stands for the ratio of mass sensitivity to dark-field illumination quality. (cid:0)n2 f − N 2(cid:1) f − n2 nc N teff (n2 c) aaniαsca (9) FOMFM = D2 λ4 This expression was obtained by dividing equation (3) by (5). We omitted the protein related parameters and the analyte efficiency since for most applications these are fixed. By further excluding the diameter of the mologram [geometrical design parameter], one obtains the figure of merit of the waveguide FOMWG = λ4N teff (n2 c) aaniαsca (10) The dependency on wavelength to fourth power can be misleading. The choice of wavelength heavily affects the scattering leakage aaniαsca. Other parameters such as N and teff also depend on the wavelength. Therefore, one should be careful when making predictions from Eq. (10). Experimentally, aaniαsca can be determined by measur- ing the intensity in the focal plane and the power in the f − N 2(cid:1) (cid:0)n2 f − n2 nc 12 waveguide and applying Eq. (5). Alternatively, aaniαsca can be estimated from the measured roughness proper- ties with the help of Eq. (7), if volume scattering is negligible compared to surface roughness scattering [see SM Section VIII E]. The relative importance of volume to surface scattering can be investigated by measuring the ratio of the scattered intensity with two different cover media, because volume scattering will be hardly affected by the change in cover medium. We performed these characterization for our waveguide and found a scattering leakage of aani,rαsca,r = 3.12 /m in air and surface rough- ness scattering to be dominant over volume scattering [see SM Section VIII E]. From this we computed a figure of merit of our waveguide of FOMWG = 1.27 · 1030 /m4. The figure of merit for molography for a mologram of di- ameter 400 µm on this waveguide is FOMFM = 2 · 1023 /m2. In summary, the current high refractive index Ta2O5 waveguide is already a good choice for molography. Mainly thanks to the high field on the surface, which leads to a strong signal and also to its negligible vol- ume scattering. This compensates for the fairly large surface scattering due to the substantial index contrast. Still, since surface scattering is dominant, we expect that the waveguide could be further optimized. The options to reduce the surface scattering include to diminish the RMS of the surface roughness, which was determined with atomic force microscopy to be 0.6 nm for our waveg- uide [see SM Section VIII E], or to adapt the waveguide parameters using Eqs. (7) and (10). However, this has to be performed with care, since Eq. (7) only considers surface roughness scattering. For other waveguides, such as thick low refractive index waveguides, volume scatter- ing is most likely the dominant scattering source and the decrease in sensitivity for low refractive index waveguides is substantial. Nevertheless, once the scattering leakage is assessed experimentally, Eq. (10) allows a straightfor- ward comparison of different waveguides. IV. LIMITS OF DETECTION AND RESOLUTION After having described the possible sources of back- ground light and formulated the figure of merits we will use this knowledge to analyze the limits of detection and resolution of focal molography. The proper limit of de- tection for a specific assay is elaborate [51, 52]. Yet, it is not practically feasible to compare different sensing platforms at the assay level since this would require a standard assays to be performed with each of them. To establish the detection limit in endpoint measurements for molography, we define it as a confidence level for false positives, namely the 99.7 percentile [for the normal dis- tribution equal to µ + 3σ] or if not possible due to lack of experimental data points the 99 or 99.5 percentile. The resolution is a common benchmark parameter for sen- sors in general and for real-time label-free biomolecular interaction analysis in particular [53]. It is defined as the temporal rms [root mean square] noise of the base- line after drift correction for the duration of a typical biosensing experiment. Next, we need to find a suitable unit to compare these two quantities amongst mologra- phy and other biosensors. This is not straightforward, since most biosensors measure a change in ad-layer den- sity, yet, focal molography measures a change in ad-layer density modulation. In order to enable comparison, we propose the molographic surface mass density Γ to be this quantity for focal molography. It is defined as the entire mass in the mass modulation uniformly spread over the mologram [see SM Section I]. This quantity is calculated from the molographic signal intensity Isig normalized by a reference intensity. A suitable intensity reference for massless affinity modulations is the mean of the speckle background Ibg, since it is affected in the same man- ner as the molographic focus by the majority of phys- ical processes that cause noise or drift [see SM Section IX A for a discussion on intensity references]. Not in all bioanalytical questions it will be possible or simply too expensive to obtain completely massless affinity modula- tions [i.e. a small ligand interacting with a large immobi- lized protein]. In this case, a reference hologram will be required in order to calibrate the molographic signal in samples with varying bulk refractive index. This is not to be confused with the inherent self-referencing charac- ter of diffractometric sensors which makes them more ro- bust than referenced refractometric sensors. Contrary to these, diffractometric sensors only measure the refractive index difference between ridges and grooves. Therefore, they do not need to compensate for refractive index drifts in the entire volume of the evanescent field. Finally, we need to specify what the molographic signal Isig refers to exactly. So far, we discussed the average intensity in the Airy disk, which is related to the maximum by Imax = 4.378 · Iavg, as the potential molographic signal. Equation (11) and the remainder of this paper use a dif- ferent algorithm for the computation of the molographic signal which amounts to Isig = 2.012 · Iavg. Isig is the intensity on a single pixel of the focal plane image after convolution with a normalized Bessel kernel of the size of the expected Airy disk [see SM Section IX B]. The molo- graphic surface mass density Γ can directly be calculated from the ratio of the molographic signal and the mean intensity of the background as: (cid:115) Γ = Γ0 Isig Ibg (11) where Γ0 [see SM Section IX B for derivation] Γ0 = 0.1056 · A+ 1 Amologram ηmod[A] √ 1 FOMFM dn dc (12) can be seen as an equivalent molographic mass density [calculated from the intensity in the focal plane by using 13 the scattering strength of biological matter]. It is due to the stochastic variation of the significantly larger equiv- alent incoherent mass density of all the incoherent scat- terers. The equivalent molographic mass density would contribute the same amount of intensity as the scattering leakage to the focal plane of the mologram [average back- ground intensity in focal plane]. For the rest of this paper we assume a sinusoidal mass modulation [ηmod[A] = 0.5]. The limit of detection in terms of molographic surface mass density for a figure of merit of a given molographic system can be obtained by replacing Isig with Isig,LOD and Γ with ΓLOD in Eq. (11). The minimal detectable normalized intensity increase ∆Isig is determined by the Ibg readout scheme. We will now distinguish between two readout schemes: Endpoint detection and real-time mea- surements. We will derive the limit of detection and the limit of resolution for the two schemes,respectively. We also provide experimental evidence for our statements as well as the theoretical projection of the optimization po- tential. A. The limit of detection of endpoint measurements is determined by the statistics of the speckle background In an endpoint measurement the operator has usually no a priori knowledge of the intensity distribution of the speckles and the location of the focal spot [except the focal distance to the surface]. Therefore, the Airy disk needs a certain brightness compared to the background intensity to be detectable in a sufficiently large field of view. The detection limit is determined by the variation over many images of the ratio of the maximum pixel value to the image mean. In order to experimentally determine this value 180 images of size 280 x 210 µm2 with 110 nm pixel size on three different ZeptoMark [Zeptosens AG, Switzerland] chips were acquired and convoluted as de- scribed in SM section IX. The 99.5 % quantile of this ratio equals to 13.8 [see Fig. 23]. Hence, if the maximum pixel after convolution is 13.8 times brighter than the mean of the speckle background it is likely to stem from a coher- ent binding signal. By inserting this into Eq. (11), one obtains a detection limit in terms of molographic surface mass density of 2.6 pg/mm2 for a 400 µm mologram with 0.4 NA on our waveguide, which has a figure of merit of 1.27·1030 /m4. [Fig. 8(a)]. Therefore, only 336 fg of mat- ter yield a signal that is clearly assignable to a coherent assembly of molecules. Fig. 8(c) displays a contour plot of the two parameters that affect the detection limit of a molographic system - the diameter of the mologram and the figure of merit of the waveguide. An increase of either of these two parameters decreases the detection limit. It should be mentioned that the endpoint measurements with no a priori knowledge of the speckle intensities rep- resents an upper bound of the achievable detection limit of diffraction-limited molography. Any readout scheme of 14 FIG. 8: Detection limit of focal molography without pre-knowledge of the position of the focal spot for the standard configuration in air [diameter mologram 400 µm, focal length 900 µm, numerical aperture of the mologram 0.33 and numerical aperture of the objective 0.4]. (a) Detection limit for the standard configuration mologram. 180 background images [280 x 210 µm with 110 nm pixel size] focused 100 µm below the surface of the waveguide of three clean chips with a scattering leakage of 3.12 /m [Figs. 18 and 19] were acquired, filtered with the shape of the Airy disk [NA = 0.33] and the maximum pixel of the convoluted image extracted and summarized in the box plot (i). All measured intensity values were normalized to a standard power of 0.02 W/m in the waveguide. The solid grey line corresponds to the 99 % percentile of the maximum pixels observed in the Airy disk convoluted background images defining the smallest intensity required for the coherent signal to be discriminated against surface roughness speckles. The solid blue line represents the scattered intensity as a function of the molographic mass density [sinusoidal mass distribution and calculated from Eq. (3) with the mass modulation replaced by the molographic mass density]. The intersection point is indicated by the dashed grey line, which denotes the coherent mass that corresponds the 99 % percentile of the measured maximum background intensity for a field of view as specified above. The box plot (ii) corresponds to 12 measured biotin molograms in air. (b) Typical focal plane image of easily detectable biotin molographic foci with intensities roughly 10 times above the detection limit. (c) Detection limit in terms of molographic surface mass density for an ideal sinusoidal mass modulation as a function of the figure of merit of the waveguide [FOMWG] and the diameter of the mologram. The detection limit decreases inversely with the diameter and decreases inversely to the square root of the FOMWG. The dashed grey line indicates the FOMWG of the investigated waveguide. greater sophistication will achieve lower detection limits. Endpoint detection of vitamin B7 [biotin] In this section we demonstrate experimentally that molography can visualize a label-free mass modulation caused by a low molecular weight compound in an end- point measurement. Molograms with the sole difference between grooves and ridges being the tiny molecule vi- tamin B7 [molecular weight 227 g/mol] were fabricated [NH-biotinNH2]. A chip was illuminated with a dose of 2000 mJ/cm2, incubated with 1 mM sNHS-biotin in HBS-T buffer [10 mM HEPES, 150 mM NaCl, 0.05 % Tween20] at pH 8.0 for 15 min and flood exposed as de- scribed in [7]. The foci of these biotin molograms were easily detectable [Fig. 8(b)]. To prove that indeed bi- otin molograms and not just random speckles were mea- sured, a real-time video of the focal point was recorded while the grooves were backfilled with biotin as described in the description of Movie 1. One can clearly see the intensity gradually decreasing until the focal spot be- comes indistinguishable from the speckle background. The molographic mass density [sinusoidal modulation] calculated from the median intensity via Eq. (11) of 12 molograms amounted to 11 pg/mm2 ± 1.6 pg/mm2 [FOMWG = 0.58 · 1030 /m4, calculated from the attenu- ation constant assuming the same aani]. The uncertainty is caused by an intrinsic property of speckle statistics. Every speckle has a non-negative intensity. However, the sign of the electric field of a background speckle can be positive or negative and is unknown, since the phase in- formation cannot be measured. If the field of the speckle is negative with respect to the field of the molographic signal some additional coherent matter is required to can- cel the contribution from the roughness. Vice versa, less biological mass is required in the case when the molo- graphic focus happens to be on a positive speckle. This physical property poses an intrinsic constraint on the ac- curacy of the molographic measurement and therefore the LOQ [limit of quantification]. It has to be stressed that the detection of this low molecular weight compound was possible without any equilibration, referencing or stabilization of the sensor. The amount of bound mass was determined in a repro- ducible fashion with an accuracy below 2 pg/mm2 even if the chip is removed and reinserted. This robustness is a fundamental difference to common refractometric sen- sors, where similar detection limits can only be achieved with samples that are mounted and stabilized within the device and cannot be removed and reinserted. B. The resolution of real-time measurements is determined by the temporal noise of the speckle background In real-time detection the intensity in the focal plane is continuously monitored. Contrary to the above de- scribed end-point measurement, the location of the focal spot in the speckle background is known exactly. This could be realized by reference focal spots or by localizing the spot before backfilling as shown below. Then im- age processing can be applied to monitor the intensity at the location of the focal spot resulting in a binding trace as a function of time. Such a binding curve is the output of all real-time biosensors and the detection limit is commonly stated as the temporal RMS noise [resolu- tion] over a defined time span, typically a few minutes, of the signal before the analyte is injected. It should be noticed that this is a different definition than the 99.5 % quantile described above. The goal of this section is to demonstrate the resolution of diffraction-limited fo- cal molography theoretically and experimentally for our measurement system and compare it to SPR, the gold standard of refractometric sensing. We start our discussion by appreciating the instrumen- tal precision at which refractometric sensors are operated in order to achieve refractive index resolutions of 10−6- 10−7 and mass resolutions of 30 fg/mm2 - 1 pg/mm2 [53, 54]. If one recalls that one mono-layer of water molecules already gives rise to a signal of 300 pg/mm2 one can appreciate of this technological achievement. Only careful optimization of sensor design, referencing strate- gies and signal processing over the past three decades made this possible [55]. The performance of real-time measurement devices is commonly characterized by two metrics, namely the baseline drift and the baseline noise. The former is expressed in pg/[mm2min] or RU/min [RU = response units] whereas the latter is expressed as an RMS value in RU or pg/mm2. Nowadays, commer- cial SPR instruments achieve a baseline noise of 15-30 fg/mm2 [measured as RMS value after drift correction] 15 FIG. 9: Comparison of the resolution of focal molography to state of the art resolution of SPR. (a) Best and routine resolution of SPR instruments is indicated. Blue squares are the resolution measured for focal molography. The dark blue line is the sensitivity of molography for a FOMFM of 1023 /m2. (b) label-free detection of 500 pM [26 ng/ml] of SAv and baseline noise levels. (c) Zoom in of the first part of the binding curve for better visualization of the baseline noise. and a baseline drift of around 300 fg/[mm2/min] [27]. Over the course of a measurement SPR sensors are usu- ally limited by temperature drifts between reference and sensing channel [56]. As it will be shown below, such drifts are virtually not present in diffractometric sensors. Therefore we will compare molography to idealized SPR instruments which are limited by the baseline noise. The three most common readout modes of SPR are angle interrogation, wavelength interrogation and inten- sity interrogation [10]. Independent of the interrogation mode, the readout of the SPR signal is a measurement of a relative intensity Ireflected/Iin. The noise in the in- tensity results in a noise of the detected surface mass density, which determines the resolution of the technique [53]. The resolution can be calculated from the intensity noise and the sensitivity. RMSΓ−(cid:104)Γ(cid:105) = RMSR−(cid:104)R(cid:105) SΓ (13) The sensitivity of SPR is stated in SM Section IX C. As described in Eq. (11) molography also measures a rela- tive intensity [Isig/Ibg] such that Eq. (13) is valid. For molography, the sensitivity can be described by [see SM Section IX D for derivation] ∝(cid:112) SΓFM = 2 Γ0 FOMFM (14) Due to the quadratic nature of the sensor transfer func- tion Eq. (14) is only valid if the signal intensity is close to the reference intensity [in this case the background in- tensity]. It must be stated here that the values for the sensitivity of molography and SPR should not be com- pared since they depend on the chosen reference intensity [Ibg and Iin]. Instead the resolution [RMSΓ−(cid:104)Γ(cid:105)] can be used for comparison. With the MoloReader an intensity baseline noise of about 10 % has been measured resulting in a resolution of 90 fg/mm2 [0.9 RU] for the waveguides used in this experiment [FOM = 0.63 · 1030 /m4]. This is close to the above mentioned 30 fg/mm2 [0.3 RU] res- olution of the best SPR sensors [54]. If the waveguide background is the intensity reference, Eq. (14) can be used to improve the resolution of molography, which can be achieved by a higher FOMWG or by a larger diameter of the mologram. Further, any reduction of the intensity noise will also significantly improve it. In order to verify these findings real-time measure- ments were performed with the MoloReader [The setup is illustrated in Fig. 26.]. First, it has to be stressed that all reported experimental results are without any kind of temperature stabilization. The sole effect of tempera- ture in a molographic measurement is a slow drift of the location of the molographic spot within the focal plane, but its intensity is hardly affected by the temperature drift [Movie 2]. This movement can easily be compen- sated for by simple image registration algorithms, which was implemented in the readout algorithm [Fig. 27] [57]. As mentioned above, for a continuous measurement, the Airy disk must to be in the field of view and the optical system needs to be focused on the focal plane [see SM Section X B for the protocol to accomplish this]. Buffer baselines were acquired at a flow rate of 20 µl/min for 20 mins with a syringe pump [NE-511L, PumpSystems Inc.] and 1 ml syringes [Henke Sass Wolf GmbH]. The buffer solutions were degassed prior to use, to avoid noise caused by micro bubble formation. The exposure time of the camera was set to 500 ms and an image was acquired every 3 seconds. After 20 minutes, the syringe was ex- changed by another one containing 500 pM SAv in PBS-T [spikes in the binding trace around 22 min]. Injection was continued at a flow rate of 20 µl/min. Finally, image pro- cessing was performed in order to obtain the baselines as described in SM Section X C [Fig. 27]. 16 Three binding curves were acquired and are displayed in Fig. 9(b). The signal change due to SAv binding was detectable almost instantaneously after injection. How- ever, whereas in two measurements the binding trace rose immediately after injection, it decreased at first in the third measurement. This is an example of the molo- graphic focus lying on a background speckle with a neg- ative electric field with respect to the focus itself, as ex- plained before. The baseline noise over 20 min amounted to RMS values of 0.074, 0.094, 0.077 RU and in terms of normalized intensity 7.1, 15.6, 7.8 %. These experi- mental noise values nicely agree with the theoretical pre- diction from Eqs. (13) and (14). As mentioned before, the mass density resolution is comparable to the best reported SPR results [54]. Yet, the molographic base- lines were calculated without any baseline drift correc- tion unlike the common practice in refractometric sens- ing demonstrating the robustness and sensitivity of focal molography. Another fundamental requirement in label-free inter- action analysis is the ability to detect a distributed en- semble of molecules on a sufficiently large area. In other words, to detect low receptor [capture molecule] occu- pancies, a fact overlooked by most of todays nanosensing and single molecular detection concepts [58]. In sensitive assays the concentration of the analyte is usually sev- eral orders of magnitude lower [10 fM - 1 pM] [59] than the dissociation constant of the capture probe [10 pM - 1 nM] [60], which leads to a receptor occupancy of typically 0.1-1% [58]. The molographic focus of a 400 µm diameter mologram monitors the activity of roughly 1 billion recog- nition sites continuously and is therefore able to resolve low receptor occupancies as well as measuring a sufficient number of analyte molecules. For example, at the demon- strated resolution of 100 fg/mm2 [roughly 1 million SAv molecules per mm2] 100000 proteins are bound to one bi- otin mologram [receptor density of 3 · 1010 molecules per mm2 [11 pg/mm2]]. Furthermore, taking into account that four biotin molecules bind one SAv molecule the re- ceptor occupancy in the experiments shown in Fig. 9 can be estimated to amount to only 0.01 %. V. CONCLUSIONS In molography, the coherent arrangement of binding sites in the mologram and the resulting insensitivity to non-coherent noise sources enables robust and highly sen- sitive detection of biomolecular interactions. A quantita- tive analysis of these interactions is amenable through the analytical models presented in this paper. These models compute the amount of biological matter bound to the mologram from the intensity of the molographic focus. Their accuracy is proven by the excellent agree- ment with the presented numerical simulations and the discussed experiments. High sensitivity and a low back- ground are achieved by a waveguide providing field en- hancement and a proper dark field illumination. How- 17 Appendix B: Movie descriptions Movie 1: Real-time backfilling of biotin molograms This movie shows the real-time backfilling of a biotin mologram [NH-biotinNH2] with 1 mM sNHS-biotin at pH 8.0 in HBS-T buffer. The molographic spot fades away upon biotin binding because also the grooves are functionalized with biotin, essentially canceling the mass modulation [NH-biotinNH-biotin]. This proves that our investigated molograms in Fig. 8 were indeed biotin molograms. Movie 2: Temperature effect on the molographic spot This video shows the influence of temperature on the speckles in the focal plane image. The speckles as well as the molographic spot shift as a function of temperature but their intensity essentially remains constant. The chip was observed in PBS-T buffer without any flow and the entire chip assembly [Fig. 27(b)] was taken from the fridge before the measurement to induce a more visible temperature drift. This drift can easily be compensated by means of image registration. Movie 3: Real-time binding of 500 pM SAv to [NH- biotinNH-PEG] molograms. The movie shows the real- time binding of 500 pM SAv in PBS-T pH 7.4 [0.05 % Tween20] buffer to a [NH-biotinNH-PEG12] mologram. The movie corresponds to the light blue curve in Fig. 9(b). ever, radiation due to scattering at waveguide imperfec- tions remains the dominant source of background light for massless affinity modulations. Therefore, figures of merits were introduced to investigate the parameter de- pendencies of the signal to background ratio. They allow straightforward comparison of different molographic ar- rangements and waveguides. Two readout schemes, end- point detection and real-time measurements prove the intrinsic robustness and high sensitivity of focal molog- raphy. In an endpoint measurement, the low molecular weight compound vitamin B7 could be easily detected and the limit of detection in terms of surface mass was just a few pg/mm2 by this simple readout scheme. The more elaborate real-time measurements exhibited a res- olution below 100 fg/mm2 over 20 min without any drift correction. This is comparable to the best commercially available refractometric sensors. With further optimiza- tion, it is therefore likely that the resolution of diffrac- tometric sensors will surpass the one of refractometric devices. Yet, by only detecting the coherent signal, the coherent detection scheme has unmet advantages over any established label-free biosensor. Its unique combi- nation of robustness and high sensitivity will enable nu- merous new applications to analyze the interactions of biomolecules in their natural habitat - the crowded envi- ronments of body fluids, tissues, cells and membranes. Acknowledgments We would like to thank Lukas Novotny [ETH] for valu- able input of the theoretical treatment of dipoles near interfaces; Arens Winfried [IMT Masken und Teilungen AG, Greifensee] for the fabrication of the SiO2 covered chips; Louis Palavi [ETH] for programming the alpha ver- sion of the reader control software in course of his intern- ship; René Rietmann [Roche] for machining the fluidic parts and Stephen Wheeler [ETH] for construction and machining of the reader stage system; and finally, Roland Dreyfus [ETH] for numerous discussions on coupled mode theory. Appendix A: Author contributions A.F., Y.B., J.V. and C.Fattinger planned the exper- iments, which were performed and evaluated by A.F. A.F and C.Fattinger designed the instrumentation. Y.B. and A.F derived the analytical expressions. S.B. and A.F implemented the simulation framework which was adapted for GPU computing by C.Forró. M.L. conducted the AFM measurements and V.G. provided the founda- tions for many experiments. A.F. and Y.B. wrote the manuscript with C.Fattinger providing input for the con- tent and the structure of the work. 18 [1] N. Mohammad, M. Meem, B. Shen, P. Wang, and [24] Z. Lai, Y. Wang, N. Allbritton, G.-P. Li, and M. Bach- R. Menon, Sci. Rep. 8, 2799 (2018). [2] J. T. Early, R. Hyde, and R. L. Baron, in UV/Optical/IR Space Telescopes: Innovative Technologies and Concepts (International Society for Optics and Photonics, 2004), vol. 5166, pp. 148 -- 157. [3] Y. Park, L. Koch, K. D. Song, S. Park, G. King, and S. Choi, J. Opt. A: Pure Appl. Opt. 10, 095301 (2008). [4] R. S. Rodrigues Ribeiro, P. Dahal, A. Guerreiro, P. A. S. Jorge, and J. Viegas, Sci. Rep. 7, 4485 (2017). [5] I. Mohacsi, P. Karvinen, I. Vartiainen, V. A. Guzenko, A. Somogyi, C. M. Kewish, P. Mercere, and C. David, J. Synchrotron Radiat. 21, 497 (2014). [6] B. A. Palmer, G. J. Taylor, V. Brumfeld, D. Gur, M. Shemesh, N. Elad, A. Osherov, D. Oron, S. Weiner, and L. Addadi, Science 358, 1172 (2017). [7] V. Gatterdam, A. Frutiger, K.-P. Stengele, D. Heindl, T. Lübbers, J. Vörös, and C. Fattinger, Nat. Nanotech- nol. 12, 1089 (2017). [8] Â. Serrano, S. Zürcher, S. Tosatti, and N. D. Spencer, Macromol. Rapid Commun. 37, 622 (2016). [9] S. Fraser, J. Y. Shih, M. Ware, E. O'Connor, M. J. Cameron, M. Schwickart, X. Zhao, and K. Regnstrom, AAPS J. 19, 682 (2017). [10] Homola, Surface Plasmon Resonance Based Sensors, vol. 4 of Springer Series on Chemical Sensors and Biosensors (Springer Berlin Heidelberg, Berlin, Heidel- berg, 2006). [11] P. Kozma, F. Kehl, E. Ehrentreich-Förster, C. Stamm, and F. F. Bier, Biosensors and Bioelectronics 58, 287 (2014). [12] M. J. Cannon, G. A. Papalia, I. Navratilova, R. J. Fisher, L. R. Roberts, K. M. Worthy, A. G. Stephen, G. R. Marchesini, E. J. Collins, D. Casper, et al., Anal. Biochem. 330, 98 (2004). [13] P. Kozma, A. Hamori, K. Cottier, S. Kurunczi, and R. Horvath, Appl. Phys. B 97, 5 (2009). man, Opt. Lett. 33, 1735 (2008). [25] H. Kogelnik, The Bell System Technical Journal 48 [26] I. Poole, Basic Radio: Principles and Technology (1969). (Newnes, 1998). [27] Selection guide biacore systems, https://proteins. gelifesciences.com/, accessed: 2017-12-1. [28] D. Marcuse, Theory of Dielectric Optical Waveguides (Academic Press, 1974). [29] L. Novotny and B. Hecht, Principles of Nano-Optics (Cambridge University Press, 2011). [30] J. W. Goodman, Introduction to Fourier Optics (Roberts and Company Publishers, 2005). [31] D. Sinclair, J. Opt. Soc. Am. 37, 475 (1947). [32] S. Monneret, P. Huguet-Chantôme, and F. Flory, J. Opt. A: Pure Appl. Opt. 2, 188 (2000). [33] T. Tamir and S. T. Peng, J. Phys. D Appl. Phys. 14, 235 [34] H. Zhao, P. H. Brown, and P. Schuck, Biophys. J. 100, (1977). 2309 (2011). [35] R. Magnusson and T. K. Gaylord, J. Opt. Soc. Am., JOSA 68, 806 (1978). [36] W. Heller, J. Phys. Chem. 69, 1123 (1965). [37] H. Fischer, I. Polikarpov, and A. F. Craievich, Protein Sci. 13, 2825 (2004). [38] W. Lukosz, J. Opt. Soc. Am., JOSA 71, 744 (1981). [39] Structural Genomics Consortium, China Structural Ge- nomics Consortium, Northeast Structural Genomics Consortium, S. Gräslund, P. Nordlund, J. Weigelt, B. M. Hallberg, J. Bray, O. Gileadi, S. Knapp, et al., Nat. Methods 5, 135 (2008). [40] J. C. Dainty, in Progress in Optics, edited by E. Wolf (Elsevier, 1977), vol. 14, pp. 1 -- 46. [41] S. Miyanaga, T. Asakura, and M. Imai, Opt. Quantum Electron. 12, 23 (1980). [42] L. Zhang, K. Dammann, S. C. Bae, and S. Granick, Soft [14] M. Piliarik, M. Bocková, and J. Homola, Biosens. Bio- Matter 3, 551 (2007). electron. 26, 1656 (2010). [15] M. Piliarik and V. Sandoghdar, Nat. Commun. 5, 4495 (2014). [16] C. Fattinger, Phys. Rev. X 4, 031024 (2014). [17] S. Pasche, S. M. De Paul, J. Voros, N. D. Spencer, and M. Textor, Langmuir 19, 9216 (2003). [43] S.-Q. Liu, X.-L. Ji, Y. Tao, D.-Y. Tan, K.-Q. Zhang, and Y.-X. Fu, in Protein engineering (InTech, 2012). [44] D. W. Hahn, Department of Mechanical and Aerospace Engineering, University of Florida (2009). [45] F. Ladouceur and J. D. Love, Silica-based buried channel waveguides and devices (Chapman & Hall, 1996). [18] J. C. Love, L. A. Estroff, J. K. Kriebel, R. G. Nuzzo, and [46] F. P. Payne and J. P. R. Lacey, Opt. Quantum Electron. G. M. Whitesides, Chem. Rev. 105, 1103 (2005). 26, 977 (1994). [19] D. Falconnet, A. Koenig, F. Assi, and M. Textor, Adv. [47] G. Ames and D. Hall, IEEE J. Quantum Electron. 19, Funct. Mater. 14, 749 (2004). [20] N. Vigneswaran, F. Samsuri, B. Ranganathan, and Padmapriya, Procedia Engineering 97, 1387 (2014). [21] M. Avella-Oliver, J. Carrascosa, R. Puchades, and Á. Maquieira, Anal. Chem. 89, 9002 (2017). [22] S. Cleverley, I. Chen, and J.-F. Houle, J. Chromatogr. B Analyt. Technol. Biomed. Life Sci. 878, 264 (2010). [23] S. Lenhert, F. Brinkmann, T. Laue, S. Walheim, C. Van- nahme, S. Klinkhammer, M. Xu, S. Sekula, T. Mappes, T. Schimmel, et al., Nat. Nanotechnol. 5, 275 (2010). 845 (1983). (1992). [48] S. Miyanaga, T. Asakura, and M. Imai, Opt. Quantum Electron. 11, 205 (1979). [49] R. Ronald Louis, Ph.D. thesis, The University of Arizona [50] J. Lacey and F. P. Payne, IEE Proceedings J- [51] D. A. Armbruster and T. Pry, Clin. Biochem. Rev. 29 Optoelectronics (1990). Suppl 1, S49 (2008). [52] C. A. Holstein, M. Griffin, J. Hong, and P. D. Sampson, Anal. Chem. 87, 9795 (2015). [53] J. Homola, Chem. Rev. 108, 462 (2008). [54] M. Piliarik and J. Homola, Opt. Express 17, 16505 (2009). [55] R. B. M. Schasfoort, ed., Handbook of Surface Plasmon Resonance (The Royal Society of Chemistry, 2017). [56] A. A. Kolomenskii, P. D. Gershon, and H. A. Schuessler, Appl. Opt. 36, 6539 (1997). [57] P. Thevenaz, U. E. Ruttimann, and M. Unser, IEEE Trans. Image Process. 7, 27 (1998). [58] T. M. Squires, R. J. Messinger, and S. R. Manalis, Nat. 19 Biotechnol. 26, 417 (2008). [59] S. Surinova, R. Schiess, R. Hüttenhain, F. Cerciello, B. Wollscheid, and R. Aebersold, J. Proteome Res. 10, 5 (2011). [60] J. P. Landry, Y. Ke, G.-L. Yu, and X. D. Zhu, J. Im- munol. Methods 417, 86 (2015). [61] An error in Eq. (2) of Ref. [24] should be noted: Λ does not refer to the grating period but to the wavelength of the n=-1 wave inside the grating region in the direction normal to the waveguide surface as described in Ref. [33].
1101.1092	1	1101	2011-01-05T21:02:23	Comparative hydrodynamics of bacterial polymorphism	[ "physics.bio-ph", "cond-mat.soft", "physics.flu-dyn" ]	Most bacteria swim through fluids by rotating helical flagella which can take one of twelve distinct polymorphic shapes. The most common helical waveform is the "normal" form, used during forward swimming runs. To shed light on the prevalence of the normal form in locomotion, we gather all available experimental measurements of the various polymorphic forms and compute their intrinsic hydrodynamic efficiencies. The normal helical form is found to be the most hydrodynamically efficient of the twelve polymorphic forms by a significant margin - a conclusion valid for both the peritrichous and polar flagellar families, and robust to a change in the effective flagellum diameter or length. The hydrodynamic optimality of the normal polymorph suggests that, although energetic costs of locomotion are small for bacteria, fluid mechanical forces may have played a significant role in the evolution of the flagellum.	physics.bio-ph	physics	Comparative hydrodynamics of bacterial polymorphism Saverio E. Spagnolie∗ and Eric Lauga† Department of Mechanical and Aerospace Engineering, University of California San Diego, 9500 Gilman Drive, La Jolla CA 92093-0411. (Dated: October 27, 2018) Most bacteria swim through fluids by rotating helical flagella which can take one of twelve distinct polymorphic shapes. The most common helical waveform is the "normal" form, used during forward swimming runs. To shed light on the prevalence of the normal form in locomotion, we gather all available experimental measurements of the various polymor- phic forms and compute their intrinsic hydrodynamic efficiencies. The normal helical form is found to be the most hydrodynamically efficient of the twelve polymorphic forms by a significant margin -- a conclusion valid for both the peritrichous and polar flagellar families, and robust to a change in the effective flagellum diameter or length. The hydrodynamic op- timality of the normal polymorph suggests that, although energetic costs of locomotion are small for bacteria, fluid mechanical forces may have played a significant role in the evolution of the flagellum. PACS numbers: 47.63.-b, 47.63.Gd, 87.17.Jj, 87.23.Kg The shapes and sizes of life in all its diversity are ever changing as form meets function, intimately tuned to nature's diverse environments. Bacteria evolved to swim through fluids by rotating a single helical flagellum ("monotrichous", or polar, bacteria), or in the case of such organisms as Salmonella and Escherichia coli, several rotating helical flagella emanating from their cell membranes ("peritrichous" bacteria). Each flagellum is assembled through the polymerization of a flagellin protein, and has been met with great interest both in and outside the scientific community due to its astoundingly complex construction [1]. Due to the various possible arrangements of polymerized flagellin, it has been postulated that the flagellar filaments can take only twelve distinct polymorphic forms [2 -- 4], of which nine have been characterized experimentally [5] (Fig. 1a). The most common helical waveform is the left-handed "normal" form, used during forward swimming "runs." Upon counterclockwise (CCW [when viewed from the flagellum's distal end]) co-rotation of the flagella by rotary motors, a flagellar bundle forms behind peritrichous bacteria, driving fluid backward and propelling the cell forward. To change their swimming directions, these bacteria undergo "tumbling" events. As shown in Fig. 1b-e, a quick direction reversal to clockwise (CW) motor rotation produces a twisting torque which temporarily transforms the associated individual flagellum from a left-handed normal form to a right-handed "semi-coiled" form, leading to an unwinding of the bundle and a change in cell orientation, followed by a transition to a right-handed "curly" form which persists until the next reversal in motor direction [6 -- 8]. The other forms are not generally used for locomotion. Mechanical stresses, such as the twisting and viscous torques present during swimming, are not the only means by which the flagellar shape might shift from one waveform to another. Filaments can also transform reversibly in response to amino acid replacements, chemical or temperature changes, or the addition of alcohols or sugars [10 -- 17]. Other authors have considered the elastic rigidity of different polymorphs and its relationship to shape selection [7, 18 -- 20]. The motion of a helical body through a viscous fluid has seen 2 extensive theoretical treatment [21 -- 25]. In this letter we present a physical rationalization of the prevalence of the normal polymorphic form in bacterial swimming. We gather all available experimental measurements of the various polymorphic flagellar waveforms [26] along with the twelve theoretical forms [3], and compute the intrinsic hydrodynamic efficiency of each geometry. We show that the normal form is the most hydrodynamically efficient of the twelve polymorphic forms by a significant margin, a result true for both peritrichous and monotrichous (polar) flagellar families. This conclusion is robust as the flagellum length is varied, or its effective diameter is increased to represent a bundle of flagella. The hydrodynamic optimality of the normal helical form therefore suggests a role for fluid mechanical forces in the evolution of the flagellum. We begin with a short description of the hydrodynamics of swimming bacteria. At the exceedingly small length and velocity scales on which bacteria swim, viscous dissipation overwhelms any inertial effects, and the fluid motion is accurately described by the Stokes equations [27 -- 29]. In this regime, there is a linear relation between the net forces and torques on an immersed flagellum, (F, N), and its associated translational and rotational velocities, (U, ω) (rigid body motion is assumed). Consider a rotating helix driving a cell body, as is the case for the swimming runs of flagellated bacteria. In this case, the net forces and torques on the rotating flagellum (or flagella) must balance those of the fluid on the body. Assuming that the cell is axisymmetric about x and swims directly along this axis, we write the body's translational (swimming) velocity as U = U x and its rotational velocity as Ω = Ω x. The corresponding fluid force and torque on the cell body are denoted by F = −A0U x and N = −D0Ω x, respectively. A linear mobility relation for the flagellum may then be written as (cid:32) (cid:33)(cid:32) (cid:33) (cid:32) (cid:33) A B C D U ω = −A0U −D0Ω , (1) FIG. 1 (a) All twelve theoretical peritrichous polymorphic waveforms, including two straight forms [3]; left-handed (resp. right-handed) helices are denoted by filled (resp. empty) symbols. (b) One flagellum of an E. coli cell displays a normal waveform; (c) semi-coiled; (d) curly; (e) normal again. Adapted with permission from Turner, L., Ryu, W.S., and Berg, H.C., J. Bacteriol., 182 2793 (2000). Copyright c(cid:13) (2000), American Society for Microbiology [9]. 02040608000.20.40.60.81 NormalCurlySemi-CoiledCoiledCurlyII02040608000.20.40.60.81 02040608000.20.40.60.81 02040608000.20.40.60.81 02040608000.20.40.60.81 02040608000.20.40.60.81 02040608000.20.40.60.81 02040608000.20.40.60.81 02040608000.20.40.60.81 02040608000.20.40.60.81 BothMotAandMotBspanthecytoplasmicmembrane.MotAhasfourmembrane-spanning␣-helicalsegments(54 -- 56).Therestofthemolecule(abouttwo-thirds)isinthecytoplasm.MotBhasonemembrane-spanning␣-helicalFigure3AnE.colicellwithoneflagellarfilament,visualizedbyfluorescencemicroscopy.Therecordingwasmadeat60Hz,butonlyeveryotherfieldisshown.Thenumbersareinunitsof1/60s.WhenthemotorswitchedfromCCWtoCWafterfield2,thefilamentchangeditsshapefromnormaltosemicoiled,10,andthentocurly1,e.g.,20.WhenthemotorswitchedbacktoCCWafterfield26,thefilamentrelaxedbacktonormal,30.Initially,thecellswamtoward7o'clock.Afterthenormaltosemicoiledtransformation,itswamtoward5o'clock.Flagellarfilamentscanalsobevisualizedbydark-fieldorinterference-contrastmicroscopy(259,260),butfluorescencehastheadvantagethatonecanseethefilamentsallthewaytothesurfaceofthecellwithreasonabledepthoffield.(FromReference3,Figure6.)27THEROTARYMOTOROFBACTERIALFLAGELLAAnnu. Rev. Biochem. 2003.72:19-54. Downloaded from arjournals.annualreviews.orgby University of California - San Diego on 07/02/10. For personal use only.BothMotAandMotBspanthecytoplasmicmembrane.MotAhasfourmembrane-spanning␣-helicalsegments(54 -- 56).Therestofthemolecule(abouttwo-thirds)isinthecytoplasm.MotBhasonemembrane-spanning␣-helicalFigure3AnE.colicellwithoneflagellarfilament,visualizedbyfluorescencemicroscopy.Therecordingwasmadeat60Hz,butonlyeveryotherfieldisshown.Thenumbersareinunitsof1/60s.WhenthemotorswitchedfromCCWtoCWafterfield2,thefilamentchangeditsshapefromnormaltosemicoiled,10,andthentocurly1,e.g.,20.WhenthemotorswitchedbacktoCCWafterfield26,thefilamentrelaxedbacktonormal,30.Initially,thecellswamtoward7o'clock.Afterthenormaltosemicoiledtransformation,itswamtoward5o'clock.Flagellarfilamentscanalsobevisualizedbydark-fieldorinterference-contrastmicroscopy(259,260),butfluorescencehastheadvantagethatonecanseethefilamentsallthewaytothesurfaceofthecellwithreasonabledepthoffield.(FromReference3,Figure6.)27THEROTARYMOTOROFBACTERIALFLAGELLAAnnu. Rev. Biochem. 2003.72:19-54. Downloaded from arjournals.annualreviews.orgby University of California - San Diego on 07/02/10. For personal use only.BothMotAandMotBspanthecytoplasmicmembrane.MotAhasfourmembrane-spanning␣-helicalsegments(54 -- 56).Therestofthemolecule(abouttwo-thirds)isinthecytoplasm.MotBhasonemembrane-spanning␣-helicalFigure3AnE.colicellwithoneflagellarfilament,visualizedbyfluorescencemicroscopy.Therecordingwasmadeat60Hz,butonlyeveryotherfieldisshown.Thenumbersareinunitsof1/60s.WhenthemotorswitchedfromCCWtoCWafterfield2,thefilamentchangeditsshapefromnormaltosemicoiled,10,andthentocurly1,e.g.,20.WhenthemotorswitchedbacktoCCWafterfield26,thefilamentrelaxedbacktonormal,30.Initially,thecellswamtoward7o'clock.Afterthenormaltosemicoiledtransformation,itswamtoward5o'clock.Flagellarfilamentscanalsobevisualizedbydark-fieldorinterference-contrastmicroscopy(259,260),butfluorescencehastheadvantagethatonecanseethefilamentsallthewaytothesurfaceofthecellwithreasonabledepthoffield.(FromReference3,Figure6.)27THEROTARYMOTOROFBACTERIALFLAGELLAAnnu. Rev. Biochem. 2003.72:19-54. Downloaded from arjournals.annualreviews.orgby University of California - San Diego on 07/02/10. For personal use only.BothMotAandMotBspanthecytoplasmicmembrane.MotAhasfourmembrane-spanning␣-helicalsegments(54 -- 56).Therestofthemolecule(abouttwo-thirds)isinthecytoplasm.MotBhasonemembrane-spanning␣-helicalFigure3AnE.colicellwithoneflagellarfilament,visualizedbyfluorescencemicroscopy.Therecordingwasmadeat60Hz,butonlyeveryotherfieldisshown.Thenumbersareinunitsof1/60s.WhenthemotorswitchedfromCCWtoCWafterfield2,thefilamentchangeditsshapefromnormaltosemicoiled,10,andthentocurly1,e.g.,20.WhenthemotorswitchedbacktoCCWafterfield26,thefilamentrelaxedbacktonormal,30.Initially,thecellswamtoward7o'clock.Afterthenormaltosemicoiledtransformation,itswamtoward5o'clock.Flagellarfilamentscanalsobevisualizedbydark-fieldorinterference-contrastmicroscopy(259,260),butfluorescencehastheadvantagethatonecanseethefilamentsallthewaytothesurfaceofthecellwithreasonabledepthoffield.(FromReference3,Figure6.)27THEROTARYMOTOROFBACTERIALFLAGELLAAnnu. Rev. Biochem. 2003.72:19-54. Downloaded from arjournals.annualreviews.orgby University of California - San Diego on 07/02/10. For personal use only.BothMotAandMotBspanthecytoplasmicmembrane.MotAhasfourmembrane-spanning␣-helicalsegments(54 -- 56).Therestofthemolecule(abouttwo-thirds)isinthecytoplasm.MotBhasonemembrane-spanning␣-helicalFigure3AnE.colicellwithoneflagellarfilament,visualizedbyfluorescencemicroscopy.Therecordingwasmadeat60Hz,butonlyeveryotherfieldisshown.Thenumbersareinunitsof1/60s.WhenthemotorswitchedfromCCWtoCWafterfield2,thefilamentchangeditsshapefromnormaltosemicoiled,10,andthentocurly1,e.g.,20.WhenthemotorswitchedbacktoCCWafterfield26,thefilamentrelaxedbacktonormal,30.Initially,thecellswamtoward7o'clock.Afterthenormaltosemicoiledtransformation,itswamtoward5o'clock.Flagellarfilamentscanalsobevisualizedbydark-fieldorinterference-contrastmicroscopy(259,260),butfluorescencehastheadvantagethatonecanseethefilamentsallthewaytothesurfaceofthecellwithreasonabledepthoffield.(FromReference3,Figure6.)27THEROTARYMOTOROFBACTERIALFLAGELLAAnnu. Rev. Biochem. 2003.72:19-54. Downloaded from arjournals.annualreviews.orgby University of California - San Diego on 07/02/10. For personal use only.BothMotAandMotBspanthecytoplasmicmembrane.MotAhasfourmembrane-spanning␣-helicalsegments(54 -- 56).Therestofthemolecule(abouttwo-thirds)isinthecytoplasm.MotBhasonemembrane-spanning␣-helicalFigure3AnE.colicellwithoneflagellarfilament,visualizedbyfluorescencemicroscopy.Therecordingwasmadeat60Hz,butonlyeveryotherfieldisshown.Thenumbersareinunitsof1/60s.WhenthemotorswitchedfromCCWtoCWafterfield2,thefilamentchangeditsshapefromnormaltosemicoiled,10,andthentocurly1,e.g.,20.WhenthemotorswitchedbacktoCCWafterfield26,thefilamentrelaxedbacktonormal,30.Initially,thecellswamtoward7o'clock.Afterthenormaltosemicoiledtransformation,itswamtoward5o'clock.Flagellarfilamentscanalsobevisualizedbydark-fieldorinterference-contrastmicroscopy(259,260),butfluorescencehastheadvantagethatonecanseethefilamentsallthewaytothesurfaceofthecellwithreasonabledepthoffield.(FromReference3,Figure6.)27THEROTARYMOTOROFBACTERIALFLAGELLAAnnu. Rev. Biochem. 2003.72:19-54. Downloaded from arjournals.annualreviews.orgby University of California - San Diego on 07/02/10. For personal use only.2µm(b)(c)(d)(e)(a)BothMotAandMotBspanthecytoplasmicmembrane.MotAhasfourmembrane-spanning␣-helicalsegments(54 -- 56).Therestofthemolecule(abouttwo-thirds)isinthecytoplasm.MotBhasonemembrane-spanning␣-helicalFigure3AnE.colicellwithoneflagellarfilament,visualizedbyfluorescencemicroscopy.Therecordingwasmadeat60Hz,butonlyeveryotherfieldisshown.Thenumbersareinunitsof1/60s.WhenthemotorswitchedfromCCWtoCWafterfield2,thefilamentchangeditsshapefromnormaltosemicoiled,10,andthentocurly1,e.g.,20.WhenthemotorswitchedbacktoCCWafterfield26,thefilamentrelaxedbacktonormal,30.Initially,thecellswamtoward7o'clock.Afterthenormaltosemicoiledtransformation,itswamtoward5o'clock.Flagellarfilamentscanalsobevisualizedbydark-fieldorinterference-contrastmicroscopy(259,260),butfluorescencehastheadvantagethatonecanseethefilamentsallthewaytothesurfaceofthecellwithreasonabledepthoffield.(FromReference3,Figure6.)27THEROTARYMOTOROFBACTERIALFLAGELLAAnnu. Rev. Biochem. 2003.72:19-54. Downloaded from arjournals.annualreviews.orgby University of California - San Diego on 07/02/10. For personal use only.3 where we have written the translational and rotational velocities of each point on the flagellum as U = U x and ω = ω x, and neglected hydrodynamic interactions between the flagellum and the body. It can be shown that C = B [27, 30]. Torque balance requires that the body rotation rate Ω and the flagellar rotation rate ω are oppositely signed, so that the cell body counter-rotates with respect to the motion of the flagellum. The rotary motor at the base of the flagellum attached to the cell body therefore rotates with angular speed Ωm = ω − Ω. In order to compare the performance of various polymorphic forms, a hydrodynamic efficiency E∗ is now defined following the work of Purcell [30]. The power output of the motor, N Ωm, is compared to the least power that would be required to move the cell body at speed U by any means of propulsion, namely A0 U 2, and so E∗ = (A0 U 2)/(N Ωm). Expressions for U and N in terms of the rotation rate Ωm may be deduced from Eq. (1), and various approximations valid for the relative length and velocity scales observed in swimming bacteria may be made (for example B2 (cid:28) AD, and ω (cid:29) Ω) [30]. Assuming the ability to rescale the propeller dimensions, for a given cell body the maximum value of the swimming efficiency can then be found to be given by E = B2/(4AD); E is the intrinsic propeller efficiency, and is a function of its shape alone [30, 31]. Note that the expression for E could also be reached using a dimensional approach, as B indicates the correlation between motor torque and forward swimming, while A and D are indicative of the fluid friction (via Eq. 1); the ratio above (or factors thereof) are the only such dimensionally proper arrangements. To determine the intrinsic efficiency E for a given experimentally measured or theoretical waveform, we need only compute the three coefficients A,B and D. To do so accurately, we perform computations using a non-local slender body theory for viscous flows [32, 33]. We consider a single rigid flagellar filament of length L and circular cross-section of radius L r(s), where r(s) is dimensionless, (cid:28) 1 is the aspect ratio of the flagellum ( (cid:46) 10−2 for bacteria), and s ∈ [0, L] is the arc-length parameter. For a given translational velocity U x and rotational velocity ω x about a point x0, the fluid force f (s) on the filament is given implicitly via 8πµ[U x + ω x × (x(s) − x0)] = −Λ[f (s)] − K[f (s(cid:48))](s), where µ is the shear viscosity of the fluid, x(s) denotes the centerline position at a station s, and K[f (s(cid:48))](s) = (I + ss) (cid:90) L 0 Λ[f ](s) = [c(I + ss) + 2(I − ss)] f (s), I + R R R(s(cid:48), s) − f (s(cid:48)) − f (s) s(cid:48) − s ds(cid:48) + 0 (cid:32) (cid:90) L (cid:33) I + ss s(cid:48) − s f (s(cid:48)) ds(cid:48), (2) (3) (4) where c = − ln(2e), R(s(cid:48), s) = x(s(cid:48)) − x(s), R = R/R, s is the local unit tangent vector at the point s, and ss is a dyadic product [33, 34]. Henceforth x0 is set at the origin. In order to obtain numerically the distribution of forces, f (s), accurately to order 2, it is required that r(s) decays no slower than O(√s) 4s(L − s)/L as in Ref. [34]. The near the filament endpoints, and we have chosen for simplicity r(s) = flagellum diameter d at the midpoint s = L/2 is 2 L. The waveforms considered are modeled as perfect (πD)2 + P 2, helices with centerlines x(s) = P K s x + (D/2) [sin(2πKs) y + cos(2πKs) z], with K = 1/ P the pitch, and D the helical diameter. (cid:112) (cid:112) 4 FIG. 2 A normal waveform undergoes pure CCW rotation about the major helical axis x, with (P, D) = (2.3 µm, 0.4 µm), L = 10 µm, and d = 20 nm. (a) Velocity vectors through a cross section at s = L/2. A dark arrow indicates both the direction of rotation of the flagellar filament, as well as the location on the filament which intersects the cross-sectional plane. (b) The lengthwise velocity u (the fluid velocity through the cross-sectional plane), normalized by the velocity of the flagellum in the cross-sectional plane, ω D/2. We solve Eq. (2) numerically for f (s) using a Galerkin method [35], in which f (s) is written as a finite sum of Legendre polynomials, and Eq. (2) is required to hold under inner products against the same basis functions. The first integral in the operator K[f ] is diagonalized in this space [34, 36]. With f (s) in hand, we define F (cid:48) = x· 0 (x(s)− x0)× f (s) ds. Then, setting (U, ω) = (1, 0) we recover A = F (cid:48); setting (U, ω) = (0, 1) we recover B = F (cid:48) and D = N(cid:48). Based on the mathematical accuracy of the method, we estimate that the numerical errors in computing the fluid flow and efficiency calculations for a specified geometry are below 0.1% of the reported values. (cid:82) L 0 f (s) ds and N(cid:48) = x· (cid:82) L The velocity field, u(x), at a point x in the fluid can be recovered using the representation f (s(cid:48)) ds(cid:48), (5) (cid:32) (cid:90) L 8πµ u(x) = − 0 (cid:33) I + R R R(s(cid:48)) + 2 2 I − 3 R R R(s(cid:48))3 where now R(s(cid:48)) = x − x(s(cid:48)) [34, 36]. We show in Fig. 2a the velocity field so computed through a cross section of a normal flagellar waveform which is undergoing pure CCW rotation at rate ω. The flow is primarily restricted to the plane, rotating along with the flagellum (due to the no-slip condition there), and decaying in magnitude away from the intersection point. There is a small lengthwise fluid motion through this plane, so that fluid is slowly shuttled backward along the axis of rotation. This lengthwise velocity u is displayed in Fig. 2b, normalized by ωD/2; it is zero at the flagellum boundary (due to the no-slip condition), and increases to a maximum of u ≈ 0.2 ωD/2 on the circular helical perimeter approximately opposite the point where the flagellum intersects the vertical plane. For each of the experimentally measured waveforms reported in the literature [26] and the theoretically predicted waveforms [3], we compute the intrinsic efficiency E using the method described above. In each case we assume a flagellum length L = 10 µm and diameter d = 20 nm. Figure 3a compiles the efficiency results, further detailed in Fig. 3b-e, overlaid upon efficiency contours in the circumference-pitch (C-P) plane (C = πD). Different symbols represent the various polymorphic forms (see Fig. 1), and experimental data show averages ± one standard deviation, with peritrichous (resp. polar) data in black (b)(a)0.05−0.05−0.050.05−0.0500.05−0.05−0.04−0.03−0.02−0.0100.010.020.030.040.05y/Lz/L00.20.4x/L−0.050.050.4uωD/2y/L0.35 FIG 3. (a) Efficiency contours in the circumference-pitch (C-P) plane (P < 0 for left-handed helices, as in diagram), assuming a flagellum diameter (resp. length) of 20 nm (resp. 10 µm), combining the information from Fig. 3b-e. The two large circles distinguish the peritrichous and monotrichous (polar) flagellar families, dashed for P > 0. Data points and bars indicate the mean computed efficiency ± one standard deviation for the peritrichous (black), polar (blue), and theoretical (red, from Ref. [3]) waveforms. Dotted lines indicate the curves C = ±P . (b) Waveform geometries from experimental data for the peritrichous flagella (see Tables S1-S4 in the supplementary material and symbols from Fig. 1). Each color represents a different data set. (c) Hydrodynamic efficiencies for each of the peritrichous waveforms as a function of the mean pitch angle, as in (a). Two curves indicate the efficiencies measured continuously along the large circle in the C-P plane in (a); the dashed curve again corresponds to P > 0 (right-handed helices), and the solid curve to P < 0 (left-handed helices). (d,e) Same as in (b,c), but for the polar flagellar family [37]. The normal form in each family is the most hydrodynamically efficient of the twelve polymorphic forms by a significant margin. (resp. blue). As the helix becomes infinitely large (or as the filament becomes infinitesimally slender), Eq. (2) returns E = y2/(8y4 + 20y2 + 8), with y = P/C. In this limiting case, the efficiency-maximizing geometry has C = P (pitch angle ψ = 45◦), indicated in Fig. 3a by dotted lines, and E = 2.8%. However, at the biologically relevant length scales and aspect ratio as studied here, for a given pitch P the optimal geometry has C ≈ (7/8)P (pitch angle ψ ≈ 40◦). We plot in Fig. 3b the geometrical data in the (C-P) plane which allows the different members of the peritrichous flagellar family to be distinguished [37]. Each color represents a different data set [26]. The mean hydrodynamic efficiencies (± one standard deviation) of flagellar polymorphs in the peritrichous family are shown in Fig. 3c as a function of the average helical pitch angle, (cid:104)ψ(cid:105), for measured (black) and theoretically predicted (red) waveforms; the numerical values of the efficiencies for each waveform are noted in the supplementary material. The normal waveform is found to be the most hydrodynamically efficient of the twelve helical forms by a significant margin (with (cid:104)E(cid:105) = 0.96%) over 23% more efficient than the next most efficient forms, the curly and semi-coiled waveforms (which are both used by bacteria during change-of-orientation events [6 -- 8]). Two curves indicate the efficiencies measured along the large circle in the C-P plane in (a); the larger efficiencies are achieved along this circle when P < 0. The leftward skew of the theoretical C-P relationship is thus seen to play an important role in the left-handed normal 304050607023456789101112x 10−3Cψ90000.012E70300.012E0459000.0020.0040.0060.0080.010.0120.002(c)(b)−0.2−0.100.10.200.10.20.30.40.50P[µm]C[µm]5−32−0.2−0.100.10.200.10.20.30.40.50C[µm]5−32(d)(e)Polar(Mean±1std)30\x{FFFF}ψ\x{FFFF}[◦]0.002−4−3−2−1012340.511.522.533.544.55C[µm] 0.511.522.533.540.511.522.533.5−0.01−0.00500.0050.010.0150.0200.015(a)E05432101234−4−2−1−3Ptan(ψ)=CPNormalNormalP[µm]P[µm]Normal\x{FFFF}ψ\x{FFFF}[◦]\x{FFFF}ψ\x{FFFF}[◦]Peritrichous(Mean±1std)Peritrichous(Theory-Calladine)6 FIG. 4 (a) Change in efficiency for the peritrichous family by varying the flagellar diameter from d = 20 nm to 40 nm. Greater percentage-wise gains in efficiency is obtained for thicker propellers. (b) Varying the flagellar length L. The efficiency ordering remains nearly the same through the biologically relevant length scales. (c) The mean efficiencies of the normal, coiled, semi-coiled, curly, and curly II waveforms as functions of the mean pitch angle for three different lengths. (d) Same as (b), but for the polar flagellar family. form being more efficient than its right-handed counterparts. A different flagellar family can be distinguished by examining the circumference-pitch curve for different measurements, and is shown in Fig. 3d. These are monotrichous (polar) flagella, assembled from a different flagellin protein than peritrichous flagella, but which follow a similar polymorphic sequence of twelve forms [37]. Similarly to the peritrichous family, the normal form is the most hydrodynamically efficient one (with (cid:104)E(cid:105) = 1.03%; Fig. 3e), a 25% increase over the next most efficient shape, a right-handed curly waveform. To address the robustness of our results against geometrical variations, we changed both the flagellum diameter and length in our computations. We show in Fig. 4a the mean efficiency computed for the peritrichous waveforms as a function of the flagellum diameter, as a model for the increased effective filament size of flagellar bundles. The efficiency decreases steadily as the filament size increases, but the efficiency of each polymorph decays at a similar rate, and thus the efficiency ordering from Fig. 3c is unchanged. The greatest percentage benefit in efficiency when using the normal form is found when the flagellar diameter is large, e.g. for bundles of many flagella. Varying the flagellar length also shows that the efficiency ordering is not modified, as shown in Figs. 4b-c, and the greatest percentage increase in the efficiency of the normal form is achieved for longer filaments. We also changed the lengths used in the computations for the polar flagellar family, with the results shown in Fig. 4d, again showing no change in order. For both peritrichous and polar flagellar families, the efficiency orderings shown in Figs. 3c-e are therefore robust throughout the biologically relevant parameter space. Finally, we find that less accurate resistive force theories, which linearly relate body velocities to fluid forces, and are the most widely used approaches for modeling slender bodies in fluids [38, 39], do not 2040608000.0020.0040.0060.0080.010.0124812164681012x 10−348121600.0020.0040.0060.0080.010.01220242832364000.0020.0040.0060.0080.010.012(a)(b)(c)L=10µmL[µm]\x{FFFF}E\x{FFFF}\x{FFFF}E\x{FFFF}\x{FFFF}E\x{FFFF}L=5µmL=15µmL=10µm(d)L[µm]\x{FFFF}E\x{FFFF}d=20nmd=20nmd=20nmd[nm]0.0120900.01200.01200.0120.003204020416416\x{FFFF}ψ\x{FFFF}[◦]7 predict the efficiency ordering found using the full non-local hydrodynamics (see also Refs. [24, 40]) [26]. Hydrodynamic interactions between different parts of the helical propeller are thus essential in order to conclude on the relative efficiencies of flagellar polymorphs. In conclusion, by examining all available experimental data on the geometry of bacterial flagella, we found that both peritrichous and monotrichous bacteria employ, among the discrete number of available flagellar shapes, the hydrodynamically optimal polymorph in order to swim in viscous fluids. In contrast to simple estimates showing that locomotion accounts for a negligible portion of a bacterium's metabolic costs [41], our results suggest that fluid mechanical forces may have played a significant role in the evolution of the flagellum We thank H. C. Berg for discussions, and permission to reproduce the figure from Ref. [9]. We ac- knowledge the support of the NSF through grant CBET-0746285. Supplementary material I. Experimental data and computed efficiencies Table S1 shows a compilation of measurements from studies on various strains of Salmonella ty- phimurium, along with the data sources and the colors used to create Fig. 3b in the main text. Here we have reproduced the measured helical pitch P [µm] and the helical diameter D [µm] from the cited sources in the form (P, D). Table S2 contains similar measurements obtained for the organism Escherichia coli, also included in Fig. 3b. Fujii et al. [37] have considered measurements of a large number of organisms along with their different polymorphic measurements, which we report below as Table S3. These authors have detected different flagellar families corresponding to peritrichous (Family I), monotrichous (or polar) (Family II), lateral (Family III), and some exceptional flagellar filaments; these families are distinguished in the table, and the color schemes match those used to create Figs. 3(b,d). Family I flagellin, Family II flagellin, and Family III flagellin each lead to different circles in the circumference-pitch (C-P) plane, the first two of which are shown in Figs. 3(a,b,d). Table S4 shows the computed data for all the possible theo- rized waveforms from Calladine's model [3], and from a theoretical calculation performed by Hasegawa et al. [4], which we have included in red in Fig. 3c to show the negligible efficiencies of the more uncommon theoretical polymorphs; note that the unnamed polymorphic forms in Fig. 1a have been obtained in a laboratory setting [5], but are not generally observed in nature. Finally, Table S5 indicates the numerical values of the efficiencies plotted in Figs. 3(c,e). 8 TABLE S1. Waveform measurements of the form (helical pitch P [µm], helical diameter D [µm]) for Salmonella from the following sources (organism strain noted in parentheses if reported): a - Kamiya & Asakura (strain SJ670) [11], b - Kamiya & Asakura (SJ25) [11], c - Kamiya & Asakura (SJ30) [11], d - Darnton & Berg (SJW1103) [7], e - Iino (SW577) [42], f - Iino & Mitani (SJ30) [43], g - Hotani [15], h - Iino, Oguchi & Kuroiwa [44], i - Macnab & Ornston [12], j - Asakura [2], k - Fujii, Shabata & Aizawa [37]. Colors correspond to those in Fig. 3b. TABLE S2. Waveform measurements (P [µm], D [µm]) for E. coli.: a - Turner, Ryu & Berg [45], b - Matsuura, Kamiya & Asakura [46], c - Fujii, Shabata & Aizawa [37]. Colors correspond to those in Fig. 3b. 11.21.41.61.822.22.42.62.8311.21.41.61.822.22.42.62.83 data1data2data3data400.511.5200.20.40.60.811.21.41.61.82 00.511.5200.20.40.60.811.21.41.61.82 Source2(Normal)3(Coiled)4(Semi-Coiled)5(CurlyI)6(CurlyII)a(2.30,0.42)(0.53,1.16)(1.15,0.31)(0.89,0.2)b(2.28,0.38)(0.9,1.4)(1.16,0.51)(1.1,0.3)(0.9,0.2)c(0.93,0.32)(0.90,0.18)d(2.17,0.42)(0.79,1.06)(1.07,0.6)e(2.2,0.5)(1.1,0.4)f(2.2,0.53)(0.91,0.29)g(2.3,0.45)(0.69,1.52)(1.24,0.52)(1.14,0.3)h(2.2,0.5)(1.10,0.36)i(2.3,0.45)j(2.49,0.34)(1.36,0.23)(1.08,0.13)k(2.55,0.6)(1.29,0.5)(1.20,0.2)(1.00,0.15)WaveformMeasurements(P,D)forSalmonellafromthefollowingsources(organismstrainnotedinparenthesesifreported).Pisthehelicalpitch,andDisthehelicaldiameter.a-Kamiya&Asakura(strainSJ670)[13],b-Kamiya&Asakura(SJ25)[13],c-Kamiya&Asakura(SJ30)[13],d-Darnton&Berg(SJW1103)[22],e-Iino(SW577)[33],f-Iino&Mitani(SJ30)[6],g-Hotani[19]andWashizuHotani?,h-Iino,Oguchi&Kuroiwa[9],i-Macnab&Ornston[18],j-Asakura[1],k-Fujii,Shabata&Aizawa[31].III.FLUID-BODYINTERACTIONANDHYDRODYNAMICEFFICIENCYBacteriaswimatexceedinglysmalllengthandvelocityscales.TakingtheflagellumlengthLasacharacteristiclengthscale,andthewavespeedCasacharacteristicvelocityscale,aReynoldsnumbercharacterizingthefluidmotionisRe=CL/ν\x{FFFF}1,withν=10−2cm2/sthedynamicviscosityofthesurroundingfluid[37].Hencetheviscousdissipationvastlyoverwhelmsanyinertialeffects,andthefluidmotioniswell-capturedbysolvingtheStokesequations,∇·σ=0,∇·u=0,withuthefluidvelocity,σ=−pI+µ(∇u+∇uT)theNewtonianstresstensor,µthekinematicviscosity,andpthepressure(see[37 -- 39]).A.SlenderbodytheoryInordertocomputethemaximalhydrodynamicefficiencyforthevariouspolymorphicwaveforms,wesolveaslenderbodyapproximationtotheStokesequations.Specifically,weconsiderthefluid-bodyinteractionofasingleflagellarfilamentoflengthLandcircularcross-sectionofradius\x{FFFF}Lr(s),where\x{FFFF}\x{FFFF}1andr(s)aredimensionless,ands∈[0,L]isthearc-lengthparameter.Thenon-localslenderbodytheoryofJohnson[30]isemployed,whichgivesaveryaccuraterepresentationofthefluidvelocityeverywherewitherrorO(\x{FFFF}2).Namely,therigidbodyvelocityUandrotationalvelocityωaboutapointx0aregivenas47Source2(Normal)4(Semi-Coiled)5(CurlyI)6(CurlyII)a(2.3,0.35)(1.1,0.5)(1.0,0.25)(0.9,0.16)b(2.2,0.41)c(2.2,0.41)(0.92,0.261)TABLEII.WaveformdataforE.coli.a-Turner,Ryu&Berg[10],b-Matsuura,Kamiya&Asakura[11],c-Fujii,Shabata&Aizawa[9].Flagellum2(Normal)3(Coiled)4(Semi-Coiled)5(CurlyI)6(CurlyII)S.typhurium(2.55,0.6)(1.29,0.5)(1.20,0.2)(1.00,0.15)E.coli(2.2,0.41)(0.92,0.261)E.cartovora(2.13,0.57)(0.76,0.169)Y.enterocolitica(2.55,0.55)(1.04,0.3)P.mirabilis(1.83,0.47)(0.87,0.35)B.subtilis(2.06,0.42)(0.91,0.18)E.faecalis(2.40,0.5)(1.11,0.15)I.loihiensis(1.32,0.33)(1.18,0.48)(0.93,0.25)P.aeruginosa(1.38,0.392)(1.0,0.57)(0.92,0.22)P.syringae(1.59,0.43)(1.04,0.82)(1.49,0.668)(0.75,0.229)X.axonopodis(1.41,0.392)(1.42,0.69)V.para-haemolyticus(1.21,0.239)B.japonicumpof(0.79,0.62)(1.16,0.36)A.brasilensepof(1.08,0.341)V.para-haemolyticus(0.47,0.11)B.japonicumlaf(0.71,0.2)A.brasilenselaf(0.66,0.2)S.meliloti(2.27,0.57)(0.61,0.44)(0.49,0.29)R.sphaeroides(2.04,0.44)(0.78,0.2)C.crescentus(0.85,0.261)TABLEIII.WaveformData(P,D)fromFujii,Shabata&Aizawa[9].Source1(Hyperextended)2(Normal)3(Coiled)4(Semi-Coiled)5(CurlyI)a(1.9715,0.48)(0.2686,0.92)(1.08,0.55)(1.0945,0.29)b(1.576,0.08)(2.233,0.392)(0.901,1.026)(1.429,0.7)(1.361,0.302)6(CurlyII)78910a(0.7893,0.093)(0.6736,0.0514)(0.5805,0.027)(0.5046,0.0104)(0.9336,0.161)b(0.881,0.088)(0.733,0.048)(0.621,0.024)(0.535,0.01)(1.087,0.16)TABLEIV.a-Calladine(?),b-Hasegawaetal.(computed).11.21.41.61.822.22.42.62.8311.21.41.61.822.22.42.62.83 data1data2data3data49 TABLE S3. Waveform measurements (P [µm], D [µm]) from Fujii, Shibata & Aizawa [37], for organisms in the peritrichous, polar, and lateral flagellar families, along with a few exceptions. For polar, lateral, and exceptional flagellar families, the "Normal" form refers to small-Normal and very-small-Normal forms (see Ref. [37]). Colors correspond to those in Fig. 3b (peritrichous) and Fig. 3d (monotrichous, or polar). TABLE S4. Theoretical waveform data (P [µm], D [µm]) from: a - Calladine [3], and b - Hasegawa et al. [4]. Colors correspond to those in Fig. 3b. 11.21.41.61.822.22.42.62.8311.21.41.61.822.22.42.62.83 data1data2data3data4data5data6data7Organism2(Normal)3(Coiled)4(Semi-Coiled)5(CurlyI)6(CurlyII)PeritrichousS.typhurium(2.55,0.6)(0,1.0)(1.29,0.5)(1.20,0.2)(1.00,0.15)E.coli(2.2,0.41)(0,1.51)(0.92,0.261)E.cartovora(2.13,0.57)(0,1.20)(0.76,0.169)Y.enterocolitica(2.55,0.55)(1,1.20)(1.04,0.3)P.mirabilis(1.83,0.47)(0,1,27)(0.87,0.35)B.subtilis(2.06,0.42)(0,1.17)(0.91,0.18)E.faecalis(2.40,0.5)(0,1.25)(1.11,0.15)PolarI.loihiensis(1.32,0.33)(0,0.69)(1.18,0.48)(0.93,0.25)P.aeruginosa(1.38,0.392)(0,0.95)(1.0,0.57)(0.92,0.22)P.syringae(1.59,0.43)(1.04,0.82)(1.49,0.668)(0.75,0.229)X.axonopodis(1.41,0.392)(0,0.86)(1.42,0.69)V.para-haemolyticus(1.21,0.239)B.japonicumpof(0,1.06)(0.79,0.62)(1.16,0.36)A.brasilensepof(1.08,0.341)LateralV.para-haemolyticuslaf(0.47,0.11)B.japonicumlaf(0.71,0.2)A.brasilenselaf(0.66,0.2)ExceptionsS.meliloti(2.27,0.57)(0,1.05)(0.61,0.44)(0.49,0.29)R.sphaeroides(0,1.17)(2.04,0.44)(0.78,0.2)C.crescentus(0.85,0.261)WaveformData(P,D)fromFujii,Shibata&Aizawa[31].FamilyI:Peritrichous,FamilyII:Polar,FamilyIII:Lateral.FamilyIflagellin,FamilyIIflagellin,FamilyIIIflagellineachleadtodifferentP-Dcircularradiiandhencedescribeddifferentfamilies.WTFWITHCOILED????!!!III.FLUID-BODYINTERACTIONANDHYDRODYNAMICEFFICIENCYBacteriaswimatexceedinglysmalllengthandvelocityscales.TakingtheflagellumlengthLasacharacteristiclengthscale,andthewavespeedCasacharacteristicvelocityscale,aReynoldsnumbercharacterizingthefluidmotionisRe=CL/ν\x{FFFF}1,withν=10−2cm2/sthedynamicviscosityofthesurroundingfluid[37].Hencetheviscousdissipationvastlyoverwhelmsanyinertialeffects,andthefluidmotioniswell-capturedbysolvingtheStokesequations,∇·σ=0,∇·u=0,withuthefluidvelocity,σ=−pI+µ(∇u+∇uT)theNewtonianstresstensor,µthekinematicviscosity,andpthepressure(see[37 -- 39]).611.21.41.61.822.22.42.62.8311.21.41.61.822.22.42.62.83 data1data2data3data4data5data611.21.41.61.822.22.42.62.8311.21.41.61.822.22.42.62.83 data1data2data3data47Source2(Normal)4(Semi-Coiled)5(CurlyI)6(CurlyII)a(2.3,0.35)(1.1,0.5)(1.0,0.25)(0.9,0.16)b(2.2,0.41)c(2.2,0.41)(0.92,0.261)TABLEII.WaveformdataforE.coli.a-Turner,Ryu&Berg[10],b-Matsuura,Kamiya&Asakura[11],c-Fujii,Shabata&Aizawa[9].Organism2(Normal)3(Coiled)4(Semi-Coiled)5(CurlyI)6(CurlyII)S.typhurium(2.55,0.6)(1.29,0.5)(1.20,0.2)(1.00,0.15)E.coli(2.2,0.41)(0.92,0.261)E.cartovora(2.13,0.57)(0.76,0.169)Y.enterocolitica(2.55,0.55)(1.04,0.3)P.mirabilis(1.83,0.47)(0.87,0.35)B.subtilis(2.06,0.42)(0.91,0.18)E.faecalis(2.40,0.5)(1.11,0.15)I.loihiensis(1.32,0.33)(1.18,0.48)(0.93,0.25)P.aeruginosa(1.38,0.392)(1.0,0.57)(0.92,0.22)P.syringae(1.59,0.43)(1.04,0.82)(1.49,0.668)(0.75,0.229)X.axonopodis(1.41,0.392)(1.42,0.69)V.para-haemolyticus(1.21,0.239)B.japonicumpof(0.79,0.62)(1.16,0.36)A.brasilensepof(1.08,0.341)V.para-haemolyticus(0.47,0.11)B.japonicumlaf(0.71,0.2)A.brasilenselaf(0.66,0.2)S.meliloti(2.27,0.57)(0.61,0.44)(0.49,0.29)R.sphaeroides(2.04,0.44)(0.78,0.2)C.crescentus(0.85,0.261)TABLEIII.WaveformData(P,D)fromFujii,Shabata&Aizawa[9].Source1(Hyperextended)2(Normal)3(Coiled)4(Semi-Coiled)5(CurlyI)a(1.9715,0.48)(0.2686,0.92)(1.08,0.55)(1.0945,0.29)b(1.576,0.08)(2.233,0.392)(0.901,1.026)(1.429,0.7)(1.361,0.302)6(CurlyII)78910a(0.7893,0.093)(0.6736,0.0514)(0.5805,0.027)(0.5046,0.0104)(0.9336,0.161)b(0.881,0.088)(0.733,0.048)(0.621,0.024)(0.535,0.01)(1.087,0.16)TABLEIV.a-Calladine(?),b-Hasegawaetal.(computed).11.21.41.61.822.22.42.62.8311.21.41.61.822.22.42.62.83 data1data2data3data410 Peritrichous 1 (Hyper-extended) (cid:104)ψ(cid:105) ± σψ [degrees] 9.1 2 (Normal) 3 (Coiled) 4 (Semi-Coiled) 5 (Curly) 6 (Curly II) 7 8 9 10 Polar (Normal) (Coiled) (Semi-Coiled) (Curly) 32.4 ± 4.5 79.6 ± 3.9 55.4 ± 3.3 39.0 ± 7.7 28.4 ± 5.0 20.3 13.5 8.3 3.7 38.6 ± 4.0 68.0 58.4 ± 6.3 41.7 ± 3.2 (cid:104)E(cid:105) ± σE (measured) 9.6 · 10−3 ± 1.3 · 10−3 9.3 · 10−4 ± 6.9 · 10−4 7.8 · 10−3 ± 1.0 · 10−3 7.8 · 10−3 ± 1.1 · 10−3 5.9 · 10−3 ± 1.0 · 10−3 3.9 · 10−3 1.03 · 10−2 ± 1.0 · 10−3 7.0 · 10−3 ± 2.3 · 10−3 8.2 · 10−3 ± 0.7 · 10−3 E (theoretical) [3] 1.1 · 10−3 1.06 · 10−2 7.5 · 10−4 6.9 · 10−3 8.6 · 10−3 6.0 · 10−3 3.4 · 10−3 1.4 · 10−3 3.4 · 10−4 5.1 · 10−5 TABLE S5. Mean pitch angles and efficiencies ± one standard deviation (when available) for the peritrichous flagellar family (as in Fig. 3c) for measured and theoretical waveforms [3], and for measured waveforms from the polar (or monotrichous) family (as in Fig. 3e). 11 II. Waveform geometries and resistive force theory predictions The primary results found from comparing the hydrodynamic efficiencies of the polymorphic forms were reported in the main text. Most notably, the normal polymorphic form was found to be the most efficient waveform by a significant margin for both peritrichous and monotrichous (polar) flagellar families. Figures 3(b,d) showed the geometries of the waveforms considered in the helical circumference-pitch (C-P) plane (with C = πD). Here we provide a different standpoint from which to visualize the geometries; Fig. S1 shows the geometrical relations in pitch angle ψ vs. circumference C for the peritrichous and monotrichous (polar) flagellar families. The normal, semi-coiled, and curly forms all occupy nearby regions of parameter space in pitch angle ψ. However, compared to the other polymorphs, the normal waveforms have a significantly larger helical circumference for a given pitch angle. FIG. S1. Geometrical data for the (a) peritrichous and (b) monotrichous polar flagellar families (see Fig. 1a for symbol legend). The normal, semi-coiled, and curly forms all occupy nearby regions of parameter space in pitch angle ψ. However, compared to the other polymorphs, the normal waveforms have a significantly larger helical circumference for a given pitch angle. Finally, resistive force theories, which are valid only at O(log 1/)−1 and relate local body velocities to local fluid forces, are the most widely used approaches for modeling slender bodies in fluids, dating in the case of highly viscous flow back to the seminal work of Gray & Hancock [38]. However, we note that the local resistive force theory achieved by ignoring the non-local integral operator K[f (s(cid:48))](s) in Eq. (2) in the main text (see also Ref. [38]), and even the more appropriate resistive force theory for helical geometries due to Lighthill [39] do not predict the efficiency ordering found using the full non-local slender body theory. A related study by Chattopadhyay and Wu also suggests the importance of solving for the full nonlocal fluid interactions in such systems [40]. Figures S2(a,b) show the efficiencies computed using these local theories for the peritrichous flagellar family data. The first approximation significantly overestimates the efficiencies, and the curly and semi-coiled forms are the most efficient. The second approximation (using Lighthill's resistive coefficients) significantly underestimate the efficiencies, and again the curly and semi-coiled forms are computed to be the most efficient. Hydrodynamic interactions between different parts of the flagella, which are captured by our slender-body approach but not in resistive force theory, are thus essential in order to conclude on the relative efficiencies of flagellar polymorphs. PeritrichousPolar(b)(a)C[µm]C[µm]045900123453050700.511.522.53ψ[◦]ψ[◦]12 FIG. S2. Mean efficiencies computed using two local resistive force models; namely, (a) the local theory achieved by simply neglecting the non-local term K[f (s(cid:48))](s) in Eq. (2) in the main text, and (b) the local theory of Lighthill for helical waveforms [23]. Neither local theory predicts the correct efficiency ordering of the polymorphic forms for biologically relevant parameters. ∗ [email protected] † [email protected] [1] K. Namba and F. Vonderviszt, Q. Rev. Biophys. 30, 1 (1997). [2] S. Asakura, Adv. Biophys. 1, 99 (1970). [3] C. R. Calladine, J. Mol. Biol. 118, 457 (1978). [4] K. Hasegawa, I. Yamashita, and K. Namba, Biophys. J. 74, 569 (1998). [5] R. Kamiya, S. Asakura, and S. Yamaguchi, Nature 286, 628 (1980). [6] H. Berg, Annu. Rev. Biochem. 72, 19 (2003). [7] N. C. Darnton and H. C. Berg, Biophys. J. 92, 2230 (2007). [8] N. Darnton, L. Turner, S. Rojevsky, and H. Berg, J. Bacteriol. 189, 1756 (2007). [9] L. Turner, W. S. Ryu, and H. C. Berg, J. Bacteriol. 182, 2793 (2000). [10] E. Leifson, Atlas of Bacterial Flagellation (Academic Press, New York and London, 1960). [11] R. Kamiya and S. Asakura, J. Mol. Biol. 106, 167 (1976). [12] R. M. Macnab and M. K. Ornston, J. Mol. Biol. 112, 1 (1977). [13] H. Hotani, Biosystems 12, 325 (1980). [14] E. Hasegawa, R. Kamiya, and S. Asakura, J. Mol. Biol. 160, 609621 (1982). [15] H. Hotani, J. Mol. Biol. 156, 791 (1982). [16] H. C. Hyman and S. Trachtenberg, J. Mol. Biol. 220, 79 (1991). [17] M. Seville, T. Ikeda, and H. Hotani, FEBS Lett. 332, 260262 (1993). [18] S. Trachtenberg and I. Hammel, J. Struct. Biol. 109, 18 (1992). [19] Y. Gebremichael, G. S. Ayton, and G. A. Voth, Biophys. J. 91, 3640 (2006). [20] S. V. Srigiriraju and T. R. Powers, Phys. Rev. E. 73, 011902 (2006). [21] G. Taylor, Proc. Roy. Soc. Lond. A 211, 225 (1952). [22] A. T. Chwang and T. Y. Wu, Proc. Roy. Soc. Lond. B 178, 327 (1971). [23] J. Lighthill, Mathematical Biofluiddynamics (SIAM, Philadelphia, 1975). [24] J. J. L. Higdon, J. Fluid Mech. 90, 685 (1979). 02040608000.0050.010.0150.02020406080012345678x 10−3(b)(a)\x{FFFF}E\x{FFFF}\x{FFFF}E\x{FFFF}\x{FFFF}ψ\x{FFFF}[◦]\x{FFFF}ψ\x{FFFF}[◦]13 [25] J. Lighthill, J. Eng. Math. 30, 35 (1996). [26] See supplementary material, where we include experimental data, references, mean efficiency values, notes on waveform geometry, and resistive force theory predictions. [27] J. Happel and H. Brenner, Low Reynolds Number Hydrodynamics (Prentice Hall, Inc., Englewood Cliffs, N.J., 1965). [28] S. Childress, Mechanics of Swimming and Flying (Cambridge University Press, Cambridge U.K., 1981). [29] E. Lauga and T. R. Powers, Rep. Prog. Phys. 72, 096601 (2009). [30] E. M. Purcell, Proc. Natl. Acad. Sci. USA 94, 11307 (1997). [31] S. Chattopadhyay, R. Moldovan, C. Yeung, and X. L. Wu, Proc. Natl. Acad. Sci. USA 103, 13712 (2006). [32] J. B. Keller and S. I. Rubinow, J. Fluid Mech. 75, 705 (1976). [33] R. E. Johnson, J. Fluid Mech. 99, 411 (1980). [34] A.-K. Tornberg and M. J. Shelley, J. Comp. Phys. 196, 8 (2004). [35] K. E. Atkinson, The numerical solution of integral equations of the second kind (Cambridge University Press, Cambridge, UK, 1997). [36] T. Gotz, Ph.D. Thesis, University of Kaiserslautern, Germany (2000). [37] M. Fujii, S. Shibata, and S.-I. Aizawa, J. Mol. Biol. 379, 273 (2008). [38] J. Gray and G. J. Hancock, J. Exp. Biol. 32, 802 (1955). [39] J. Lighthill, SIAM Rev. 18, 161 (1976). [40] S. Chattopadhyay and X.-L. Wu, Biophys. J. 96, 2023 (2009). [41] E. M. Purcell, Am. J. Phys. 45, 3 (1977). [42] T. Iino, J. Gen. Microbiol. 27, 167 (1962). [43] T. Iino and M. Mitani, J. Gen. Microbiol. 44, 27 (1966). [44] T. Iino, T. Oguchi, and T. Kuroiwa, J. Gen. Microbiol. 81, 37 (1974). [45] L. Turner, W. S. Ryu, and H. C. Berg, J. Bacteriol. 182, 2793 (2000). [46] S. Matsuura, R. Kamiya, and S. Asakura, J. Mol. Biol. 118, 431 (1978).
1310.7261	1	1310	2013-10-27T22:32:09	Self-Motion Mechanism Of Chained Spherical Grains Cells	[ "physics.bio-ph", "cond-mat.soft" ]	Cells are modeled with spherical grains connected each other. Each cell can shrink and swell by transporting its fluid content to other connected neighbor while still maintaining its density at constant value. As a spherical part of a cell swells it gains more pressure from its surrounding, while shrink state gains less pressure. Pressure difference between these two or more parts of cell will create motion force for the cell. For simplicity, cell is considered to have same density as its environment fluid and connections between parts of cell are virtually accommodated by a spring force. This model is also limited to 2-d case. Influence of parameters to cell motion will be presented. One grain cell shows no motion, while two and more grains cell can perform a motion.	physics.bio-ph	physics	Submitted to The International Symposium on BioMathematics (Symomath) 2013, October 27-29, 2013, Bandung, Indonesia Self-Motion Mechanism Of Chained Spherical Grains Cells Sparisoma Viridi1* and Nuning Nuraini2 1Nuclear Physics and Biophysics Research Division, Institut Teknologi Bandung, Bandung 40132, Indonesia 2Industrial and Financial Mathematics Research Division, Institut Teknologi Bandung, Bandung 40132, Indonesia * Email: [email protected] Abstract. Cells are modeled with spherical grains connected each other. Each cell can shrink and swell by transporting its fluid content to other connected neighbor while still maintaining its density at constant value. As a spherical part of a cell swells it gains more pressure from its surrounding, while shrink state gains less pressure. Pressure difference between these two or more parts of cell will create motion force for the cell. For simplicity, cell is considered to have same density as its environment fluid and connections between parts of cell are virtually accommodated by a spring force. This model is also limited to 2-d case. Influence of parameters to cell motion will be presented. One grain cell shows no motion, while two and more grains cell can perform a motion. Keywords: spherical grain cell, self-motion mechanism, granular simulation. PACS: 87.18.Fx, 87.17.Rt, 47.85.Dh. INTRODUCTION Cell motion is an interesting field to discuss. The motion can be addressed to cytoskeletally driven contractions and expansions of the cell membrane [1], drag from fluid medium [2], rigidity of substrate [3], cytoplasmic streaming and membrane protrusions and retractions [4], and myosin-based contractility and transcellular adhesions [5]. Simple mechanism based on only pressure difference is proposed in this work for a 2-d spherical grains based cell. CELL MODEL A cell is modeled by a 2-d spherical grain or iR connected spherical grains. Each grain has radius at time t and always the same density ρ . Cell is located in a fluid with the same density as the cell. It is also assumed that cell motion does not change its depth in fluid, so that hydrostatic pressure suffered by 0p . Total force per length 0p for a single grain at time t will the cell remains the same as due to the pressure be = F i 1 Rp 02 i 2 ∫ θd , (1) where θ is angular components in polar coordinate system. A single spherical cell makes integration of Equation (1) zero. For a cell consisted of two connected grains, there is intersection length L between two grains, which is assumed to have always the same value. The length L will produce a non-zero net force per length for both grains. ρ p0 ρ L Rj Ri radius FIGURE 1. Cell consists of two connected grains with jR and also intersection length L . iR and Left grain in Figure 1 will have net force per length iF to right, while right grain will have net force per jF to left. Both this net force per length is due 0p . During changing radius jR total area of the cell is maintained iR and constant. It means that to surrounding pressure length of Submitted to The International Symposium on BioMathematics (Symomath) 2013, October 27-29, 2013, Bandung, Indonesia ( ππ + 2 R R i j 2 − A L ) 0 = , d dt (2) d dt    − 11 N ∑ 2 = 1 i 2 π R i + 1 2 2 π R i + 1 − A + iiL 1,,  =  0 , (9) A L = seg + A i , A seg , j , (4) = Ri ( 10 + R )t ωδsin , (10) that where LA is intersection area. Equation (2) indicates jR depend on each other, so that only iR and one variable can be considered a free variable. Intersection area can be obtained through equating the two circle equations located at ( )i i yx , )j ( j y x , with condition that ( x i + RR i and ) 2 ) 2 (3) < − + − y i x j , j y j ( so that the two grains still overlap to each other. From Figure 1 it can be seen that L is a chord of both LA can found from circles. Intersection area with ( )1I )2I and ( x 2I , y x 1I , y points of the two circle equations. The area in triangle is are the intersection A sec , i − A seg i , = 1 2 RL i 2 − 2 L . 1 4 (5) Angle iα between two radii iR is obtained from Then, α i =   asin  L 2 R i    . (6) 2 A sec 1 α= i 2 Substituting Equations (6) and (7) into Equation (5) will produce (7) R i . i , A seg i , = 1 2 2 R i   asin  L 2 R i  −  1 2 RL i 2 − 2 L .(8) 1 4 Substitution of Equation (8) into Equation (4), and the result into Equation (2) will give the dynamics of the cell. For cell consisted of three or more 2-d spherical grains, formulation in Equation (4) can be generalized into with N number of 2-d spherical grains. SIMULATION Since one-grain cell is trivial case and the result can be deduced logically, it will not be simulated. Case of two- and three-grain cell will be performed in simulations. Case of two-grain cell radius Considered there are two circles with identical = = R 0R R i with a function that is modulated through time j . jR is not free, since it is connected to with 0R<δ iR through 2 π R i − 1 2 2 R i   asin  L 2 R i    + 1 2 RL i 2 − 2 L 1 4 + π R 2 j − 1 2 2 R j asin     L 2 R j     + 1 2 RL 2 j − 2 L = c 1 4 , (11) where c is a constant. Equation (11) can be solved jR from known value of iR numerically to obtain from Equation (10). Net force per length for each grain is calculated   1 1 ααθ ∈ − , using Equation (1) with   2 2   1 1 απαπθ − +∈ ,   2 2 for the left grain and the right for grain. Submitted to The International Symposium on BioMathematics (Symomath) 2013, October 27-29, 2013, Bandung, Indonesia Case of two-grain cell Motions prediction For this case, it is actually similar to the previous one, instead that there is additional middle grain which has two α, left and right. The relation in Equation (11) is then expanded into Motion predictions of cell consisted of one-, two-, and three-linear-grains are given in Figures 2 – 4. 1 2 RL i 2 − 2 L 1 4 2   asin    asin   2  +    +  L R i L 2 R j 2 π R i − 1 2 2 R i + π R 2 j − R j 2 + π R k − 1 2 2 R k   asin  L 2 R k    + 1 2 RL k 2 − 2 L = c 1 4 RL 2 j − 2 L 1 4 ,(12) FIGURE 2. Motion prediction for cells consists of one 2-d spherical grain: no motion. Equation (12) will be more difficult to solve than Equation (11) since there is two independent variables, iR and jR , if kR is considered a dependent variable. e.g. Approximation In a case that Ri << Equations (11) dan (12) L can be simplified through approximation, e.g. Equation (12) can be turned into c 2 ≈ π R i + 1 4 LR i + π R 2 j + and for Equation (12) c 2 ≈ π R i + 1 4 LR i + π R 2 j + 2 + π R k + 1 4 LR k 1 4 1 2 LR , j (13) LR j . (14) RESULTS AND DICSUSSION In this section prediction and simulation results are presented and discussed. FIGURE 3. Motion prediction for cells consists of two 2-d spherical grains: oscillation motion. FIGURE 4. Motion prediction for cells consists of two 2-d spherical grains: motion in one direction. Submitted to The International Symposium on BioMathematics (Symomath) 2013, October 27-29, 2013, Bandung, Indonesia As predicted, Figure 5 shows an oscillating net force per length that will move the cell also in oscillation motion. In Figure 6 it can be seen that the force is not symmetry for positive and negative values, as a matter a fact it always negative, this difference can induce a one direction motion. For this case other ω is chosen to be half as the previous one. This result shows that the cell will move to left in some backward and forward motion. It is interesting to investigate non-linier triangle grains configuration, e.g. circular or configuration for more-grain cell, that will be conducted in the next work. CONCLUSION Motion of cell consisted of connected grains has been simulated. One-grain cell shows no motion, two- grain cell performs oscillation motion, and three-grain cell exhibits one-direction motion. More-grain cell and not-linear combinations are subject for next investigation. ACKNOWLEDGMENTS Riset Institut Inovasi Kelompok Keahlian Teknologi Bandung (RIK-ITB) in year 2013 with contract number 248/I.1.C01/PL/2013 supports this work partially. REFERENCES 1. J. C. M. Mombach and J. A. Glazier, Phys. Rev. Lett. 76, 3032 (1996). 2. T. W. Secomb, R. Hsu, and A. R. Pries, Am. J. Physiol. Heart Circ. Physiol. 274, H1016 (1998). 3. C.-M. Lo, H.-B. Wang, M. Dembo, and Y.-I. Wang, Biophys. J. 79, 114 (2000). 4. W. Alt and M. Dembo, Math. Biosci. 156, 207 (1999). 5. D. E. Discher, P. Janmey, and Y.-I. Wang, Sience 310, 1139 (2005). One-grain cell can not move since it whatever shrinks or swells net force per length is always zero, as it is illustrated in Figure 2. Two-grain cell exhibits an oscillating motion, that the cell can move forward and backward but not only in one direction. Then, this configuration can not make a motion also, as it is given in Figure 3. Three-grain or more-grain cell can perform a one direction motion, but with cell shrinking and swelling not in the same order and rate. The same ones will only produce the oscillating motion. An example of one direction motion is shown in Figure 4. Simulation results Parameters used in the simulation are given in following Table 1. TABLE 1. Simulation parameters. Parameters Value 0R δ ω L 0p 1.0 0.2 0.1 1.0 6.2832 1.2E-02 6.0E-03 F 0.0E+00 -6.0E-03 -1.2E-02 0.0 0.5 1.0 t FIGURE 5. Net force per length for Case of two-grain cell. 1.5 2.0 0.0E+00 -2.0E-02 F -4.0E-02 -6.0E-02 0.0 0.5 1.0 t FIGURE 6. Net force per length for Case of three-grain cell. 1.5 2.0
1512.05451	1	1512	2015-12-17T03:14:11	Predicting 3D structure, flexibility and stability of RNA hairpins in monovalent and divalent ion solutions	[ "physics.bio-ph", "cond-mat.soft", "physics.chem-ph", "physics.comp-ph" ]	A full understanding of RNA-mediated biology would require the knowledge of three-dimensional (3D) structures, structural flexibility and stability of RNAs. To predict RNA 3D structures and stability, we have previously proposed a three-bead coarse-grained predictive model with implicit salt/solvent potentials. In this study, we will further develop the model by improving the implicit-salt electrostatic potential and involving a sequence-dependent coaxial stacking potential to enable the model to simulate RNA 3D structure folding in divalent/monovalent ion solutions. As compared with the experimental data, the present model can predict 3D structures of RNA hairpins with bulge/internal loops (<77nt) from their sequences at the corresponding experimental ion conditions with an overall improved accuracy, and the model also makes reliable predictions for the flexibility of RNA hairpins with bulge loops of different length at extensive divalent/monovalent ion conditions. In addition, the model successfully predicts the stability of RNA hairpins with various loops/stems in divalent/monovalent ion solutions.	physics.bio-ph	physics	Predicting 3D structure, flexibility and stability of RNA hairpins in monovalent and divalent ion solutions Ya-Zhou Shi1, Lei Jin1, Feng-Hua Wang2, Xiao-Long Zhu3, and Zhi-Jie Tan1* 1Department of Physics and Key Laboratory of Artificial Micro & Nano-structures of Ministry of Education, School of Physics and Technology, Wuhan University, Wuhan 430072, China 2Engineering Training Center, Jianghan University, Wuhan 430056, China 3Department of Physics, School of Physics & Information Engineering, Jianghan University, Wuhan Running title: A coarse-grained model for RNA structure 430056, China ABSTRACT A full understanding of RNA-mediated biology would require the knowledge of three-dimensional (3D) structures, structural flexibility and stability of RNAs. To predict RNA 3D structures and stability, we have previously proposed a three-bead coarse-grained predictive model with implicit salt/solvent potentials. In this study, we will further develop the model by improving the implicit-salt electrostatic potential and involving a sequence-dependent coaxial stacking potential to enable the model to simulate RNA 3D structure folding in divalent/monovalent ion solutions. As compared with the experimental data, the present model can predict 3D structures of RNA hairpins with bulge/internal loops (<77nt) from their sequences at the corresponding experimental ion conditions with an overall improved accuracy, and the model also makes reliable predictions for the flexibility of RNA hairpins with bulge loops of different length at extensive divalent/monovalent ion conditions. In addition, the model successfully predicts the stability of RNA hairpins with various loops/stems in divalent/monovalent ion solutions. Keywords: RNA; coarse-grained model; 3D structure prediction; flexibility; stability; ion electrostatics * To whom correspondence should be addressed. Email: [email protected] 1 I. INTRODUCTION In early years, RNA has been considered as an intermediary in transcription and translation (1). However, in the recent two decades, RNAs have been shown to perform other crucial functions such as catalyzing biological reactions and controlling gene expression (2,3). Understanding and utilizing the functions would require the comprehensive knowledge of RNA structures and dynamics (4-7). Although RNA sequences are being discovered rapidly, only limited RNA three-dimensional (3D) structures have been determined through experimental methods such as X-ray crystallography, nuclear magnetic resonance (NMR) spectroscopy and cryo-electron microscopy (4,8). Simultaneously, for high efficiency and low cost, some computational models have been developed for predicting 3D structures or thermodynamics of RNAs (8-18). Some models based on fragment assembly, sequence alignment and secondary structure are highly successful in predicting 3D structures for even large RNAs (19-39), e.g., MC-Fold/MC-Sym pipeline (30). However, these models are primarily designed to predict folded structures and would not give reliable predictions for the dynamic and thermodynamic properties of RNAs in three dimensions (16-18). Simultaneously, some other models have been developed aiming to predict RNA dynamics and thermodynamics. The Go-like coarse-grained (CG) model of TIS can predict folding thermodynamics for hairpins and pseudoknots (40,41). Another CG model of oxRNA can capture the thermodynamic and mechanical properties of RNA structures with pair-wise interaction potentials (42). But neither of the TIS and oxRNA could give reliable predictions for 3D structures of RNAs from the sequences (16,18). Although the three-bead CG model of iFoldRNA (43) and the six/seven-bead CG model of HiRE-RNA (16,44) can predict 3D structures of small RNAs including pseudoknots, the parameters of the two models may need further validation or adjustment for predicting thermodynamic and dynamic properties of RNAs (18,43,44). Furthermore, since RNAs are highly charged polyanionic polymers, RNA structures can be sensitive to ion conditions and temperature (15,45-53). However, none of the above models could predict 3D structures and thermodynamics of RNAs over a wide range of ion concentrations and temperature from the sequences (16-18). Very recently, to predict 3D structures and thermal stability of RNAs, we have developed a CG model with three beads placed on the existing atoms P, C4’ and N9 for the purine (or N1 for the pyrimidine) (18,54), respectively. Combined with an implicit-salt/solvent force field and the Monte Carlo (MC) simulated annealing algorithm, the model can not only predict native-like 3D structures of small RNAs from their sequences at a high salt (e.g., 1M NaCl), but also give reliable predictions on the stability for RNA hairpins over a wide range of sequences and monovalent ion concentrations as compared with extensive experimental data (54). However, compared with monovalent ions (e.g., Na+), divalent ions such as Mg2+ can play a more special role in the stability and dynamics of RNA structures (55-60). For example, Mg2+ is about 1000 times more efficient in inducing the tertiary 2 structures folding of Tetrahymena thermophila ribozyme (45,49). Although a recent structure-based model with an explicit treatment of Mg2+ and an implicit treatment of K+ can well capture the ion atmosphere around RNAs (61-63), there is still lack of a model for predicting 3D structures and stability for RNAs in divalent/monovalent ion solutions from the sequences. In the present work, we will further develop our previous model to enable it to predict the 3D structure and stability of RNAs in the presence of divalent ions. Additionally, the functions of RNAs may be not only related to the static 3D structures, but also influenced by the flexibility and stability of their structures (4-7,64-66). The flexibility of RNAs is rather important in the recognition with protein and in gene regulation (64-67). For example, the transactivator response element (TAR) for the transactivator (TAT) protein of the human immunodeficiency virus (HIV) can undergo large conformational changes through the small bulge during the binding of TAT proteins (66,67). Due to the polyanionic nature of RNAs, their flexibility would strongly depend on metal ions such as Mg2+ and loops (67-69). To examine the effects of metal ions (e.g., Mg2+) and loops on RNA flexibility, we will select HIV-1 TAR and HIV-2 TAR variants as two paradigms in the present work. In order to predict the 3D structures and flexibility of RNA hairpins with bulge loops such as HIV TAR variants in divalent/monovalent ion solutions, we will introduce a new implicit electrostatic potential and an indispensable coaxial stacking interaction between two helices at the junction in the present work. With the present model, we will predict the flexibility of HIV-1 TAR and HIV-2 TAR variants in divalent/monovalent ion solutions to understand the effects of salt and bulge loop. In this work, we will essentially develop the model to simulate RNA folding in divalent/monovalent ion solutions through improving the implicit-salt electrostatic potential and involving a parameterized coaxial stacking potential. Afterwards, we will firstly show the present CG model can predict 3D structures of RNAs with bulge/internal loops at given ionic conditions with higher accuracy. Secondly, the model will be employed to investigate the effects of divalent/monovalent salts and bulge length on the flexibility of HIV TAR variant RNAs. Finally, the model will be used to quantitatively examine the stability of various RNA hairpins in divalent ion solutions. Throughout the paper, all the predictions will be compared with the extensive experimental data. II. MODEL AND METHODS A. Coarse-grained structural model Since CG models generally allow considerable extension of the accessible size and time scale in simulations of biological systems (70-74), we have proposed a three-bead CG model for RNAs where three beads stand for phosphate, sugar and base, respectively (54). The backbone phosphate (P) bead and sugar (C) bead are placed respectively at P and C4’ atom positions, while the base (N) 3 beads are placed at N9 position for purine or N1 for pyrimidine; see Fig. 1. The P, C and N beads are treated as the spheres with van der Waals radii of 1.9 Å, 1.7 Å and 2.2 Å, respectively (54,75). B. Force field In our CG model, the implicit-solvent/salt force field includes eight energy potentials (54): The function forms for the eight energy potentials are described detailly in Supporting Material, and in the following we only introduce them briefly except for the electrostatic interaction Uel and the coaxial stacking interaction Ucs. The first three terms in Eq. 1 are the bonded potentials for covalent bonds (Ub), bond angles (Ua), and dihedral angles (Ud), respectively. The bonded potentials whose function forms have been shown in Ref. 54, were initially parameterized by the statistical analysis on the available 3D structures of RNA molecules in Protein Data Bank (PDB, http://www.rcsb.org/pdb/home/home.do) (54,75-77). Since lots of native structures in PDB are mostly A-form helix, the statistical parameters from these structures would not be reasonable to describe the nature of RNA free chains (77). Therefore, for bonded potentials, two sets of parameters are calculated for single-strands/loops and stems, named as Paranonhelical and Parahelical, respectively; see Supporting Material and Ref. 54 for details. The former are used to describe the folding of an RNA from a free chain, and the latter are only used for stems during structure refinement after the folding process. The remaining terms of Eq. 1, namely the nonbonded potentials, describe various pairwise nonbonded interactions. The excluded volume Uexc between CG beads is modeled by a purely repulsive Lennard-Jones potential (54,75). Ubp in Eq. 1 is employed to capture base-pairing interaction between Watson-Crick (G-C, A-U) and wobble (G-U) base pairs (43,54,78-80). Ubs in Eq. 1 is a temperature-dependent base-stacking potential which works between nearest-neighbor base pairs. The strength of Ubs was derived from the combined analysis of available sequence-dependent thermodynamic parameters (78-80) and the MC algorithm, and the details are shown in Ref. 54. The electrostatic interaction Uel in Eq. 1, which is a newly refined term for the effect of divalent ions, is taken into account with the combination of the Debye-Hückel (DH) approximation and the concept of counterion condensation (CC) (78,81-83): The summation is over all the phosphate beads and rij is the distance between two phosphate beads i and j. ε0 is the permittivity of vacuum. ε(T) is an effective temperature-dependent dielectric constant (40,54,58). lD is the Debye length of ionic solution. Beyond our previous model (54), the effect of pure divalent ions and the competition between monovalent and divalent ions are also taken into account in the present model to study RNA folding in pure and mixed divalent ion solutions. Based 4 . (1)badexcbpbselcsUUUUUUUUU20. (2)4PijDrlNelijijQeTUeron the CC theory (83), for a pure salt solution containing only one species of salt such as NaCl or MgCl2, the reduced charge fraction Q could be written as Q = b/(vlB) (40,54), where v is the cation valence. b is the phosphate-phosphate spacing of an RNA and lB is the Bjerrum length (40,54). For a mixed Na+/Mg2+ ion solution, we assume , where and represent the contribution fractions from Na+ and Mg2+, respectively. can be approximately calculated by the empirical formula previously derived from the tightly bound ion (TBI) model which could account for the divalent ion-RNA interactions (15,60,84,85) Here, (15,60). [Na+] and [Mg2+] are the corresponding bulk concentrations in molar (M), and N is the chain length. Ucs in Eq. 1 is newly introduced to model the coaxial stacking interaction at RNA junctions (4,86,87), and the coaxial stacking interaction between two discontinuous neighbour helices with interfaced base-pairs i-j and k-l can be given by (86,87) where Gi-j,k-l is the sequence-dependent base-stacking strength. Gi-j,k-l is approximately taken as the stacking strength between the corresponding nearest-neighbour base-pairs in an uninterrupted helix (79,86,87). rik (or rjl) is the distance between two interfaced bases i(j) and k(l) of two stems and a represents the extent of distance constraint. rcs is the optimum distance between two coaxially stacked stems. a and rcs are directly obtained from the statistical analysis on the known structures in PDB database (see Fig. S1 in Supporting Material). Here, we only include the cases that there is one base or less in at least one of the single-stranded chains between two neighbour helices (see Fig. S1 in Supporting Material), since the noncanonical base pairs would be generally formed when there are more than one bases in each side between two stems (3-4,86). The detailed descriptions of the potentials in Eq. 1 and all the parameters for the potentials have been described in Supporting Material; see also Ref. 54 for the building process of force field. C. Simulation algorithm We use the MC simulated annealing algorithm, which can effectively avoid the trap in local energy minima (34,54,88), to search near-native conformations for an RNA at a given solution condition. Based on a random chain generated for an RNA sequence, the MC simulated annealing algorithm is performed from an initial high temperature to the target temperature (e.g., 298K) at a fixed ion condition. In the folding process, the Paranonhelical of bonded parameters and the efficient 5 2NaNaNaMg(1)QfQfQNafNa(1)fNaf2Na[Na]. (3)[Na][Mg]xf8.164.8/5.2ln[Na]xN22()()-,--,-1(cid:160){[112}, ][](4)2cstjlcsikcsNarrarrcsijklijklUGeepivot moves for RNA chain as well as the standard Metropolis algorithm are used to sample conformations of a free RNA chain (54,75,78). With gradual cooling of the system, the initial 3D near-native structures would be folded at room temperature for RNAs. After the MC annealing process with the bonded parameters of Paranonhelical for the whole RNA chain, the secondary structure and native-like 3D structures are predicted from a given sequence. To better capture the geometry of helical parts, the further structure refinement is performed for higher accuracy of 3D structures as follows: based on the final 3D structure predicted by the preceding annealing process, another MC simulation (generally 1×106 steps) is performed at the corresponding ion condition and room temperature, with the bonded parameters of Parahelical and Paranonhelical for the base-pairing regions (stems) and loops/single-strands, respectively (54). As a result, an ensemble of refined 3D structures (about eight thousand structures) would be obtained over the last ~8×105 MC steps. The predicted 3D structures are evaluated by their RMSD’s calculated over C beads from the corresponding atoms C4’ in the native structure in PDB, though other parameters can also be used to evaluate the predicted structures such as the TM-score and the base interaction network fidelity (89,90). Since the present model generally predicts a series of native-like structures during refinement process, we will use mean RMSD (the averaged value over the whole structure ensemble in refinement process) and minimum RMSD (corresponds to the structure closest to the native one in refinement process) to evaluate the reliability of predictions on 3D structures (29,43,54). Although RNA hairpins are mostly A-form helix, since helical stems may deform from the standard A-form one (48) and the relative orientation between stems at a junction may change (67) upon ionic conditions, we still use the RMSD of the whole RNA including stem and loop to evaluate the performance of 3D structure prediction (89). III. RESULTS AND DISCUSSION In the following, we will employ the CG model to predict the 3D structures of 32 RNA hairpins most of which are with bulge/internal loops (≤77nt) at the respective experimental ion (Na+/Mg2+) conditions from their sequences. Afterwards, the CG model will be used to predict the flexibility and stability of RNA hairpins at extensive divalent/monovalent ion conditions. Our predictions will be compared with the extensive experiments. A. Predicting RNA 3D structures at the respective experimental ion conditions Beyond our previous work focused on predicting RNA structures at 1M NaCl (54), we will predict the 3D structure of 32 RNA hairpins most of which are with bulge/internal loops. The 3D structures of these RNAs have been determined by NMR method at certain ion conditions; see Table I. For each RNA, we will make two separate predictions using the present model at the experimental ion condition and its previous version at 1M NaCl (54), respectively. Table I summarizes the major 6 information of the RNA molecules, the corresponding experimental monovalent/divalent ion conditions and the predictions from our model and MC-Fold/MC-Sym pipeline (30). 1. In monovalent solutions Firstly, we employ the present model to predict 3D structures for RNA hairpins (with bulge/internal loops) at the corresponding monovalent ionic conditions listed in their PDB files. For example, the structure of stem loop IIa (PDB code: 1u2a) is determined by NMR in the buffer including 10mM KH2PO4, 50mM KCl, 15mM NaCl and 0.5mM EDTA (91). We predict the 3D structures of the stem loop IIa from its sequence in the solution of 75mM monovalent salt. As shown in Table I, the mean RMSD of the predicted 3D structures is ~2.1 Å, which is obviously smaller than that (~2.6 Å) of the structures predicted by the previous version of our model at 1 M NaCl. For the 28 tested RNA hairpins in the respective monovalent salt solutions, the overall mean RMSD between the structures predicted by the present model at the respective experimental monovalent ionic conditions and the experimental structures is 3.37 Å, a smaller value than that (3.66 Å) predicted by the previous version of the model at 1 M NaCl (54); see Table I. 2. In divalent solutions In addition, one important feature of the present model is the combination between the CC theory and the results from the TBI model, and the present model can be employed to simulate RNA folding in mixed monovalent/divalent ion solutions. Four RNA hairpins with bulge/internal loops (PDB code: 1yn2, 2l5z, 2aht and 1p5o) have been determined by NMR in solutions containing Mg2+ and the corresponding monovalent/divalent ion conditions are listed in Table I. The mean RMSDs for these RNAs predicted by the present model in the corresponding experimental mixed ion solutions are 4.1 Å, 4.0 Å, 3.2 Å and 11.0 Å, respectively, which are apparently smaller than the values (4.3 Å, 4.6 Å, 3.9 Å and 13.0 Å) predicted by our previous model regardless of salt effect and the coaxial stacking interactions (54). 3. Salt vs coaxial stacking As shown in Table I, for the 32 hairpins, the present model can predict a visibly lower overall mean RMSD than our previous model (3.64 Å vs 4.02 Å; the P-value of <0.01 from the two-tailed Student’s t-test (12)), which suggests that the involvement of monovalent/divalent salt and coaxial stacking in the present model is effective for predicting RNA 3D structures in ion solutions. To clarify the contributions of the two improvements, we further perform two additional predictions for each hairpin using the present model with coaxial stacking at 1M NaCl and the present model without coaxial stacking at the experimental ion conditions, respectively. We find that the two improvements would both make positive contributions to the overall improved predictions and the involvement of salt effect has stronger contribution than that of coaxial stacking, as shown in Fig. S2 7 in Supporting Material. Due to the high charge density of RNA backbone, the native-like structures of hairpins could be slightly stretched at low salt (48), and the difference between conformations at low and high salts should not be ignored (see Fig. S2). Although the coaxial stacking potential would only have slight effect on hairpins with internal/bulge loops whose bases are stacking into neighbor helices (e.g., 1jur and 1lc6), it would be indispensable for the formation of coaxially stacked states which are common in RNAs with bulges (e.g., 2kuu and 1p5o), and in the HIV-1 and HIV-2 TAR variants; see the subsection of “flexibility of RNAs with bulges”. 4. Comparisons with MC-Fold/MC-Sym pipeline The MC-Fold/MC-Sym pipeline is a web server (http://www.major.iric.ca/MC-Pipeline/) for RNA secondary and tertiary structure prediction (30). To test the present model, we also make comparisons with the MC-Fold/MC-Sym pipeline. The RMSDs of the top 1 structures predicted by the pipeline online server (option: return the best 100 secondary structures and model_limit = 1000 or time_limit = 12 h) are calculated over C4’ atoms from the corresponding atoms in the experimental structures. As shown in Table I and Fig. 2, the overall predictions of 32 structures from the present model (overall mean RMSD = 3.64Å) appear visibly better than those from the MC-Fold/MC-Sym pipeline (overall mean RMSD = 4.12Å), which is also suggested by the P-value of <0.01 from the two-tailed Student’s t-test (12). Table I and Fig. 2 also show that the present model can give reliable prediction for relatively large RNAs (>45nt (54)). For the 48-nt and 68-nt RNAs (PDB codes: 2kuu and 2mqt), the present model gives much better predictions than those predicted by the MC-Fold/MC-Sym pipeline, while for the 77-nt RNA (PDB code: 1p5o), the two models give the similar predictions (see Table I and Fig. 2). This may be attributed to that the present model still ignores the possible noncanonical base pairs, which are abundant in the 77-nt RNA. B. Flexibility of RNAs with bulges RNAs are highly flexible biomolecules that can undergo dramatic conformational changes to fulfil their diverse functions, e.g., the structural reorganization of riboswitches or the hammerhead ribozyme (2,3). Generally, large conformational transitions of RNA structures are induced by the binding of ions, proteins or ligands (45-47,65-67). For example, HIV TAR RNAs, which can bind TAT protein through conformational changes in viral replication, have been the important paradigms for studying RNA dynamics (66,67). In the following, we will employ the present model to study the flexibility of HIV-1/HIV-2 TAR variants at different ion conditions and make comparisons with the experimental data. The sequences of the HIV TAR variants and their secondary structures predicted by the present model are shown in Fig. 3. HIV-1 TAR variant with a 3-nt bulge is used to examine the conformational changes at different ion conditions with the present model, and HIV-2 TAR variants with different length of polyU (polyA) bulges are used to study the conformational changes induced by various bulges. Our predictions are compared with the existing experiments for HIV-1 8 TAR variant at extensive Na+/Mg2+ concentrations and for HIV-2 TAR variant with various length bulges at 5mM [Na+] with/without 2mM [Mg2+], respectively (67,68). 1. Bending versus monovalent/divalent salts Recent studies have shown that RNA flexibility is strongly coupled to its ion condition (45-47,67-69,75,92). As Casiano-Negroni et al have experimentally measured, the bending angle at the junction of bulge is strongly dependent on salt (67). Here, we use the present model to evaluate the bending at the bulge junction of the HIV-1 TAR variant (see Fig. 3) over a broad range of [Na+] as well as [Mg2+]. As shown in Figs. 4a and 4b, the inter-helical bend angle at the junction of bulge decreases apparently with the increase of [Na+] (and [Mg2+]), and our predictions agree well with the corresponding experimental data (67). Such decrease of bending angle with the increase of [Na+]/[Mg2+] is understandable. It comes from the competition between electrostatic repulsion which plays a dominate role at low salt and coaxial stacking interaction at the bulge which plays a dominate role at high salt. The comparison between Fig. 4a and Fig. 4b shows that Na+ and Mg2+ induce a similar structural transition from a bent state at low ion concentrations to a coaxially stacked state at high salt, while Mg2+ is much more efficient in inducing such structure transition due to the higher ionic charge (55-58). To further clarify the effect of coaxial stacking, we use the present model without coaxial stacking to predict the 3D structures for HIV-1 TAR variant at extensive Na+/Mg2+ concentrations. The comparison of inter-helical bend angles predicted by the present model with and without coaxial stacking indicates that the involvement of coaxial stacking can effectively capture the coaxially stacked conformations closer to the experiments especially at high salt (67), as shown in Fig. S3 in Supporting Material. 2. Bending versus bulge length Zacharias and Hagerman have experimentally measured the bending angle for long RNA helices induced by bulges of various length (n=1 to 6) and base composition (An and Un series) (68). For simplicity, we predict the 3D structures of HIV-2 TAR variant (see Fig. 3) with different bulge length at 5mM NaPO4 in the absence/presence of 2mM Mg2+ to study the bulge-induced bending of RNAs. As shown in Figs. 5a and 5b, the bend angle at the bulge increases with the increase of the number of nucleotides in the bulge, which is in good accordance with the experimental data for the bulge of Un (68). As shown in Fig. 5a, the severe inter-helical bending and its sharp increase for longer bulge come from the more extended structures with larger end-to-end distance for longer bulges and the strong electrostatic repulsion between helices which can offset the interhelical stacking interactions at low salt concentrations. Higher salt (2mM Mg2+) would reduce the electrostatic repulsion more strongly, and thus would promote the interhelical stacking and reduce the bending angles; see Fig. 5b. However, as shown in Figs. 5a and 5b, the predictions on bending 9 angle are apparently smaller than the experimental data for HIV-2 TAR variant with An bulge. This is because polyU behaves like a random coil, while polyA would exhibit strong intra-chain self-stacking (69) which is not accounted for in the present model. Such self-stacking would enhance the rigidity of single-stranded chain and consequently cause the large bending angle at the bulge. Recently, Mustoe et al have also developed a CG model of TOPRNA, which treats RNAs as collections of semirigid helices linked by freely rotatable single strands (74). The model could nearly reproduce experimental bending angles of HIV-2 TAR with more than 2 nt bulges at low salt concentrations, while it did not give reliable predictions for HIV-2 TAR with polyU bulge at high salt as well as HIV-2 TAR with 1 nt bulge, possibly due to the ignorance of salt effect/coaxial stacking and the overestimate of bulge rigidity for polyU (74). 3. Charactering global structural fluctuation The global size of an RNA can be characterized by its radius of gyration Rg, the fluctuation of which can properly reflect the RNA structural flexibility (64,93). Based on the conformational ensemble of an RNA, we have calculated the variance of Rg by ; see Fig. S4 and Fig. S5 in Supporting Material, as well as the mean RMSD from the time-averaged reference structure. As shown in Figs. 4c and 4d, and RMSD of the HIV-1 TAR variant increase with the increase of salt concentration, and such increase becomes saturated at high salts. This is because that the higher salt can reduce the electrostatic repulsion in the RNA and would favor the conformational fluctuation. At high salts, the coaxial stacking between the two stems can be formed and consequently (and RMSD) becomes saturated. Furthermore, as shown in Figs. 5c and 5d, (and RMSD) of the HIV-2 TAR variant increases for longer bulge length in 5mM NaPO4 solution in the absence/presence of Mg2+. This is because RNA single-stranded loops are distinctively more flexible than duplexes (69,75,93). 4. Charactering local structural fluctuation Furthermore, we have calculated the root-mean-square-fluctuation (RMSF) of backbone beads to analyze the local flexibility along an RNA chain (64). The RMSF for the i-th C-bead is calculated as: where means the average over time t. ri(t) is the position of the i-th C-bead at time t, and ri 0 is the time-averaged reference position of the i-th C-bead. Fig. 6 shows the RMSF of each C-bead of HIV-1 TAR variant at 1M NaCl. The nucleotides near 10 2R22()RggRR2R2R2R20(), (5)iiitSFtRMrrtthe 5’ and 3’ ends exhibit stronger conformational fluctuations, since the terminal base pairs have less spatial constraints (93). As shown in Fig. 6 and Fig. 7, the bulge and hairpin loops are more flexible than the stems, which comes from the significantly higher flexibility of unpaired single-stranded chains, since the previous theoretical and experimental studies have suggested that the persistence length of stem is ~60 times higher than that of single-strand chain (64-69,75,93). When the length of bulges increases from 1 to 8, the fluctuations of nucleotides especially at the bulge are apparently enhanced; see Fig. 7. This is because longer bulge has higher flexibility and naturally causes the higher flexibility of the whole RNA. Additionally, Fig. 7 shows that the addition of Mg2+ could increase the local flexibility of RNAs, especially for HIV-2 TAR variant with longer bulge. For the RNAs with small bulges, the salt effect on local flexibility would not be very obvious; see also Fig. S6. It is reasonable since the addition of Mg2+ would bring the strong electrostatic screening and consequently increase the flexibility of RNAs, especially at bulges. But for RNAs with small bulges (≤ ~3nt), the coaxial stacking would be formed between two stems and the addition of Mg2+ would only have slight effect on their flexibility. C. RNA hairpin stability in divalent ion solutions RNA folded structure is stabilized by the interplay of diverse interactions such as base pairing/stacking and electrostatic interactions. RNA stability at high salt concentrations (e.g., 1M NaCl) can be predicted with a relatively simple nearest-neighbor model (79,80). However, the nearest-neighbor model cannot predict the 3D structure of RNAs at an arbitrary temperature, and cannot predict RNA stability at ionic conditions departing from 1M NaCl. However, due to the polyanionic nature, RNA stability is very sensitive to the ionic condition (55-58,94-99), and Mg2+ ions are particularly efficient in stabilizing RNA tertiary structure (55-57). To address the effect of Mg2+ in RNA stability, we will employ the present model to study the stability of various RNA hairpins in divalent and mixed divalent/monovalent ion solutions. Generally, a hairpin is either in folded state at low T, or in unfolded state at high T, or in bistability at middle T around the melting temperature Tm (50-52,54). Based on the equilibrium value of the number of base pairs at each temperature T, to obtain the Tm of a hairpin, the fraction of denatured base pairs f(T) can be calculated and fitted to a two-state model (54,79), where dT is an adjustable parameter (54). 1. In pure divalent solutions Recently, the loop-size dependence of the stability of an RNA hairpin (denoted as R0) has been experimentally investigated in 2.5mM [Mg2+] (94). Fig. 8a shows the sequences of hairpins R0 as 11 ()/1()1, (6)1mTTdTfTewell as the secondary structures predicted by the present model. Fig. 8a also shows the predicted Tm for R0 in 2.5mM [Mg2+] as a function of the loop size (m=4~34 nt), which is in good accordance with experimental data (94). Due to the larger conformational entropy for longer loop, the hairpin stability would decrease when the hairpin loop becomes longer. In addition, we have studied the stability of hairpins with the same loop while with different stems in 0.7mM Mg2+ solutions. Hairpins R1~R4 are four similar RNAs, whose stems are slightly different in length or sequence; see Fig. 8b. As shown in Fig. 8b, Tm’s of four hairpins predicted by the present model at 0.7mM MgCl2 are in good agreement with the experimental data (95). The addition of G-C or C-G base-pair (from R1 to R2 and then to R3) can dramatically stabilize the RNA hairpins due to the strong base pairing/stacking interactions. The difference between Tm’s of R3 and R4 indicates that RNA stability is sensitive to sequence-dependent base pairing/stacking. The good agreement between our predictions and the experiments (94,95) suggests that the present model can well describe the folding stability of small RNAs in pure Mg2+ solutions. 2. In mixed divalent/monovalent solutions With the present model, we have also examined the stability of RNA hairpins R5 and R6 in mixed K+/Mg2+ solutions, and the secondary structures for R5 and R6 predicted by the present model are shown in Figs. 8c and 8d. As shown in Figs. 8c and 8d for Tm’s of R5 and R6 at a fixed [K+] (100mM), in addition to the general trend of increased stability for higher [Mg2+], the competition between K+ and Mg2+ also leads to the following behavior of RNA stability. At low [Mg2+] (≤0.1mM), the stability of RNAs is dominated by (~100mM) K+ and the Tm’s are close to that at pure corresponding [K+]. As [Mg2+] is increased, Mg2+ ions begin to play a role and the stability of RNAs begins to increase markedly due to the efficient role of Mg2+ in stabilizing RNAs. At very high [Mg2+], Mg2+ would become dominated and Tm becomes saturated (60). As shown in Figs. 8c and 8d, the agreements with experimental data (98,99) indicate that the combination of CC theory and the results from the TBI model could give good description for the competition between monovalent and divalent ions in stabilizing RNA hairpins, and the present model can give good predictions for the stability of RNA hairpins in mixed divalent/monovalent solutions. IV. CONCLUSIONS In this work, we have developed our CG model to predict 3D structures and structural properties of RNAs with bulge/internal loops in the presence of divalent and monovalent ions. The major extensions of our CG model include the improvement of the electrostatic potential to implicitly consider the effect of divalent ions and the inclusion of the coaxial stacking at two-way junction. The improved CG model has been employed to examine the effects of divalent/monovalent salt on the 3D structures, flexibility and stability of RNA hairpins with two-way junction. 12 Firstly, we have employed the present model to predict 3D structures for 32 RNAs (≤ 77nt) at their respective monovalent/divalent salt conditions in which the RNA structures have been experimentally determined by NMR method, and the overall mean RMSD of 3.64 Å between the predictions and the experimental structures is visibly smaller than those from predictions by the MC-Fold/MC-Sym pipeline and by the previous version of our model at 1M NaCl. Secondly, we have studied the flexibility of RNA hairpins with varying bulge loops at extensive [Na+]’s and [Mg2+]’s, and the predicted bending angles at bulge for HIV-1 and HIV-2 TAR variants are in good agreement with the available experimental data for different salt conditions as well as the different lengths of bulge loops. Thirdly, we have predicted the stability for RNA hairpins in divalent and mixed divalent/monovalent ion solutions, and the predictions agree well with the experimental data. Therefore, the present model can provide the ensemble of probable 3D structures at extensive divalent/monovalent ion conditions and can make reliable predictions on the structural properties such as flexibility and stability for small RNAs. Despite of the extensive agreement between our predictions and the experiments, the present model still involves some approximations and simplifications. First, in the present model, the effect of divalent and monovalent salts is implicitly accounted for by the combination of CC theory and TBI model. The good agreement with experimental data suggests that the combination of CC theory and TBI model can well capture the efficient role of divalent ions over monovalent ions, though the more extensive experimental validation for larger RNAs is still required. Also, the present model ignores the effect of specific ion binding, which might become important for large RNAs with complex structures (55-57). Of course, the more accurate treatment on salt is to explicitly consider the metal ions (75,97), which would bring huge computation cost. Very recently, Hayes et al have proposed a generalized Manning counterion condensation model in an alternative way to reproduce the ion atmosphere around RNAs through the explicit representation of Mg2+ and implicit treatment of K+ (61-63). Second, the present CG model only considers the canonical Watson-Crick (C-G, A-U) and wobble G-U base pairing, and ignores the possible noncanonical base pairs due to the lack of the experimental thermodynamic parameters (79,80). The noncanonical base pairing can be involved in the present model with the corresponding thermodynamic parameters, which would further improve the accuracy of structure prediction for RNAs with loops (29,30). Third, although our model can predict 3D structures for RNAs beyond hairpins, e.g., small pseudoknots (Ref. 54), it is still with challenge for the model at the present version to accurately and efficiently predict 3D structures of large RNAs with complex structures. Nevertheless, we are currently extending the model to predict 3D structures and stability for extensive RNA pseudoknots, and complex structures, while the 3D structure prediction for large RNAs from sequences may possibly require certain experimental constraints (34,39). Finally, the 3D structure predicted by the present model is at coarse-grained level, and consequently it is still required to develop the present model to reconstruct the all-atomistic structures based on the CG predictions. Nevertheless, the present model can be a reliable predictive 13 model for 3D structure ensemble of small RNAs in divalent/monovalent solutions and at arbitrary temperatures. SUPPORTING MATERIAL More details of force-field and six figures are available at xxxx. AUTHOR CONTRIBUTIONS Z.J.T. and Y.Z.S. designed the research; Y.Z.S., L.J., and F.H.W. performed the research; Z.J.T., Y.Z.S., and X.L.Z. analyzed data; and Y.Z.S. and Z.J.T. wrote the article. ACKNOWLEDGEMENTS We are grateful to Profs. Shi-Jie Chen (Univ Missouri), Yang Zhang (Univ Michigan) and Wenbing Zhang (Wuhan Univ) for valuable discussions. This work was supported by the National Key Scientific Program (973)-Nanoscience and Nanotechnology (No. 2011CB933600), the National Science Foundation of China grants (11175132, 11374234 and 11575128), and the Program for New Century Excellent Talents (Grant No. NCET 08-0408). REFERENCES 1. Crick, F. 1970. Central dogma of molecular biology. Nature. 227:561-563. 2. Doherty, E. A., and J. A. Doudna. 2001. Ribozyme structures and mechanisms. Annu. Rev. Biophys. Biomol. Struct. 30:457-475. 3. Edwards, T. E., D. J. Klein, and A. R. Ferre-d’Amare. 2007. Riboswitches: small-molecule recognition by gene regulatory RNAs. Curr. Opin. Chem. Biol. 17:273-279. 4. Tinoco, I., Jr., and C. Bustamante. 1999. How RNA folds. J. Mol. Biol. 293:271-281. 5. Hall, K. B., 2012. Spectroscopic probes of RNA structures and dynamics. Methods Mol. Biol. 875:67-84. 6. Zhang, W., and S. J. Chen. 2002. RNA hairpin-folding kinetics. Proc. Natl. Acad. Sci. USA 99:1931-1936. 7. Gong, S., Y. Wang, and W. Zhang. 2015. Kinetic regulation mechanism of pbuE riboswitch. J. Chem. Phys. 142:015103. 8. Sim, A. Y., P. Minary, and M. Levitt. 2012. Modeling nucleic acids. Curr. Opin. Struct. Biol. 22:273-278. 9. Rother, K., M. Rother, M. Boniecki, T. Puton, and J. M. Bujnicki. 2011. RNA and protein 3D structure modeling: similarities and differences. J. Mol. Model 17:2325-2336. 10. Laing, C., and T. Schlick. 2011. Computional approaches to RNA structure prediction, analysis, and design. Curr. Opin. Struct. Biol. 21:306-318. 11. Cruz, J. A., M. F. Blanchet, M. Boniecki, J. M. Bujnicki, S. J. Chen, S. Cao, R. Das, F. Ding, N. V. Dokholyan, S. C. Flores, L. Huang, C. A. Lavender, V. Lisi, F. Major, K. Mikolajczak, D. J. Patel, A. Philips, T. Puton, J. 14 Santalucia, F. Sijenyi, T. Hermann, K. Rother, M. Rother, A. Serganov, M. Skorupski, T. Soltysinski, P. Sripakdeevong, I. Tuszynska, K. M. Weeks, C. Waldsich, M. Wildauer, N. B. Leontis, and E. Westhof, 2012. RNA-Puzzles: A CASP-like evalution of RNA three-dimensional structure prediction. RNA. 18:610-625. 12. Hajdin, C. E., F. Ding, N. V. Dokholyan, and K. M. Weeks. 2010. On the significance of an RNA tertiary structure prediction. RNA. 16:1340-1349. 13. Shapiro, B. A., Y. G. Yingling, W. Kasprzak, and E. Bindewald. 2007. Bridging the gap in RNA structure prediction. Curr. Opin. Struct. Biol. 17:157-165. 14. Zhang, Y. 2008. Progress and challenges in protein structure prediction. Curr. Opin. Struct. Biol. 18:342-348. 15. Tan, Z. J., W. Zhang, Y. Z. Shi, and F. H. Wang. 2015. RNA folding: structure prediction, folding kinetics and ion electrostatics. Adv. Exp. Med. Biol. 827:143-183. 16. Cragnolini, T., P. Derreumaux, and S. Pasquali. 2015. Ab initio RNA folding. J. Phys. Condens. Mat. 27:233102. 17. Bailor, M. H., A. M. Mustoe, C. L. Brooks, and H. M. Al-Hashimi. 2011. Topological constraints: using RNA secondary structure to model 3D conformation, folding pathways, and dynamic adaptation. Curr. Opin. Struct. Biol. 21:296–305. 18. Shi, Y. Z., Y. Y. Wu, F. H. Wang, and Z. J. Tan. 2014. RNA structure prediction: progress and perspective. Chin. Phys. B 23:078701. 19. Massire, C., and E. Westhof. 1998. MANIP: an interactive tool for modelling RNA. J. Mol. Graph. Model. 16:197-205. 20. Jossinet, F., T. E. Ludwig, and E. Westhof. 2010. Assemble: an interactive graphical tool to analyze and build RNA architectures at 2D and 3D levels. Bioinformatics 26:2057-2059. 21. Martinez, H. M., J. V. Maizel, and B. A. Shapiro. 2008. RNA 2D3D: a program for generating, viewing, and comparing 3-dimensional models of RNA. J. Biomol. Struct. Dyn. 25:669-683. 22. Rother, M., K. Rother, T. Puton, and J. M. Bujnicki. 2011. ModeRNA: A tool for comparative modeling of RNA 3D structure. Nucleic Acids Res. 39:4007-4022. 23. Flores, S. C., and R. B. Altman. 2010. Turning limited experimental information into 3D models of RNA. RNA. 16:1769-1778. 24. Popenda, M., M. Szachniuk, M. Antczak, K. J. Purzycka, P. Lukasiak, N. Bartol, J. Blazewicz, and R. W. Adamiak. 2012. Automated 3D structure composition for large RNAs. Nucleic Acids Res. 40:e112. 25. Zhao, Y., Z. Gong, and Y. Xiao, 2011. Improvements of the hierarchical approach for predicting RNA tertiary structure. J. Biomol. Struct. Dyn. 28:815-826. 26. Zhao, Y., Y. Huang, Z. Gong, Y. Wang, J. Man, and Y. Xiao. 2012. Automated and fast building of three-dimensional RNA structures. Sci. Rep. 2:734. 27. Huang, Y., S. Liu, D. Guo, L. Li, and Y. Xiao. 2013. A novel protocol for three-dimensional structure prediction of RNA-protein complexes. Sci. Rep. 3:1887. 28. Wang, J., Y. Zhao, C. Zhu, and Y. Xiao. 2015. 3dRNAscore: a distance and torsion angle dependent evalution function of 3D RNA structure. Nucleic Acids Res. 43:e63. 29. Das, R., and D. Baker. 2007. Automated de novo prediction of native-like RNA tertiary structures. Proc. Natl. Acad. Sci. USA 104:14664-14669. 30. Parisien, M., and F. Major. 2008. The MC-Fold and MC-Sym pipeline infers RNA structure from sequence 15 data. Nature. 452:51-55. 31. Bida, J.P., and L. J. Maher. 2012. Improved prediction of RNA tertiary structure with insights into native state dynamics. RNA. 18:385-393. 32. Zhang, J., Y. Bian, H. Lin, and W. Wang. 2012. RNA fragment modeling with a nucleobase discrete-state model. Phys. Rev. E 85:021909. 33. Zhang, J., J. Dundas, M. Lin, M. Chen, W. Wang, and J. Liang. 2009. Prediction of geometrically feasible three-dimensional structures of pseudoknotted RNA through free energy estimation. RNA. 15:2248-2263. 34. Seetin, M. J., and D. H. Mathews. 2011. Automated RNA tertiary structure prediction from secondary structure and low-resolution restraints. J. Comput. Chem. 32:2232-2244. 35. Tan, R. K. Z., A. S. Petrov, and S. C. Harvey. 2006. YUP: A molecular simulation program for coarse-grained and multiscaled models. J. Chem. Theory Comput. 2:529-540. 36. Jonikas, M. A., R. J. Radmer, A. Laederach, R. Das, S. Pearlman, D. Herschlag, and R. B. Altman. 2009. Coarse-grained modeling of large RNA molecules with knowledge-based potentials and structural filters. RNA. 15:189-199. 37. Cao, S., and S. J. Chen. 2011. Physics-based de novo prediction of RNA 3D structures. J. Phys. Chem. B 115:4216-4226. 38. Xu, X., P. Zhao, and S. J. Chen. 2014. Vfold: A web server for RNA structure and folding thermodynamics prediction. PLoS One 9:e107504. 39. Xia, Z., D. R. Bell, Y. Shi, and P. Ren. 2013. RNA 3D structure prediction by using a coarse-grained model and experimental data. J. Phys. Chem. B 117:3135-3144. 40. Denesyuk, N., and D. Thirumalai. 2013. Coarse-grained model for predicting RNA folding thermodynamics. J. Phys. Chem. B 117:4901–4911. 41. Hyeon, C., and D. Thirumalai. 2005. Mechanical unfolding of RNA hairpins. Proc. Natl. Acad. Sci. USA 102:6789-6794. 42. Sulc, P., F. Romano, T. E. Ouldridge, J. P. K. Doye, and A. A. Louis. 2014. A nucleotide-level coarse-grained model of RNA. J. Chem. Phys. 140:235102. 43. Ding, F., S. Sharma, P. Chalasani, V. V. Demidov, N. E. Broude, and N. V. Dokholyan. 2008. Ab initio RNA folding by discrete molecular dynamics: from structure prediction to folding mechanisms. RNA. 14:1164-1173. 44. Pasquali, S., and P. Derreumaux. 2010. HiRE: A high resolution coarse-grained energy model for RNA. J. Phys. Chem. B 114:11957-11966. 45. Woodson, S. A. 2005. Metal ions and RNA folding: a highly charged topic with a dynamic future. Curr. Opin. Struct. Biol. 9:104-109. 46. Chen, S. J. 2008. RNA folding: conformational statistics, folding kinetics, and ion electrostatics. Annu. Rev. Biophys. 37:197-214. 47. Lipfert, J., S. Doniach, R. Das, and D. Herschlag. 2014. Understanding nucleic acid-ion interactions. Annu. Rev. Biochem. 83:813-841. 48. Manning, G. S. 2014. The response of DNA length and twist to changes in ionic strength. Biopolymers. 103:223-226. 49. Takamoto, K., Q. He, S. Morris, M. R. Chance, and M. Brenowitz. 2002. Monovalent cations mediate 16 formation of native tertiary structure of the Tetrahymena thermophila ribozyme. Nat. Struct. Biol. 9:928-933. 50. Zhang, Y., J. Zhang, and W. Wang. 2011. Atomistic analysis of pseudoknotted RNA unfolding. J. Am. Chem. Soc. 133:6882-6885. 51. Bian, Y., J. Zhang, J. Wang, J. Wang, and W. Wang. 2015. Free energy landscape and multiple folding pathways of an H-Type RNA pseudoknot. PLoS One 10:e0129089. 52. Xu, X. J., and S. J. Chen. 2012. Kinetic mechanism of conformational switch between bistable RNA hairpins. J. Am. Chem. Soc. 134:12499-12507. 53. Chen, J., and W. Zhang. 2012. Kinetic analysis of the effects of target structures on siRNA efficiency. J. Chem. Phys. 137:225102. 54. Shi, Y. Z., F. H. Wang, Y. Y. Wu, and Z. J. Tan. 2014. A coarse-grained model with implicit salt for RNAs: predicting 3D structure, stability and salt effect. J. Chem. Phys. 141:105102. 55. Pabit, S. A., J. L. Sutton, H. Chen, and L. Pollack. 2013. Role of ion valence in the submillisecond collapse and folding of a small RNA domain. Biochemistry 52:1539-1546. 56. Leipply, D., and D. E. Draper. 2011. Effects of Mg2+ on the free energy landscape for folding a purine riboswitch RNA. Biochemistry. 50: 2790-2799. 57. Meisburger, S. P., S. A. Pabit, and L. Pollack. 2015. Determining the locations of ions and water around DNA from X-ray scattering measurements. Biophys. J. 108:2886-2895. 58. Tan, Z. J., and S. J. Chen. 2006. Nucleic acid helix stability: effects of salt concentration, cation valence and size, and chain length. Biophys. J. 90:1175-1190. 59. Tan, Z. J., and S. J. Chen. 2010. Predicting ion binding properties for RNA tertiary structures. Biophys. J. 99:1565-1576. 60. Tan, Z. J., and S. J. Chen. 2007. RNA helix stability in mixed Na+/Mg2+ solution. Biophys. J. 92:3615–3632. 61. Hayes, R. L., J. K. Noel, A. Mandic, P. C. Whitford, K. Y. Sanbonmatsu, U. Mohanty, and J. N. Onuchic. 2015. Generalized manning condensation model captures the RNA ion atmosphere. Phys. Rev. Lett. 114:258105. 62. Hayes, R. L., J. K. Noel, P. C. Whitford, U. Mohanty, K. Y. Sanbonmatsu, and J. N. Onuchic. 2014. Reduced Model Captures Mg2+-RNA Interaction Free Energy of Riboswitches. Biophys. J. 106:1508-1519. 63. Hayes, R. L., J. K. Noel, U. Mohanty, P. C. Whitford, S. P. Hennelly, J. N. Onuchic, and K. Y. Sanbonmatsu. 2012. Magnesium fluctuations modulate RNA dynamics in the SAM-I riboswitch. J. Am. Chem. Soc. 134:12043-12053. 64. Hagerman, P. J. 1997. Flexibility of RNA. Annu. Rev. Biophys. Biomol. Struct. 26:139-156. 65. Herschlag, D., B. E. Allred, and S. Gowrishankar. 2015. From static to dynamic: the need for structural ensembles and a predictive model of RNA folding and function. Curr. Opin. Struct. Biol. 30:125-133. 66. Mouzakis, K. D., E. A. Dethoff, M. Tonelli, H. M. Al-Hashimi, and S. E. Butcher. 2015. Dynamic motions of the HIV-1 frameshift site RNA. Biophys. J. 108:644-654. 67. Casiano-Negroni, A., X. Sun, and H. M. Al-Hashimi. 2007. Probing Na+-induced changes in the HIV-1 TAR conformational dynamics using NMR residual dipolar couplings: new insights into the role of counterions and electrostatic interactions in adaptive recognition. Biochemistry. 46:6525-6535. 68. Zacharias, M., and P. J. Hagerman. 1995. Bulge-induced bends in RNA: quantification by transient electric birefringence. J. Mol. Biol. 247:486-500. 69. Chen, H., S. P. Meisburger, S. A. Pabit, J. L. Sutton, W. W. Webb, L. Pollack. 2012. Ionic strength-dependent 17 persistence lengths of single-stranded RNA and DNA. Proc. Natl. Acad. Sci. U. S. A. 109:799-804. 70. de Pablo, J. J. 2011. Coarse-Grained Simulations of Macromolecules: From DNA to Nanocomposites. Annu. Rev. Phys. Chem. 62:555. 71. Zhou, H. X. 2014. Theoretical frameworks for multiscale modeling and simulation. Curr. Opin. Struct. Biol. 25:67–76. 72. Kikot, I. P., A. V. Savin, E. A. Zubova, M. A. Mazo, E. B. Gusarova, L. I. Manevitch, and A. V. Onufriev. 2011. New coarse-grained DNA model. Biophysics, 56:387-392. 73. Daily, M. D., B. N. Olsen, P. H. Schlesinger, D. S. Ory, and N. A. Baker. 2014. Improved coarse-grained modeling of cholesterol-containing lipid bilayers. J. Chem. Theory Comput. 10:2137-2150. 74. Mustoe, A. M., H. M. Al-Hashimi, and C. L. Brooks. 2014. Coarse grained models reveal essential contributions of topological constraints to the conformational free energy of RNA bulges. J. Phys. Chem. B 118:2615-2627. 75. Wang, F. H., Y. Y. Wu, and Z. J. Tan. 2013. Salt contribution to the flexibility of single-stranded nucleic acid of finite length. Biopolymers. 99:370-381. 76. Huang, S. Y., and X. Q. Zou. 2014. A knowledge-based scoring function for protein-RNA interactions derived from a statistical mechanics-based iterative method. Nucleic Acids Res. 42:e55. 77. Huang, S. Y., and X. Q. Zou. 2011. Statistical mechanics-based energy scoring function for structural model selection in protein structure prediction. Proteins 79:2648–2661. 78. Zhang, Y., H. Zhou, and Z. Ouyang. 2001. Stretching single-stranded DNA: Interplay of electrostatic, base-pairing, and base-pair stacking interactions. Biophys. J. 81:1133-1143. 79. Xia, T., J. SantaLucia, M. E. Burkand, R. Kierzek, S. J. Schroeder, X. Jiao, C. Cox, and D. H. Turner. 1998. Thermodynamic parameters for an expanded nearest-neighbor model for formation of RNA duplexes with Watson-Crick base pairs. Biochemistry 37:14719-14735. 80. Mathews, D. H., J. Sabina, M. Zuker, and D. H. Turner. 1999. Expended sequence dependence of thermodynamic parameters improves prediction of RNA secondary structure. J. Mol. Biol. 288:911-940. 81. Thomas, D. G., J. Chun, Z. Chen, G. Wei, and N. A. Baker. 2013. Parameterization of a geometric flow implicit solvation model. J. Comput. Chem. 34:687-695. 82. Ren, P., J. Chun, D. G. Thomas, M. Schnieders, M. Marucho, J. Zhang, and N. A. Baker. 2012. Biomolecular electrostatics and solvation: a computational perspective.” Q. Rev. Biophys. 45:427-491. 83. Manning, G. S. 1978. The molecular theory of polyelectrolyte solutions with applications to the electrostatic properties of polynucleotides. Q. Rev. Biophys. 11:179-246. 84. Tan, Z. J., and S. J. Chen. 2005. Electrostatic correlations and fluctuations for ion binding to a finite length polyelectrolyte. J. Chem. Phys. 122:44903. 85. Tan, Z. J., and S. J. Chen. 2011. Salt contribution to RNA tertiary structure folding stability. Biophys. J. 101:176-187. 86. Walter, A. E., D. H. Turner, J. Kim, M. H. Lyttle, P. Müller, D. H. Mathews, and M. Zuker. 1994. Coaxial stacking of helixes enhances binding of oligoribonucleotides and improves predictions of RNA folding. Proc. Natl. Acad. Sci. 91:9218–9222. 87. Walter, A. E., and D. H. Turner. 1994. Sequence dependence of stability for coaxial stacking of RNA helixes with Watson-Crick base paired interface. Biochemistry 33:12715–12719. 18 88. Kirkpatrick, S., C. D. Gelatt, and M. P. Vecchi. 1983. Optimization by simulated annealing. Science. 220:671-680. 89. Parisien, M., J. A. Cruz, E. Westhof, and F. Major. 2009. New metrics for comparing and assessing discrepancies between RNA 3D structures and models. RNA 15:1875-1885. 90. Zhang, Y., and J. Skolnick. 2004. Scoring function for automated assessment of protein structure template quality. Proteins. 57:702-710. 91. Stallings, S. C., and P. B. Moore. 1997. The structure of an essential splicing element: stem loop IIa from yeast U2 snRNA. Structure. 5:1173-1185. 92. Tan, Z. J., and S. J. Chen. 2008. Electrostatic free energy landscapes for DNA helix bending. Biophys. J. 94:3137-3149. 93. Wu, Y. Y., L. Bao, X. Zhang, and Z. J. Tan. 2015. Flexibility of short DNA helices with finite-length effect: from base pairs to tens of base pairs. J. Chem. Phys. 142:125103. 94. Kuznetsov, S. V., C. C. Ren, S. A. Woodson, and A. Ansari. 2008. Loop dependence of the stability and dynamics of nucleic acid hairpins. Nucleic Acids Res. 36:1098-1112. 95. Sehdev, P., G. Crews, and A. M. Soto. 2012. Effect of helix stability on the formation of loop-loop complexes. Biochemistry. 51:9612-9623. 96. Tan, Z. J., and S. J. Chen. 2008. Salt dependence of nucleic acid hairpin stability. Biophys. J. 96:738-752. 97. Wu, Y. Y., Z. L. Zhang, J. S. Zhang, X. L. Zhu, and Z. J. Tan. 2015. Multivalent ion-mediated nucleic acid helix-helix interactions: RNA versus DNA. Nucleic Acids Res. 43:6156-6165. 98. Nixon, P. L., C. A. Theimer, and D. P. Giedroc. 1999. Thermodynamics of stabilization of RNA pseudoknots by cobaltIII hexamine. Biopolymers. 50:443–458. 99. Nixon, P. L., and D. P. Giedroc. 1998. Equilibrium unfolding (folding) pathway of a model H-type pseudoknotted RNA: the role of magnesium ions in stability. Biochemistry. 37:16116–16129. 19 FIGURES AND TABLES FIGURE 1. (a) Our coarse-grained representation for one fragment of an RNA superposed on an all-atom representation. Namely, three beads are located at the atoms of phosphate (P), C4’ (C), and N1 for pyrimidine or N9 for purine (N), respectively. The structure is shown with the PyMol (http://www.pymol.org). (b) The schematic representation for base-pairing (dashed line) and base-stacking (dash-dotted line). FIGURE 2. (a) Comparison of the RMSDs of predicted RNA 3D structures between the present/previous version of our model and the MC-Fold/MC-Sym pipeline. For each of the 32 tested RNAs, we use the MC-Fold/MC-Sym pipeline online tool (http://www.major.iric.ca/MC-Fold/) (30) to test the accuracy of MC-Fold/MC-Sym and calculate the RMSD for the top 1 predicted structure over C4’ atom in the backbone. The RMSDs of structures predicted by the present model at the experimental ion conditions (see in Table I) and by its previous version (54) at 1M NaCl are calculated over C beads from the corresponding C4’ atoms in the native structures. 20 FIGURE 3. The sequences and the secondary structures predicted by the present model for HIV-1 TAR variant with a 3-nt bulge and HIV-2 TAR variants with different length of polyU (polyA) bulges, which are used to study the flexibility of RNAs. FIGURE 4. (a, b) The experimental (Ref. 67) and predicted inter-helical bend angle as functions of [Na+] (a) and [Mg2+] (b) for HIV-1 TAR variant (see Fig. 3). The corresponding typical 3D structures predicted by the present model are shown with the PyMol (http://www.pymol.org). (c, d) The variances of radius of gyration and the RMSDs from time-averaged reference structures as functions of [Na+] (c) and [Mg2+] (d) for HIV-1 TAR variant (see Fig. 3). 21 2R FIGURE 5. (a, b) The experimental (Ref. 68) and predicted inter-helical bend angles as functions of bulge length at 5mM NaPO4 without (a) or with (b) 2mM Mg2+ for HIV-2 TAR variant (see Fig. 3). The corresponding typical 3D structures predicted by the present model are shown with the PyMol (http://www.pymol.org). (c, d) The variances of radius of gyration and the RMSDs from time-averaged reference structures as functions of bulge length at 5mM NaPO4 without (c) or with 2mM Mg2+ (d) for HIV-2 TAR variant (see Fig. 3). FIGURE 6. The RMSF for C-beads along HIV-1 TAR variant (see Fig. 3) in 1M NaCl solution. The sequences of bulge and hairpin loops are in italics. 22 2R FIGURE 7. The RMSF for C-beads along HIV-2 TAR variant (see Fig. 3) with different bulge length at 5mM NaPO4 with/without 2mM Mg2+. The length n of bulge loop is 1 (a), 3 (b), 5 (c) and 8 (d), respectively. FIGURE 8. (a) The experimental (Ref. 94) and predicted melting temperature as functions of length of hairpin loop for RNA hairpin R0 at 2.5mM [Mg2+]. (b) The experimental (Ref. 95) and predicted melting temperatures of four RNA hairpins with different stems at 0.7mM [Mg2+]. (c, d) The experimental (Ref. 98 and Ref. 99) and predicted melting temperature Tm’s as functions of [Mg2+] for two RNA hairpins R5 (c) and R6 (d) in the presence of 0.1 M [K+]. The sequences and the secondary structures predicted by the present model are also shown in (a), (b), (c) and (d). 23 Table I. The 32 RNA molecules for 3D structure prediction in this work. RNAs PDBa Length Type d (nt) [1+/2+]e (mM) RMSDpred. f (Å) (mean/minimum) 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 2y95b 2lp9 b 1j4y b 1yn2 b 1z30 b 1u2ac 1qwa b 1d0u b 17rac 1jur b 1osw b 2ro2 b 1bgzc 1s34 b 1lc6 b 2kez b 1m82 b 2l5z b 2aht b 1f6x b 1xsh b 1nbr b 2jwv b 1yne b 1jo7 b 2lwk b 2jxv b 2kpv b 1zc5 b 2kuu b 2mqt b 1p5o b 14 16 17 17 18 20 21 21 21 22 22 23 23 23 24 24 25 26 27 27 27 29 29 31 31 32 33 34 41 48 68 77 H B H H H H B B B B I H B&I B I I B I B B B B I B B&I B&I I B&I B B B&I B&I 100/0 65/0 20/0 55/40 50/0 75/0 5/0 50/0 20/0 100/0 25/0 12.4/0 20/0 35/0 50/0 50/0 20/0 50/5 100/6 100/0 100/0 20/0 50/0 10/0 10/0 50/0 20/0 20/0 10/0 20/0 10/0 2.0/1.0 2.6/1.1 3.9/1.9 4.1/2.3 1.7/0.9 2.1/1.1 2.8/1.5 3.8/1.5 3.7/1.3 2.9/1.5 3.7/1.5 2.8/1.4 3.4/2.1 3.5/1.7 3.7/2.0 3.4/1.5 2.1/1.2 4.0/2.6 3.2/1.3 2.8/1.5 3.4/1.7 3.3/2.1 4.5/2.0 2.6/1.3 4.0/2.3 3.9/1.7 3.2/1.9 4.1/1.9 3.5/1.8 4.5/2.3 6.4/3.8 100/5 11.0/8.7 g pred. RMSD0 (Å) 2.2 2.7 4 4.3 1.9 2.6 3 3.7 4 2.9 4 2.9 4.6 4 4.1 3.6 2.4 4.6 3.9 2.9 3.6 3.7 5 2.7 4.6 4.2 3.9 4.3 3.4 5.1 6.8 13 h RMSDMC-Sym (Å) 2 2.8 4.6 2.9 2.6 3 4.6 3.7 4.9 3.4 4 2.6 4.8 2 5.7 2.9 2.2 5.2 3.8 2.7 2.9 3.1 4.9 3.9 4.9 3.3 4.9 4.1 5.9 5.5 7.1 10.8 a The 3D structures of these RNA hairpins have been determined by NMR method at certain ion conditions. b, c The hairpins are experimentally determined afterb and beforec the year of 2000, respectively. d H, B, I, B&I represent RNA hairpins without bulge/internal loops (H), RNA hairpins with bulges (B), RNA hairpins with internal loops (I), and RNA hairpins with bulge and internal loops (B&I), respectively. e The salt conditions of solutions in which RNA structure have been experimentally determined. f The mean/minimum RMSDs are calculated over C beads of the structures predicted by the present model from the corresponding atoms C4’ of the native structures. g The mean RMSDs are calculated over C beads of structures predicted by our previous model at 1M NaCl from the corresponding atoms C4’ of the native structures. h The RMSD is calculated over the C4’ atoms of the top 1 structure for each RNA predicted by the MC-Fold/MC-Sym pipeline (http://www.major.iric.ca/MC-Fold/) (30) from the native structure. 24
1502.07166	2	1502	2016-01-20T15:43:29	Solitary electromechanical pulses in Lobster neurons	[ "physics.bio-ph", "q-bio.NC" ]	Investigations of nerve activity have focused predominantly on electrical phenomena. Nerves, however, are thermodynamic systems, and changes in temperature and in the dimensions of the nerve can also be observed during the action potential. Measurements of heat changes during the action potential suggest that the nerve pulse shares many characteristics with an adiabatic pulse. First experiments in the 1980s suggested small changes in nerve thickness and length during the action potential. Such findings have led to the suggestion that the action potential may be related to electromechanical solitons traveling without dissipation. However, they have been no modern attempts to study mechanical phenomena in nerves. Here, we present ultrasensitive AFM recordings of mechanical changes on the order of 2 - 12 {\AA} in the giant axons of the lobster. We show that the nerve thickness changes in phase with voltage change. When stimulated at opposite ends of the same axon, colliding action potentials pass through one another and do not annihilate. These observations are consistent with a mechanical interpretation of the nervous impulse.	physics.bio-ph	physics	Solitary electromechanical pulses in Lobster neurons. Alfredo Gonzalez-Perez1, Lars D. Mosgaard1, Rima Budvytyte1, Edgar Villagran-Vargas3, Andrew D. Jackson2 and Thomas Heimburg1,∗ 1Niels Bohr Institute, University of Copenhagen, Blegdamsvej 17, 2100 Copenhagen Ø, Denmark 2Niels Bohr International Academy, University of Copenhagen, Blegdamsvej 17, 2100 Copenhagen, Denmark 3Facultad de Ciencias, Universidad Aut´onoma del Estado de M´exico. Instituto Literario 100, Colonia Centro, 50000, Toluca, M´exico ABSTRACT Investigations of nerve activity have focused predominantly on electrical phenomena. Nerves, however, are thermodynamic systems, and changes in temperature and in the dimensions of the nerve can also be observed during the action potential. Measurements of heat changes during the action potential suggest that the nerve pulse shares many characteristics with an adiabatic pulse. First experiments in the 1980s suggested small changes in nerve thickness and length during the action potential. Such findings have led to the suggestion that the action potential may be related to electromechanical solitons traveling without dissipation. However, they have been no modern attempts to study mechanical phenomena in nerves. Here, we present ultrasensitive AFM recordings of mechanical changes on the order of 2 -- 12 A in the giant axons of the lobster. We show that the nerve thickness changes in phase with voltage change. When stimulated at opposite ends of the same axon, colliding action potentials pass through one another and do not annihilate. These observations are consistent with a mechanical interpretation of the nervous impulse. Keywords: action potential, atomic force microscopy, nerve pulse collision, heat, capacitance ∗corresponding author, [email protected]. Introduction It is not generally appreciated that the size of the nervous im- pulse is remarkably large. A myelinated motor neuron has a pulse velocity on the order of 100 m/s. Given a typical pulse duration of 1 ms, the resulting size of the nerve pulse is 10 cm. Propagation velocities in slow, non-myelinated fibers are re- duced to 1-10 m/s, indicating a pulse size of 1-10 mm. Thus, nerve pulses are macroscopic phenomena spanning a signif- icant fraction of the total axon length. In some cases they can even be larger than small neurons such as interneurons that are only a few 100 µm long (1). In the past, the activity of nerves has conventionally been considered to be a purely electrical phenomenon produced by the flux of ions and the charging of the membrane capacitor (2). The shear size of a nerve pulse suggests that the macroscopic thermodynamic properties of the nerve membrane or of the entire nerve ought to be taken into account. It is to be expected that the state of the nerve cell depends not only on electrochemical potentials and the conjugated flux of ions but also on all other thermody- namic forces including variations in lateral pressure (resulting in changes of membrane area and thickness) and temperature (resulting in heat flux). It is therefore not surprising that, dur- ing the action potential, one finds changes not only in voltage but also in thickness (3 -- 6), length (5, 7) as well changes in membrane temperature (8 -- 11). The change of thickness of a single squid axon was found to be on the order of 1 nm, and temperature changes range between 1 -- 100 µK depending on the specimen. While both mechanical and thermal signals are very small, they are found to be in phase with voltage changes. The heat signal can be blocked by neurotoxins such as tetrodotoxin (11). This strongly suggests that these ther- modynamic phenomena are correlated with the voltage pulse and do not represent independent secondary phenomena. From a thermodynamic viewpoint, one can consider two extreme cases of possible dynamic changes associated with the action potential: 1. Purely dissipative processes during which entropy increases, such as the flow of ions along con- centration gradients. Such processes form the basis of the Hodgkin-Huxley model for the action potential. 2. Adia- batic processes that do not dissipate heat and thus conserve entropy. These phenomena are rather governed by the laws of analytical mechanics. They play an important role for dy- namic properties such as sound propagation. In this context, the heat changes observed in nerves are of fundamental inter- est. It was found that heat is released during an initial phase of the action potential and that this heat is reabsorbed in the final phase of the action potential. During the nerve pulse no heat (or only very little) is dissipated, and the entropy of the membrane is basically conserved (9, 11). Thus, thermal measurements suggest that the action potential is an adiabatic phenomenon reminiscent of a sound wave. As early as 1912, the striking absence of heat production led Hill to conclude: 'This suggests very strongly ... that the propagated nervous impulse is not a wave of irreversible chemical breakdown, but a reversible change of a purely physical nature' (12). In con- trast, the contemporary understanding of the nerve pulse is based on the flow of ions along gradients through ion channel proteins (2) and therefore assumes that it is of a dissipative nature. Hodgkin himself compared it to the 'burning of a fuse of gunpowder' (13). There is thus a disagreement between electrophysiological models and some thermodynamics find- ings. These problems are not easily resolved and merit careful attention. The observed reversible change in temperature as well as mechanical changes seen in optical and mechanical experi- ments led to the proposal that the action potential is a con- sequence of an electromechanical pulse or soliton (14). A condition for the existence of such a soliton is the existence of an order transition in the membrane from solid to liquid slightly below physiological temperature. Such transitions have been found in various biological membranes (15). It is thought that the soliton consists of a region of ordered lipid membrane traveling in the otherwise liquid membrane with a speed somewhat less than the speed of sound in the mem- brane (14, 16). This is shown schematically in Fig. 1. The 1 6 1 0 2 n a J 0 2 ] h p - o i b . s c i s y h p [ 2 v 6 6 1 7 0 . 2 0 5 1 : v i X r a difference in membrane thickness between the solid and the liquid is of order 1 nm (≈17% of the total membrane thick- ness). The associated reversible change in energy is related to the latent heat of the membrane transition and thus to the reversible heat production found in nerves. Since the mem- brane changes its thickness, changes in membrane voltage of order 50mV are to be expected as a consequence of changes in its capacitance (17, 18). Thus, the soliton is of an elec- tromechanical or piezoelectric nature. An intrinsic feature of such solitons is that two colliding pulses pass through each other without dissipation (16) rather than annihilating as ex- pected in Hodgkin-Huxley-like pulses due to the refractory period. In fact, the penetration of colliding nerve pulses was seen recently in nerves from earth worm and the ventral cord of lobster (19). The soliton model treats the pulse as a longi- tudinal compressional density change that is strongly affected by the presence of a phase transition in the membrane. The velocity of the pulse is closely related to the sound velocity in a liquid lipid membrane. It predicts pulse velocities close to 100 m/s, very similar to those in myelinated nerves. In the soliton model, the pulse velocity is not related to the axon radius, while it depends on the square root of the radius in the HH-model. However, there is evidence that the radius dependence in real nerves deviates from the latter behavior. For instance, Goldman (20) found that the 4-fold stretching of a single neuron (equivalent to a 4-fold decrease in radius) did not lower the pulse velocity. In fact, upon stretching the pulse velocity first increased and then stayed constant over a significant range of axon radii. Figure 1: Schematic representation of a mechanical soliton traveling in a nerve fiber (according to (14)). Small local changes in thickness are caused by pressure- induced order transitions in the membrane. Red re- gions correspond to ordered lipids in the otherwise liq- uid (green) lipid membrane. In a living nerve, the spatial extension of the pulse is much larger than shown here. Changes in voltage, thickness and temperature in a soli- ton are all associated with a single phenomenon, and it is not appropriate to consider some of these changes as side effects of another dominant process. They are rather different as- pects of the same phenomenon as seen by different instru- mentation. There exists clear evidence that electromechanical pulses can travel on lipid monolayers close to the LE-LC tran- sition with velocities very close to those of non-myelinated nerves (21, 22). However, there is a striking lack of experi- ments that actually demonstrate the mechanical nature of the nerve pulse. For this reason, the thermodynamic and mechan- ical interpretation of the nerve pulse has been widely ignored. Materials and methods Materials. In our experiments we used lobster, Homarus americanus, that was obtained from a local supplier that im- ported the animals from Canada. We used a lobster saline solution adapted from Evans et al. (23) 462 mM NaCl, 10 mM KCl, 25 mM CaCl2, 8 mM MgCl2, 10 mM TRIS and 11 mM Glucose, adjusted to pH 7.4 with NaOH. All chemicals used in the preparation were purchased from Sigma-Aldrich. Recording of the electromechanical action potential. The electrical signal was recorded using a Powerlab 8/35 (ADIn- struments Europe, Oxford, UK). It was preamplified with a differential amplifier DP-304 from Warner Instrument Corpo- ration with a gain x1000 using a 3 kHz lowpass and a 10 Hz highpass filter. The mechanical displacement was obtained using an AFM, NanoWizard II from JPK (Germany). The sig- nal was fed into the PowerLab 8/35 and analyzed simultane- ously with the electrical recording without any preamplifica- tion. The AFM was mounted on the top of an inverted micro- scope Olympus IX71 (Olympus Corporation, Japan) placed on an anti-vibration table TS-150 (low power, from Herzan LLC, USA). The full setup containing the AFM, optical mi- croscope as well as the anti-vibration table was placed in- side a Faraday cage in order to avoid external electrical noise. We used tipless cantilevers from Mikromasch Europe (type HQ:CSC37 and HQ:CSC38). The resonance frequency is 20 kHz for the HQ:CSC37 cantilever and the force constant is 0.3 N/m. The resonance frequency is 10 kHz for the HQ:- CSC38 cantilever and the force constant is 0.03 N/m. Thus, the resonance frequencies are outside of the range of the ex- pected signal (1-2 s). Both cantilevers were used under the same experimental conditions. The recording signal fre- quency was 40 kHz for both, the electrical and mechanical measurement. All the experiments were performed at room temperature of about 22 C. Nerve chamber. We used two different nerve chambers. For the collision experiment the nerve chamber is composed by an array of 21 stainless steel electrodes in a longitudinal cav- ity covered by a lid in order to isolate the nerve once ex- tracted. The lid also allows to keep the nerve in a saturated va- por atmosphere to prevent the ventral cord from drying. The nerve chamber is a 7 x 2.5 cm block by 1 cm height made on Plexiglass that contains two longitudinal perforation of 1.5 cm length by 0.5 cm wide and 0.5 cm width (19). The large longitudinal aperture contains an array of 21 perforations to allocate the stainless steel electrodes. The array was placed about 0.25 cm from the top of the chamber. The distance be- tween two consecutive electrodes is of 0.25 cm. The stainless steel electrodes have a length of about 3.4 cm and a diam- eter of 0.5 mm and were fixed in the perforation along the chamber by using a rubber replica casting system (Reprorub- ber, Islandia, NY) composed by a base and a catalyst from Flexbar Machine Corp (Islandia, NY). For the AFM experi- ment we used a variation of the previous chamber with half size thickness and with a small open area in the middle with- out electrodes that allows access to the AFM cantilever. Sample preparation. The Lobster, Homarus americanus, was anesthetized by keeping the animal in the freezer for about 30 min. Once removed from the freezer the animal was placed 2 Figure 2: Outline of the mechanical experiment. A. Image of the two lobster connectives including the brain (top) and the subesophageal ganglion (bottom). B. Detail of a lobster connective with the lateral giant axon exposed. C. Schematic representation of the cross-section of a lobster connective containing a medial giant axon, a lateral giant axon and several small fibers. The sheath surrounding the connective is cut in the longitudinal direction. D. After opening the sheats, the two giant axons are exposed to the outside solution. Mechanical signals are recorded with a tipless AFM cantilever. The open connective is placed on a cell containing several pairs of electrodes to stimulate the nerve and to record the signal. The mechanical response is measured on a planar support in the middle of the cell. on the dissecting table and the legs were removed. A second cut was made at the beginning of the abdomen to separate the abdominal part from the thorax. In the thorax, we removed the carapace and made the extraction dorsally. The ventral cord is contained in a tube-like structure formed by the ex- oskeleton of the lobster and is easy to remove after break- ing the shell with scissors. The thoracic segment of the ven- tral cord including the brain was extracted as a single piece. Special care was taken to remove the circumesophageal con- nectives from their location surrounding the esophagus. The brain, circumesophageal connectives and subesophageal gan- glia was removed from the rest of the ventral cord as placed in a petri-dish with lobster saline solution. The basic anatom- ical features as well as the basic steps in the preparation are described in the literature (24). The two circumesophageal connectives were severed at the level of the brain and the subesophageal ganglia (Fig. 2A). From each circumesophageal connective we remove the external connective sheath to expose the main median giant axon (with diameters between 150 and 200 µm, cf. Fig. 2B and C). The preparation was transferred to the AFM nerve chamber with the medial giant axon facing the AFM can- tilever (Fig. 2D). For the collision experiment the external connective layer was not removed ensuring a longer survival of the nerve. One of the giant axons was cut at one or two po- sition to rule out the possibility that propagation of pulses in opposite directions occurs in different axons. The total length of the connectives used by us was ≈ 4-5 cm. Collision experiments. The collision experiment was per- formed independently using a PowerLab 26T from ADIn- struments. The instrument possesses an internal bio-amplifier that allows the recording of small electrical potential on the 3 order of microvolts. The bio-amplifier contains two recording channels (further description see ADInstruments webpage). We used the Labchart software from ADInstruments in or- der to record the signals coming from the ventral cord. The recording frequency was 40 kHz. All experiments were per- formed at room temperature of about 22 C. In some experiments we cut the LG axon of the connec- tive at two locations in order to observe a collision of pulses in the MG axon only (cf. Fig. 4, right). In order to make sure that the same axon was cut twice, we performed the following tests: 1. The intact connective (before removing the sheath and cut- ting the LG axon), we observed two signals from each side corresponding to the action potentials in the MG and LG ax- ons. 2. We performed on single cut at the one end of LG axon. Then, connective was tested again by stimulating at one end of axon before the cut and recording after cut. Only a single peak originating from the MG axon was observed. When placing the stimulation electrode behind the cut, two peaks from the LG and MG axons could be observed. 3. A second cut was made close to the other end of connective. When placing the stimulation electrodes before the first cut and the recording electrodes after the second cut, only one action potential originating from the MG axon was observed, while LG signal was absent. Ethical. The work described in this article has been carried out in accordance with the policy on the use of animals of the Society for Neuroscience. Figure 3: Vertical displacements as recorded by AFM. Left: Differential voltage change (A), integrated sig- nal (B) and corresponding mechanical displacement in the same neuron (C). Right: Mechanical recordings from six different preparations all yield mechanical changes between 0.2 and 1.2 nm. All recordings have a similar shape with small differences due to the size of the nerve and the precise positioning of the AFM cantilever. Mechanical and electrical signals were not measured at exactly the same location. In this figure, they have been temporally aligned. Results Mechanical changes Here, we present the results of experiments on the mechanical changes in single axons from lobster connectives. The ventral cord of the lobster is made of two connective strands contain- ing one lateral and one median giant axons each (LG and MG axons, respectively). Additionally, they contain several small nerve fibers. The action potentials of the giant axons can be clearly distinguished from those of the small fibers. The lat- ter yield much smaller signals and are excited only at higher voltages. The MG axon displays a larger peak amplitude and a larger conduction velocity as compared to the LG axon. We recorded the mechanical response of the giant axons using atomic force microscopy (AFM). Since the vertical dis- location of the axonal membrane is expected to be small, the surface of individual giant axons must be accessed directly by the AFM cantilever. In order to achieve this, the sheath sur- rounding the connective was cut open. This provides direct access to the single axons (a single exposed LG axon is shown in Fig. 2 B together with the opened connective). The exper- imental setup is described schematically in Figs. 2C and D, and described in detail in the Materials and Methods section. In brief, a tipless cantilever is placed on a single axon exposed from the opened connective. The open connective is placed on top of a cell with 21 electrodes that allow us to stimulate the nerve and to measure the electrical response. In the center of the cell, the nerve is placed on a support for the mechani- cal measurement. We stimulated the nerve periodically with a pair of electrodes at one end. We monitor only the vertical displacement of the cantilever without scanning in the x-y plane. The sampling rate is 40 kHz, and the response time of the AFM setup is about 1 ms. The experimental results are shown in Fig. 3 (left panels). The nerve signal is recorded by two electrodes separated by a distance less than the width of the nerve pulse. Thus, in effect, the first derivative of the true voltage change is measured (Fig. 3A). The integral of this signal is roughly proportional to the true voltage change (Fig. 3B). The bottom trace is the AFM signal (Fig. 3C). In or- der to achieve a good signal-to-noise ratio we averaged over approximately 100 individual action potentials, excited at in- tervals of 0.2 seconds. In the experiment shown in Fig. 3C, the vertical displacement was about 2 A. Fig. 3D) shows me- chanical signals from six different lobster specimens. We find displacements between 2 and 12 A lasting between ∼2 and 4 ms. Thus, the mechanical changes in these neurons can be measured in a consistent and reproducible manner. We found these mechanical signals consistently in all connective prepa- rations studied. Electrical recordings demonstrated that more than one single axon was stimulated in the recordings with larger voltage amplitudes. The stimulation artifact visible in 4 Figure 4: Pulse collision experiment in the median axon of the lobster connective. The top panels represent a schematic drawing of the connective, the position of the electrodes and of the cuts in the lateral giant axon. Left: The lateral giant axon was cut at one position on the orthodromic side of the recording electrodes. Trace A. The nerve was stimulated in orthodromic direction with the two red electrodes The signal was recorded with the two blue electrodes. Trace B. The nerve was stimulated in antidromic direction with the two green electrodes and recoded with the blue electrodes.Trace C. The nerve was stimulated from both sides. Both pulses can be measured with the blue electrodes, suggesting that they did not annihilate upon collision. The dashed line is the sum of the orthodromic and the antidromic signals from traces A and B and is given for comparison. Right: A similar experiment in which the lateral giant axon was cut at two positions left and right of the recording electrodes. In this experiment, both orthodromic and antidromic pulses could be seen at the recording electrodes suggesting that the did not annihilate upon collision. The grey-shaded regions display the stimulation artifact. For experimental details see also (19). the electrical recordings had no influence on the mechanical recording. This suggests that the measured displacement is not due to a direct influence of voltage on the cantilever. All of our recordings were averaged over ≈ 100 pulses. Our measurements displayed some variance of the thickness change between 0.15-1.2 nm, which is partially related to the quality of the contact between the cantilever and the neu- ronal membrane.Therefore, the displacement of 0.15 nm as in Fig. 3C should be regarded as a lower limit of the true dis- location. Further, the mechanical amplitude depends on how many neurons fire at the same time. Most recordings dis- played a displacement around 1 nm (partially shown in Fig. 3D). A thickness change of 1nm is very significant because it is close to the thickness difference between a liquid and a solid membrane with the correct sign. This is exactly the change expected in the electromechanical soliton model. It is important to notice that we find that the mechanical changes are proportional to the voltage changes (see Fig. 3 B and C). This is consistent with an electromechanical in- terpretation of the coupling between voltage and membrane thickness as put forward by (18) and (25). However, this re- sult deviates from the data by Iwasa and Tasaki (3, 4) who find that the mechanical trace resembles the first derivative of the voltage trace as it is typically obtained in extracellular recordings with two electrodes on the neuron (personal com- munication with K. Iwasa, cf. Fig. 3A). In this respect, our data are not merely a confirmation of previous data but ac- tually provide new evidence. This will be discussed in more detail in the Discussion section. Pulse collision In a recent publication we showed that action potentials trav- eling in opposite direction in some nerves (earth worm, lob- ster ventral cords) can pass through each other upon colli- sion (19). Whether this is a generic feature of nerves is not clear. In a very early experiment Tasaki showed that colliding pulses in single axons from sciatic nerves of toads annihilate (26). Penetration of pulses speak in favor of a mechanical mechanism for nerve pulse propagation. In contrast, annihi- lation favors a dissipative ion-channel based view. Here, we present evidence for pulse penetration in single axons from the connectives that were used in the mechanical experiment. We stimulated an action potential in orthodromic 5 and in antidromic directions. We performed experiments on about 20 lobsters. The voltage was chosen such that only one axon was exited. In about 6 experiments, we severed one axon of the connective in order to eliminate the possi- bility that pulses in opposite directions traveled in different axons (indicated in Fig. 4, top left). In an additional 3 exper- iments, we cut one axon at two different locations (indicated in Fig. 4, top right, see Materials and Methods section for de- tails). In these experiments, there exists only one giant axon between the three pairs of electrodes. Stimulation was made using pairs of electrodes at the ends of the axon. Propagating pulses were recorded with a pair of electrodes at a position closer to the antidromic side (indicated in Fig. 4, top left) or closer to the orthodromic side (indicated in Fig. 4, top right). The intact giant axon (the MG fiber) was stimulated in the orthodromic direction (trace A), in the antidromic direction (trace B) and in both directions (trace C). When increasing the voltage it became evident that only one giant axon could be stimulated. Some residual activity from the small fibers could sometimes be seen. In the experiment shown in Fig. 4, left), the antidromic signal arrives at the recording site prior to the orthodromic signal. When stimulated at both ends, the re- sulting trace looked similar to the sum of single orthodromic and single antidromic pulses indicating that the pulses pen- etrated without major perturbation. This is consistent with previous findings in worm axons and other axons in lobster (19). In the experiment shown in Fig. 4, right, the antidromic signal arrives at the recording site after the orthodromic sig- nal. After collision, both pulses can still be seen, indicating that they passed through each other. We have shown that the same events can also be seen when nerves are inserted into capillaries and the nerves are complete surrounded by saline solution (data not shown). Discussion Here, we have reported mechanical changes in single axons from lobster connectives. These are the first AFM recordings of an action potential in a single axon, and they are probably the most sensitive recordings of mechanical changes in a sin- gle axon so far. We have also reported collision experiments in the single axons that show that action potentials traveling from the two ends of an axon can pass through each other without being annihilated. We discuss below why such find- ings are important. Mechanical recordings of nerve pulses were pioneered by Iwasa & Tasaki in the early 1980s (3, 4). They reported dis- placements of the membrane of the squid axon on the order of 1 nm. While the magnitude of the displacement in our ex- periments is of the same order, the functional form of their displacement differs from ours. In our data the displacement is proportional to the voltage change in the axon, whereas the data by Iwasa & Tasaki are in fact proportional to the first derivative of the voltage. Thus, in their experiments the mechanical trace resembles the one shown in Fig. 3A, while our data resemble the traces shown in Fig. 3B. Our re- sults are consistent with data by Kim et al. (6) on synapse bundles, where the mechanical changes were also found to be proportional to the voltage change. Optical recordings 6 of dimensional changes in nerves also indicate that displace- ments are proportional to voltage (6, 27). This difference in functional form is important because Iwasa's & Tasaki's data are not consistent with the concept of electrostriction, while the data presented here are actually in good agreement with this concept (see Appendix A for details). For this reason, the data by Iwasa & Tasaki are also not consistent with the soliton model put forward in (14). The origin of the devia- tion of the early mechanical data from other recordings is not quite clear. Later direct recordings of mechanical changes induced by voltage include (28), who determined movement of membranes induced by voltage changes across HEK cell membranes by AFM (they find a displacement proportional to voltage), and (29), who measured mechanical changes in the soma of rat PC12 cells with a piezo-electric ribbon (reporting a force generated by a membrane proportional to voltage). The latter publication pointed out that AFM experiments on single neurons are in fact very difficult. These reports justify the need of studies such as presented here. In his textbook from 1964 Alan L. Hodgkin recommended: "In thinking about the physical basis of the action potential perhaps the most important thing to do at the present moment is to consider whether there are any unexplained observations which have been neglected in an attempt to make the experi- ments fit into a tidy pattern" (13). Hodgkin was especially concerned about the temperature changes measured during the action potential (8). While Hodgkin's concerns reflect good scientific praxis, neither the measurement of heat nor mechanical changes associated with the action potential have received the attention that they merit. Here, we have demon- Our experimental findings are embedded into a discus- sion about the validity of the Hodgkin-Huxley model, and the possibility of the existence of thermodynamic phenomena. Both the mechanical changes reported here, and the thermal changes described in the introduction, are in fact not included in the Hodgkin-Huxley model as it is presently understood. One has to distinguish between the HH-model from 1952 (2) (or modern adaptations of it as in (30)) that is written in a precise mathematical language and a broader electrochemical viewpoint loosely referring to the function of ion channels. The Hodgkin-Huxley model itself does not explicitly contain either mechanical changes or temperature changes. The only thermodynamic term in the HH equation is the charging of a capacitor with constant capacitance. The latter ad hoc as- sumption excludes the possibility of mechanical changes in the HH-model, because a thickness change of the membrane will by necessity lead to changes in its capacitance. There- fore, the HH-model cannot be complete. An electrochemi- cal theory that quantitatively contains the mechanical changes and the temperature changes does not yet exist. The implica- tions of the mechanical and thermal findings are discussed in depth in Appendices A and B. The reversible heat changes discussed in the introduction are not consistent with the Hodgkin-Huxley model either be- cause 1. the magnitude of the heat change is too large (by at least a factor of two ) and 2. because a heat change at the site of the membrane will only be positive upon discharge of the membrane capacitor, but not negative upon charging. This was pointed out by Howarth et al. (9). Heat changes are dis- cussed in more detail in Appendix B. strated that mechanical signals in individual axons from lob- ster connectives propagate in phase with electrical signals. This is in agreement with very early recordings in squid by Iwasa and Tasaki (3, 4). The vertical amplitudes vary from 0.2-1.2 nm. We have also shown that in our nerve prepara- tion pulses traveling in opposite directions pass through each other without significant distortion. This is expected for elec- tromechanical pulses or solitons. It is, however, unexpected within the framework of a Hodgkin-Huxley mechanism (see also (19)). There exists one old (somewhat unclear) paper from 1949 (26) that reported pulse annihilation in single ax- ons from the sciatic nerve of toads. To our knowledge, this is the only other study besides (19) and the present study in- vestigating pulse collisions in a single axon. Since we report on a different nerve, we cannot claim that this is not possible. However, it would be interesting to study the origin of the deviating result in different preparations. While our findings do not prove any particular mechanism for the action poten- tials, they do suggest that it is advisable to study changes in thermodynamic variables in addition to voltage and current. Recently, there have been various reports, both theoreti- cal and experimental, regarding the possibility of mechanical pulse propagation in artificial systems close to transitions and in nerves (14, 16, 17, 21, 22, 31 -- 33). Heimburg and Jack- son (14) argued that, close to the phase transitions found in biological tissue, electromechanical solitons with properties similar to those of the action potential can travel along the nerve axons. El Hady and Machta (25) recently argued that, in general, the change of charge on the membrane capacitor leads to electrostrictive forces that must alter the membrane dimensions (see also (18)). In (34) it was shown that such an electromechanical picture can provide important insights regarding the mechanism of general anesthesia, which is then seen as the familiar physical chemical effect of anesthetics on melting transitions in biomembranes. It seems likely that the neglect of non-electrical effects has been motivated by the relatively narrow and exclusively electrical focus of the ac- cepted framework provided by Hodgkin and Huxley (2). The study of the mechanical nature of the nerve pulse promises important insights into the underlying thermodynamic nature of the action potential and its control by thermodynamic vari- ables distinct from voltage. Acknowledgments: This work was supported by the Villum Foundation (VKR 022130). Appendix Changes in membrane thickness in the HH- model and the soliton theory Any voltage applied across a membrane will change the mem- brane dimensions (18, 25, 35). Furthermore, any dimensional change of the membrane will change its capacitance (18). This effect is called piezoelectricity or electrostriction. It is independent of the origin of the voltage change. Thus, volt- age changes introduced by currents through a protein as as- sumed in the Hodgkin-Huxley model can also induce thick- ness changes. This statement, however, does not imply that such mechanical changes are contained in the HH-model or even consistent with it. The differential equation describing pulse propagation in the Hodgkin-Huxley model is given by (cid:88) i a 2Ri ∂2V ∂x2 = Cm ∂V ∂t + gi(V − Ei) . (1) Here, a is the radius of the axon, Ri is the specific resistance of the internal medium, the gi are the conductances of the membrane for ion species i and the Ei are the corresponding Nernst potentials. Ic = Cm(dV /dt) is the capacitive cur- rent assuming constant capacitance Cm. However, if mem- brane dimensions can change, the capacitive current is cor- rectly given by (18) Ic = Cm ∂V ∂t + V ∂Cm dt . (2) Only the second term on the right hand side allows for di- mensional changes of the membrane. One might hope that this term was small and could be neglected. However, as shown in (18), a change in membrane thickness such as the one observed here leads to a change in capacitance of ≈50%. A quick back-of-the-envelope estimate suggests that the two terms of eq. (2) have a roughly equal order of magnitude. This means that the second term cannot be neglected. Thus, the HH model is at least incomplete. It contains neither the possibility of changes in capacitance, nor electrostrictive forces, nor any other work term not contained in the charging of a capacitor with constant capacitance. If capacitance changes were allowed, the HH equations would assume a different form. In particular, the second term of eq. (2) and the elas- tic constants of the membrane would have to be introduced. This requires a thermodynamic theory for the changes in ca- pacitance. This is clearly not contained in the present under- standing of the HH-model, and this problem is not easily fixed without introducing detailed knowledge about the thermody- namics of the membrane. In contrast, the soliton model takes precisely such matters into account since it is a hydrodynamic theory based on the changes of the elastic constants and di- mensional changes of the membrane during the nervous im- pulse. The soliton model therefore provides a thermodynamic basis for the changes in capacitance and for piezoelectricity. Since the membrane capacitance changes proportionally with the membrane thickness, voltage changes will be proportional to the membrane thickness. This is the case for the data re- ported here, but not for the data by Iwasa & Tasaki (3, 4). Throughout our text we refer to eq. (1) when we dis- cuss the Hodgkin-Huxley model. It represents a very specific mindset with well-defined mathematics. There exist modern adaptations for more complicated scenarios (30) that also as- sume constant capacitance. We do not refer to more gen- eral pictures of the nerve membrane somehow containing ion channels that are not associated to a clearly worked-out math- ematical form. Heat production in nerves One of the striking properties of the nerve pulse is the ob- servation of reversible heat production during the action po- tential. A first phase of heat release is followed by a second 7 phase of heat reabsorption of nearly equal magnitude. We dis- cuss below how heat exchange arises in the Hodgkin-Huxley model and in the soliton model. We further argue that the heat changes are inconsistent with the HH model but con- sistent with a view where the action potential consists of an adiabatic pulse in the membrane. Hodgkin and Huxley (2) modelled the nerve by equivalent cir- cuits containing resistors (the ion channels), capacitors (the membrane) and a battery (the concentration differences of Na+ and K+ ions between inside and outside). In such an electrical view, the resistors heat up when a current flows independent of the direction of the currents (Joule heating). There is no mechanism to cool the membrane. Looking at thermodynamic analogues of the HH-model, the charging and discharging of a capacitor is caused by the flow of ions from on side to the other, which can be considered as ideal gases that expand through semipermeable walls. The expansion of an ideal gas does not lead to a change in the energy of the ions. Therefore, the work W done by the ions on the capacitor is equivalent to the heat Q that is absorbed from the reservoir (W=-Q). Thus, the charging of a capacitor leads to the absorption of heat from the reservoir that is exactly 2 Cm(V − V0)2 (35). equal to the free energy on the capacitor 1 Upon discharging the capacitor, this heat is released locally at the site of the membrane. Thus, as expected for an ideal gas, when reversible work is done on a nerve, there is no net heat production after the pulse. This scenario, called the con- denser theory was discussed in depth by (9). They made a de- tailed summary of their findings and concluded amongst oth- ers: "The condenser theory, according to which the positive heat represents the dissipation of electrical energy stored in the membrane capacity, while the negative heat results from the recharging of the capacity, appears unable to account for more than half of the observed temperature changes." This implies that the observed heat changes are simply too large to be consistent with the charging of the membrane alone. In the condenser theory, the heat is absorbed in the bulk (i.e., the battery) but released locally. Therefore, the condenser theory is not isentropic, and the membrane itself is not adiabatic on the time scale of the nerve pulse. A thermocouple placed di- rectly on the membrane would see an increase of the temper- ature upon discharge of the capacitor but no decrease in tem- perature upon charging it as a consequence of ion flows. It is essential in these thermodynamic analogies to the HH-model that the changes of ion-concentrations in the bulk provide the energy for charging the membrane, and not some energy of the membrane itself. It should also be noted, that the ther- modynamic reinterpretation of the HH-model made above is not identical to the HH-model because the latter is based on equivalent circuits. I.e., it is a purely electrical theory. Howarth et al. conclude that "It seems probable that the greater part of the initial heat results from changes in the entropy of the nerve membrane when it is depolarized and re- polarized". This is in fact the view of the soliton model. It requires a density change of the membrane from a liquid to a solid state of the membrane. During the first phase of the pulse, the latent heat of the transition is released into the en- vironment of the nerve membrane. During the second phase, the membrane returns to the liquid state, and the latent heat is reabsorbed. This is exactly the chain of events seen in experimental heat recordings. In (14), the transient heat re- leased during the action potential was explained be the en- ergy change of the membrane during a compression. It was found that the heat release is qualitatively and quantitatively consistent with the heat recordings by (8 -- 11). In the soliton model, the reversible heat release is due to the work neces- sary to compress a charged membrane. Thus, it contains both the charging of the capacitor and the mechanical work per- formed in order to change the membrane density. Since the physiological membrane is in its liquid state slightly above a melting in the membrane, the soliton model requires that local cooling of the membrane from physiological tempera- ture is able to trigger a nerve pulse, while heating inhibits the nerve pulse. This was in fact found by (36). The au- thors argued that this provides evidence in favor of a phase transition in the membrane. In contrast, a more recent study reported that heating by infrared pulses can trigger nerve ac- tivity (37). The authors explained this effect rather by heating of the electrolyte (rather than of the membrane) and a subse- quent change in the electrostatics of the electric double layer. They suggested that the heating of the buffer generates a volt- age change triggering a pulse. Thus, these two studies may well address different phenomena. References 1. Goodman, C. 1974. Anatomy of locust ocellar interneurons - constancy and variability. J. Comp. Physiol. 95:185 -- 201. 2. Hodgkin, A. L., and A. F. Huxley. 1952. A quantitative description of membrane current and its application to conduction and excitation in nerve. J. Physiol. London 117:500 -- 544. 3. Iwasa, K., and I. Tasaki. 1980. Mechanical changes in squid giant-axons associated with production of action potentials. Biochem. Biophys. Re- search Comm. 95:1328 -- 1331. 4. Iwasa, K., I. Tasaki, and R. C. Gibbons. 1980. Swelling of nerve fibres associated with action potentials. Science 210:338 -- 339. 5. Tasaki, I., K. Kusano, and M. Byrne. 1989. Rapid mechanical and ther- mal changes in the garfish olfactory nerve associated with a propagated impulse. Biophys. J. 55:1033 -- 1040. 6. Kim, G. H., P. Kosterin, A. Obaid, and B. M. Salzberg. 2007. A me- chanical spike accompanies the action potential in mammalian nerve terminals. Biophys. J. 92:3122 -- 3129. 7. Wilke, E., and E. Atzler. 1912. Experimentelle Beitrage zum Problem der Reizleitung im Nerven. Pflugers Arch. 146:430 -- 446. 8. Abbott, B. C., A. V. Hill, and J. V. Howarth. 1958. The positive and negative heat production associated with a nerve impulse. Proc. Roy. Soc. Lond. B 148:149 -- 187. 9. Howarth, J. V., R. Keynes, and J. M. Ritchie. 1968. The origin of the ini- tial heat associated with a single impulse in mammalian non-myelinated nerve fibres. J. Physiol. 194:745 -- 793. 10. Howarth, J. 1975. Heat production in non-myelinated nerves. Phil. Trans. Royal Soc. Lond. 270:425 -- 432. 11. Ritchie, J. M., and R. D. Keynes. 1985. The production and absorption of heat associated with electrical activity in nerve and electric organ. Quart. Rev. Biophys. 18:451 -- 476. 12. Hill, A. V. 1912. The absence of temperature changes during the trans- mission of a nervous impulse. J. Physiol. London 43:433 -- 440. 8 13. Hodgkin, A. L., 1964. The conduction of the nervous impulse. Liverpool University Press, Liverpool, UK. 34. Heimburg, T., and A. D. Jackson. 2007. The thermodynamics of general anesthesia. Biophys. J. 92:3159 -- 3165. 35. Mosgaard, L. D., K. Zecchi, and T. Heimburg. capacitive properties of polarized membranes. 10.1039/C5SM01519G. 2015. Mechano- Soft Matter DOI: 36. Kobatake, Y., I. Tasaki, and A. Watanabe. 1971. Phase transition in membrane with reference to nerve excitation. Adv. Biophys. 208:1 -- 31. 37. Shapiro, M. G., K. Homma, S. Villareal, C.-P. Richter, and F. Bezanilla. Infrared light excites cells by changing their electrical capaci- 2012. tance. Nat. Communications 3:736. 14. Heimburg, T., and A. D. Jackson. 2005. On soliton propagation in biomembranes and nerves. Proc. Natl. Acad. Sci. USA 102:9790 -- 9795. 15. Heimburg, T., 2007. Thermal biophysics of membranes. Wiley VCH, Berlin, Germany. 16. Lautrup, B., R. Appali, A. D. Jackson, and T. Heimburg. 2011. The stability of solitons in biomembranes and nerves. Eur. Phys. J. E 34:57. 17. Heimburg, T., and A. D. Jackson. 2007. On the action potential as a propagating density pulse and the role of anesthetics. Biophys. Rev. Lett. 2:57 -- 78. 18. Heimburg, T. 2012. The capacitance and electromechanical coupling of lipid membranes close to transitions. the effect of electrostriction. Biophys. J. 103:918 -- 929. 19. Gonzalez-Perez, A., R. Budvytyte, L. D. Mosgaard, S. Nissen, and T. Heimburg. 2014. Penetration of action potentials during collision in the median and lateral giant axons of invertebrates. Phys. Rev. X 4:031047. 20. Goldman, L. 1964. The effects of stretch on cable and spike parameters of single nerve fibres; some implications for the theory of impulse. J. Physiol. London 175:425 -- 444. 21. Griesbauer, J., S. Bossinger, A. Wixforth, and M. F. Schneider. 2012. Propagation of 2d pressure pulses in lipid monolayers and its possible implications for biology. Phys. Rev. Lett. 108:198103. 22. Griesbauer, J., S. Bossinger, A. Wixforth, and M. F. Schneider. 2012. Simultaneously propagating voltage and pressure pulses in lipid mono- layers of pork brain and synthetic lipids. Phys. Rev. E 86:061909. 23. Evans, P. D., E. A. Kravitz, B. R. Talamo, and B. G. Wallace. 1976. The association of octopamine with specific neurones along lobster nerve trunks. The Journal of physiology 262:51 -- 70. 24. Wright, E. B., and J. P. Reuben. 1958. A comparative study of some excitability properties of the giant axons of the ventral nerve cord of the lobster, including the recovery of excitability following an impulse. J. Cell Comp. Physiol. 51:13 -- 28. 25. El Hady, A., and B. B. Machta. 2015. Mechanical surface waves ac- company action potential propagation. Nat. Communications 6:6697. 26. Tasaki, I. 1949. Collision of two nerve impulses in the nerve fiber. Biochim. Biophys. Acta 3:494 -- 497. 27. Salzberg, B. M. 1989. Optical recording of voltage changes in nerve terminals and in fine neuronal processes. Ann. Rev. Physiol. 51:507 -- 526. 28. Zhang, P.-C., A. M. Keleshian, and F. Sachs. 2001. Voltage-induced membrane movement. Nature 413:428 -- 432. 29. Nguyen, T. D., N. Deshmukh, J. M. Nagarah, T. Kramer, P. K. Purohit, M. J. Berry, and M. C. McAlpine. 2012. Piezoelectric nanoribbons for monitoring cellular deformations. Nat. Nanotechnol. 7:587 -- 593. 30. Howells, J., L. Trevillion, H. Bostock, and D. Burke. 2012. The voltage dependence of Ih in human myelinated axons. J. Physiol. 590:1625 -- 1640. 31. Kaufmann, Lipid membrane. K., 1989. http://membranes.nbi.dk/Kaufmann/pdf/Kaufmann book5 org.pdf, Caruaru. 32. Shrivastava, S., and M. F. Schneider. Evidence for two- dimensional solitary sound waves in a lipid controlled interface and its implications for biological signalling. J. R. Soc. Interface 11:20140098. 2014. 33. Shrivastava, S., K. H. Kang, and M. F. Schneider. 2015. Solitary shock waves and adiabatic phase transition in lipid interfaces and nerves. Phys. Rev. E 91:012715. 9
1905.09805	1	1905	2019-05-23T17:55:00	Mechanical interplay between cell shape and actin cytoskeleton organization	[ "physics.bio-ph", "cond-mat.soft", "q-bio.CB" ]	We investigate the mechanical interplay between the spatial organization of the actin cytoskeleton and the shape of animal cells adhering on micropillar arrays. Using a combination of analytical work, computer simulations and in vitro experiments, we demonstrate that the orientation of the stress fibers strongly influences the geometry of the cell edge. In the presence of a uniformly aligned cytoskeleton, the cell edge can be well approximated by elliptical arcs, whose eccentricity reflects the degree of anisotropy of the cell's internal stresses. Upon modeling the actin cytoskeleton as a nematic liquid crystal, we further show that the geometry of the cell edge feeds back on the organization of the stress fibers by altering the length scale at which these are confined. This feedback mechanism is controlled by a dimensionless number, the anchoring number, representing the relative weight of surface-anchoring and bulk-aligning torques. Our model allows to predict both cellular shape and the internal structure of the actin cytoskeleton and is in good quantitative agreement with experiments on fibroblastoid (GD$\beta$1,GD$\beta$3) and epithelioid (GE$\beta$1, GE$\beta$3) cells.	physics.bio-ph	physics	Mechanical interplay between cell shape and actin cytoskeleton organization Koen Schakenraad,1, 2 Jeremy Ernst,1 Wim Pomp,3, 4 Erik H. J. Danen,5 Roeland M. H. Merks,2, 6 Thomas Schmidt,3 and Luca Giomi1, ∗ 1Instituut-Lorentz, Leiden University, P.O. Box 9506, 2300 RA Leiden, The Netherlands 2Mathematical Institute, Leiden University, P.O. Box 9512, 2300 RA Leiden, The Netherlands 3Kamerlingh Onnes-Huygens Laboratory, Leiden University, P.O. Box 9504, 2300 RA Leiden, The Netherlands 4Division of Gene Regulation, The Netherlands Cancer Institute, P.O. Box 90203, 1006 BE Amsterdam, The Netherlands 5Leiden Academic Center for Drug Research, Leiden University, P.O. Box 9502, 2300 RA Leiden, The Netherlands 6Institute of Biology, Leiden University, P.O. Box 9505, 2300 RA Leiden, The Netherlands We investigate the mechanical interplay between the spatial organization of the actin cytoskeleton and the shape of animal cells adhering on micropillar arrays. Using a combination of analytical work, computer simulations and in vitro experiments, we demonstrate that the orientation of the stress fibers strongly influences the geometry of the cell edge. In the presence of a uniformly aligned cytoskeleton, the cell edge can be well approximated by elliptical arcs, whose eccentricity reflects the degree of anisotropy of the cell's internal stresses. Upon modeling the actin cytoskeleton as a nematic liquid crystal, we further show that the geometry of the cell edge feeds back on the organization of the stress fibers by altering the length scale at which these are confined. This feedback mechanism is controlled by a dimensionless number, the anchoring number, representing the relative weight of surface-anchoring and bulk-aligning torques. Our model allows to predict both cellular shape and the internal structure of the actin cytoskeleton and is in good quantitative agreement with experiments on fibroblastoid (GDβ1,GDβ3) and epithelioid (GEβ1, GEβ3) cells. I. INTRODUCTION Mechanical cues play a vital role in many cellular pro- cesses, such as durotaxis [1, 2], cell-cell communication [3], stress-regulated protein expression [4] or rigidity- dependent stem cell differentiation [5, 6]. Whereas me- chanical forces can directly activate biochemical signaling pathways, via the mechanotransduction machinery [7], their effect is often mediated by the cortical cytoskele- ton, which, in turn, affects and can be affected by the geometry of the cell envelope. By adjusting their shape, cells can sense the mechan- ical properties of their microenvironment and regulate traction forces [8 -- 10], with prominent consequences on bio-mechanical processes such as cell division, differenti- ation, growth, death, nuclear deformation and gene ex- pression [11 -- 16]. On the other hand, the cellular shape itself depends on the mechanical properties of the en- vironment. Experiments on adherent cells have shown that the stiffness of the substrate strongly affects cell morphology [17, 18] and triggers the formation of stress fibers [19, 20]. The cell spreading area increases with the substrate stiffness for several cell types, including car- diac myocytes [17], myoblasts [18], endothelial cells and fibroblasts [19], and mesenchymal stem cells [21]. In our previous work [22] we have investigated the shape and traction forces of adherent cells [23] character- ized by a highly anisotropic actin cytoskeleton. Using a ∗ Corresponding author: [email protected] contour model of cellular adhesion [8, 23 -- 26], we demon- strated that the edge of these cells can be accurately ap- proximated by a collection of elliptical arcs obtained from a unique ellipse, whose eccentricity depends on the de- gree of anisotropy of the contractile stresses arising from the actin cytoskeleton. Furthermore, our model quantita- tively predicts how the anisotropy of the actin cytoskele- ton determines the magnitudes and directions of traction forces. Both predictions were tested in experiments on highly anisotropic fibroblastoid and epithelioid cells [27] supported by stiff microfabricated elastomeric pillar ar- rays [28 -- 30], finding good quantitative agreement. Whereas these findings shed light on how cytoskele- tal anisotropy controls the geometry and forces of adher- ent cells, they raise questions on how anisotropy emerges from the three-fold interplay between isotropic and di- rected stresses arising within the cytoskeleton, the shape of the cell envelope and the geometrical constraints in- troduced by focal adhesions. It is well known that the orientation of the stress fibers in elongated cells strongly correlates with the polarization direction of the cell [31 -- 34]. Consistently, our results indicate that, in highly anisotropic cells, stress fibers align with each other and with the cell's longitudinal direction (see, e.g., Fig. 1A) [22]. However, the physical origin of these alignment mechanisms is less clear and inevitably leads to a chicken- and-egg causality dilemma: do the stress fibers align along the cell's axis or does the cell elongate in the di- rection of the stress fibers? The biophysical literature is not scarce of cellular pro- cesses that might possibly contribute to alignment of stress fibers with each other and with the cell edge. Me- chanical feedback between the cell and the extracellu- lar matrix or substrate, for instance, has been shown to play an important role in the orientation and alignment of stress fibers [21, 35 -- 38]. Molecular motors have also been demonstrated to produce an aligning effect on cy- toskeletal filaments, both in mesenchymal stem cells [39] and in purified cytoskeletal extracts [40], where the ob- servation is further corroborated by agent-based simula- tions [41]. A similar mechanism has been theoretically proposed for microtubules-kinesin mixtures [42]. Stud- ies in microchambers demonstrated that steric interac- tions can also drive alignment of actin filaments with each other and with the microchamber walls [43 -- 45]. Theoret- ical studies have highlighted the importance of the stress fibers' assembly and dissociation dynamics [35, 46], the dynamics of focal adhesion complexes [47, 48], or both [49, 50]. Also the geometry of actin nucleation sites has been shown to affect the growth direction of actin fila- ments, thus promoting alignment [51, 52]. Finally, me- chanical coupling between the actin cytoskeleton and the plasma membrane has been theoretically shown to lead to fiber alignment, as bending moments arising in the membrane result into torques that reduce the amount of splay within the filaments [53]. Despite such a wealth of possible mechanisms, it is presently unclear which one or which combination is ultimately responsible for the observed alignment of stress fibers between each other and with the cell's longitudinal direction. Analogously, it is unclear to what extent these mechanisms are sensi- tive to the specific mechanical and topographic properties of the environment, although some mechanisms rely on specific environmental conditions that are evidently ab- sent in certain circumstances (e.g., the mechanical feed- back between the cell and the substrate discussed in Refs. [35, 37, 54] relies on deformations of the substrate and is unlikely to play an important role in experiments per- formed on stiff micro-pillar arrays). In this paper we investigate the interplay between the anisotropy of the actin cytoskeleton and the shape of cells adhering to stiff microfabricated elastomeric pillar arrays [28 -- 30]. Rather than pinpointing a specific align- ment mechanism, among those reviewed above, we focus on the interplay between cell shape and the spatial or- ganization of the actin cytoskeleton. This is achieved by means of a phenomenological treatment of the stress fiber orientation based on the continuum description of nematic liquid crystals, coupled with a contour model of the cell edge [22]. The paper is organized as follows: in Sec. II we review in detail our contour model of cells with an anisotropic cytoskeleton [22]. We study how the anisotropy affects the curvature of and tension in the cell cortex, as well as the forces that the cell exerts on the substrate. In Sec. III we further generalize this approach by studying the mechanical interplay between the orien- tation of the actin cytoskeleton, modeled as a nematic liquid crystal under confinement of the cell edge, and the shape of the cell, and we compare our results to experi- mental data on highly anisotropic cells. In both sections 2 FIG. 1. (A) A cell with an anisotropic actin cytoskeleton (fibroblastoid) cultured on a stiff microfabricated elastomeric pillar array [22]. The color scale indicates the measured ori- entation of the actin stress fibers. (B) Schematic represen- tation of a contour model for the cell in (A). The cell con- tour consists of a collection of concave cellular arcs (red lines) that connect pairs of adhesion sites (blue dots). These arcs are parameterized as curves spanned counterclockwise around the cell by the arclength s, and are entirely described via the tangent unit vector T = (cos θ, sin θ) and the normal vector N = (− sin θ, cos θ), with θ the turning angle. The vector n describes the local orientation of the stress fibers. we assume that the coordinates of the adhesion sites are constant in time and known. A theoretical description of the dynamics of these adhesion sites, as a result of focal adhesion dynamics, is beyond the scope of this study and can be found, for example, in Refs. [47, 48]. II. EQUILIBRIUM CONFIGURATION OF THE CELL BOUNDARY Many animal cells spread out after coming into con- tact with a stiff adhesive surface. They develop trans- membrane adhesion receptors at a limited number of ad- hesion sites that lie mainly along the cell contour (i.e., focal adhesions [55]). These cells are then essentially flat and assume a typical shape characterized by arcs which span between the sites of adhesion, while forces are mainly contractile [23]. This makes it possible to de- scribe adherent cells as two-dimensional contractile films, whose shape is unambiguously identified by the posi- tion r = (x, y) of the cell contour [8, 22, 24 -- 26, 56, 57]. Fig. 1B illustrates a schematic representation of the cell (fibroblastoid) in Fig. 1A, where the cell contour consists of a collection of curves, referred to as "cellular arcs", that connect two consecutive adhesion sites. These arcs are parameterized by the arclength s as curves spanned counterclockwise around the cell, oriented along the tan- gent unit vector T = ∂sr = (cos θ, sin θ), with θ = θ(s) the turning angle of the arc with respect to the hori- zontal axis of the frame of reference. The normal vec- tor N = ∂sr⊥ = (− sin θ, cos θ), with r⊥ = (−y, x), is directed toward the interior of the cell. The equation describing the shape of a cellular arc is obtained upon BFocal adhesionctinStress fibers90°45°0°- 45°- 90°OrientationA10 μmbalancing all the conservative and dissipative forces ex- perienced by the cell contour. These are: ξt∂tr = ∂sFcortex + ( Σout − Σin) · N , (1) where t is time, ξt is a (translational) friction coefficient, Σout and Σin are the stress tensors on the two sides of the cell boundary and Fcortex is the stress resultant along the cell contour [8, 22, 23, 25, 26, 56]. The temporal evolu- tion of the cell contour is then dictated by a competition between internal and external bulk stresses acting on the cell boundary and the contractile forces arising within the cell cortex. We assume the dynamics of the cell contour to be overdamped. The stress tensor can be modeled upon taking into account isotropic and directed stresses. The latter are constructed by treating the stress fibers as contractile force dipoles, whose average orientation θSF is parallel to the unit vector n = (cos θSF, sin θSF) (see Fig. 1B). This gives rise to an overall contractile bulk stress of the form [58, 59]: Σout − Σin = σ I + αnn , (2) where I is the identity matrix, σ > 0 embodies the mag- nitude of all isotropic stresses (passive and active) ex- perienced by the cell edge and α > 0 is the magnitude of the directed contractile stresses and is proportional to the local degree of alignment between the stress fibers, in such a way that α is maximal for perfectly aligned fibers, and vanishes if these are randomly oriented. In Sec. III we will explicitly account for the local orientational order of the stress fibers using the language of nematic liquid crystals. The degree of anisotropy of the bulk stress is thus determined by the ratio between the isotropic con- tractility σ and the directed contractility α. Finally, the tension within the cell cortex is modeled as Fcortex = λT , where line tension λ embodies the contractile forces aris- ing from myosin activity in the cell cortex. This quantity varies, in general, along an arc and can be expressed as a function of the arclength s. In the presence of anisotropic bulk stresses, in particular, λ(s) cannot be constant, as we will see in Sec. II A. The force balance condition, Eq. (1), becomes then ξt∂tr = ∂s(λT ) + σN + α(n · N )n . (3) A. Equilibrium cell shape and line tension In this section we describe the position of the cell boundary under the assumption that the timescale re- quired for the equilibration of the forces in Eq. (3) is much shorter than the typical timescale of cell migration on the substrate (i.e., minutes). Under this assumption, ∂tr = 0 and Eq. (3) can be cast in the form: 0 = (∂sλ) T + (λκ + σ)N + α(n · N )n , (4) 3 where we have used ∂sT = κN , with κ = ∂sθ the curva- ture of the cell edge. A special situation is obtained when the cytoskeleton is purely isotropic. In this case, first dis- cussed by Bar-Ziv et al. in the context of cell pearling [24] and later expanded by Bischofs and coworkers [8, 25], α = 0 and Eq. (4) reduces to (∂sλ) T + (λκ + σ)N = 0 . (5) Solving this equation in the two orthogonal directions T and N leads to two important insights for isotropic cells. First, in the presence of strictly isotropic bulk forces, the line tension λ must be constant along a single arc at equilibrium (i.e., ∂sλ = 0). Second, bulk and pe- ripheral contractility, that pull the cell edge inward and outward along the normal direction, compromise along an arc of constant curvature κ = −σ/λ, with the nega- tive sign of κ indicating that the arcs are curved inwards. This mechanism is analogous to the Young-Laplace law for fluid interfaces, where the isotropic bulk contractil- ity σ plays the role of the pressure difference and the line tension λ that of surface tension. The cell edge is then described by a sequence of circular arcs, whose radius R = 1/κ = λ/σ depends on the local cortical tension λ of the arc. The latter depends on the local myosin concentration and, in general, varies from arc to arc. The possibility of a geometry-dependent corti- cal tension, originating from the elasticity of the trans- verse fibers, was also explored in Refs. [8, 25], to ac- count for an observed correlation between the radius and the length of the cellular arcs. Both geometry-dependent and geometry-independent models successfully describe the shape of adherent cells in the presence of strictly isotropic forces. However, as we showed elsewhere [22], these isotropic models are not suited for describing cells that develop strong directed forces due to the anisotropic cytoskeleton originating from actin stress fibers [60, 61]. The presence of an anisotropic cytoskeleton is modeled by a nonzero directed stress, α > 0. By virtue of Eq. (4), this results into the fact that the cortical tension λ is no longer constant along the cellular arcs, since the directed stress has, in general, a non-vanishing tangential compo- nent (i.e., n · T (cid:54)= 0). In order to highlight the physical mechanisms described by Eq. (4), we introduce a number of simplifications in the remainder of this section. These will be lifted in Sec. III, where we will consider the most general scenario. First, because the orientation of the stress fibers typically varies only slightly along a single arc [22], we assume the orientation of the stress fibers, θSF, to be constant along a single cellular arc, but dif- ferent from arc to arc. Furthermore, we assume that the contractile bulk stresses α and σ are uniform throughout the cell. This allows us to solve Eq. (4) analytically for a given arc. Without loss of generality, we orient the refer- ence frame such that the stress fibers are parallel to the y−axis. Thus, θSF = π/2 and n = y. Since α, σ and n are constant along an arc, Eq. (4) can be expressed as a total derivative and integrated directly. This yields λT + (σ I + αnn) · r⊥ = C1 , (6) where C1 = (C1x, C1y) is an integration constant. De- composing Eq. (6) into x− and y−directions yields λ cos θ = C1x + σy λ sin θ = C1y − (α + σ)x . (7a) (7b) Next, taking the ratio of Eqs. (7), using tan θ = dy/dx and integrating, we obtain a general solution for the shape of the cellular arc subject to a non-vanishing isotropic stress (i.e., σ (cid:54)= 0), namely (x − xc)2 + (y − yc)2 = C2 , 1 γ (8) where C2 is another integration constant and we have set xc = C1y σ + α , yc = − C1x σ , γ = . σ σ + α √ (8) describes an ellipse centered at (xc, yc) and Eq. √ γC2 and whose minor and major semi-axis are a = b = C2. The constant γ quantifies the anisotropy of the bulk stress: γ = 1 corresponds to the isotropic case, whereas γ = 0 represents purely anisotropic bulk con- tractility. Using again Eqs. (7), we further obtain an expression for the line tension λ as a function of x and y: λ2 = σ2(y − yc)2 + (σ + α)2(x − xc)2 . (9) Using Eqs. (7) and (8), this can be also expressed as a function of the turning angle θ, namely λ2 σ2 = C2 1 + tan2 θ 1 + γ tan2 θ . (10) This expression highlights the physical meaning of the integration constant C2. The right-hand side of Eq. (10) attains its minimal value (C2) where θ = 0, hence when the tangent vector is perpendicular to the stress fibers (i.e., n · T = 0). Thus C2 = λ2 min/σ2, where λmin is the minimal tension withstood by the cortical actin. Substi- tuting C2 in Eq. (10) yields the tension as a function of the turning angle: (cid:115) λ(θ) = λmin 1 + tan2 θ 1 + γ tan2 θ . (11) Eq. (11) describes how the line tension deviates from the minimum value λmin for nonzero θ in order to balance the tangential component of the directed stress [Eq. (4)]. The maximum value of the line tension is found at θ = √ π/2, where the stress fibers are parallel to the arc, and is given by λmax = λmin/ γ. We note that these variations in line tension along a single arc do not necessarily have to be regulated by the cell. Instead, they could simply be a result of passive mechanical forces in a way very similar to the space-dependent tension in a simple cable hanging under gravity. 4 FIG. 2. Schematic representation of a cellular arc, de- scribed by Eq. (13), for θSF = π/2. A force balance be- tween isotropic stress, directed stress and line tension results √ in the description of each cell edge segment (red curve) as part of an ellipse of aspect ratio a/b = γ. The cell exerts forces F0 and F1 on the adhesion sites (blue). The vector d = d(cos φ, sin φ) describes the relative position of the two adhesion sites, d⊥ = d(− sin φ, cos φ) is a vector perpendicu- lar to d, and θ is the turning angle of the cellular arc. The coordinates of the ellipse center (xc, yc) and the angular co- ordinates of the adhesion sites along the ellipse ψ0 and ψ1 are given in Appendix A. Although the minimal line tension λmin could, in prin- ciple, be arc-dependent, for example if the cell cortex displays substantial variations in the myosin densities [25], here we approximate λmin as a constant. This ap- proximation is motivated by the fact that our previous in vitro observations of anisotropic epithelioid and fibro- blastoid cells did not identify a correlation between the arc length and curvature [22], which, on the other hand, is expected if λmin was to vary significantly from arc to arc [25]. Hence, α, σ and λmin represent the material parameters of our model. Substituting C2 in Eq. (8) yields an implicit represen- tation of the plane curve approximating individual cellu- lar arcs, namely σ2 γλ2 min (x − xc)2 + σ2 λ2 min (y − yc)2 = 1 . (12) This equation describes an ellipse centered at the point (xc, yc) and oriented along the y−direction, whose minor and major semi-axes are a = λmin γ/σ and b = λmin/σ respectively (Fig. 2). For arbitrary stress fiber orienta- tion θSF, Eq. (12) can be straightforwardly generalized in the form: √ [(x − xc) cos θSF + (y − yc) sin θSF]2 σ2 λ2 min + σ2 γλ2 min [−(x − xc) sin θSF + (y − yc) cos θSF]2 = 1 . (13) Isotropic stressDirected stress5 Consistently, the major axis of the ellipse is always par- allel to the stress fibers, hence tilted by an angle θSF with respect to the x−axis (Fig. 2). The direct relation be- tween the contractile forces arising from the cytoskeleton and the shape of the cell is highlighted by the dimension- less parameter γ = σ/(σ + α): on the one hand, γ defines the anisotropy of the contractile bulk stress, on the other hand it dictates the anisotropy of the cell shape. The co- ordinates of the center of the ellipse (xc, yc) and the an- gular coordinates of the adhesion sites along the ellipse, ψ0 and ψ1 in Fig. 2, can be calculated using standard algebraic manipulation and are given in Appendix A. One of the most striking consequences of the ellipti- cal shape of the cellular arcs is that the local curva- ture is no longer constant, consistent with experimen- tal observations on epithelioid and fibroblastoid cells in [22]. This can be calculated from Eq. (12) using Ref. κ = (d2y/dx2)/[1+(dy/dx)2] 3 2 and expressed in terms of the turning angle θ using Eqs. (7) and (11). This yields (cid:18) 1 + γ tan2 θ (cid:19) 3 2 1 + tan2 θ κ = − 1 γb , (14) with the negative sign indicating that the arcs are curved inwards. A cellular arc thus attains its maximal (mini- mal) absolute curvature, where θ = 0 (θ = π/2) and the stress fibers are parallel (perpendicular) to the arc tangent vector, namely (cid:16) (cid:17) √ π 2 θ = = − κmin = κ κmax = κ (θ = 0) = − 1 γb γ b . , (15a) (15b) Consistent with experimental evidence [22], the radius of curvature of arcs perpendicular to stress fibers is smaller than the radius of curvature of arcs parallel to the stress fiber direction. For the experimental methods and a more detailed comparison between theory and experiment, see Ref. [22]. We stress that, as long as the contractile stresses arising from the actin cytoskeleton are roughly uniform across the cell (i.e., α, σ and λmin are constant), all cellular arcs are approximated by a unique ellipse (see, e.g., Fig. 3). Thus, regions of the cell edge having higher and lower local curvature correspond to different portions of the same ellipse, depending on the relative orientation of the local tangent vector and the stress fibers. B. Traction forces With the expressions for shape of the cellular arcs [Eq. (13)] and cortical tension [Eq. (11)] in hand, we can now calculate the traction forces exerted by the cell via the focal adhesions positioned at the end-points of a given cellular arc (Fig. 2). Calling these F0 and F1 and re- calling the the cell edge is oriented counter-clockwise, we have F0 = −λ(θ0)T (θ0) and F1 = λ(θ1)T (θ1), where θ0 FIG. 3. A cell with an anisotropic actin cytoskeleton on a microfabricated elastomeric pillar array (fibroblastoid [22], same cell as in Figure 1A), with a unique ellipse (white) fitted to its arcs. The actin, cell edge, and micropillar tops are in the red, green, and blue channels respectively. The endpoints of the arcs (cyan) are identified based on the forces that the cell exerts on the pillars. For a detailed description of the experimental methods, see Ref. [22]. Scale bar is 10 µm. and θ1 are the turning angles at the end-points of the arc. For practical applications, it is often convenient to express the position of the adhesion sites in terms of their relative distance d = d(cos φ, sin φ) (Fig. 2). This yields sin φ n , (16a) (cid:34)(cid:32) F0 = λmin sin φ + (cid:32) d 2b + − 1 γ d 2b F1 = λmin (cid:34)(cid:32) d 2b + sin φ − ρ b (cid:32) (cid:33) ρ b cos φ n⊥ cos φ + (cid:33) cos φ ρ b n⊥ (cid:33) (cid:35) (cid:33) (cid:35) − 1 γ d 2b cos φ − ρ b sin φ n , (16b) where n⊥ = (sin θSF,− cos θSF) and the length scale ρ is defined as (cid:115) ρ = b2 (cid:18) 1 + tan2 φ (cid:19) 1 + γ tan2 φ − 1 γ (cid:18) d (cid:19)2 2 . (17) The total traction force exerted by the cell can be cal- culated by summing the two forces associated with the arcs joining at a given adhesion site, while taking into account that the the orientation n of the stress fibers is generally different from arc to arc. Another interesting quantity is obtained by adding the forces F0 and F1 from the same arc. Although these two forces act on two different adhesion sites, their sum represents the total net force that a single cellular arc exerts on the substrate. This is given by F0 + F1 = dσ sin φ n⊥ − d(σ + α) cos φ n , = −(cid:16) (cid:17) · d⊥ , σ I + α nn (18) where d⊥ = d(− sin φ, cos φ) (Fig. 2). Eq. (18) is the force resulting from the integrated contractile bulk stress [see Eq. (1)] and is independent of the line tension λmin. C. Mechanical invariants We conclude this section by highlighting two mechan- ical invariants, local quantities that are constant along a cellular arc, thus showing the intimate relation between the geometry of the cell and the mechanical forces it ex- erts on the environment. From Eqs. (16) we find F 2⊥ + γF 2(cid:107) = const., (19) where F(cid:107) and F⊥ are the components of the force, par- allel and perpendicular to n, at any point along a same cellular arc. Furthermore, by inspection of Eqs. (14) and (11) we observe that λ3κ = −λ2 min(α + σ) = const . (20) From this, we find that the normal component of the cortical force, λκ [see Eq. (4)], is then given by (cid:18) λmin (cid:19)2 λ λκ = − (α + σ) . (21) This relation is an analog of the Young-Laplace law for our anisotropic system. In the isotropic limit, α = 0 and λmin = λ, thus we recover the standard force-balance expression λκ = −σ. Eq. (21) shows that the normal force λκ decreases with increasing line tension λ, because an increase in line tension is accompanied by an even stronger decrease in the curvature κ. 6 liquid crystals [62]. This setting can account for the me- chanical feedback between the orientation of the stress fibers and the cellular shape and allows us to predict both these features starting from the position of the ad- hesions sites alone. A treatment of the dynamics of focal adhesions is beyond the scope of this paper and can be found elsewhere, e.g., in Refs. [47, 48]. As mentioned in the Introduction, experimental ob- servations, by us [22] and others [31 -- 34], have indicated that stress fibers tend to align with each other and with the cell's longitudinal direction. As we discussed, several cellular processes might contribute to these alignment mechanisms, such as mechanical cell-matrix feedback [21, 35 -- 38], motor-mediated alignment [39 -- 42], steric in- teractions [43 -- 45], stress fiber formation and dissociation [35, 46, 49, 50], focal adhesion dynamics [47 -- 50], the ge- ometry of actin nucleation sites [51, 52], or membrane- mediated alignment [53], but it is presently unclear which combination of mechanisms is ultimately responsible for the orientational correlation observed in experiments. Our phenomenological description of the actin cytoskele- ton allows us to focus on the interplay between cellular shape and the orientation of the stress fibers, without the loss of generality that would inevitably result from selecting a specific alignment mechanism among those listed above. A. Dynamics of the stress fibers The actin cytoskeleton is modeled as a nematic liq- uid crystal confined within the cellular contour. This is conveniently represented in terms of the two-dimensional nematic tensor: (cid:18) (cid:19) ninj − 1 2 Qij = S where δij is the Kronecker delta and S = (cid:112)2 tr Q2 is (22) δij , the so called nematic order parameter, measuring the amount of local nematic order. In the context of the actin cytoskeleton, a low S value might result from either a low local density of stress fibers, or from a high density of randomly oriented stress fibers. In the standard {x, y} Cartesian basis, Eq. (22) yields (cid:20)Qxx Qxy Qxy −Qxx (cid:21) Q = (cid:20)cos 2θSF = S 2 sin 2θSF sin 2θSF − cos 2θSF . (23) (cid:21) III. INTERPLAY BETWEEN ORIENTATION OF THE CYTOSKELETON AND CELLULAR SHAPE In this Section we generalize our approach by allowing the orientation of the stress fibers to vary along indi- vidual cellular arcs. This is achieved by combining the contour model for the cell shape, reviewed in Sec. II, with a continuous phenomenological model of the actin cytoskeleton, rooted into the hydrodynamics of nematic Just like the dynamics of the cell edge, the dynamics of the cytoskeleton are assumed to be overdamped, thus: ∂tQij = − 1 ξr δFcyto δQij , (24) where ξr is a rotational friction coefficient, dictating the rate of the relaxational dynamics, and Fcyto is the Landau-de Gennes free-energy [62]: (cid:90) (cid:20) Fcyto = K 2 ∇ Q2 + dA (cid:73) 1 (cid:104) δ2 tr Q2(tr Q2 − 1) ( Q − Q0)2(cid:105) ds tr + W 2 (cid:21) . (25) The first integral in Eq. (25) corresponds to a stan- dard mean-field free-energy, favoring perfect nematic or- der (i.e., S = 1), while penalizing gradients in the orien- tation of the stress fibers and their local alignment. K is the orientational stiffness of the nematic phase in the one elastic constant approximation. The length-scale δ sets the size of the boundary layer in regions where the order parameter drops to zero to compensate a strong distortion of the nematic director n, such as in proxim- ity of topological defects. The second integral, which is extended over the cell contour, is the Nobili-Durand an- choring energy [63] and determines the orientation of the stress fibers along the edge of the cell, with the tensor Q0 representing their preferential orientation. Experimental evidence form our (Fig. 3 and Ref. [22]) and other's work (e.g., Refs. [31 -- 34]), suggests that, in highly anisotropic cells, peripheral stress fibers are preferentially parallel to the cell edge. The same trend is recovered in experiments with purified actin bundles confined in microchambers [43, 44]. In the language of Landau-de Gennes theory, this effect can be straightforwardly reproduced by set- ting Q0,ij = TiTj − 1 2 δij , (26) with T the tangent unit vector of the cell edge, under the additional assumption that nematic order is promi- nent along the cell contour (i.e., S = 1). The constant W > 0 measures the strength of this parallel anchoring. To ensure the free-energy to be minimal at equilibrium (i.e., when ∂tQij = 0), we solve Eq. (24) with Neumann boundary conditions: KN · ∇Qij − 2W (Qij − Q0,ij) = 0 . (27) B. The dynamics of the cell edge The relaxational dynamics of the cell contour are gov- erned, in our model, by Eq. (3), which is now lifted from the assumption that the orientation n of the stress fibers is uniform along individual cellular arcs. Furthermore, the contractile stress given by Eq. (2) can now be gener- alized as: (cid:18) (cid:19) Σout − Σin = σ + 1 2 α0S I + α0 Q , (28) with α0 a constant independent on the local order pa- rameter. Comparing Eqs. (2) and (28) yields α = α0S, thus the formalism introduced in this Section allows us to 7 explicitly account for the effect of the local orientational order on the amount of contractile stress exerted by the stress fibers. Next, we decompose Eq. (3) along the normal and tangent directions of the cell contour. Since the cells considered here are pinned at the adhesion sites and the density of actin along the cell contour is assumed to be constant, tangential motion is suppressed, i.e., T · ∂tr = 0. Together with Eq. (28) this yields: 0 = ∂sλ + α0 T · Q · N , ξtN · ∂tr = λκ + σ + α0S + α0 N · Q · N . 1 2 (29a) (29b) Eq. (29a) describes then the relaxation of tension λ within the cell edge, given its shape, whereas Eq. (29b) describes the relaxation of the cellular shape itself. The variations in the cortical tension might result from a reg- ulation of the myosin activity or simply form a passive response of the cortical actin to the tangential stresses. Integrating Eq. (29a) then yields the cortical tension along an arc: λ(s) = λ(0) − α0 (cid:90) s ds(cid:48) T · Q · N , (30) 0 where Q = Q(s) varies, in general, along an individual cellular arc. The integration constant λ(0), which rep- resents the cortical tension at one of the adhesion sites, is related to the minimal tension λmin withstood by the cortical actin which we used, in Sec. II, as material pa- rameter of the problem. In practice, we first calculate λ over an entire arc using a arbitrary guess for λ(0). Then, we apply a uniform shift in such a way that the minimal λ value coincides with λmin. Combining the dynamics of the cell contour and that of the cell bulk, we obtain the following coupled differential equations: (cid:21) (cid:20) (cid:20) λκ + σ + α0S + α0 N · Q · N N , (31a) ∂tr = ∂t Q = 1 ξt K ξr 1 2 (cid:21) δ2 (S2 − 1) Q ∇2 Q − 2 . (31b) These are complemented by the condition that r is fixed in a number of specific adhesion sites, the boundary con- dition given by Eq. (27) for the nematic tensor Q and the requirement that mins λ(s) = λmin on each arc. IV. NUMERICAL RESULTS Eqs. (31) are numerically solved using a finite differ- ence integration scheme with moving boundary. As we detail in Appendix B, the time-integration is performed iteratively using the forward Euler algorithm by alternat- ing updates of the cell contour and of the bulk nematic tensor. This process is iterated until both the cell shape and the orientation reach a steady state. To highlight the physical meaning of our numerical re- sults, we introduce two dimensionless numbers, namely the contractility number, Co, and the anchoring number, An. Co is defined as the ratio between the typical dis- tance between two adhesion sites d and the major axis of the ellipse approximating the corresponding cellular arc (b = λmin/σ, see Sec. II A): Co = σd λmin , (32) and provides a dimensionless measure of the cell contrac- tion (thus of the cell average curvature). The anchoring number, on the other hand, is defined as the ratio be- tween a typical length scale R in which the internal cell structure is confined (not necessarily equal to the dis- tance d) and the length scale K/W , which sets the size of the boundary layer where Q crosses over from its bulk configuration to Q0: An = W R K . (33) This number expresses the ratio between the anchoring energy, which scales as W R [i.e., last term in Eq. (25)], and the bulk energy, which scales as K, thus indepen- dently on cell size [i.e., first term in Eq. (25)]. Hence, An represents the competition between boundary align- ment (with strength W ) and bulk alignment (strength K) within the length scale of the cell R. For An (cid:28) 1, bulk elasticity dominates over boundary anchoring and the orientation of the stress fibers in the bulk propagates into the boundary, resulting into a uniform orientation throughout the cell and large deviations from parallel an- choring in proximity of the edge. Conversely, for An (cid:29) 1, boundary anchoring dominates bulk elasticity and the orientation of the stress fibers along the cell edge propa- gates into the bulk, leading to non-uniform alignment in the interior of the cell. To get insight on the effect of the combinations of these dimensionless parameters on the spatial organiza- tion of the cell, we first consider the simple case in which the adhesion sites are located at the corners of squares and rectangles (Sec. IV B we consider more realistic adhesion geometries and compare our nu- merical results with experimental observations on highly anisotropic cells. IV A). In Sec. A. Rectangular cells Fig. 4 shows the possible configurations of a model cell whose adhesion sites are located at the vertices of a square, obtained upon varying An and Co, while keeping γ = σ/(σ + α0) constant. The thick black line represents the cell boundary, the black lines in the interior of the cells represent the orientation field n of the stress fibers 8 FIG. 4. Configurations of cells whose adhesion sites are lo- cated at the vertices of a square. The thick black line repre- sents the cell boundary, the black lines in the interior of the cells represent the orientation field n = (cos θSF, sin θSF) of the stress fibers and the background color indicates the local nematic order parameter S. The vertical axis corresponds to the anchoring number An = W R/K and the horizontal axis to the contractility number Co = σd/λmin. The cells shown correspond to values of An = 0, 1, 10 and Co = 0, 0.125, 0.25, where we take both d and R equal to the length of the square side. The ratios σ/(σ+α0) = 1/9, λmin∆t/(ξtR2) = 2.8·10−6, and K∆t/(ξrR2) = 2.8· 10−6, and the parameters δ = 0.15R, Narc = 20, and ∆x = R/19 are the same for all cells. The number of iterations is 5.5 · 105. For definitions of ∆t, ∆x, and Narc, see Appendix B. and the background color indicates the local nematic or- der parameter S. As expected, for low Co values (left column), cells with high An exhibit better parallel anchoring than cells with small An values, but lower nematic order parameter in the cell interior. Interestingly, the alignment of stress fibers with the walls in the configuration with large An value (top left) resembles the configurations found by Deshpande et al. [35, 46], who specifically accounted for the assembly and dissociation dynamics of the stress fibers. More strikingly, the structure reported in the top left of Fig. 4 appears very similar to those found in exper- iments of actin filaments in cell-sized square microcham- bers [43, 44], simulations of hard rods in quasi-2D con- finement [43], and results based on Frank elasticity [64], where the tendency of the nematic director to align along the square edges competes with that of aligning along the diagonal. For large Co values (right column of Fig. 4), the cell deviates from the square shape. Interestingly, An and Co do not influence the geometry of the cell independently. For constant Co, i.e., for fixed stress fiber contractility, increasing An leads to higher tangential alignment of the stress fibers with the cell edge, thus increasing An de- creases the contractile force experienced by the cell edge, which is proportional to (n · N )2 [Eq. (31a)]. Finally, we note that all configurations in Fig. 4 dis- play a broken rotational and/or chiral symmetry. For An = 0 the stress fibers are uniformly oriented, but any direction is equally likely. For non-zero An, the stress fibers tend to align along either of the diagonals (with the same probability) to reduce the amount of distortion. Upon increasing Co, chirality emerges in the cytoskeleton and in the cell contour (see, e.g., the cell in the middle of the right column in Fig. 4). In light of the recent in- terest in chiral symmetry breaking in single cells [65] and in multicellular environments [66], we find it particularly interesting that chiral symmetry breaking can originate from the natural interplay between the orientation of the stress fibers and the shape of the cell. To conclude this section, we focus on four-sided cells whose adhesion sites are located at the vertices of a rect- angle and explore the effect of the cell aspect ratio (i.e., height/width). Fig. 5 displays three configurations with aspect ratio increasing from 1 to 2. Upon increasing the cell aspect ratio, the mean orientation of the stress fibers switches from the diagonal (aspect ratio 1) to longitudi- nal (aspect ratio 2), along with an increase in the order parameter in the cell bulk, as can be seen in Fig. 5 by the slightly more red-shifted cell interior. This behavior originates from the competition between bulk and bound- ary effects. Whereas the bulk energy favors longitudi- nal alignment, as this reduces the amount of bending of the nematic director, the anchoring energy favors align- ment along all four edges alike, thus favoring highly bent configurations at the expense of the bulk elastic energy. When the aspect ratio increases, the bending energy of the bulk in the diagonal configuration increases, whereas the longitudinal state only pays the anchoring energy at the short edges, hence independently on the aspect ra- tio. Therefore, elongating the cell causes the stress fibers to transition from tangential to longitudinal alignment, with a consequent increase of the nematic order parame- ter. Interestingly, these observations are consistent with the findings of experiments on actin filaments in cell-sized microchambers [43, 44]. B. Cells on micropillar arrays In order to validate our model experimentally, we com- pare our numerical results with experiments on fibro- blastoid and epithelioid cells [27] plated on stiff micropil- lar arrays [28 -- 30]. The cells are imaged using spinning disk confocal microscopy (see, e.g., Fig. 6A) and the images are then processed in order to detect the orienta- 9 FIG. 5. Effect of the aspect ratio, ranging from 1 to 2, of the cell on cytoskeletal organization for cells whose four ad- hesion sites are located at the vertices of a rectangle. The thick black line represents the cell boundary, the black lines in the interior of the cells represent the orientation field n = (cos θSF, sin θSF) of the stress fibers and the background color indicates the local nematic order parameter S. The sim- ulations shown are performed with An = W R/K = 1 where R is equal to the short side of the rectangle, and Co = σd/λmin equal to 0.125, 0.1875, and 0.25 respectively, where d is equal to the long side of the rectangle. The ratios σ/(σ +α0) = 1/9, λmin∆t/(ξtR2) = 2.8·10−6, and K∆t/(ξrR2) = 2.8·10−6, and the parameters δ = 0.15R and ∆x = R/19 are the same for all cells. Narc = 20, 30, 40 from left to right and the number of iterations is 5.5 · 105. For definitions of ∆t, ∆x, and Narc, see Appendix B. tion of the stress fibers. Upon coarse-graining the local gradients of the image intensity, we reconstruct both the nematic director n (black lines, representing the orien- tation of the stress fibers) and order parameter S (back- ground color, representing the degree of alignment), as visualized in Fig. 6B. See Appendix C for more detail on the experimental methods and image processing. Consistent with our results on rectangular cells (Fig. 5), the stress fibers align parallel to the cell's longitu- dinal direction and perpendicularly to the cell's shorter edges. Furthermore, the nematic order parameter is close to unity in proximity of the cell contour, indicating strong orientational order near the cell edge, but drops in the interior. This behavior is in part originating from the lower density of stress fibers around the center of mass of the cell, and in part from the presence of ±1/2 nematic disclinations away from the cell edge. These topologi- cal defects naturally arise from the tangential orienta- tion along the boundary. Albeit not uniform through- out the whole cell contour, thus not sufficient to impose hard topological constraints on the configuration of the director in the bulk (i.e., Poincar´e-Hopf theorem), this forces a non-zero winding of the stress fibers, which in 10 turn is accommodated via the formation of one or more disinclinations. As a consequence of the concave shape of the cell boundary, these defects have most commonly strength −1/2. To compare our theoretical and experimental results, we extract the locations of the adhesion sites from the experimental data (see Ref. [22]) and use them as in- put parameters for the simulations. In Figs. 6C-E we show results of simulations of the cell in Figs. 6A,B for increasing An values, thus decreasing magnitude of the length scale K/W . Here, we take the length scale R = 23.6 µm to be the square root of the cell area and we use constant values for the ratios λmin/σ = 14.7 µm and σ/(σ + α0) = 0.40 as found by an analysis of the el- liptical shape of this cell [22]. Fig. 6C shows the results of a simulation where bulk alignment dominates over boundary alignment (An = 0.33, K/W = 71 µm), result- ing in an approximately uniform cytoskeleton oriented along the cell's longitudinal direction. The nematic or- der parameter is also approximatively uniform and close to unity. For larger An values (Fig. 6D, An = 1.7 and K/W = 14 µm), anchoring effects become more promi- nent, resulting in distortions of the bulk nematic director and the emergence of a −1/2 disclination in the bottom left side of the cell. Upon further increasing An (Fig. 6E, An = 8.0 and K/W = 2.9 µm), the −1/2 topological de- fect moves towards the interior, as a consequence of the increased nematic order along the boundary. This results in a decrease in nematic order parameter in the bulk of the cell, consistent with our experimental data. A qualitative comparison between our in vitro (Fig. 6B) and in silico cells (Fig. 6E), highlights a number of striking similarities, such as the overall structure of the nematic director, the configuration of the order parame- ter (large along the cell edge and in the thin neck at the bottom right of the cell) and the occurrence of a −1/2 disclination on the bottom-left side. In order to make the comparison quantitative and infer the value of the phenomenological parameters introduced in this Section, we have further analyzed the residual function N(cid:88) i=1 tr(cid:2)( Qsim,i − Qexp,i)2(cid:3) , 1 2 (34) ∆2 = 1 N expressing the departure of the predicted configurations of the nematic tensor, Qsim, from the experimental ones, Qexp. The index i identifies a pixel in the cell and N is the total number of pixels common to both numer- ical and experimental configurations. By construction, 0 ≤ ∆2 ≤ 1, with 0 representing perfect agreement. Fig. 7 shows a plot of ∆2 versus the anchoring num- ber An for the cell shown in Fig. 6. Consistent with the previous qualitative comparison, the agreement is best at large An values, indicating a substantial pref- erence of the stress fibers for parallel anchoring along the cell edge. For the cell in Fig. 6, ∆2 is minimized for An = 8.0 (∆2 = 0.027), corresponding to a boundary FIG. 6. Validation of our model to experimental data. (A) Optical micrograph of a fibroblastoid cell (same cell as in Figs. 1 and 3) [22]. (B) Experimental data of cell shape and coarse-grained cytoskeletal structure of this cell. The white line represents the cell boundary, black lines in the interior of the cells represent the orientation field n = (cos θSF, sin θSF) of the stress fibers and the back- ground color indicates the local nematic order parameter S. (C-E) Configurations obtained from a numerical solution of Eqs. (31) using the adhesion sites (cyan circles) of the experimental data as input and with various anchoring number (An) values. This is calculated from Eq. (33), with R = 23.6 µm, taken from the square root of the cell area. The corresponding values of the length scale K/W are 71 µm (C), 14 µm (D) and 2.9 µm (E) respectively. The values for λmin/σ = 14.7 µm and σ/(σ + α0) = 0.40 are found by an analysis of the elliptical shape of this cell [22]. The ratios λmin∆t/ξt = 1.2 · 10−3µm2 and K∆t/ξr = 1.2 · 10−3µm2, and the parameters δ = 11 µm, Narc = 20, and ∆x = 1.1 µm are the same for figures (C-E). The number of iterations is 2.1·106. For definitions of ∆t, ∆x, and Narc, see Appendix B. 11 FIG. 7. Residual function ∆2, defined in Eq. (34), as a func- tion of the anchoring number An (Eq. (33) with R = 23.6µm) for the cell displayed in Fig. 6. The error bars display the standard deviation obtained using jackknife resampling. For large An values the residual flattens, indicating that the actual value of An becomes unimportant once the anchoring torques, which determine the tangential alignment of the stress fibers in the cell's periphery, outcompete the bulk elastic torques. The minimum (∆2 = 0.027) is found for An = 8.0. layer K/W = 2.9 µm. The corresponding numerically calculated configuration is shown in Fig. 6E. However, the flattening of ∆2 for large An values implies that the actual value of An becomes unimportant once the anchor- ing torques, which determine the tangential alignment of the stress fibers in the cell's periphery, outcompete the bulk elastic torques. Therefore, we conclude that the cell illustrated in Fig. 6 is best described by An >∼ 5, corre- sponding to a boundary layer K/W <∼ 5 µm. The same analysis presented above has been repeated for five other cells (Fig. 8). The first column shows the raw experimental data, the second column shows the coarse-grained experimental data, and the third column shows the simulations. For these we used the values of λmin/σ and σ/(σ + α0) obtained from a previous analy- sis of the cell morphology [22] and the An values found by a numerical minimization of ∆2. Also for these cells ∆2 flattens for large An values, and we estimate An >∼ 3 and K/W <∼ 7 µm. Similar to the cell in Fig. 6, we ob- serve that the overall structure of the nematic director, including the emergence of −1/2 topological defects, is captured well by our approach, although there is no pre- cise correspondence between the locations of the defects in the experiments and the theory. V. DISCUSSION AND CONCLUSIONS In this article we investigated the spatial organization of cells adhering on a rigid substrate at a discrete number of points. Our approach is based on a contour model for FIG. 8. Comparison of experimental data on five anisotropic cells with the results of computer simulations. (A-E) Epithelioid (A,B,E) and fibroblastoid (C,D) cells on a micro- fabricated elastomeric pillar array [22]. (F-J) Experimental data of cell shape and coarse-grained cytoskeletal structure on a square lattice of these cells. The white line represents the cell boundary, the black lines in the interior of the cells represent the orientation field n = (cos θSF, sin θSF) of the stress fibers and the background color indicates the local nematic order parameter S. (K-O) Simulations with the adhesion sites of the experimental data as input. The values for λmin/σ = 12.6; 15.7; 18.0; 10.8; 13.4 µm and σ/(σ + α0) = 0.75; 0.25; 0.46; 0.95; 0.52 are found by an analy- sis of the elliptical shape of these cells [22]. The values of An = 4.4; 4.1; 19; 4.6; 4.7, where R = 17.3; 24.4; 39.9; 24.9; 25.3 µm is defined as the square root of the cell area, are de- termined by minimizing ∆2, with the minima given by ∆2 = 0.016; 0.058; 0.057; 0.034; 0.037. These An values correspond to K/W = 3.9; 5.9; 2.1; 5.4; 5.4 µm. The ratios λmin∆t/ξt = 1.2· 10−3µm2 and K∆t/ξr = 1.2· 10−3µm2, and the parameters δ = 11 µm, Narc = 20, and ∆x = 1.1 µm are the same for all cells. The number of iterations is 2.1 · 106. For definitions of ∆t, ∆x, and Narc, see Appendix B. the cell shape [8, 23 -- 26] coupled with a continuous phe- nomenological model for the actin cytoskeleton, inspired by the physics of nematic liquid crystals [62]. This ap- proach can be carried out at various levels of complexity, offering progressively insightful results. Assuming that the orientation of the stress fibers is uniform along in- dividual cellular arcs (but varies from arc to arc), it is possible to achieve a complete analytical description of the geometry of the cell, in which all arcs are approx- imated by different portions of a unique ellipse. The eccentricity of this ellipse depends on the ratio between the isotropic and directed stresses arising in the actin cy- toskeleton, and the orientation of the major axis of this ellipse is parallel to the stress fibers. This method fur- ther allows to analytically calculate the traction forces exerted by the cell on the adhesion sites and compare them with traction force microscopy data. Lifting the assumption that the stress fibers are uni- formly oriented along individual cellular arcs allows one to describe the mechanical interplay between cellular shape and the configuration of the actin cytoskeleton. Using numerical simulations and inputs from experi- ments on fibroblastoid and epithelioid cells plated on stiff micropillar arrays, we identified a feedback mech- anism rooted in the competition between the tendency of stress fibers to align uniformly in the bulk of the cell, but tangentially with respect to the cell edge. Our ap- proach enables us to predict both the shape of the cell and the orientation of the stress fibers and can account for the emergence of topological defects and other dis- tinctive morphological features. These predictions are in good agreement with the experimental data and fur- ther offer an indirect way to estimate quantities that are generally precluded to direct measurement, such as the cell's internal stresses and the orientational stiffness of the actin cytoskeleton. The success of this relatively simple approach is re- markable given the enormous complexity of the cytoskele- ton and the many physical, chemical, and biological mechanisms associated with stress fiber dynamics and alignment [21, 35 -- 53]. Yet, the agreement between our theoretical and experimental results suggests that, on the scale of the whole cell, the myriad of complex mechanisms that govern the dynamics of the stress fibers in adherent cells can be effectively described in terms of simple en- tropic mechanisms, as those at the heart of the physics of liquid crystals. In addition, our analysis demonstrates that chiral sym- metry breaking can originate from the natural interplay between the orientation of the stress fibers and the shape of the cell. A more detailed investigation of this mech- anism is beyond the scope of this study, but will repre- sent a challenge in the near future with the goal of shed- ding light on the fascinating examples of chiral symmetry breaking observed both in single cells [65] and tissues [66]. In the future, we plan to use our model to investigate the mechanics of cells adhering to micropatterned sub- strates that impose reproducible cell shapes [67], with 12 special emphasis to the interplay between cytoskeletal anisotropy and the geometry of the adhesive patches [68]. These systems are not new to theoretical research, but previous studies have focused on either the cytoskeleton [49] or on cell shape [69], rather then on their interac- tion. Furthermore, our framework could be extended to study the role of cytoskeletal anisotropy in cell motility, for instance by taking into account the dynamics of fo- cal adhesions [47, 48], biochemical pathways in the actin cytoskeleton [70], or cellular protrusions and retractions [71]. Finally, our approach could be extended to compu- tational frameworks such as vertex models, cellular Potts models, or phase field models [72], and could provide a starting point for exploring the role of anisotropy in mul- ticellular environments such as tissues [73 -- 80]. AUTHOR CONTRIBUTIONS L.G. designed this study with the help of T.S., R.M.H.M. and E.H.J.D. K.S. and L.G. developed the theoretical framework. K.S. and J.E. developed the code, and J.E. performed the simulations. J.E. and W.P. car- ried out the analysis of the experimental data. K.S. and L.G. wrote the article. All authors commented on the manuscript. ACKNOWLEDGMENTS This work was for supported by funds from the Scientific Research Netherlands Organisation (NWO/OCW), as part of the Frontiers of Nanoscience program (L.G.), the Netherlands Organization for Scientific Research (NWO-FOM) within the program on Barriers in the Brain (W.P. and T.S.; No. FOM L1714M), the Netherlands Organization for Scientific Research (NWO-ALW and NWO-ENW) within the In- novational Research Incentives Scheme (R.M.H.M.; Vidi 2010, No. 864.10.009 and Vici 2017, No. 865.17.004), and the Leiden/Huygens fellowship (K.S.). Appendix A: Angular coordinates of the adhesion sites With reference to the schematic representation of Fig. 2, the coordinates of the center of the ellipse can be ex- pressed as: xc = yc = d 2 d 2 cos φ − γρ sin φ , sin φ + ρ cos φ , (A1a) (A1b) where the length scale ρ is defined in Eq. (17). From Eqs. (A1), standard algebraic manipulations allow us to express the angular coordinate ψ of the adhesion points in the frame of the ellipse (Fig. 2a), namely tan ψ0 = tan ψ1 = d sin φ + 2ρ cos φ d cos φ − 2γρ sin φ d sin φ − 2ρ cos φ d cos φ + 2γρ sin φ , . (A2a) (A2b) Appendix B: Numerical methods The starting point of our numerical simulations are the positions of the adhesion sites. These are directly deter- mined from the experiments by detecting the pillars along the cell contour that are subject to the largest traction forces (see Ref. [22] for more details). These adhesion sites are fixed during the simulation. Cellular arcs are parameterized in terms of a discrete number of vertices connected by straight edges in a chain-like manner. The bulk of the cell, representing the cytoskeleton, is defined as the region enclosed by the cell edge, and is discretized as a regularly spaced two-dimensional square lattice with Q defined at every lattice point. At t = 0 the cell consists of an irregular polygon en- closing a random configuration of the nematic tensor. Evidently, this bears no resemblance to a real cell, but reduces the risk of a possible bias in the final configura- tion. The time-integration is performed by alternating updates of the cell contour and of the bulk nematic ten- sor. This process is iterated until both the cell edge and the cell bulk reach a steady-state. Our code is available upon request. The configurations of the cytoskeleton in Figs. 4, 5, 6 and 8 have been visualized with Mathematica Version 11.3 (Wolfram Research, Champaign, IL) using the line integral convolution tool. 1. Numerical implementation of the cell bulk Eq. (31b) is numerically solved via a finite-difference scheme. Time integration is performed using the forward Euler method, whereas spatial derivatives are calculated using the centered difference approximation. In order to calculate derivatives at lattice points located in proxim- ity of the edge, we use the boundary conditions, specified in Eq. (27), to express the values of Qxx and Qxy in a number of ghost points located outside the cells. This is conveniently done upon identifying three possible scenar- ios, illustrated in Fig. 9. 1) There is a single ghost point on the x− or y−axis (Fig. 9A). 2) There are two ghost points, one on each axis (Fig. 9B). 3) There are two ghost points on the same axis and possibly a third one on the other axis (Fig. 9C). In the following, we explain how to address each of these cases. 1) Using the centered difference approximation for the first derivative yields the following expression of the ne- 13 FIG. 9. Schematic overview of the three geometrical situa- tions described in Appendix B. (A) There is a single ghost- point on the x− or y−axis. (B) There are two ghost points, one on each axis. (C) There are two ghost points on the same axis and possibly a third one on the other axis. matic tensor at a ghost point located at (x ± ∆x, y) or (x, y ± ∆y), with ∆x = ∆y the lattice spacing: Qij(x ± ∆x, y) = Qij(x ∓ ∆x, y) ± 2∆x ∂xQij(x, y) , Qij(x, y ± ∆y) = Qij(x, y ∓ ∆y) ± 2∆y ∂yQij(x, y) . (B1a) (B1b) The derivative with respect to x in Eq. (B1a) can (27), upon taking N = ±x, be calculated from Eq. where the plus (minus) sign correspond to a ghost point located on the left (right) of the central edge point. Thus N · ∇Qij = ±∂xQij. Analogously, the deriva- tive with respect to y in Eq. (B1b), is approximated as N · ∇Qij = ±∂yQij, where the plus (minus) sign corre- sponds to a ghost point located below (above) the central edge point. Combining this with Eq. (27), yields: Qij(x ± ∆x, y) = Qij(x ∓ ∆x, y) − 4∆x W K [Qij(x, y) − Q0,ij(x, y)] , (B2a) Qij(x, y ± ∆y) = Qij(x, y ∓ ∆y) − 4∆y W K [Qij(x, y) − Q0,ij(x, y)] . (B2b) The tensor Q0,ij is evaluated via Eq. (26) using the local orientation of the cell edge. 2) If a given lattice point is linked to ghost points in both the x− and y−directions, we evaluate equation (B2) for both directions independently as explained in the pre- vious paragraph. 3) If a given lattice point is linked to two ghost points in either the x− or y−direction, we employ a forward or backward finite difference approximation for the first spa- tial derivative of Qij to evaluate Qij at the ghost points. This yields: Qij(x ± ∆x, y) = Qij(x, y) − 2∆x W K [Qij(x, y) − Q0,ij(x, y)] , (B3a) BInternal grid pointACGhost pointCentral grid pointQij(x, y ± ∆y) = Qij(x, y) − 2∆y W K [Qij(x, y) − Q0,ij(x, y)] . (B3b) Finally, if the given lattice point is also linked to a ghost point on the other axis, this is evaluated independently using Eq. (B2). The lattice points enclosed by the cells continuously change during the course of a simulation, as a conse- quence of the relaxation of the cell shape. In case of in- clusion of a new lattice point that was previously located outside the cell, the associated Qxx and Qxy values are generated by averaging over the nearest neighbours (hor- izontally and vertically, not diagonally) that were inside the cell during the previous time step. 2. Numerical implementation of the cell edge To calculate the line tension λ, Eq. (30) is discretized as follows: λk = λ0−α0 k(cid:88) n=1 ∆sn Tn·(cid:68) Qn (cid:69)·Nn , k = 1, 2 . . . Narc , (B4) where λ0 is the line tension at the adhesion site at s = 0 (position r0) and λk is the line tension at vertex k (position rk). Narc is the total number of vertices in which cellular arcs are discretized, and λNarc represents the line tension at the other adhesion site. Furthermore, ∆sn = rn − rn−1, Tn = (rn − rn−1)/∆sn, Nn = T ⊥ and n (cid:69) (cid:68) Qn = Qn + Qn−1 2 , (B5) with Qn and Qn−1 the nematic tensor at the vertices n and n − 1. These are set equal to Q at the closest bulk lattice point inside the cell among the four points delimiting the unit cell containing the edge vertices n and n−1 respectively. If none of these is inside the cell, we set Qxx,n = Qxy,n = 0. The quantity λ0 is calculated in such a way that the minimal λ value along an arc equates the input parameter λmin, representing the minimal tension withstood by the cortical actin. Next, the position of the vertices rk, k = 0, 1 . . . Narc is updated upon integrating Eq. (31a) using the forward Euler method. The curvature and normal vector at ver- tex k, κk and Nk, are found by constructing a circle with radius R through vertices k − 1, k, and k + 1. The vector from vertex k to the center of the circle is then equated to ±RNk, with the sign such that Nk is an inward pointing normal vector, and κk = ±1/R, with a negative sign for a concave shape and a positive sign for a convex shape. Along each arc, r0 and rNarc represent the positions of the adhesion sites and are kept fixed during simulations. 14 Appendix C: Experimental data analysis The coarse-grained experimental data displayed in Figs. 6B and 8F-J are obtained using the following four- step procedure. 1) Epithelioid GE11 and fibroblastoid GD25 cells ex- pressing either α5β1 or αvβ3 (GDβ1, GDβ3, GEβ1 and GEβ3) [81] are cultured on microfabricated elastomeric pillar arrays [28 -- 30]. The resulting cells are imaged using spinning disk confocal microscopy and displayed in Figs. 6A and 8A-E. For details, see Ref. [22]. 2) The locations of the cell interior and the cell edge are found by applying a low-pass filter on the images using Matlab. The interior of the cell is then sampled by overlaying a square lattice of 512 × 512 pixels (1 pixel = 0.138× 0.138 µm2) on the microscope field-of-view (Figs. 6A and 8A-E). 3) For all pixels that are inside the cell, we calcu- late the nematic tensor using ImageJ with the Orien- tationJ plugin [82] in the following way. Given the in- tensity I(x0, y0) of the image at the point (x0, y0), we define the symmetric 2 × 2 matrix J = (cid:104)∇I∇I(cid:105), where (cid:104)···(cid:105) = (cid:82) w(x, y)dx dy (··· ) represents a weighted aver- age with w(x, y) a Gaussian with a standard deviation of five pixels (0.69 µm) centered at (x0, y0). The J matrix can be expressed as: J = (Λmin − Λmax) Λmax + Λmin (cid:19) (cid:18) I , + eminemin − 1 2 I 2 (C1) where Λmax and Λmin are the largest and smallest eigen- values of J , emin the eigenvector corresponding to Λmin, and I the two-dimensional identity matrix. The J matrix is then used to estimate the average stress fiber direction u: (cid:104)∇I∇I(cid:105) (cid:104)∇I2(cid:105) = I − (cid:104)uu(cid:105) . (cid:28) Here, the quantity I −(cid:104)uu(cid:105) reflects that the largest gra- dients in intensity are perpendicular to the orientation of the stress fibers and (cid:104)∇I2(cid:105) = tr J = Λmax + Λmin. Combining Eqs. (C1) and (C2), we obtain uu − 1 2 I Λmax − Λmin Λmax + Λmin = eminemin − 1 2 I . (C3) Comparing this with the definition of the nematic tensor [62]: (C2) (cid:19) (cid:29) (cid:19) Q = uu − 1 2 I = S nn − 1 2 I , (C4) (cid:18) (cid:18) (cid:29) (cid:28) we find: S = Λmax − Λmin Λmax + Λmin , n = (cos θSF, sin θSF) = emin . (C5) 4) The data are further coarse-grained in blocks of 8 × 8 pixels corresponding to regions of size 1.104 × 1.104 µm2 in real space. This results in a new 64 × 64 lattice. The value of the nematic tensor in the new coarse-grained pix- els is obtained from an average over those of the original 15 8 × 8 pixels located inside the cell. In turn, the coarse- grained pixels are considered inside the cell if more than half of the original pixels were inside the cell. [1] C. M. Lo, H. B. Wang, M. Dembo, and Y. L. Wang, "Cell movement is guided by the rigidity of the substrate.," Biophysical journal, vol. 79, no. 1, pp. 144 -- 152, 2000. [2] R. D. Sochol, A. T. Higa, R. R. R. Janairo, S. Li, and L. Lin, "Unidirectional mechanical cellular stimuli via micropost array gradients," Soft Matter, vol. 7, no. 10, p. 4606, 2011. [3] C. A. Reinhart-King, M. Dembo, and D. A. Hammer, "Cell-cell mechanical communication through compliant substrates," Biophysical Journal, vol. 95, no. 12, pp. 6044 -- 6051, 2008. [4] Y. Sawada, M. Tamada, B. J. Dubin-Thaler, O. Cherni- avskaya, R. Sakai, S. Tanaka, and M. P. Sheetz, "Force sensing by mechanical extension of the Src family kinase substrate p130Cas.," Cell, vol. 127, no. 5, pp. 1015 -- 26, 2006. [5] A. J. Engler, S. Sen, H. L. Sweeney, and D. E. Discher, "Matrix elasticity directs stem cell lineage specification.," Cell, vol. 126, no. 4, pp. 677 -- 689, 2006. [6] B. Trappmann, J. E. Gautrot, J. T. Connelly, D. G. T. Strange, Y. Li, M. L. Oyen, M. A. Cohen Stuart, H. Boehm, B. Li, V. Vogel, J. P. Spatz, F. M. Watt, and W. T. S. Huck, "Extracellular-matrix tethering reg- ulates stem-cell fate.," Nature materials, vol. 11, no. 7, pp. 642 -- 649, 2012. [7] T. Panciera, L. Azzolin, M. Cordenonsi, and S. Piccolo, "Mechanobiology of yap and taz in physiology and dis- ease," Nature Reviews Molecular Cell Biology, vol. 18, p. 758, 2017. [8] I. Bischofs, S. Schmidt, and U. Schwarz, "Effect of Adhe- sion Geometry and Rigidity on Cellular Force Distribu- tions," Physical Review Letters, vol. 103, no. 4, pp. 1 -- 4, 2009. [9] M. Ghibaudo, L. Trichet, J. Le Digabel, A. Richert, P. Hersen, and B. Ladoux, "Substrate topography in- duces a crossover from 2D to 3D behavior in fibroblast migration.," Biophysical journal, vol. 97, no. 1, pp. 357 -- 368, 2009. [10] D. A. Fletcher and R. D. Mullins, "Cell mechanics and the cytoskeleton," Nature, vol. 463, no. 7280, pp. 485 -- 492, 2010. [11] N. Minc, D. Burgess, and F. Chang, "Influence of cell geometry on division-plane positioning," Cell, vol. 144, no. 3, pp. 414 -- 426, 2011. [12] M. Versaevel, T. Grevesse, and S. Gabriele, "Spatial co- ordination between cell and nuclear shape within mi- cropatterned endothelial cells," Nature Communications, vol. 3, p. 671, 2012. [13] C. S. Chen, M. Mrksich, S. Huang, G. M. Whitesides, and D. E. Ingber, "Geometric control of cell life and death," Science, vol. 276, no. 5317, pp. 1425 -- 1428, 1997. [14] N. Jain, K. V. Iyer, A. Kumar, and G. V. Shiv- ashankar, "Cell geometric constraints induce modular gene-expression patterns via redistribution of hdac3 regu- lated by actomyosin contractility," Proceedings of the Na- tional Academy of Sciences, vol. 110, no. 28, pp. 11349 -- 11354, 2013. [15] R. McBeath, D. M. Pirone, C. M. Nelson, K. Bhadri- raju, and C. S. Chen, "Cell shape, cytoskeletal tension, and rhoa regulate stem cell lineage commitment," Devel- opmental Cell, vol. 6, no. 4, pp. 483 -- 495, 2004. [16] K. A. Kilian, B. Bugarija, B. T. Lahn, and M. Mrk- sich, "Geometric cues for directing the differentiation of mesenchymal stem cells," Proceedings of the National Academy of Sciences, vol. 107, no. 11, pp. 4872 -- 4877, 2010. [17] A. Chopra, E. Tabdanov, H. Patel, P. A. Janmey, and J. Y. Kresh, "Cardiac myocyte remodeling medi- ated by n-cadherin-dependent mechanosensing," Amer- ican Journal of Physiology-Heart and Circulatory Physi- ology, vol. 300, no. 4, pp. H1252 -- H1266, 2011. [18] A. J. Engler, M. A. Griffin, S. Sen, C. G. Bonnemann, H. L. Sweeney, and D. E. Discher, "Myotubes differen- tiate optimally on substrates with tissue-like stiffness," The Journal of Cell Biology, vol. 166, no. 6, pp. 877 -- 887, 2004. [19] 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 Motility, vol. 60, no. 1, pp. 24 -- 34, 2005. [20] F. Grinnell, "Fibroblast-collagen-matrix contraction: growth-factor signalling and mechanical loading," Trends in Cell Biology, vol. 10, no. 9, pp. 362 -- 365, 2000. [21] A. Zemel, F. Rehfeldt, A. E. X. Brown, D. E. Dis- cher, and S. A. Safran, "Cell shape, spreading symmetry, and the polarization of stress-fibers in cells," Journal of Physics: Condensed Matter, vol. 22, no. 19, p. 194110, 2010. [22] W. Pomp, K. Schakenraad, H. E. Balcıoglu, H. van Hoorn, E. H. J. Danen, R. M. H. Merks, T. Schmidt, and L. Giomi, "Cytoskeletal anisotropy controls geome- try and forces of adherent cells," Physical Review Letters, vol. 121, no. 17, p. 178101, 2018. [23] U. S. Schwarz and S. A. Safran, "Physics of adher- ent cells," Reviews of Modern Physics, vol. 85, no. 3, pp. 1327 -- 1381, 2013. [24] R. Bar-Ziv, T. Tlusty, E. Moses, S. A. Safran, and A. Ber- shadsky, "Pearling in cells: A clue to understanding cell shape," Proceedings of the National Academy of Sciences, vol. 96, no. 18, pp. 10140 -- 10145, 1999. [25] I. B. Bischofs, F. Klein, D. Lehnert, M. Bastmeyer, and U. S. Schwarz, "Filamentous Network Mechanics and Active Contractility Determine Cell and Tissue Shape," Biophysical Journal, vol. 95, no. 7, pp. 3488 -- 3496, 2008. [26] L. Giomi, "Contour models of cellular adhesion," in Cell migrations: causes and function (S. Zapperi and C. A. M. La Porta, eds.), p. in press, Berlin: Springer, 2019. [27] E. H. J. Danen, P. Sonneveld, C. Brakebusch, R. Fassler, and A. Sonnenberg, "The fibronectin-binding integrins α5β1 and αvβ3 differentially modulate RhoAGTP load- ing, organization of cell matrix adhesions, and fibronectin fibrillogenesis," The Journal of Cell Biology, vol. 159, no. 6, pp. 1071 -- 1086, 2002. [28] J. L. Tan, J. Tien, D. M. Pirone, D. S. Gray, K. Bhadri- raju, and C. S. Chen, "Cells lying on a bed of micronee- dles: An approach to isolate mechanical force," Proceed- ings of the National Academy of Sciences, vol. 100, no. 4, pp. 1484 -- 1489, 2003. [29] 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 mechanosens- ing mechanism for cellular adaptation to substrate stiff- ness," Proceedings of the National Academy of Sciences, vol. 109, no. 18, pp. 6933 -- 6938, 2012. [30] H. Van Hoorn, R. Harkes, E. M. Spiesz, C. Storm, D. Van Noort, B. Ladoux, and T. Schmidt, "The nanoscale archi- tecture of force-bearing focal adhesions," Nano Letters, vol. 14, no. 8, pp. 4257 -- 4262, 2014. [31] T. Vignaud, L. Blanchoin, and M. Th´ery, "Directed cytoskeleton self-organization," Trends in Cell Biology, vol. 22, no. 12, pp. 671 -- 682, 2012. [32] B. Ladoux, R.-M. M`ege, and X. Trepat, "Frontrear polar- ization by mechanical cues: From single cells to tissues," Trends in Cell Biology, vol. 26, no. 6, pp. 420 -- 433, 2016. [33] N. T. Lam, T. J. Muldoon, K. P. Quinn, N. Rajaram, and K. Balachandran, "Valve interstitial cell contractile strength and metabolic state are dependent on its shape," Integrative Biology, vol. 8, no. 10, pp. 1079 -- 1089, 2016. [34] S. K. Gupta, Y. Li, and M. Guo, "Anisotropic mechanics and dynamics of a living mammalian cytoplasm," Soft Matter, vol. 15, no. 2, pp. 190 -- 199, 2019. [35] V. S. Deshpande, R. M. McMeeking, and A. G. Evans, "A model for the contractility of the cytoskeleton includ- ing the effects of stress-fibre formation and dissociation," Proceedings of the Royal Society of London A: Math- ematical, Physical and Engineering Sciences, vol. 463, no. 2079, pp. 787 -- 815, 2007. [36] S. Walcott and S. X. Sun, "A mechanical model of actin stress fiber formation and substrate elasticity sensing in adherent cells," Proceedings of the National Academy of Sciences, vol. 107, no. 17, pp. 7757 -- 7762, 2010. [37] 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 Physics, vol. 6, p. 468, 2010. [38] N. Nisenholz, M. Botton, and A. Zemel, "Early-time dy- namics of actomyosin polarization in cells of confined shape in elastic matrices," Soft Matter, vol. 10, no. 14, pp. 2453 -- 2462, 2014. [39] M. Raab, J. Swift, P. D. P. Dingal, P. Shah, J.-W. Shin, and D. E. Discher, "Crawling from soft to stiff matrix polarizes the cytoskeleton and phosphoregulates myosin- ii heavy chain," The Journal of Cell Biology, vol. 199, no. 4, pp. 669 -- 683, 2012. [40] V. Schaller, C. Weber, C. Semmrich, E. Frey, and A. R. Bausch, "Polar patterns of driven filaments," Nature, vol. 467, p. 73, 2010. [41] P. Kraikivski, R. Lipowsky, and J. Kierfeld, "Enhanced ordering of interacting filaments by molecular motors," Physical Review Letters, vol. 96, no. 25, p. 258103, 2006. [42] S. Swaminathan, F. Ziebert, D. Karpeev, and I. S. Aranson, "Motor-mediated alignment of microtubules in semidilute mixtures," Physical Review E, vol. 79, no. 3, p. 036207, 2009. 16 [43] M. Soares e Silva, J. Alvarado, J. Nguyen, N. Georgoulia, B. M. Mulder, and G. H. Koenderink, "Self-organized patterns of actin filaments in cell-sized confinement," Soft Matter, vol. 7, no. 22, pp. 10631 -- 10641, 2011. [44] J. Alvarado, B. M. Mulder, and G. H. Koenderink, "Alignment of nematic and bundled semiflexible poly- mers in cell-sized confinement," Soft Matter, vol. 10, no. 14, pp. 2354 -- 2364, 2014. [45] S. Deshpande and T. Pfohl, "Hierarchical self-assembly of actin in micro-confinements using microfluidics," Biomi- crofluidics, vol. 6, no. 3, p. 034120, 2012. [46] V. S. Deshpande, R. M. McMeeking, and A. G. Evans, "A bio-chemo-mechanical model for cell contractility," Pro- ceedings of the National Academy of Sciences, vol. 103, no. 38, pp. 14015 -- 14020, 2006. [47] V. S. Deshpande, M. Mrksich, R. M. McMeeking, and A. G. Evans, "A bio-mechanical model for coupling cell contractility with focal adhesion formation," Journal of the Mechanics and Physics of Solids, vol. 56, no. 4, pp. 1484 -- 1510, 2008. [48] E. G. Rens and R. M. H. Merks, "Cell shape and duro- taxis follow from mechanical cell-substrate reciprocity: A multiscale model of cell behavior." In preparation. [49] A. Pathak, V. S. Deshpande, R. M. McMeeking, and A. G. Evans, "The simulation of stress fibre and focal adhesion development in cells on patterned substrates," Journal of The Royal Society Interface, vol. 5, no. 22, pp. 507 -- 524, 2008. [50] W. Ronan, V. S. Deshpande, R. M. McMeeking, and J. P. McGarry, "Cellular contractility and substrate elas- ticity: a numerical investigation of the actin cytoskele- ton and cell adhesion," Biomechanics and Modeling in Mechanobiology, vol. 13, no. 2, pp. 417 -- 435, 2014. [51] A.-C. Reymann, J.-L. Martiel, T. Cambier, L. Blanchoin, R. Boujemaa-Paterski, and M. Th´ery, "Nucleation geom- etry governs ordered actin networks structures," Nature Materials, vol. 9, p. 827, 2010. [52] G. Letort, A. Z. Politi, H. Ennomani, M. Th´ery, F. Ned- elec, and L. Blanchoin, "Geometrical and mechanical properties control actin filament organization," PLOS Computational Biology, vol. 11, no. 5, pp. 1 -- 21, 2015. [53] A. P. Liu, D. L. Richmond, L. Maibaum, S. Pronk, P. L. Geissler, and D. A. Fletcher, "Membrane-induced bundling of actin filaments," Nature Physics, vol. 4, p. 789, 2008. [54] S. Thomopoulos, G. M. Fomovsky, and J. W. Holmes, "The development of structural and mechanical anisotropy in fibroblast populated collagen gels," Journal of Biomechanical Engineering, vol. 127, no. 5, pp. 742 -- 750, 2005. [55] K. Burridge and M. Chrzanowska-Wodnicka, "Focal Ad- hesions, Contractility, and Signaling," Annual Review of Cell and Developmental Biology, vol. 12, no. 1, pp. 463 -- 519, 1996. [56] S. Banerjee and L. Giomi, "Polymorphism and bistability in adherent cells," Soft Matter, vol. 9, no. 21, pp. 5251 -- 5260, 2013. [57] L. Giomi, "Softly constrained films," Soft Matter, vol. 9, no. 34, p. 8121, 2013. [58] T. J. Pedley and J. O. Kessler, "Hydrodynamic Phenom- ena in Suspensions of Swimming Microorganisms," An- nual Review of Fluid Mechanics, vol. 24, no. 1, pp. 313 -- 358, 1992. [59] R. A. Simha and S. Ramaswamy, "Hydrodynamic Fluc- tuations and Instabilities in Ordered Suspensions of Self- Propelled Particles," Physical Review Letters, vol. 89, no. 5, p. 058101, 2002. [60] S. Pellegrin and H. Mellor, "Actin stress fibres.," Journal of cell science, vol. 120, no. 20, pp. 3491 -- 3499, 2007. [61] K. Burridge and E. S. Wittchen, "The tension mounts: Stress fibers as force-generating mechanotransducers," The Journal of Cell Biology, vol. 200, no. 1, pp. 9 -- 19, 2013. [62] P. G. de Gennes and J. Prost, The Physics of Liq- uid Crystals. International Series of Monogr, Clarendon Press, 1995. [63] M. Nobili and G. Durand, "Disorientation-induced disor- dering at a nematic-liquid-crystal -- solid interface," Phys. Rev. A, vol. 46, no. 10, pp. R6174 -- R6177, 1992. [64] J. Galanis, D. Harries, D. L. Sackett, W. Losert, and R. Nossal, "Spontaneous patterning of confined granular rods," Physical Review Letters, vol. 96, no. 2, p. 028002, 2006. [65] Y. H. Tee, T. Shemesh, V. Thiagarajan, R. F. Hari- adi, K. L. Anderson, C. Page, N. Volkmann, D. Hanein, S. Sivaramakrishnan, M. M. Kozlov, and A. D. Bershad- sky, "Cellular chirality arising from the self-organization of the actin cytoskeleton," Nature Cell Biology, vol. 17, p. 445, 2015. [66] G. Duclos, C. Blanch-Mercader, V. Yashunsky, G. Sal- breux, J.-F. Joanny, J. Prost, and P. Silberzan, "Spon- taneous shear flow in confined cellular nematics," Nature Physics, vol. 14, no. 7, pp. 728 -- 732, 2018. [67] M. Th´ery, "Micropatterning as a tool to decipher cell morphogenesis and functions," Journal of Cell Science, vol. 123, no. 24, pp. 4201 -- 4213, 2010. [68] M. Th´ery, A. P´epin, E. Dressaire, Y. Chen, and M. Bor- nens, "Cell distribution of stress fibres in response to the geometry of the adhesive environment," Cell Motility, vol. 63, no. 6, pp. 341 -- 355, 2006. [69] P. J. Albert and U. S. Schwarz, "Dynamics of cell shape and forces on micropatterned substrates predicted by a cellular potts model," Biophysical Journal, vol. 106, no. 11, pp. 2340 -- 2352, 2014. [70] A. F. M. Mar´ee, V. A. Grieneisen, and L. Edelstein- Keshet, "How cells integrate complex stimuli: The ef- fect of feedback from phosphoinositides and cell shape on cell polarization and motility," PLOS Computational Biology, vol. 8, no. 3, pp. 1 -- 20, 2012. [71] F. J. Segerer, F. Thuroff, A. Piera Alberola, E. Frey, and J. O. Radler, "Emergence and persistence of collective cell migration on small circular micropatterns," Physical Review Letters, vol. 114, no. 22, p. 228102, 2015. 17 [72] P. J. Albert and U. S. Schwarz, "Modeling cell shape and dynamics on micropatterns," Cell Adhesion & Migration, vol. 10, no. 5, pp. 516 -- 528, 2016. [73] M. Eastwood, V. C. Mudera, D. A. McGrouther, and R. A. Brown, "Effect of precise mechanical loading on fibroblast populated collagen lattices: morphological changes.," Cell Motility and the Cytoskeleton, vol. 40, no. 1, pp. 13 -- 21, 1998. [74] D. W. J. Van der Schaft, A. C. C. Van Spreeuwel, H. C. Van Assen, and F. P. T. Baaijens, "Mechanoregulation of Vascularization in Aligned Tissue-Engineered Muscle: A Role for Vascular Endothelial Growth Factor," Tissue Engineering Part A, vol. 17, no. 21-22, pp. 2857 -- 2865, 2011. [75] O. Wartlick, P. Mumcu, F. Julicher, and M. Gonzalez- Gaitan, "Understanding morphogenetic growth control - lessons from flies.," Nature Reviews Molecular Cell Biol- ogy, vol. 12, no. 9, pp. 594 -- 604, 2011. [76] R. F. M. Van Oers, E. G. Rens, D. J. LaValley, C. A. Reinhart-King, and R. M. H. Merks, "Mechanical Cell- Matrix Feedback Explains Pairwise and Collective En- dothelial Cell Behavior In Vitro," PLoS Computational Biology, vol. 10, no. 8, p. e1003774, 2014. [77] P. Santos-Oliveira, A. Correia, T. Rodrigues, T. M. Ribeiro-Rodrigues, P. Matafome, J. C. Rodr´ıguez- Manzaneque, R. Sei¸ca, H. Girao, and R. D. M. Travasso, "The Force at the Tip - Modelling Tension and Prolifer- ation in Sprouting Angiogenesis," PLoS Computational Biology, vol. 11, no. 8, pp. 1 -- 20, 2015. [78] D. S. Vijayraghavan and L. A. Davidson, "Mechanics of neurulation: From classical to current perspectives on the physical mechanics that shape, fold, and form the neural tube," Birth Defects Research, vol. 109, no. 2, pp. 153 -- 168, 2017. [79] T. B. Saw, A. Doostmohammadi, V. Nier, L. Kocgozlu, S. Thampi, Y. Toyama, P. Marcq, C. T. Lim, J. M. Yeo- mans, and B. Ladoux, "Topological defects in epithelia govern cell death and extrusion," Nature, vol. 544, p. 212, 2017. [80] D. L. Barton, S. Henkes, C. J. Weijer, and R. Sknepnek, "Active vertex model for cell-resolution description of ep- ithelial tissue mechanics," PLOS Computational Biology, vol. 13, no. 6, pp. 1 -- 34, 2017. [81] H. E. Balcioglu, H. van Hoorn, D. M. Donato, T. Schmidt, and E. H. J. Danen, "The integrin expres- sion profile modulates orientation and dynamics of force transmission at cell-matrix adhesions.," Journal of Cell Science, vol. 128, no. 7, pp. 1316 -- 1326, 2015. [82] OrientationJ, orientation." "http://bigwww.epfl.ch/demo/
1602.01666	1	1602	2016-02-04T13:21:51	Entrainment dominates the interaction of microalgae with micron-sized objects	[ "physics.bio-ph", "cond-mat.soft", "cond-mat.stat-mech", "physics.flu-dyn" ]	The incessant activity of swimming microorganisms has a direct physical effect on surrounding microscopic objects, leading to enhanced diffusion far beyond the level of Brownian motion with possible influences on the spatial distribution of non-motile planktonic species and particulate drifters. Here we study in detail the effect of eukaryotic flagellates, represented by the green microalga Chlamydomonas reinhardtii, on microparticles. Macro- and micro-scopic experiments reveal that microorganism-colloid interactions are dominated by rare close encounters leading to large displacements through direct entrainment. Simulations and theoretical modelling show that the ensuing particle dynamics can be understood in terms of a simple jump-diffusion process, combining standard diffusion with Poisson-distributed jumps. This heterogeneous dynamics is likely to depend on generic features of the near-field of swimming microorganisms with front-mounted flagella.	physics.bio-ph	physics	Entrainment dominates the interaction of microalgae with micron-sized objects 1Physics Department, University of Warwick, Gibbet Hill Road, Coventry CV4 7AL, United Kingdom Raphael Jeanneret1, Vasily Kantsler1, Marco Polin1∗ (Dated: October 2, 2018) The incessant activity of swimming microorganisms has a direct physical effect on surrounding microscopic objects, leading to enhanced diffusion far beyond the level of Brownian motion with pos- sible influences on the spatial distribution of non-motile planktonic species and particulate drifters. Here we study in detail the effect of eukaryotic flagellates, represented by the green microalga Chlamydomonas reinhardtii, on microparticles. Macro- and micro-scopic experiments reveal that microorganism-colloid interactions are dominated by rare close encounters leading to large displace- ments through direct entrainment. Simulations and theoretical modelling show that the ensuing particle dynamics can be understood in terms of a simple jump-diffusion process, combining stan- dard diffusion with Poisson-distributed jumps. This heterogeneous dynamics is likely to depend on generic features of the near-field of swimming microorganisms with front-mounted flagella. Introduction navigate Swimming microorganisms commonly through fluids characterised by a variety of suspended microparticles, with which they inevitably interact. From malarial parasites meandering around densely packed red blood cells and infecting them [1], to protists which regulate primary production in oceans and lakes by grazing on sparsely distributed microalgae and bacteria [2, 3], these interactions can have important biological and ecological implications [4]. Understanding the basic mechanisms of such interactions is also timely, as a vast number of plastic micro- and nano-particles of anthropogenic origin scattered throughout the world's oceans (up to ∼ 105 /m3) appear to be easily ingested by microorganisms, which then recycle plastics within the marine food web with currently unknown consequences [5]. From a physics perspective, these systems are particularly appealing. Abstracted as binary suspen- sions [6, 7], where an active, self-propelled species interacts with a passive, thermalised one, they represent a naturally occurring category of out of equilibrium stochastic systems driven by energy produced directly within the bulk, rather than transmitted through the system's boundaries. Easily accessible experimentally, and amenable to detailed quantitative modelling [8], they are uniquely placed to provide benchmark tests for theories of out-of-equilibrium statistical mechanics [9 -- 11]. Currently best characterised are the so-called bacterial baths. Wu and Libchaber [12], and more recently [13 -- 16], showed that colloidal particles within bacterial suspensions perform a persistent random walk [17 -- 19] leading to a diffusivity up to ∼ 10× the thermal value [16]. This increase is proportional to bacterial concentration, at least in the dilute limit, and can be described by introducing a particle-size-dependent effective temperature [16], which captures well also the colloids' velocity fluctuations [16]. At the same time, coarse graining bacteria-microparticle interactions in terms of an effective temperature is insufficient to account for other exquisitely nonequilibrium dynamical properties, like motility-induced pair interactions [8] or coupling between enhanced translational and rotational diffusion [20]. This shortfall is particularly striking for larger mesoscopic particles, where a careful choice of shape can result in directed translation [21, 22], and rotation [23, 24]. Microparticles' interactions with the other major class of microorganisms, eukaryotic flagellates, is distinctly less explored. Almost exclusively larger than bacteria (∼ 10 − 100 µm), these species probe a new and biologically relevant physical regime, where the microparticles' size is significantly smaller than that of the microorganisms. Working with the microalga Chlamydomonas reinhardtii, often studied as model eukaryotic microswimmer, the pioneering study of Leptos et al. [25] (followed by [26]) reported an increase in particle diffusivity of magnitude similar to the bacterial case, but due to a diffusively scaling fat-tailed distribution of displacements later shown to be in agreement with estimates based on the far field flows generated by swimming [27, 28]. Here we revisit the behaviour of colloids within a sus- pension of Chlamydomonas reinhardtii (CR), taken as representative of eukaryotic flagellates, and reveal that their overall dynamics is in fact dominated by rare but dramatic entrainment events. These jumps underpin the surprisingly large diffusivity we observe directly in both sedimentation and collective spreading experiments, (cid:38) 40× previously reported values. The colloids' behav- ior, alternating entrainments and periods of standard en- hanced diffusion, is fundamentally different from the per- sistent random walk common with bacterial suspensions. Through microscopic experiments, simulations and ana- lytical modelling we show instead that this dynamics is well captured by a simple jump-diffusion model reminis- cent of actomyosin-dependent molecular clustering [29]. 6 1 0 2 b e F 4 ] h p - o i b . s c i s y h p [ 1 v 6 6 6 1 0 . 2 0 6 1 : v i X r a 2 FIG. 1. Effect of swimming algae on particle diffusivity. a) Semi-log plot of the normalised density profiles of 1µm-PS colloids along the gravity direction for different cells' concentrations. Full lines are best exponential fits to the data. Curves have been shifted apart along the y-axis for clarity. Colorbar: CR concentration Nc in units of 106 cells/ml. b) Gravitational length lg,ef f − l0 as a function of Nc extracted from the fits in panel a). The solid orange line is the best linear fit to the data, lg,ef f − l0 = ((62.8 ± 5)Nc) µm (Nc in units of 106 cells/ml). c) Effective microparticle diffusivity (Def f − D0) as a function of Nc for the three experiments. Diffusivities are rescaled by the ratio of the average CR speeds (cid:104)v(cid:105)S /(cid:104)v(cid:105)j where j stands for sedimentation (S), spreading (CS) or tracking (2D) experiments. Solid lines are best linear fits to the data. Orange squares: sedimentation experiment (slope αS = 1.71 ± 0.14 (µm2/s)/(106 cells/ml); all other values in the same units); green triangles: spreading experiment (αCS = 1.62 ± 0.14); blue circles: quasi-2D experiment (α2D = 1.67 ± 0.13). Orange circles and dashed line: direct tracking in the sedimentation experiment (αT = 0.074±0.014). Black dashed line: fit to the experimental diffusivity obtained by direct tracking in [25] (αL = 0.041). A close-up on the low-concentration values is available in [36]. Experiments A full description of the experimental procedures can be found in the Methods section. Briefly, wild type CR strain CC125 was grown axenically in Tris-Acetate- Phosphate medium [30] at 21◦C under continuous fluo- rescent illumination. Cells were harvested in the expo- nential phase (∼ 5× 106cells/ml), concentrated by gentle spinning and then resuspended to reach the desired con- centration. Polystyrene microparticles (1 µm diameter) were then added to the suspension, and their diffusivity at different CR concentrations (Nc) was measured from three different sets of experiments: mapping the par- ticles' sedimentation profile at steady state; measuring the relaxation dynamics of an inhomogeneous distribu- tion of particles; and by long-timescale direct tracking of individual colloids' dynamics in a thin Hele-Shaw cell. Except for the sedimentation experiments, the medium was density matched with the colloids using Percoll [31]. This increased its viscosity as reflected in the slower av- erage swimming speed, (cid:104)v(cid:105), of the algae. We measured (cid:104)v(cid:105)S = 81.8 ± 3.5 µm/s, (cid:104)v(cid:105)CS = 40.9 ± 3.5 µm/s, and (cid:104)v(cid:105)2D = 49.1± 2.5 µm/s for the sedimentation, spreading and tracking experiments respectively. Macroscopic diffusion Macroscopic diffusion experiments coarse-grain over the colloids' microscopic dynamics and ensure the direct measurement of their effective transport properties, in the spirit of Jean Perrin's seminal work on sedimenta- tion equilibrium [32]. We begin by characterizing the colloids' steady-state sedimentation profile at increasing CR concentrations, always within the dilute regime (vol- ume fractions (cid:46) 0.15%). Fig. 1a shows that even when the algae are present, the microparticles' distributions are still in excellent agreement with simple exponential pro- files, but crucially with different effective gravitational lengths lg,ef f (Fig. 1b). The exponential profiles are a standard consequence of the balance between the parti- cles' sedimentation speed vsed, here due to a δρ = 50 g/l density mismatch, and their effective diffusivity, combin- ing passive and active processes. Boltzmann-like distri- butions are indeed what should be expected even in these out-of-equilibrium systems, at least for small enough vsed [33], akin to what has been observed for active colloids alone [34]. The characteristic length lg,ef f , experimen- tally observed to be proportional to the concentration of algae, allows us to measure the concentration-dependent effective diffusivity as Def f = vsed lg,ef f (Fig. 1c, or- ange squares). We obtain Def f = D0 + αSNc, where D0 = 0.40 ± 0.01 µm2/s is the thermal diffusivity, the algal concentration Nc is in units of 106 cells/ml, and αS = 1.71± 0.14 (µm2/s)/(106 cells/ml) (slopes α will be expressed in these units throughout the paper). Within the same experiments, however, the diffusivity can also be inferred microscopically from direct short-duration tracking of microparticles' trajectories in the bulk, as pre- viously done in [25]. These measurements return a differ- Z(µm)050100150N(Z)/N010-210-1100101012Nc(106cells/ml)012lg,eff−l0(µm)020406080100120140160Nc(106cells/ml)0510⟨v⟩S⟨v⟩j(Deff−D0)(µm2.s−1)0246810121416a)b)c)3 FIG. 2. Microparticle behavior within the active suspension. a) Typical microparticle trajectory (∼ 210 s) in the quasi-2D experiment at Nc = 4.84 ± 0.13 × 106 cells/ml. Color represents instantaneous speed (colorbar unit: µm/s). The trajectory shows three types of dynamics: Brownian motion and loop-like perturbations (yellow-green blobs) followed by rare and large Inset: representative trajectory of a purely Brownian particle in the same setup, lasting ∼ 210 s. b) A jumps (red lines). representative entrainment event: as the cell swims from the left to the right of the panel, it drives the colloid along the dashed line. Scalebar: 20 µm. The event lasts 1.16 s. ent estimate, Def f = D0 +αT Nc (Fig. 1c, orange circles), with a slope αT = 0.074± 0.014 in reasonable agreement with previous results (αL = 0.041 in [25]) but more than 40-times smaller than the sedimentation value αS. The surprisingly large value of αS, never previously reported for any type of microswimmer, calls for an independent verification. It was tested here at the macroscopic level by following the diffusive spreading of a uniform band of density-matched colloids within a microfluidic device filled with a uniform concentration of cells (for experi- mental details see the Methods section and [36]). The band's profile, initially tight around the middle third of a 2 mm-wide, 60 µm-thick channel and running along its full length (∼ 10 mm), spreads with a characteristically diffusive dynamics [36] which enables to measure directly the effective diffusivity Def f at different Nc values. The results are shown in Fig. 1c (green triangles) after being multiplied by the ratio (cid:104)v(cid:105)S /(cid:104)v(cid:105)CS of the cells' swim- ming speeds to account for their slower motion in the density matched medium. As before, Def f depends lin- early on cell concentration, with a slope αCS = 1.62±0.14 in remarkable agreement with the sedimentation value. Microscopic entrainment and diffusion The quantitative agreement between steady-state and time-dependent macroscopic measurements suggests that our understanding of the microscopic interaction between particles and microorganisms, based on the effect of the swimmers' far field flows [27, 28, 37] and leading to αT ((cid:28) αS), is missing a key element. The crucial mi- croscopic insight is provided by long-time tracking of the particles, here individually followed for ∼ 200 s within a 26 µm-thick Hele-Shaw cell, at a range of Nc values. Fig. 2a and supplementary movie S1 [36] show a typi- cal colloidal trajectory in this quasi-2D geometry. The dynamics, which leads to a dramatically larger spread- ing than simple Brownian motion (Fig. 2a inset), results from the combination of three different effects of well- separated magnitudes. The weakest component is stan- dard Brownian motion, which dominates the dynamics when the algae are more than ∼ 25 µm [25] away from the colloid. At closer separation, but before close contact, the far-field flows of the microorganisms induce loop-like trajectories [18, 37 -- 40]. Originally reported in [25], these loops provide the concentration-dependent contribution to the particles' diffusivity previously estimated as αT Nc [25]. Finally, particles within the near field can be occa- sionally entrained by the algae over distances up to tens of microns (see Fig. 2b), giving rise to the sudden jumps in the trajectory highlighted as red solid lines in Fig. 2a. These jumps, a fundamental feature of the dynamics never observed previously, are the microscopic origin of the term αSNc, the unexpectedly large contribution to the particle diffusivity observed in the macroscopic exper- iments. For our 1 µm-diameter tracers, the jump length L is (mostly) exponentially distributed with a character- istic length LJ = 7.5 ± 0.5 µm, above a threshold value LT = 7.5 ± 0.5 µm (Fig. 3a). The average displacement is (cid:104)L(cid:105) = 12.6 ± 0.5 µm, although we did record values of L up to ∼ 70 µm. Jumps are fast (Fig. 3a, inset), lasting on average τ = 1.5 s (τ = 1.7s if considering only jumps with L ≥ LT ) but they are rare: as shown in Fig. 2b and the supporting high-speed movie S2 [36], the algae need to meet a microparticle almost head-on in order to X(µm)-2002040Y(µm)-40-30-20-100102030400.1110100a)b)4 FIG. 3. Microscopic characterization of particle dynamics. a) Probability distribution function (PDF) of the end-to-end length L of the jumps. Above LT = 7.5 µm, the distribution is well fitted by an exponential function with characteristic length LJ = 7.5 µm (solid red line). Inset: PDF of the duration τ of the jumps. The average is τ = 1.7 s when considering only jumps of length L ≥ LT (solid red line). b) Mean time interval between consecutive jumps, (cid:104)∆TJ(cid:105), as a function of Nc. The red solid line is the hyperbolic fit used in the simulations, (cid:104)∆TJ(cid:105) = ((68.2 ± 8)/Nc) s (Nc in units of 106 cells/ml). Inset: PDF of the time interval ∆TJ between consecutive jumps at Nc = (1.56 ± 0.10) × 106 cells/ml. Black solid line: exponential fit with characteristic time (31.9 ± 4) s. Note that these distributions provide a biased measure of the mean waiting time (cid:104)∆TJ(cid:105), which should be estimated instead from the average number of jump events as discussed in [36]. c) Effective diffusivities, Def f − D0, from the quasi-2D experiment (blue circles) and from the simulations: red circles/solid red line for τ = 1.7 s (slope 1.03× α2D); red diamonds/dashed red line for τ = 0.1 s (slope 1.22 × α2D). Inset: continuation of the simulated Def f − D0 curves to very high cells concentrations shows saturation to a τ -dependent value. entrain it (∼ 65% of all jumps have initial impact param- eter ≤ 2 µm). As a consequence, even at the highest cell concentration probed, Nc (cid:39) 5 × 106 cells/ml, the aver- age time between two consecutive jumps is rather long, (cid:104)∆TJ(cid:105) (cid:38) 16 s. Because (cid:104)∆TJ(cid:105) is dictated by the cell- microparticle encounter rate, for a purely random process one would expect (cid:104)∆TJ(cid:105) ∝ 1/Nc : Fig. 3b shows that this hypothesis is clearly supported experimentally. Further support comes from the analysis of individual inter-jump intervals, which appear to be exponentially distributed as predicted for a simple Poisson process (Fig. 3b inset). Computing the mean-square displacement from these long colloidal trajectories offers a microscopic estimate of the effective particle diffusivity Def f . The results are shown in Fig. 1c (blue circles) after multiplication by the swimming speeds ratio (cid:104)v(cid:105)S /(cid:104)v(cid:105)2D: Def f is linear with cell concentration, with a slope α2D = 1.67 ± 0.13 in excellent quantitative agreement with both macroscopic values. This agreement confirms that the properties of the colloids' microscopic dynamics, which includes im- portant but rare jumps, have indeed been probed appro- priately. Numerical simulations The agreement between the macroscopic and micro- scopic diffusivity measurements highlights the impor- tance of entrainment events, and suggests a combination of jumps and (far-field-enhanced) diffusion as a plausible minimal model of microparticles' dynamics, at least over the medium-to-long timescales we are interested in. Con- sequently, we model the stochastic trajectory of a colloid within a quasi-2D suspension of microalgae, (X(t), Y (t)), as (1) (cid:112) (cid:112) dX(t) = dY (t) = 2DW J dWX (t) + L cos(θ) dP (t) . 2DW J dWY (t) + L sin(θ) dP (t) This combines standard Wiener processes dWX,Y (t) lead- ing to diffusion with diffusivity DW J , with a Poisson process for the jumps, dP (t), characterised by the av- erage time interval (cid:104)∆TJ (Nc)(cid:105) between jumps (Fig. 3b, solid red line). Far-field effects are included at a coarse- grained level by choosing DW J = D0 + αW J Nc, where αW J = 0.20 ± 0.03. This is the effective diffusivity mea- sured from the microscopic experimental tracks when the entrainment events have been removed. Notice that αW J > αT due to our conservative choice for what con- stitutes a jump, which in the simulation we draw from an exponential fit to the part of the experimental distribu- tion strictly above LT (Fig. 3a, solid red line). As a re- sult, the compounded effect of shorter jumps is included within the coefficient αW J . The jumps' orientation θ is uniformly distributed to give an isotropic process, and their duration τ = 1.7 s is constant. This dynamics, simulated with a simple acceptance- rejection method, produces trajectories very similar to the experimental ones [36]. The motion is characterised by an effective diffusivity in excellent agreement with the L(µm)0204060PDF(L)00.010.020.030.040.050.060.07Nc(106cells/ml)024⟨∆TJ⟩(s)020406080100120140160Nc(106cells/ml)02468⟨v⟩S⟨v⟩2D(Deff−D0)(µm2.s−1)0510152002004000200400a)b)c)τ(s)02PDF(τ)00.020.040.060.08∆TJ(s)0200PDF(∆TJ)00.10.20.30.4values from the microscopic experiments (see Fig. 3c, red and blue circles respectively), and (cid:38) 5× larger than DW J at the corresponding cell concentration. The simulations, then, highlight the striking influence that jumps have on particle dynamics, despite their rarity. At the same time, they allow us to explore easily parameter values that have not yet been probed experimentally. For example, Fig. 3c (red diamonds and dashed red line) shows that within the experimental range of cell concentrations, even a drastic reduction of the entrainment duration τ to 0.1 s has only a minimal effect on particle diffusivity. The exact value of τ , however, will have a major influence as soon as (cid:104)∆TJ (Nc)(cid:105) ∼ τ , leading to a plateau in the effective dif- fusivity as the cell concentration grows above a threshold c (τ ) ∝ 1/τ (Fig. 3c inset), at least as long as collective N∗ effects do not modify our single-particle picture. Analytical theory The experimental and simulation results can be tied together further through a simple continuum theory for the dynamics of microparticles, which rationalises the de- pendence of the effective diffusivity Def f (Nc) on the con- centration of algae. Here we extend to two dimensions the one-dimensional approach developed in [42]. Briefly, we consider two populations whose densities at position (x, y) and time t are ρd(x, y, t) and ρb(x, y, t, φ), corre- sponding respectively to particles diffusing with diffu- sivity DW J , and to particles moving ballistically with a constant velocity u in the direction φ. Particles switch from diffusion to directed motion and vice versa with constant transition rates λd (= 1/(cid:104)∆TJ(cid:105)) and λb (= 1/τ ) respectively. The system then obeys the following set of equations: ∂ρd ∂t ∂ρb ∂t = DW J ∆ρd − λdρd + λb = −u cos(φ) − u sin(φ) ∂ρb ∂x ρbdφ 0 ∂ρb ∂y + λd 2π ρd − λbρb (2) which can be easily solved by Fourier-Laplace transform [36], yielding the time evolution of the particles' mean square displacement and hence their diffusivity. At long timescales this is given by (cid:90) 2π 5 in units of 106 cells/ml we have γ = 299±35 mm2. At low concentrations, where λd (cid:28) λb (or equally (cid:104)∆TJ(cid:105) (cid:29) τ ), Def f becomes Def f = D0 + αW J Nc + (cid:104)L(cid:105)2 2 γvNc, (4) which is independent of the jumps' duration τ , here equivalent to independence on u, as suggested by the simulations. Eq. (4) recovers the clear division between thermal, far-field and entrainment contributions to the diffusivity that we previously discussed in the context of microscopic experiments. Notice that the contribu- tion from the jumps, by far the most important in our experiments, is simply what should be expected if we interpreted the colloidal trajectory as a freely jointed chain where bonds with exponentially distributed length of mean (cid:104)L(cid:105) are added at a rate λd = γvNc. The coefficient αW J , representing far-field effects, is proportional to the speed v of the microalgae [25, 27, 38], leading to an overall contribution (Def f − D0) which scales with the active flux vNc. Fig. 1c shows indeed that the diffusivity curves from different experiments col- lapse when rescaled by the corresponding velocities. In turn, then, the distribution of jump lengths should be in- dependent of the average velocity of the microorganisms, as expected at low Reynolds numbers. Within the model, however, this proportionality is limited to sufficiently low cell concentrations. As Nc increases and (cid:104)∆TJ(cid:105) becomes closer to τ , nonlinearities become important, and even- tually Def f plateaus to a τ -dependent value as seen in the simulations (Fig. 3c, inset). Finally, the short timescale limit of the particles' mean square displacement returns Def f = xd DW J , where xd is the fraction of particles that are in the diffusing state [36]. Estimating xd (cid:39) (cid:104)∆TJ(cid:105)/((cid:104)∆TJ(cid:105) + τ ), we can as- sume xd (cid:39) 1 within the whole range of cell concentra- tions probed experimentally. Short timescale tracking of microparticles, then, will inevitably return Def f (cid:39) DW J rather than the full expression in Eq. (4). This is the reason for the small diffusivity reported in [25], which we also observe from direct short-duration tracking of mi- croparticles in the sedimentation experiments. λdu2 Discussion Def f = DW J λb λd + λb + 2λb(λd + λb) D0 + αW J Nc + = 1 + (cid:104)L(cid:105) u γvNc (cid:104)L(cid:105)2 2 γvNc (3) , where we wrote DW J = D0 + αW J Nc, λb = u/(cid:104)L(cid:105) and λd = γvNc as the frequency of entrainment events is pro- portional to the product of the speed v and concentration Nc of microorganisms, also called "active flux" [14]. For our quasi-2D experiments, v = (cid:104)v(cid:105)2D, and measuring Nc Proposed as a theoretical possibility [18, 38, 39], par- ticle entrainment by microorganisms has never been ob- served previously. Lack of experimental evidence ques- tioned its existence, as well as the importance within particle-microswimmer interactions. Here we have shown experimentally not only that entrainment of microparti- cles by microorganisms exists, but also that these rare but large events can dominate particle dynamics, leading in the present case to a diffusivity more than 40× larger than previously reported. Simulation and analytical re- sults support a jump-diffusion process as a good min- imal model for the medium-to-long timescales dynam- ics of the colloids. The simplified continuum model we discuss provides a theoretical support for the observed dependence of the experimental diffusivities on the so- called "active flux" vNc, already introduced in the bac- terial context [14, 15, 41]. At the same time, it clarifies that long-duration particle tracking is necessary to sam- ple correctly the microscopic dynamics and recover the real long timescale impact of microorganisms. Although we cannot yet pinpoint the specific mecha- nism leading to entrainment, the structure of the near- field flow is likely to play a crucial role. Entrainment requires almost head-on collisions, which is facilitated by the type of stagnation point found (on average) in front of the cell [28, 45]. After reaching the cell apex, the en- trained particles slide down the sides of the cell body along a high-shear region almost co-moving with the mi- croorganism, and are eventually left behind having spent slightly more than half of the jump in the front part of the cell (54± 9%). Eukaryotic microswimmers with multiple front-mounted flagella will share much of the near-field flow structure found in Chlamydomonas: entrainment is then likely to be a generic feature of this whole class of microorganisms. Several of these species are predators [2, 3], and prey on cells of size similar to our plastic parti- cles. Front-mounted flagella, then, would spontaneously lead to contact with the prey at a predictable location on the cell body within easy reach of the flagella, and therefore facilitate the ingestion of both natural preys and environmental microplastics. Methods Cell culture Cultures of CR strain CC125 were grown axenically in a Tris-Acetate-Phosphate medium [30] at 21◦ C un- der continuous fluorescent illumination (100 µE/m2s, OS- RAM Fluora). Cells were harvested at ∼ 5.106 cells/ml in the exponentially growing phase, then centrifuged at 800 rpm for 10 minutes and the supernatants replaced by DI-water (sedimentation experiment) or by a Per- coll solution (tracking and spreading experiments; Percoll Plus, Sigma) already containing the desired concentra- tion of PS colloids (Polybead R(cid:13) Microspheres, diameter d = 1 ± 0.02 µm). Microfluidics and Microscopy The Percoll solution (38.5% vol/vol) was made to density-match the beads for the quasi-2D and spread- ing experiments, while preserving the Newtonian nature 6 of the flows [31]. The appropriate solution was injected into either 185 µm (sedimentation experiment) or 26 µm (quasi-2D tracking experiment) thick PDMS-based mi- crofluidic channels previously passivated with 0.15% w/w BSA solution in water. The microfluidic devices were then sealed using photocurable glue (Norland NOA-68) to prevent evaporation. Regarding the spreading experi- ment, the band of colloids was initiated in a 3-arms fork- shape channel (60 µm thick) by injecting at the same flow-rate (using a PHD 2000 Harvard Apparatus syringe pump and high-precision Hamilton 50 µl and 100µl gas- tight syringes) the CR+beads Percoll solution in the cen- tral arm and the CR Percoll solution only in the two sided arms [36]. Colloids were observed under either bright-field (sedimentation experiment) or phase contrast (tracking and spreading experiments) illumination on a Nikon TE2000-U inverted microscope. A long-pass filter (cutoff wavelength 765 nm) was added to the optical path to prevent phototactic response of the cells. Stacks of 200 images at 60× magnification were acquired at 10 fps (camera Pike F-100B, AVT) layer by layer by manually moving the plane of focus in order to reconstruct the den- sity profiles for the sedimentation experiment. We used a 40× oil immersion objective (Nikon CFI S Fluor 40× oil) combined with an extra 1.5× optovar magnification. The condenser iris was completely opened to minimise the depth of field. Regarding the quasi-2D experiment, several movies of 2 × 104 images were recorded at 25 fps (camera Pike F-100B, AVT) using a 20× phase contrast objective (Nikon LWD ADL 20×F). Particles trajecto- ries were then digitised using a standard Matlab particle tracking algorithm [43]. Finally the spreading dynamics of the band of colloids was probed by recording the sys- tem for a few hours at 10× magnification (Nikon ADL 10xI) and low frame rate (0.5 fps) using a Nikon D5000 DSLR camera. The colloids were then featured using the same Matlab particle tracking algorithm [43] in order to reconstruct the density profiles. In all experiments, the concentration of algae was determined in situ by imaging the system under dark-field illumination at low magnifi- cation (objective Nikon CFI Plan Achromat UW 2×). Numerical Simulations As described previously, the parameters of the simu- lation correspond to the red lines in Fig. 3a,b. For each algae concentration, 1000 trajectories of 2000 s have been simulated using a time-step δt = 0.004 s, which is ten times smaller than the acquisition period in the quasi- 2D experiment. At each time step a random walk with diffusivity DW J is performed. To simulate the Poisson process, we draw at each time step a random number in the open interval ]0; 1[. If this number is within the cen- tered closed interval [(1−δt/(cid:104)∆TJ(cid:105))/2; (1+δt/(cid:104)∆TJ(cid:105))/2], then we perform a jump that lasts τ = 1.7 s (or 0.1 s) with a length taken out of the distribution P DF (L), along the direction corresponding to θ, which is uniformly dis- tributed in [−π, π[. This technique allows to simulate ac- curately the Poisson process [44]. We stress that within the simulation, once the particle has entered a jump, it will escape it after exactly τ = 1.7 s (or 0.1 s). This is done in order to approximate the experiments, which show that the distribution function of jumps' duration is tighter around the mean than that of jumps' lengths. No- tice that during the jumps the random walk component is switched off. After the jump has been performed, a new random number is picked in order to choose between a random step or a jump and continue the jump-diffusion process. Acknowledgement We gratefully acknowledge discussions with D. Pushkin in the initial stages of this project. Author contribution RJ, VK, MP designed the study; RJ performed the experiments, analytical and numerical modelling; RJ and MP analysed the results; RJ, VK, MP wrote the manuscript. ∗ [email protected] [1] Heddergott, N. et al. Trypanosome motion represents an adaptation to the crowded environment of the verterbrate bloodstreal. PLoS Pathog. 8, e1003023 (2012). [2] Montagnes, D. J. S. et al. Selective feeding behaviour of key free-living protists: avenues for continued study. Aquat. Microb. Ecol. 53, 83-98 (2008). [3] Wetherbee, R. and Andersen, R. Flagella of a chryso- phycean alga play an active role in prey capture and se- lection. Direct observations on Epiphysis pulchra using image enhanced video microscopy. Protoplasma. 166, 1- 7 (1992). [4] Katija, K. Biogenic inputs to ocean mixing. J. Exp. Biol. 215, 1040-1049 (2012). [5] Wright, S. L. Thompson, R. C. and Galloway, T. S. The physical impact of microplastics on marine organisms: a review. Env. Pol. 178, 483-492 (2013). [6] Mallory, S. A. Valeriani, C. and Cacciuto, A. Induced activation of a passive tracer in an active bath. Phys. Rev. E 90, 032309 (2014). [7] Kummel, F. Shabestari, P. Lozano, C. Volpe, G. and Bechinger, C. Formation, compression and surface melt- ing of colloidal clusters by active particles. Soft Matter 11, 6187 (2015). [8] Angelani, L. Maggi, C. Bernardini, M. L. Rizzo, A. and di Leonardo, R. Effective interactions between colloidal particles suspended in a bath of swimming cells. Phys. Rev. Lett. 107, 138302 (2011). 7 [9] Maggi, C. Paoluzzi, M. Pellicciotta, N. Lepore, A. Ange- lani, L. and di Leonardo, R. Generalized energy equipar- tition in harmonic oscillators driven by active baths. Phys. Rev. Lett. 113, 238303 (2014). [10] Koumakis, N. Maggi, C. and di Leonardo, R. Directed transport of active particles over asymmetric energy bar- riers. Soft Matter 10, 5695 (2014). [11] Takatori, S. C. and Brady, J. F. A theory for the phase behavior of mixtures of active particles. Soft Matter 11, 7920-7931 (2015). [12] Wu, X.-L. and Libchaber, A. Particle diffusion in a quasi- two-dimensional bacterial bath. Phys. Rev. Lett. 84, 13 (2000). [13] Valeriani, C. Li, M. Novosel, J. Arlt, J. and Marenduzzo, D. Colloids in a bacterial bath: simulations and experi- ments. Soft Matter 7, 5228 (2011). [14] Mino, G. L. et al. Enhanced diffusion due to active swim- mers at a solid surface. Phys. Rev. Lett. 106, 048102 (2011). [15] Jepson, A. Martinez, V. A. Schwarz-Linek, J. Morozov, A. and Poon, W. C. K. Enhanced diffusion of non swim- mers in a three-dimensional bath of motile bacteria. Phys. Rev. E 88, 041002(R) (2013). [16] Patteson, A. E. Gopinath, A. Purohit, P. K. and Arratia, P. E. Particle diffusion in active fluids is non-monotonic in size. ArXiv e-prints 1505.05803 (2015). [17] Underhill, P. T. Hernandez-Ortiz, J. P. and Graham, M. D. Diffusion and spatial correlations in suspensions of swimming particles. Phys. Rev. Lett. 100, 248101 (2008). [18] Morozov, A. and Marenduzzo, D. Enhanced diffusion of tracer particles in dilute bacterial suspensions. Soft Mat- ter 10, 2748 (2014). [19] Hinz, D. F. Panchenko, A. Kim, T.-Y. and Fried, E. Motility versus fluctuations in mixtures of self-motile and passive agents. Soft Matter 10, 9082-9089 (2014). [20] Peng, Y. Lai, L. Tai, Y.-S. Zhang, K. Xu, X. and Cheng, X. Diffusion of an ellipsoid in bacterial suspen- sions. ArXiv e-prints 1509.05893 (2015). [21] Angelani, L. and di Leonardo, R. Geometrically biased random walks in bacteria-driven micro-shuttles. New J. Phys. 12, 113017 (2010). [22] Kaiser, A. Peshkov, A. Sokolov, A. ten Hagen, B. Lowen, H. and Aranson, I. S. Transport powered by bacterial turbulence. Phys. Rev. Lett. 112, 158101 (2014). [23] Angelani, L. di Leonardo, R. and Ruocco, G. Self-starting micromotros in a bacterial bath. Phys. Rev. Lett. 102, 048104 (2009). [24] di Leonardo, R. et al. Bacterial ratchet motors. Proc. Nat. Acad. Sci. 107, 9541-9545 (2010). [25] Leptos, K. C. Guasto, J. S. Gollub, J. P. Pesci, A. I. and Goldstein, R. E. Dynamics of enhanced tracer diffusion in suspensions of swimming eukaryotic microorganisms. Phys. Rev. Lett. 103, 198103 (2009). [26] Kurtuldu, H. Guasto, J. S. Johnson, K. A. and Gollub, J. P. Enhancement of biomixing by swimming algal cells in two-dimensional films. Proc. Nat. Acad. Sci. 108, 10391- 10395 (2011). [27] Thiffeault, J.-L. Distribution of particle displacements due to swimming microorganisms. Phys. Rev. E 92, 023023 (2015). [28] Drescher, K. Goldstein, R. E. Michel, N. Polin, M. and Tuval, I. Direct measurement of the flow field around swimming microorganisms. Phys. Rev. Lett. 105, 168101 (2010). [29] Das, A. Polley, A. and Rao, M. Phase segregation of pas- sive advective particles in an active medium. ArXiv e- prints 1506.06928 (2015). [30] Rochaix, J. D. Mayfield, S. Goldschmidt-Clermont, M. and Erickson, J. M. in Plant Molecular Biology: A Prac- tical Approach, edited by C. H. Schaw (IRL Press, Ox- ford, England, 1988), pp. 253-275. [31] Gachelin, J. Mino, G. Berthet, H. Lindner, A. Rousse- let, A. and Clement, E. Non-Newtonian viscosity of Es- cherichia coli suspensions. Phys. Rev. Lett. 110, 268103 (2013). [32] Perrin, J. Molecular reality. Ann. Chim. Phys. 8, 1 (1909). [33] Tailleur, J. and Cates, M. E. Sedimentation, trapping, and rectification of dilute bacteria. Eur. Phys. Lett. 86, 60002 (2009). [34] Palacci, J. Cottin-Bizonne, C. Ybert, C. and Bocquet, L. Sedimentation and effective temperature of active col- loidal suspensions. Phys. Rev. Lett. 105, 088304 (2010). [35] Rafaı , S. Jibuti, L. and Peyla, P. Effective viscosity of microswimmer suspensions. Phys. Rev. Lett. 104, 098102 (2010). [36] See [URL to be inserted by publisher] for Supplementary Material and Supplementary Movies. 8 [37] Dunkel, J. Putz, V. B. Zaid, I. M. and Yeomans, J. M. Swimmer-tracer scattering at low Reynolds number. Soft Matter 6, 4268-4276 (2010). [38] Lin, Z. Thiffeault, J.-L. and Childress, S. Stirring by squirmers. J. Fluid Mech. 669, 167-177 (2011). [39] Pushkin, D. O. and Yeomans, J. M. Stirring by swimmers in confined microenvironments. J. Stat. Mech. 4, P04030 (2014). [40] Mathjissen, A. J. T. M. Pushkin, D. O. and Yeomans, J. M. Tracer trajectories and displacement due to a micro- swimmer near a surface. J. Fluid Mech. 773, 498-519 (2015). [41] Mino, G. L. Dunstan, J. Rousselet, A. Clement, E. and Soto, R. Induced diffusion of tracers in a bacterial sus- pension: theory and experiments. J. Fluid Mech. 729, 423-444 (2013). [42] M´endez, V. Campos, D. and Bartumeus, F. Stochastic Foundations in Movement Ecology (Springer Complexity, 2014), Ch. 3.3. code downloaded http://people.umass.edu/kilfoil/downloads.html. [43] The can be at [44] Hanson, F. B. Applied Stochastic Processes and Control for Jump-Diffusions: Modeling, Analysis, and Computa- tion (SIAM, 2007), Ch. 1. [45] Guasto, J.S. Johnson, K. A. and Gollub, J. P. Oscilla- tory flows induced by microorganisms swimming in two dimensions, Phys. Rev. Lett. 105, 168102 (2010).
1001.1109	1	1001	2010-01-07T16:28:17	3D Corporate Tourism: A Concept for Innovation in Nanomaterials Engineering	[ "physics.bio-ph", "physics.soc-ph" ]	Nature's materials are complex, multifunctional, hierarchical and responsive and in most instances functionality on the nanoscale is combined with performance on the macroscale. Materials engineers have just started to produce complex nanomaterials. Biomimicry and biomimetics deal with knowledge transfer from nature to technology. Inspired by the 'Biomimicry and Design Workshops' and the 'Biomimicry Innovation Method' by the US based Biomimicry Guild, '3D Corporate Tourism', a solution based approach to innovation in nanomaterials research, is proposed. The three main pillars of this integrated concept are discover, develop and design. Biologists, research and development engineers as well as designers jointly work in an environment with high inspirational potential and construct first prototypes and designs on site. This joint approach yields new links, networks and collaborations between communities of thinkers in different countries in order to stimulate and enhance creative and application oriented problem solving for society.	physics.bio-ph	physics	3D Corporate Tourism: A Concept for Innovation in Nanomaterials Engineering Ille C Gebeshuber* Institute of Microengineering and Nanoelectronics (IMEN), Universiti Kebangsaan Malaysia, 43600 UKM, Bangi, Selangor, Malaysia and Institute of Applied Physics, Vienna University of Technology, Wiedner Hauptstrasse 8–10/134, 1040 Vienna, Austria E–mail: [email protected], [email protected] FAX: +60 3 8925 0439 *Corresponding author Burhanuddin Y. Majlis Institute of Microengineering and Nanoelectronics (IMEN), Universiti Kebangsaan Malaysia, 43600 UKM, Bangi, Selangor, Malaysia FAX: +60 3 8925 0439 E–mail: [email protected] Abstract Nature´s materials are complex, multi–functional, hierarchical and responsive and in most instances functionality on the nanoscale is combined with performance on the macroscale. Materials engineers have just started to produce complex nanomaterials. Biomimicry and biomimetics deal with knowledge transfer from nature to technology. Inspired by the “Biomimicry and Design Workshops” by the US based Biomimicry Guild, “3D Corporate Tourism”, a solution–based approach to innovation in nanomaterials research, is proposed. The three main pillars of this integrated concept are discover, develop and design. Biologists, research and development engineers as well as designers jointly work in an environment with high inspirational potential and construct first prototypes and designs on site. This joint approach yields new links, networks and collaborations between communities of thinkers in different countries in order to stimulate and enhance creative and application–oriented problem solving for society. Keywords bioinspiration, biomimetics, production methods, natural materials, biological engineering, biomimicry, innovation, nanomaterials, nanotechnology, nanofabrication, design, corporate tourism Biographical Notes Ille C. Gebeshuber is expert in Nanotechnology and Biomimetics. Since 2009 she has been Contract Professor at the Institute of Microengineering and Nanoelectronics (IMEN), Universiti Kebangsaan Malaysia (UKM). Her permanent professorship affiliation is with the Vienna University of Technology (Austria) where she is habilitated for experimental physics. She is cofounder of the Vienna–based Center of Excellence for Biomimetics. Prof. Gebeshuber is Associate Editor of the UK based Journal of Mechnical Engineering Science (IMechE), Editorial Board Member of various scientific journals and editor of a book on biomimetics by Springer Scientific Publishing. She is highly active in Science Outreach and Board Member of UKM s Permata Pintar programm, identifying and promoting Malaysian geniuses at early age. Burhanuddin Y. Majlis is professor of microelectronics at Universiti Kebangsaan Malaysia. He is senior member of the Institution of Electrical and Electronics Engineer (IEEE) and was Chairman of IEEE Electron Devices Malaysia Chapter from 1994 to 2006. He is the founder chaiman of Malaysia Nanotechnology Association that was established in 2007. He initiated research in microfabrication and microsensors at UKM in 1995 and has also initiated research in GaAs technology with Telekom Malaysia. In 2001 he stared research in MEMS with substantial research funding of US$10 million from the Malaysian Ministry of Science, Technology and Innovation. His current interests are design and fabrication of MEMS sensors, RFMEMS, BioMEMS and microenergy. He has published four text books in electronics and one book on Integrated Circuits Fabrication Technology for undergraduate courses and more than 300 academic research papers. Now he is the founder director of IMEN. Acknowledgements The Austrian Society for the Advancement of Plant Sciences funded part of this work via the Biomimetics Pilot Project “BioScreen”. Living in the tropics and exposure to high species diversity at frequent excursions to the tropical rainforests is a highly inspirational way to do biomimetics. Profs. F. Aumayr, H. Störi and G. Badurek from the Vienna University of Technology are acknowledged for enabling ICG three years of research in the inspiring environment in Malaysia. 3D Corporate Tourism: A Concept for Innovation in Nanomaterials Engineering Abstract Nature´s materials are complex, multi–functional, hierarchical and responsive and in most instances functionality on the nanoscale is combined with performance on the macroscale. Materials engineers have just started to produce complex nanomaterials. Biomimicry and biomimetics deal with knowledge transfer from nature to technology. Inspired by the “Biomimicry and Design Workshops” and the “Biomimicry Innovation Method” by the US based Biomimicry Guild, “3D Corporate Tourism”, a solution–based approach to innovation in nanomaterials research, is proposed. The three main pillars of this integrated concept are discover, develop and design. Biologists, research and development engineers as well as designers jointly work in an environment with high inspirational potential and construct first prototypes and designs on site. This joint approach yields new links, networks and collaborations between communities of thinkers in different countries in order to stimulate and enhance creative and application–oriented problem solving for society. Keywords bioinspiration, biomimetics, production methods, natural materials, biological engineering, biomimicry, innovation, nanomaterials, nanotechnology, nanofabrication, design, corporate tourism 1. Introduction Natural materials are complex on various levels of hierarchy, from the nanometer length scale via the microscale to the macroscale (Aizenberg et al., 2005; Fratzl and Weinkamer, 2007). Materials engineers have just started to produce complex nanomaterials (Lao, Wen, and Ren, 2002; Cao, 2007; Tao et al., 2007), and there is still a long way to go to reach the natural “best practice” examples in terms of precision, functionality and efficiency of production (Ryadnov, 2009). Due to nanotechnology´s increasingly, transdisciplinary nature, inter– and inherently collaborations across fields prove successful (Gebeshuber, 2007; Gebeshuber and Drack, 2008; Porter and Youtie, 2009; Gebeshuber et al., 2010). Such inter– and transdisciplinary approaches are highly useful for innovation. For companies it is not interesting if the material is a nanomaterial or not, what is interesting for them is that the material performs as intended and can be produced at reasonable costs. Therefore, a solution–based approach to innovation in materials research seems to be more rewarding and beneficial, for the companies as well as for the scientists and engineers who develop and produce such materials on the prototype level. Bioinspirations seems to be especially rewarding for innovation regarding nanomaterials, since most of the materials in nature are hierarchical, with important functionalities on the nanoscale. One example for this is recent research on biosynthesis of nanoparticles: Various microbes such as bacteria, yeast and fungi produce inorganic nanostructures and metallic nanoparticles with properties similar to chemically–synthesised materials, while exercising strict control over size, shape and composition of the particles (see e.g., Sastry et al. 2003; Mandal et al. 2006; Gericke and Pinches 2006a; Gericke and Pinches 2006b). Examples include the formation of magnetic nanoparticles by magnetotactic bacteria (Roh et al., 2001), the production of silver nanoparticles by the bacterium Pseudomonas stutzeri (Klaus et al., 1999) and the fungi Fusarium oxysporum (Senapati et al. 2004) and Verticillium sp. (Mohanpuria, Rana and Yadav, 2008), synthesis of nano–scale, semi–conducting CdS crystals in the yeast Schizosaccharomyces pombe (Kowshik et al., 2003), the synthesis of gold nanoparticles (Gericke and Pinches, 2006b) and the formation of palladium nanoparticles using sulphate reducing bacteria in the presence of an exogenous electron donor (Yong et al., 2002). Various “best practices” for novel nanomaterials can be found in nature. Abstraction and understanding of their design principles is the prerequisite for successful transfer to technology. Nature´s materials are like a treasure box for the nanomaterials engineer. There are functional materials, hierarchical materials, functional gradient materials, responsive materials, materials with an expiration date (i.e., they only work for a certain amount of time) and biodegradable materials (Meyers et al., 2007). One intriguing example for the exqusite materials engineering in nature is biomineralisation. More than 60 different biominerals such as are produced by organisms, including carbonates, phosphates, arsenates, chlorides and silfides (Weiner and Dove, 2003; Sigel, Sigel and Sigel, 2008; Behrens and Baeuerlein, 2009). A biomineral consists of an inorganic part (about 97 vol%), and about 3 vol% of organic matrix. The way of production of biominerals varies substantially from current materials synthesis: In the production of biominerals, properties such as shape, size and crystallinity as well as isotopic and trace element compositions are controlled by the organic matrix (Weiner and Dove, 2003). Figure 1 shows an example for biomineralised SiO2 in the diatom Ellerbeckia arenaria. Diatoms are unicellular microalgae with a cell wall consisting of a siliceous skeleton enveloped by a thin organic case (Round, Crawford and Mann, 1990). The cell walls of each diatom form a pillbox-like shell consisting of two parts that fit within each other. These microorganisms vary greatly in shape, ranging from box- shaped to cylindrical; they can be symmetrical as well as asymmetrical and exhibit an amazing diversity of nanostructured frameworks. These biogenic hydrated silica structures have elaborate shapes, interlocking devices, and, in some cases, hinged structures (Gebeshuber and Crawford, 2006), making them interesting for emerging micro- and nanoelectromechanical systems (MEMS and NEMS, Gebeshuber et al., 2009). Figure 1. Biomineralised amorphous silica from the diatom Ellerbeckia arenaria. Scalebar: 1µm. This article aims at mapping new frontiers in emerging and developing technology areas in nanomaterials research and innovation. A novel way to foster and promote innovative thinking in the sciences is introduced, considering the need for synergy and collaboration between biology, engineering, materials science and nanotechnology rather than segmentation and isolation: Supported by specially trained biologists, research and development engineers as well as designers apply the Biomimicry Innovation Method (© Biomimicry Guild, Helena, MT, USA 2008; Gebeshuber et al., 2009) in an environment with high inspirational potential and discover, develop and design complex nanomaterials inspired by nature. Directly at the site of this research, first prototypes and designs are constructed, and first detailed investigations take place. The three main pillars of this approach are discover, develop and design – the integrated concept is therefore termed “3D Corporate Tourism” (for logo, see Figure 2, for basic concept see Figure 3). This concept has been inspired by the “Biomimicry and Design Workshops” (offered for one week per year, location: rainforest in Peru or in Costa Rica) by the US based Biomimicry Guild. Companies such as Boeing, Colgate–Palmolive, General Electric, Levi’s, NASA, Nike and Procter and Gamble have already used their services. Janine Benyus, the founder of the Biomicry Guild, states: “When we take designers and engineers and architects on these workshops – these are people who everyday are inventing. They usually get their inspiration by looking at other human inventions. We got them outside, where they were surrounded in Costa Rica by the most amazingly sustainable system that one can imagine, and so much variety. Each of those organisms was solving amazing challenges, but they were solving them in very, very different ways. ….. There was somebody there who was working on packaging, and for them it was a revelation because here’s this packaging, which was the exoskeleton of the beetle. It breathes, it signals, it creates its beautiful color without toxic chemicals, it’s waterproof, it’s manufactured in a completely nontoxic way, and it’s abrasion–resistant. All the things they would love to have in packaging, and yet it’s made of one material that is completely recyclable.” (Benyus, 2009). Figure 2. Logo of “3D Corporate Tourism”. With “3D Corporate Tourism”, the successful concept of the “Biomimicry and Design Workshops” is developed further into a complete niche tourism concept. The outcome of such a joint effort are – besides the research results, developments and designs – new links, networks and collaborations between communities of thinkers in different countries in order to stimulate and enhance creative and application–oriented problem solving for society. 2. Materials and Methods 2.1. Method Research and development engineers as well as designers from companies work at “3D Corporate Tourism” headquarters on the three pillars of the concept: discover, develop, design. The Biomimicry Innovation Method (BIM, © Biomimicry Guild, Helena, MT, USA 2008; Gebeshuber et al., 2009) is applied to identify high–potential biological systems, processes and materials that shall inspire novel outcomes. BIM is an innovation method that seeks sustainable solutions by emulating Nature's time– tested patterns and strategies. The goal is to create products, processes, and policies – new ways of living – that are well adapted to life on earth over the long haul. The steps in BIM are as follows: Identify function, biologise the question, find nature’s best practices and generate product ideas. Identify function: The biologists distil challenges posed by engineers/natural scientists/architects and/or designers to their functional essence. In the case of nanomaterial development and innovation, self–repairing structures, tough and yet flexible materials, photoactive materials such as luminescent fungi, and the natural recycling system that is at work in the rainforest (the waste of one species is nurturing another species, a concept that has as “Waste to Wealth” already entered current science and economy (Kathiravale and Muhd Yunus, 2008)). Biologise the question: In the next step, these functions are translated into biological questions such as “How does Nature provide strong and yet flexible materials?” or “How does Nature convert energy?” or “How does Nature self–repair structures?” The basic question is “What would Nature do here?” Find Nature’s best practices: Screens of the relevant literature in scientific databases as well as entering a highly inspiring environment with the biologised questions in mind (task–oriented visit) are used to obtain a compendium of how plants, animals and ecosystems solve the challenges in question. The inspiring environments should preferably be habitats with high species diversity, e.g., rain forests or coral reefs. Thereby a compendium of how plants, animals and ecosystems solve the specific challenge is obtained. Generate process/product ideas: From these best practices (according to the Biomimicry Guild, 90% of them are usually new to clients) ideas for cost effective, innovative, life–friendly and sustainable products and processes are generated. First designs and investigations are already turned into prototypes right at the site. In general the projects are structured in three phases (discover, develop and design, Figures 3 and 4). 2.1.1. Discovery Phase The discovery phase aims to define the key elements that are needed for a solution and to approach nature for inspiration, here the support of experts will be important as the way to look at nature and to isolate segments with solution potential has to be acquired. Here nature will be observed with a ‘function view’ (What works like we need it?). 2.1.2. Development Phase The development phase is probably the most important phase and comprises all activities that are undertaken to convert the solution of nature into human technology. Here top–engineering and prototyping support will be crucial. Nature will be screened if needed with ‘process view’ (How is the solution achieved in detail? Which process steps are there?). Figure 3. The three pillars of the “3D Corporate Tourism” concept, starting from the product challenge and resulting in the final product ideas. To achieve the best possible results it is suggested to insert a normal holiday (one or two weeks with family in a Rainforest Resort) between the development and the design phase; that way the team members can relax and rethink the project results before finalizing the design. This period is also a kind of time reserve in case the prototyping cannot be finished in time, which is very likely the case for more complicated designs. 2.1.3. Design Phase The design phase focuses on the optimization and final design of the defined solution. The final design will be further improved by a second inspiration loop with nature (‘design view’: Why do the components of nature work together this well? Where is the detail optimization?). Here experts and designers will help the team to prepare a top–quality presentation for their corporate headquarters. 2.2. Project Team The project team consists of the visiting engineers, developers and designers from the respective companies, and specifically trained biologists and engineers trained in biomimetics from the host country. The visitors either arrive with a specific problem they would like to deal with, or the theme(s) of the project are determined on site, after initial introduction to the respective approaches, joint discussions and brainstorming sessions, and inspirational task–oriented visits to the rainforest. There is also the possibility of MSc and PhD students to join the projects and contribute their knowledge and ideas. Information flow is free between the members of the individual projects, and intellectual property rights are secured by standard documents that are signed at the beginning of the project. 2.3. Environment The “3D Corporate Tourism” headquarters are small lodges located directly inside the rainforest or close to coral reefs. They have sufficient standard for the clientele from the companies and are equipped with amenities such as CAD machines, state of the art communication facilities, scientific and patent databases and prototyping machinery. This distinguishes them from the currently available rainforest biology field stations. The lodges are utilised by one project team at one time, and joint social activities shall foster discussion in the free time. Figure 4. Details of the three main pillars of the “3D Corporate Tourism” concept. 3. Results and Discussion The rainforest serves as nanotechnology lab, and combined with the “3D Corporate Tourism” headquarters provides a complete solution for work from the concept phase to first prototypes (Figure 5). In this way, collaborations between local scientists and engineers with the international clientele are initiated, bidirectional knowledge transfer takes place and new knowledge is generated. The high species variety in the rainforest, with nature´s “best practices” everywhere aids to relate structure with function in natural materials, structures and processes and helps to increase awareness about the natural resources surrounding us. With the concept of “3D Corporate Tourism” the potential of the virgin rainforests is used in a sustainable way, without exploiting the natural resources or removing anything else from the jungle apart from ideas. In this way, the value of the virgin forests is increasing in the minds of policy makers and threshold countries have the opportunity to contribute highly valued inputs to the international research and development elite, as well as train their local experts in very important future technologies. The possibility to perform first investigations directly on–site, and subsequent deeper and more detailed investigations at the home institution fosters collaborations and results in synergistic effects across borders. 4. Conclusion The “3D Tourism” concept seems to be especially interesting for development of novel functional nanomaterials and contributes to overcoming the first two of the three gaps between the world of ideas, inventors, innovators and investors as well as the market as introduced by Gebeshuber, Gruber and Drack, 2009 for accelerated scientific and technological breakthroughs to improve the human condition (Figure 5). Figure 5. The Three–Gaps–Theory as proposed by Gebeshuber, Gruber and Drack, 2009. Image © 2009 Professional Engineering Publishing, UK. Image reproduced with permission. References Aizenberg, J., Weaver, J.C., Thanawala, M.S., Sundar, V.C., Morse, D.E. and Fratzl, P. (2005) ´Skeleton of Euplectella sp.: Structural hierarchy from the nanoscale to the macroscale´, Science, Vol. 309, pp.275–278. Behrens, P. and Baeuerlein, E. (2009) Handbook of biomineralization: Biomimetic and bioinspired chemistry, Weinheim, Germany: Wiley–VCH. Benyus, J. (2009). Living Green, Episode 26: Biomimicry: Janine Benyus is Honored by TIME. Obtained through the Internet: http://personallifemedia.com/podcasts/224– living–green/episodes/3351–biomimicry–janine–benyus–honored, [accessed 20/12/2009]. Cao, H. (2007) ´Constructing advanced nanostructures: a key step towards nanodevices´, In: Sabatini, D.M. (ed.), Leading Edge Nanotechnology Research Developments, (pp.1–3), New York: Nova Science Publishers. Esichaikul, R., Macqueen, M. and Gebeshuber I.C. (2010) ´3D Corporate Tourism: Application–oriented problem solving in tropical rainforests´. Paper presented at the 8th Asia–Pacific CHRIE Conference. June 3–6, 2010. Bangkok, Thailand. Fratzl, P. and Weinkamer, R. (2007) ´Nature’s hierarchical materials´, Progress in Materials Science, Vol. 52, No. 8, pp.1263–1334. Gebeshuber, I.C. (2007) ´Biotribology inspires new technologies´, Nano Today, Vol. 2, No. 5, pp.30–37. Gebeshuber, I.C. and Drack, M. (2008) ´An attempt to reveal synergies between biology and engineering mechanics´, Proc. IMechE Part C: J. Mech. Eng. Sci., Vol. 222, No. 7, pp.1281–1287. Gebeshuber, I.C., Aumayr, M., Hekele, O., Sommer, R., Goesselsberger, C.G., Gruenberger, C., Gruber, P., Borowan, E., Rosic, A. and Aumayr, F. (2010) ´Bacilli, green algae, diatoms and red blood cells – how nanobiotechnological research inspires architecture´, In: Zhou, Y. (ed.), Bio–Inspired Nanomaterials and Nanotechnology, (Chapter IX), New York, USA: Nova Science Publishers. https://www.novapublishers.com/catalog/product_info.php?products_id=10764 [accessed 20/12/2009]. Gebeshuber, I.C., Gruber, P. and Drack, M. (2009) ´A gaze into the crystal ball – biomimetics in the year 2059´, Proceedings of the Institution of Mechanical Engineers Part C: Journal of Mechanical Engineering Science, Vol. 223, No. 12, pp.2899–2918. Gebeshuber, I.C., Stachelberger, H., Ganji, B.A., Fu, D.C., Yunas, J. and Majlis, B.Y. (2009) ´Exploring the innovational potential of biomimetics for novel 3D MEMS, Advanced Materials Research, Vol. 74, pp.265–268. Gericke, M. and Pinches, A. (2006b) `Microbial production of gold nanoparticles´, Gold Bulletin, Vol. 39, No. 1, pp.22–28. Gericke, M. and Pinchesa, A. (2006a) ´Biological synthesis of metal nanoparticles´, 16th International Biohydrometallurgy Symposium, Hydrometallurgy, Vol. 83, No. 1–4, pp.132–140. Kathiravale, S. and Muhd Yunus, M.N. (2008) ´Waste to wealth´, Asia Europe Journal, Vol. 6, No. 2, pp. 359–371. Klaus, T., Joerger, R., Olsson, E. and Granqvist, C–G. (1999) ´Silver–based crystalline nanoparticles, microbially fabricated´, Proceedings of the National Academy of Sciences USA, Vol. 96, No. 24, pp. 13611–13614. Kowshik M., Deshmukh N., Vogel W., Urban J., Kulkarni S.K., and Paknikar K.M. (2003) their semiconductor CdS nanoparticles, synthesis of ´Microbial characterization, and their use in the fabrication of an ideal diode´, Biotechnology and Bioengineering, Vol. 78, pp.583–588. Lao, J.Y., Wen, J.G. and Ren Z.F. (2002) ´Hierarchical ZnO nanostructures´, Nano Letters, Vol. 2, No. 11, pp.1287–1291. Mandal, D., Bolander, M.E., Mukhopadhyay, D., Sarkar, G. and Mukherjee, P. (2006) ´The use of microorganisms for the formation of metal nanoparticles and their application´, Applied Microbiology and Biotechnology, Vol. 69, No. 5, pp. 485–92. Meyers, M.A., Chena, P.–Y., Lina, A. Y.–M. and Sekia, Y. (2008) ´Biological materials: Structure and mechanical properties´, Progress in Materials Science, Vol. 53, No. 1, pp.1–206. Mohanpuria, P., Rana, N.K. and Yadav, S.K. (2008) ´Biosynthesis of nanoparticles: technological concepts and future applications´, Journal of Nanoparticle Research, Vol. 10, No. 3, pp.507–517. Porter, A.L. and Youtie, J. (2009) ´How interdisciplinary is nanotechnology?´, Journal of Nanoparticle Research, Vol. 11, No. 5, pp.1023–1041. Roh, Y., Lauf, R.J., McMillan, A.D., Zhang, C., Rawn, C.J., Bai, J., and Phelps, T.J. (2001) ´Microbial synthesis and the characterization of some metal–doped magnetite´, Solid State Communications, Vol. 118, No. 10, pp.529–534. Round, F.E., Crawford, R.M. and Mann, D.G. (1990) The diatoms: biology and morphology of the genera, Cambridge, UK: Cambridge University Press. Ryadnov, M. (2009) Bionanodesign: Following natures touch (RSC Nanoscience and Nanotechnology), Cambridge, UK: Royal Society of Chemistry Publications Sastry, M., Ahmad, A., Islam, N.I., Kumar, R. (2003) ´Biosynthesis of metal nanoparticles using fungi and actinomycete´, Current Science, Vol. 85, pp.162–170. Senapati, S., Mandal, D., Ahmad, A., Khan, M.I., Sastry, M. and Kumar, R. (2004) ´Fungus mediated synthesis of silver nanoparticles: a novel biological approach´, Indian Journal of Physics and Proceedings of the Indian Association for the Cultivation of Science A, Vol. 78, No. 1, pp. 101–105. Sigel, A., Sigel, H. and Sigel, R.K.O. (eds.) (2008) Biomineralization: From nature to application (Metal Ions in Life Sciences), Vol. 2, Chichester, UK: Wiley. Tao, Y., Zohar, H., Olsen, B.D. and Segalman, R.A. (2007) ´Hierarchical nanostructure control in rod−coil block copolymers with magnetic fields´, Nano Letters, Vol. 7, No. 9, pp.2742–2746. Weiner, S. and Dove, P.M. (2003) ´An overview of biomineralization processes and the problem of the vital effect´, Reviews in Mineralogy and Geochemistry, Vol. 54, No. 1, pp.1–29. Yong, P., Rowsen, N.A., Farr, J.P.G., Harris, I.R. and Macaskie, L.E. (2002) ´Bioreduction and biocrystallization of palladium by Desulfovibrio desulfuricans NCIMB 8307´, Biotechnology and Bioengineering, Vol. 80, No. 4, pp.369–379.
1612.07645	1	1612	2016-12-22T15:28:22	First free-flight flow visualisation of a flapping-wing robot	[ "physics.bio-ph", "cs.RO" ]	Flow visualisations are essential to better understand the unsteady aerodynamics of flapping wing flight. The issues inherent to animal experiments, such as poor controllability and unnatural flapping when tethered, can be avoided by using robotic flyers. Such an approach holds a promise for a more systematic and repeatable methodology for flow visualisation, through a better controlled flight. Such experiments require high precision position control, however, and until now this was not possible due to the challenging flight dynamics and payload restrictions of flapping wing Micro Air Vehicles (FWMAV). Here, we present a new FWMAV-specific control approach that, by employing an external motion tracking system, achieved autonomous wind tunnel flight with a maximum root-mean-square position error of 28 mm at low speeds (0.8 - 1.2 m/s) and 75 mm at high speeds (2 - 2.4 m/s). This allowed the first free-flight flow visualisation experiments to be conducted with an FWMAV. Time-resolved stereoscopic Particle Image Velocimetry (PIV) was used to reconstruct the 3D flow patterns of the FWMAV wake. A good qualitative match was found in comparison to a tethered configuration at similar conditions, suggesting that the obtained free-flight measurements are reliable and meaningful.	physics.bio-ph	physics	First free-flight flow visualisation of a flapping-wing robot Matej Kar´asek1, Mustafa Percin1, Torbjørn Cunis2, Bas W. van Oudheusden1, Christophe De Wagter1, Bart D.W. Remes1 and Guido C.H.E. de Croon1 1Faculty of Aerospace Engineering, Delft University of Technology, Kluyverweg 1, 2629 HS Delft, Netherlands 2Department of Systems Control and Flight Dynamics, ONERA - The French Aerospace Lab, Centre Midi-Pyr´en´ees, 31055 Toulouse, France E-mail: [email protected], [email protected], [email protected], [email protected] Abstract Flow visualisations are essential to better understand the unsteady aerodynamics of flapping wing flight. The issues inherent to animal exper- iments, such as poor controllability and unnatural flapping when tethered, can be avoided by using robotic flyers. Such an approach holds a promise for a more systematic and repeatable methodology for flow visualisation, through a better controlled flight. Such experiments require high preci- sion position control, however, and until now this was not possible due to the challenging flight dynamics and payload restrictions of flapping wing Micro Air Vehicles (FWMAV). Here, we present a new FWMAV- specific control approach that, by employing an external motion tracking system, achieved autonomous wind tunnel flight with a maximum root- mean-square position error of 28 mm at low speeds (0.8 - 1.2 m/s) and 75 mm at high speeds (2 - 2.4 m/s). This allowed the first free-flight flow vi- sualisation experiments to be conducted with an FWMAV. Time-resolved stereoscopic Particle Image Velocimetry (PIV) was used to reconstruct the 3D flow patterns of the FWMAV wake. A good qualitative match was found in comparison to a tethered configuration at similar condi- tions, suggesting that the obtained free-flight measurements are reliable and meaningful. Keywords: Flapping wing, PIV, free flight, flapping flight, Micro Air Vehicles, control 1 1 Introduction Flapping flight, the only form of powered aerial locomotion in nature, involves unsteady aerodynamic phenomena that remain to be fully understood, especially at small scales and low Reynolds numbers. Such understanding would be of great benefit in the development of flapping-wing Micro Air Vehicles (FWMAVs); the performance of the current designs [1, 2, 3, 4] remains far inferior compared to the extreme manoeuvrability, agility and flight efficiency of their biological counterparts [5, 6, 7, 8]. Despite an intense research in the fluid dynamics modelling techniques over the past decades, reliable and accurate models applicable to an arbitrary flap- [9, 10, 11, 12], can ping wing are missing. Simpler, quasi-steady models, e.g. successfully predict the general trends of the sub-flap forces, provided that their force coefficients are based on empirical data. Some studies capture the ge- ometry of a deformable flapping wing during flapping, which is used as input for numerical fluid dynamics simulations [13, 14, 15]. While these models do provide some insight into the flow details, in most cases they still cannot pre- dict the aerodynamic forces to a sufficient level of accuracy, as comparison to force-balance measurements reveals [16]. A proper numerical treatment requires coupling of models of fluid dynamics with structural dynamics of the wing in or- der to capture the wing deformations under the load of aerodynamic forces [17]. Such models require a high computational effort while a further challenge can be an accurate identification of the structural parameters of the true wing. Thus, so far, reliable flow field data has been obtained by experimental techniques. In biological fliers, the flow visualisation can be carried out either with teth- ered animals, or in-flight [18]. Tethering [19, 20, 21, 22] typically allows for higher quality flow visualisation results, as the relative position and orientation of the animal and the measurement region can be precisely adjusted, result- ing into a higher resolution data [18]. However, tethering usually also leads to unnatural wing movements so such measurements may not be representative of free flight. Therefore, there would be a strong preference to perform flow visualisation under free flight conditions. Free flight measurements were conducted in a flight arena with hovering hummingbirds [23, 24, 25, 26] and in a wind tunnel (to represent the forward flight condition) with bumblebees [27], bats [28, 29], moths [30] or humming- birds [31]. Here, the challenge is to make the animal fly at the desired position with respect to the measurement region. This typically requires intensive train- ing and food sources, such as nectar feeders, are used to attract the animal. Nevertheless a successful measurement always requires some degree of luck due to the unpredictable behaviour of the animal. To increase the likeliness of a useful measurement, researchers typically opt for a larger measurement region, which has a trade off of lower resolution and thus less flow details captured by the measurements [18]. Free flight experiments with flapping-wing robots would be attractive for multiple reasons. Apart from being able to quantify the effect of inherent body oscillations (present only in free-flight) on the air flow, flapping-wing robots can 2 be programmed, meaning that the air flow could be investigated also during (controlled and reproducible) manoeuvres. Moreover, it would be possible to investigate the effect of small parameter changes, such as wing span, wing aspect ratio, etc. in a structured manner. However, until now, flow visualisation ex- periments with robotic flappers were conducted in a tethered condition, because precise position control necessary for successful flow measurements posed con- siderable challenges. Most of the studies used purposely built experimental flap- ping devices with model wings [32, 33, 34, 35, 36, 37, 38], while only a few works studied flight-capable FWMAVs in a tethered configuration [39, 40, 41, 42]. To make the free flight flow visualisation feasible, the FWMAV needs to fly with high position accuracy. For the forward flight condition, autonomous preci- sion wind tunnel flight has already been achieved with quadrotors [43] and fixed wings [44], but FWMAVs are much more challenging to control, because of more complex dynamics and stricter weight and size restrictions on on-board comput- ers and sensors. Our previous effort achieved the first successful autonomous wind-tunnel flight of a FWMAV [45], but further improvements were still neces- sary to achieve the position and flight state stability necessary to perform such in-flight flow visualisation experiments. Figure 1: Free flight PIV measurement of a FWMAV (a) and traditional mea- surement in a tethered setting used for comparison (b). A novel FWMAV- specific control approach was developed in order to achieve sufficient position accuracy and stability necessary for successful in-flight PIV measurements. The photos were taken with a reduced laser power compared to the real measure- ment. In this work, we present a methodology with which we have performed the first flow visualisation of a freely flying FWMAV (figure 1a). A main component of the methodology is a novel FWMAV specific controller, which controls the MAV position in the wind tunnel, with high accuracy, through feedback from an on-board inertial measurement unit (IMU) and an external motion tracking system. In this first free-flight flow visualisation effort, a time resolved stere- ographic Particle Image Velocimetry (PIV) method was used to measure the wake behind the FWMAV, similar to our previous experiments with a tethered 3 configuration [40]. Thanks to the achieved control accuracy and repeatability, future analysis of flow at different locations and with different PIV methods is now possible. In addition to the challenging free-flight PIV measurements, reference ex- periments were carried out under similar flight conditions, but with the same FWMAV tethered in a fixed position in the wind tunnel (figure 1b). The pur- pose of these latter tests was to provide a comparison and assessment of the in-flight measurements, as our past study revealed differences between in-flight force estimates and clamped force balance measurements [46]. These differences, observed mainly in the direction of the stroke plane, were partly attributed to the dynamic oscillations that are present in the free flight but are restricted in the clamped measurement. The paper is organised as follows: the experimental setup, the control algo- rithms and the PIV processing techniques are presented in section 2. Section 3 reports on the results of precision position control of the FWMAV as well as on the flow visualisation results. Finally, section 4 provides conclusions and discusses potential future tests and improvements. 2 Methods 2.1 Experimental setup Figure 2: The DelFly II FWMAV used in the tests. (a) Description of the main components. (b) The important phases of the flapping motion, including the 'clap' and 'peel', which help enhance the lift production and efficiency. For reliable tracking, the DelFly was equipped with four active IR-LED markers, three placed on the tail and one on the nose. 4 The experiments were carried out with the DelFly II MAV (further called simply the DelFly), a well-studied flapping-wing platform developed at TU Delft [47]. The DelFly, displayed in figure 2 (a), is a biplane design with flexible wings (280 mm wing span) arranged in a cross configuration, moving in opposite sense while flapping. Once per wing beat cycle, the wings clap together as they meet and peel apart again, see figure 2 (b). This clap-and-peel mechanism has a positive effect on the overall thrust production and efficiency [35, 48]. Due to its conventional tail with horizontal and vertical tail surfaces, the DelFly is inherently stable and its two control surfaces, the rudder and the elevator, are only used for steering. The DelFly has a large flight envelope, which ranges from near hover flight (≈ 0 m/s, vertical body orientation), to fast forward flight (≈ 7 m/s, nearly horizontal body orientation). For faster speeds, the centre of gravity needs to be shifted forward for inherent stability [49]. For the experiments described here, the DelFly was equipped with a Lisa/S autopilot board [50], which includes a 6 DOF MEMS inertial measurement unit (IMU) for on-board attitude estimation (Invensense MPU 6000) and a 72 MHz ARM CPU capable of running Paparazzi open source autopilot system [51]. The autopilot board was attached to the fuselage with a soft foam mount in order to isolate the high frequency vibration. Further components include an Mi-3A speed controller (flashed with BL heli firmware) driving the main brush- less motor (customised design with 28 turns per winding [4]), two Super Micro linear servo actuators for the tail control surfaces, a DelTang Rx31 receiver for the radio link and an ESP8266 Esp-09 WiFi module for the datalink between the autopilot and the ground station. The system was powered by a 180 mAh single cell LiPo battery (Hyperion G3 LG325-0180-1S). The overall mass of around 23 grams allowed for flight endurance between 2 to 6 minutes, depending on the flight speed. The experiments were conducted in the Open Jet Facility wind tunnel at TU Delft, see figure 3. This return-type, low speed wind tunnel has a large open test section with a cross-section of 2.8 m × 2.8 m, providing enough space for the proposed free flight experiments. During the tests, the wind tunnel was operated at speeds ranging from 0.8 m/s to 2.4 m/s. The wind tunnel room was equipped with an OptiTrack motion tracking system (NaturalPoint, Inc.) consisting of 12 OptiTrack Flex 13 motion tracking cameras (resolution 1280 px × 1024 px, 120 fps). The system was primarily used for tracking the test aircraft position and heading, but provided also the positions and orientations of the measurement plane and the high speed cameras of the PIV system. For reliable tracking, the DelFly MAV was equipped with four active IR LED markers placed on its body according to figure 2 (a). Reflective markers were used on the remaining objects (calibration plate, PIV cameras). The flow visualisation technique chosen for the experiments presented here is that of time-resolved stereoscopic Particle Image Velocimetry (PIV). The PIV system consists of a high speed laser and two high speed cameras which acquired images (1024 px × 1024 px) at a rate of 5 kHz. Based on our prior experience in similar experiments with a clamped FWMAV [40, 41], we have opted for per- forming measurements in the wake behind the DelFly in order to avoid problems 5 Figure 3: A schematic sketch of the experimental setup. The wind tunnel room was equipped with 12 OptiTrack Flex 13 motion tracking cameras for FWMAV tracking. The stereoscopic PIV setup consisted of 2 Photron FastCam SA 1.1 high speed cameras mounted at a relative angle of around 40◦. A high speed Mesa-PIV double-pulse laser illuminated the measurement plane located about 150 mm downstream of the FWMAV tail. Its beam was expanded to form a ≈ 2 mm wide laser sheet. Prior to the measurements, the room was filled with water-glycol based fog of droplets in order to achieve homogenous seeding of the flow. Illustration by Sarah Gluschitz (CC BY-ND 4.0). associated with laser reflections on the shiny surfaces of the wing and that of the wings blocking the camera view. The measurement plane was set normal to the free flow, behind the DelFly tail and an advective approach (Taylors hy- pothesis) was applied to reconstruct an estimate of the three dimensional wake configuration. A similar approach has been used in a variety of animal studies [28, 52, 53, 54]. The DelFly was controlled by the on-board autopilot, which was steering it towards a desired position set-point based on feedback from the external mo- tion tracking system. An operator was monitoring on-line the position errors and triggered the measurement at a convenient moment. He would also repeat the measurement in case the errors were too large. Additional IR LEDs, fixed with respect to the ground and detected by the tracking system, were turned 6 on together with the trigger signal to the PIV system, which served as a time stamp for time synchronisation of the tracking and PIV data sets. The simulta- neous application of the free-flight FWMAV control and the PIV measurements required to ensure that the optical motion tracking operation was not adversely affected by the laser light and the seeding fog introduced for the PIV experi- ments. 2.2 Control To ensure successful PIV measurements a high precision position control needs to be achieved, so that the wake of the DelFly stays within the measurement re- gion. At the same time, because we are interested in free steady flight, the thrust and power should not vary (dramatically) during the measurement. These are two opposing requirements: the wind tunnel will always have some remaining turbulence that the controller should respond to, but if tuned too aggressively, the power will vary significantly and the controller may even respond to the inherent flapping induced body rocking. The size of the PIV measurement region (170 mm × 170 mm) was chosen to be slightly larger than the half span of the DelFly (140 mm) so that the wake of the right half wings could be captured (a symmetry of left and right half wings was assumed). Because the dominant flow structures are observed behind the wing tips, we have estimated that a successful measurement can be carried out if the root mean square (rms) position error remains below 25 mm in all directions for a time course of 2 seconds (a single PIV measurement takes approximately 1 second). In order to meet these requirements, we designed a novel FWMAV-specific control scheme. The tests presented here cover the flight speeds between 0.8 m/s and 2.4 m/s, which corresponds to body pitch angles between approximately 70◦ and 30◦, re- spectively. The large range of body pitch throughout the flight envelope affects the way the DelFly is controlled: in near hover flight (body almost vertical) a change of flapping frequency will affect mostly the climbing/descending while elevator deflection ζ will have a dominant effect on the body pitch and sub- sequently the forward speed. In fast forward flight (body nearly horizontal) the control is inverted: flapping frequency change has a dominant effect on for- ward speed while pitching the body through elevator deflections affects mainly climbing/descending. A rudder deflection η will initially cause a banked turn, but will result in a pure heading change once the rudder returns back to its neutral position. This is due to a positive dihedral angle of the MAV providing inherent stability around the roll axis. Such behaviour can be observed over the whole flight envelope, but the rudder effectiveness will vary with airspeed. Thus, control of FWMAV is extremely challenging as it needs to consider all these effects. A general block diagram of the designed control system is in figure 4. The wind tunnel generates uniform airflow with a constant speed. The DelFly flies relative to the moving air and is controlled by an on-board autopilot, which steers the vehicle based on feedback from the on-board IMU (used for attitude 7 estimation) and from an external motion tracking system that provides position and heading information (with respect to ground). The tracking system data, captured at 120 Hz, is transmitted via LAN network to the ground control station and sent further, with a rate of 30 Hz, to the autopilot using a wireless WiFi data-link. The same link is also used for telemetry that can be viewed on-line on the ground station. Speed setpoint Wind tunnel DelFly FWMAV Motion with respect to air Commands Attitude Gyro rates Autopilot Motion with respect to ground Motion Tracking Cameras, 120 fps WiFi Telemetry Position & Heading Uplink, 30 Hz Ground station Tracking Software Ground segment Figure 4: Block diagram of the control system. The DelFly FWMAV is con- trolled by an on-board autopilot that uses feedback from an on-board inertial measurement unit and an external motion tracking system, which measures the FWMAV position and orientation with respect to the wind tunnel axes. A pro- prietary software (Motive 1.9) processes the camera data and sends the position and heading to the ground station. A WiFi uplink is used to transmit this information on-board at a rate of 30 Hz. 2.2.1 Axis system The body position x is expressed in the ground fixed system aligned with the wind tunnel: the xw-axis points opposite the wind velocity vector, zw points down and yw completes the right-handed Cartesian system, see Fig. 5. The body-fixed coordinate system is defined by the body's main axes: the xf -axis points along the fuselage towards the nose, the zf -axis points opposite to the direction of the vertical stabiliser, and the yf -axis points starboard. Its origin is placed at the centre of gravity. Because the external motion tracking system measures the position of the geometrical centre of the four LED markers, we used that value as an approximation of the centre of mass position. The body attitude Φ is described by roll Φ, pitch Θ and yaw Ψ angles, which define the rotation around the xf , yf and zf axes, respectively. The aircraft velocity ~V is (in steady-state) pointing opposite to the wind 8 (a) Longitudinal system (side view) (b) Lateral system (top view) VW xf V F zf H W FD ϴ xw zw xw yw V VW VK VW yf Figure 5: Definition of the axis systems. Two frames, wind-tunnel-fixed w and body-fixed f , are introduced to define the body position in the wind-tunnel and the body attitude angles, respectively. Consistent with the aerospace conven- tion, the z axis is pointing downwards. Panel (a) displays the side view with the longitudinal system parameters, assuming steady flight against the free stream VW . Panel (b) shows the top view with the lateral system parameters. Due to no roll control authority, displacement in the yw direction is achieved through heading Ψ. velocity vector ~VW , that is we have ~V = −~VW and there is no motion relative to ground (~VK = ~V + ~VW = ~0, see the longitudinal system in figure 5 a)). Height H = −zw is used as a measure of vertical position. The lift force FL and thrust force FF are oriented normal and parallel to the wind velocity VW , respectively. They act at the centre of gravity and, in steady state, compensate the weight W and drag force FD. Figure 5 b) shows the lateral system for the case of non-zero heading Ψ. In such case the aircraft will move relative to ground with a non-zero velocity ~VK = ~V + ~VW . 2.2.2 Control overview In the wind-tunnel experiment, the desired flight path is simply a (ground-) fixed way-point. Since the flight dynamics of the DelFly are still being investigated and the linearised models identified so far are only valid at a single operating condition [12], no reliable model that would cover the whole flight envelope was available. Thus, we employed a traditional aerospace control approach with control loops in a cascade arrangement, as implemented in the open-source Paparazzi UAV System [51]. However, an additional speed-thrust control block was added in between the standard guidance and attitude control blocks to take care of varying thrust and lift produced at different body speeds (and body 9 attitudes), see figure 6. Thus, the guidance control determines the desired body accelerations and heading based on the position error from the set-point. The commanded accelerations are transformed into the desired thrust and pitch by the speed-thrust control block. While novel to FWMAVs, a similar solution was used in the transitioning phase of hybrid UAVs [55, 56]. Finally, the attitude control loop determines the rudder and elevator deflections necessary to achieve the desired body pitch and heading. Figure 6: Block diagram of the cascade control approach consisting of guidance, speed-thrust and attitude controllers. Guidance block commands the desired heading Ψ and accelerations V c A, H c based on the current position error. The speed-thrust block determines the combination of pitch Θc and thrust T c com- mands that leads, at the wind tunnel speed VW , to the desired accelerations. The attitude block controls the attitude through the rudder and elevator com- mands, ηc and ζ c, respectively. 2.2.3 Guidance control The guidance control is decentralised, i.e. we control the forward position xw, height H = −zw and lateral position yw in separate loops. In the longitudinal loops (forward + vertical) we assume the DelFly is always aligned with the wind tunnel axis, i.e. it is flying opposite to the wind direction ~VW . Ordinary PD controllers are used in the longitudinal and vertical loops to determine the desired accelerations V c and H c, which are commanded to the inner loops. Since the DelFly has no roll control authority, the lateral position yw is controlled through heading Ψc. To compensate for steady state errors, an integral gain was introduced to the lateral loop. The accelerations and heading commanded to the inner loops are thus de- termined as V c = kdx∆ xw + kpx∆xw H c = −kdz∆ zw − kpz∆zw Ψc = kdy∆ yw + kpy∆yw + kiy Z ∆yw dt , (1) (2) (3) 10 where kp, kd and ki denotes the P, D and I gains, respectively, and ∆ stands for the position error from the set-point. The P and D gains of the longitudinal and vertical loops were selected based on the desired closed loop behaviour (assuming the plant to be a second order system with no inherent damping). The gains of the lateral loop were tuned during the flight tests. All the gain values, as used for the experiments in section 3.1, are summarised in table 2. 2.2.4 Speed-thrust control Since the generation of lift and thrust is highly coupled, a suitable combination of pitch angle Θ and throttle command T (controlling the flapping frequency f ) that will result in the desired accelerations in the longitudinal and vertical direc- tions needs to be found. This is the role of the speed-thrust control (figure 7), which consists of a feedforward and feed-back part. The feedforward control se- lects the necessary combination of Θ and T based on a linear model constructed from wind tunnel force measurements data. The feedback part improves the performance by correcting for model uncertainties, change of performance over time as well as external disturbances. Feedforward control Using wind tunnel data obtained with a clamped DelFly for various wind speeds VW , pitch angles Θ, and throttle commands T (the data were collected during an experiment described in [57]), a linear relationship be- tween the pitch angle and throttle and the measured thrust and lift forces can be found by first-order Taylor linearisation FL(cid:21) = (cid:20)FF 0(VW ) (cid:20)FF FL0(VW )(cid:21) + A(VW )(cid:20)Θ − Θ0(VW ) T − T0(VW )(cid:21) , (4) where Θ0(VW ), T0(VW ) denote the equilibrium condition of no acceleration for wind speed VW , resulting into a thrust FF 0(VW ) that is equal to drag at VW and lift FL0(VW ) that is equal to the DelFly weight. A(VW ) is the matrix of force derivatives (evaluated at the equilibrium Θ0(VW ), T0(VW )) A(VW ) = (cid:20) ∂FF ∂Θ (VW ) ∂FL ∂Θ (VW ) ∂FF ∂T (VW ) ∂FF ∂T (VW )(cid:21) . By inverting equation (4) we get (cid:20)Θ T(cid:21) = (cid:20)Θ0(VW ) T0(VW )(cid:21) + A(VW )−1(cid:20)∆FF (VW ) ∆FL(VW )(cid:21) , (5) (6) where ∆FF (VW ) and ∆FL(VW ) are the differences of thrust and lift from the equilibrium values FF 0(VW ) and FL0(VW ), respectively. Assuming the horizon- tal and vertical systems are decoupled and neglecting any inherent damping, VA = ∆FF /m, H = the body acceleration is a result of only the forces applied, 11 (a) Semi-adaptation (b) Correction Figure 7: The two-phase semi-adaptive control approach. At the start of each flight, an adaptation loop with gain γ is used to adapt the assumed equilib- rium conditions T0 and Θ0, until any potential position drift caused by model uncertainties of the feedforward control is removed. At time t = tA, when the equilibrium is approached, the operator switches to the correction phase and the adapted equilibrium conditions T ∗ 0 are kept. In this phase the accel- eration set-points from the outer guidance loop are tracked and an additional feedback loop is employed to compensate for disturbances, model uncertainties and performance changes due to decreasing battery voltage. 0 and Θ∗ 12 ∆FL/m. Thus, the desired pitch angle and thrust can be found from the accel- eration set-points as (cid:20)Θc T c(cid:21) = (cid:20)Θ0(VW ) T0(VW )(cid:21) + mA(VW )−1(cid:20) V sp H sp(cid:21) . (7) The matrix of aerodynamic force derivatives has been derived for different wind speeds, see table 1; switching between the different values is done manually based on the wind tunnel set-point, which remains constant throughout the tests. Table 1: Equilibrium conditions and aerodynamic force derivatives for various wind speeds, based on wind tunnel measurements described in [57]. ∂FL ∂T ∂FL ∂Θ ∂FF ∂T ∂FF ∂Θ VW (m/s) 0.8 1.2 2.5 Θ0 (◦) 65.9 47.2 30.5 T0 (%) 86.8 78.0 68.5 (mN/◦) (mN/%) (mN/◦) (mN/%) -5.2 -2.8 -5.5 1.4 2.4 2.2 0.8 0.8 4.9 3.7 3.4 3.2 Semi-adaptation In ideal case, setting the throttle and elevator to the equi- librium values Θ0(VW ), T0(VW ) should result in steady state flight at speed VW . However, because the wind tunnel data used to derive the matrices of force derivatives A(VW ) were obtained with another DelFly, and because the performance of the DelFly can deteriorate over time, the DelFly will typically drift (with respect to ground). To correct for the drift, each flight started with an adaptation phase, during which the assumed equilibrium conditions Θ0(VW ), T0(VW ) are being adapted, via a position feedback, until a steady flight is reached 0(VW ) (cid:20)Θ∗ 0 (VW )(cid:21) = (cid:20)Θ0(VW ) ∆H(cid:21) , T0(VW )(cid:21) + mA(VW )−1γ(cid:20)∆xw T ∗ (8) where the ∗ superscript denotes the adapted equilibrium conditions, ∆xw and ∆H are errors from the position where the adaptation was started and γ is a positive gain. The proof that such feedback leads to a stable equilibrium is in the Appendix of [58]. Feedback control Once the equilibrium is found, the operator switches to a correction phase, where an additional feedback loop with a PI controller is added to the feedforward control to compensate for disturbances, model uncertainties and performance changes due to decreasing battery level (cid:20) V fb H fb(cid:21) = (cid:20)kpF ∆ V kpV ∆ H(cid:21) +(cid:20)kiF R ∆ V dt kiV R ∆ Hdt(cid:21) . 13 (9) The integrated error is calculated asR ∆ V dt = R V spdt−VA,R ∆ Hdt = R H spdt− H. Because A contains a guess of direction of the force derivatives, we add the correction before the feedforward control. The combined control law results into (cid:20)Θc T c(cid:21) = (cid:20)Θ∗ T ∗ 0(VW ) 0 (VW )(cid:21) + mA−1(VW )(cid:18)(cid:20) V sp H sp(cid:21) +(cid:20)kpF ∆ V kpV ∆ H(cid:21) +(cid:20)R kiF ∆ V dt R kiV ∆ H dt(cid:21)(cid:19) . (10) 2.2.5 Attitude control In the inner most loop, we used the Integer-quaternion implementation of the attitude stabilisation algorithm of the Paparazzi UAV system [51], which con- trolled the body pitch and yaw via elevator and rudder deflections, respectively. The PID gains were tuned manually prior to the wind tunnel tests. For faster speeds a feed-forward term kf f was used in the yaw loop. The gain values are summarised in table 2. Table 2: Wind speed dependent gain values of all the control loops. The high speed gains are significantly different because the FWMAV gets close to the limit of inherent stability at these speeds. VW kpx kdx kpy kiy kdy kpz kdz kpv kiv kpf kif γ kpΘ kiΘ kdΘ kpΨ kiΨ kdΨ kf f Ψ Low speed High speed ≈ 0.8 m/s ≈ 1.2 m/s ≈ 2 to 2.4 m/s Guidance control 6 0.75 0.75 4.8 0.6 1.2 1 2 1 2 6 0.98 3 Speed-thrust control 0 2 0 2 0.3 0.5 1 0.3 2.5 Attitude control 1.75 0.034 0.031 1.5 0 0.031 5 2.53 0.25 0.063 1.63 0.019 0.094 0 14 2.3 PIV measurement setup and processing High-speed Stereo-PIV measurements were performed at a spanwise-oriented plane approximately 150 mm downstream from the DelFly tail. Note that only one side of the wake was imaged, due to field of view size restrictions, however, the wake is assumed to be nominally symmetric with respect to the centre plane. The flow was illuminated with a double-pulse Nd:YLF laser (Mesa-PIV) with a wavelength of 527 nm. The laser sheet with a thickness of 2 mm was kept at a fixed position, while the DelFly position was varied based on the measurement case. The flow was seeded with a water-glycol based fog of droplets with a mean diameter of 1 µm, which is produced by a SAFEX fog generator. The complete measurement room was filled with the fog beforehand in order to achieve a homogeneous seeding of the flow. Images of tracer particles were captured with two high-speed Photron FastCam CMOS cameras which allows to achieve a maximum resolution of 1024 × 1024 pixels at a data rate of up to 5.4 kHz. Each camera was equipped with a Nikon 60 mm focal objective with numerical aperture f/4 and mounted with Scheimpflug adaptors. The cameras were placed with an angle of 40◦ with respect to each other. A schematic overview of the PIV measurement setup can be seen in figure 3. A field of view of 170 mm × 170 mm was captured with a magnification factor of approximately 0.12 at a digital resolution of 6 pixels/mm. Single-frame images were recorded at a rate of 5 kHz for approximately a second, yielding a data ensemble size of about 5000 images. The associated time interval of 0.2 ms between individual frames corresponds to an out-of-plane displacement of 0.4 mm, based on the free stream velocity. This is sufficiently low with respect to the laser sheet thickness to allow for an accurate correlation of subsequent particle images. In order to increase the signal-to-noise ratio in the images, two laser cavities were shot in each single camera frame with 1 µs time delay in between, ensuring frozen particle images. The commercial software Davis 8.0 (LaVision) was used in data acquisition, image pre-processing, stereoscopic correlation of the images, and further vector post-processing. The pre-processed single-frame images were interrogated using windows of final size of 64 × 64 pixels with two refinement steps and an overlap factor of 75% resulting in approximately 4800 vectors with a spacing of 2.9 mm in each direction. A spatio-temporal reconstruction was performed for the initial interpretation of the wake structure. For this purpose, the time-series measurements performed in the single static measurement plane (i.c., around 150 mm downstream of the DelFly tail) were employed to generate a quasi-three-dimensional representation of the wake structure by using a passive convection model (Taylors hypothesis). This implies that the data of the measurement plane is translated with the free-stream velocity U∞ = VW (with an assumption of non-deforming wake and neglecting the induced velocities). More precisely, from the time-resolved velocity data u(x, y, z, t) recorded at the fixed streamwise position x = x0, the volumetric representation of the instantaneous flow field at a specific time t = t0 is computed as u(x, y, z, t0) ≈ u(x0, y, z, t0 − x/U∞). This approach results in a spatial resolution of 0.4 - 0.5 mm in the streamwise direction for the given 15 image recording rate of 5 kHz and for the considered free-stream velocity range (2 - 2.4 m/s). 3 Results 3.1 Position control The following section shows the performance of the position control, in steady state as well as in response to a step input. During the PIV measurement, the flying DelFly should ideally stay at a prescribed constant position. Thus, our primary goal was to achieve high precision steady flight around a fixed set-point rather than fast tracking performance of a moving set-point. 3.1.1 Step commands The sequence of step commands in all three wind tunnel frame axes is captured, for a wind tunnel set-point VW = 1.2 m/s, in figure 8. Apart from the position, we also recorded the body attitude (motion tracking system) and commands to the attitude loop and to the motor speed controller (WiFi telemetry). From the position graphs we can see that similar rise times, between 3 and 6 seconds, were achieved in all the three directions. A slight overshoot and longer settling was observed especially in the lateral direction, since it was controlled indirectly, through the change of heading. In longitudinal manoeuvres, it is the speed thrust control block (section 2.2.4) that determines the necessary combination of throttle and elevator com- mands, based on current wind tunnel set-point. In figure 8 (VW = 1.2 m/s), the vertical manoeuvre is dominated by a throttle change, while forward manoeu- vre also requires pitching the body via the elevator. Because this block is based on experimental data obtained with a slightly different aircraft, some coupling remains when forward step is commanded, nevertheless the feedback control damps these effects out. The lateral position is sensitive to both changes in ver- tical and forward directions, which is an inherent property of the aircraft, but again the feedback control will steer the vehicle back to the set-point through a heading change controlled by rudder deflection. From the command plots we can further observe that the throttle command increases over time. This is due to the battery voltage, which is decreasing as the battery gets discharged, and due to the integrator action, which responds by increasing the throttle command in order to keep the flapping frequency constant. A comparison of measured body pitch with the pitch command con- firms that the attitude loop manages to follow the pitch set-point. A post flight telemetry analysis showed that the decreasing trend in the yaw command is a result of a slow drift of the on-board heading estimate. While heading from the tracking system should be used for correcting the drift typical when integrating gyroscope readings, a small drift remained and was again corrected for by the integrators in the control loops. The heading measured by the tracking system remained close to zero, i.e. aligned with the wind tunnel axis. 16 ) m ( x 1.2 1 0.8 0.6 0.4 ) m ( y 0.6 0.4 0.2 0 -0.2 2.5 2 1.5 50 0 -50 ) m ( z - ) g e d ( l l o r ) g e d ( h c t i p ) g e d ( w a y ) % ( d m c . v e E l ) % ( d m c . d u R ) % ( d m c . t o r h T 320 340 360 380 400 420 440 460 480 500 320 340 360 380 400 420 440 460 480 500 320 340 360 380 400 420 440 460 480 500 300 320 340 360 380 400 420 440 460 480 500 60 50 40 300 320 340 360 380 400 420 440 460 480 500 20 0 -20 300 320 340 360 380 400 420 440 460 480 500 -20 -40 -60 10 0 -10 -20 90 85 80 75 320 340 360 380 400 420 440 460 480 500 320 340 360 380 400 420 440 460 480 500 320 340 360 380 420 400 time (s) 440 460 480 500 Figure 8: Response to a sequence of step commands in all three directions at VW = 1.2 m/s. The position set-points are displayed in black dashed lines, blue lines show the unfiltered tracking and telemetry data. The dotted vertical lines mark the time stamps of the step commands. 17 3.1.2 Steady state The results of steady flight with a fixed position set-point, performed at wind tunnel set-points VW = 0.8 m/s, 1.2 m/s, 2.0 m/s and 2.4 m/s, are in figure 9. Each panel shows the difference from the set-point in all three wind tunnel axes. The highlighted parts (thicker line, red colour) show the segments, where a successful PIV measurement could be performed according to our estimations, i.e. where the root mean square error (rms) remains below 25 mm in all three axes for the next 2 seconds. The rms errors over the whole measurement are summarised, together with mean body attitude angles, mean marker tracking errors and their respective standard deviations, in table 3. All the data was measured by the motion tracking system. Table 3: Position precision and attitude at various wind tunnel set-points. The attitude and mean marker tracking error are represented as mean ± standard deviation over the interval displayed in figure 9. VW (m/s) rmsx (mm) rmsy (mm) rmsz (mm) Φ (◦) Θ (◦) Ψ (◦) 0.8 1.2 2.0 2.4 18 15 75 65 28 25 45 43 9 11 44 42 4.5 ± 3.4 2.3 ± 3.1 1.5 ± 2.7 1.7 ± 2.9 68.9 ± 0.8 50.5 ± 0.8 33.8 ± 1.4 28.7 ± 1.1 0.3 ± 3.4 -3.1 ± 1.7 -3.4 ± 1.5 -3.6 ± 1.4 Track. err. (mm) 0.73 ± 0.17 0.45 ± 0.11 0.76 ± 0.30 0.59 ± 0.19 It can be immediately noted that the performance at low speeds is much better than at high speeds. Originally, prior to the PIV test session, we tuned the controller for speeds ranging from 0.8 m/s to 1.2 m/s, where the inherent stability of the DelFly is the most pronounced. However, during the PIV session we observed that the quality of captured data was worse than expected. The direction of the wake structures was dominated by the flapping-induced flow (aligned with the body fuselage that is pitched by 50.5◦ to 68.6◦ at slow speeds) and this resulted into a considerable angle between the measurement plane normal and the wake axis. Therefore, within the time constraints of the wind tunnel slot available for the PIV tests, we quickly tuned the controller also for high speeds (2.0 m/s and 2.4 m/s shown here), where the body pitch is much lower (33.8◦ to 28.7◦). This was much more challenging, because the DelFly (in the configuration used for the tests) is already very hard to fly at these speeds without any stability augmentation. At low speeds (0.8 m/s and 1.2 m/s), the position fix was very good. The best results were achieved in the vertical zw direction, where the DelFly stayed within ± 25 mm most of the time and the corresponding rms error was around 10 mm. In the forward direction, an accuracy better than ± 50 mm was achieved most of the time, with an rms error of below 20 mm. Most oscillations were observed in the lateral direction, yet a large part of the time the aircraft was also within ± 50 mm from the set-point, which can be seen from the rms values that remain below 30 mm. According to the estimated criteria (rms below 25 mm for the next 2 seconds) a successful PIV measurement could be started 18 (a) (b) (c) (d) Figure 9: Position difference from the set-point for various wind tunnel speeds: (a) 0.8 m/s, (b) 1.2 m/s, (c) 2 m/s, (d) 2.4 m/s. Root-mean-square (rms) position error values over the whole measurement are displayed in the top left corner of each subplot. The segments where a successful PIV measurement could be performed according to our estimations (i.e. the rms error remains below 25 mm in all three axes for the next 2 seconds) are highlighted by a thicker red line. Significant position accuracy decrease can be observed for high speeds (2 m/s and 2.4 m/s), where the FWMAV is operated close to its inherent stability limit. Besides, the gain tuning for high speeds may not have been optimal due to time constraints during the wind-tunnel slot dedicated to the flow visualisation measurements. 19 49% and 65% of the time for 0.8 m/s and 1.2 m/s, respectively, meaning the waiting time of the operator monitoring the MAV position and triggering the PIV measurement would be very short. At high speeds, the DelFly control is much more challenging as explained earlier. The rms error values were about 45 mm in forward and lateral directions and around 70 mm for forward direction. Despite a worse overall performance (the rms values were computed over several minutes of flight), there were seg- ments of several seconds (highlighted parts) when the platform was very close to the set-point for at least 2 seconds in all the axes (3% and 4.8% of the total time for 2.0 m/s and 2.4 m/s, respectively). This gave us enough opportunities to trigger and perform successful PIV measurements also at lower body pitch angles, where the flow patterns of the wake move almost normally to the vertical measurement plane, yielding higher quality flow measurements. The mean attitude captured at different speeds (table 3) reveals that the roll and yaw angles were not exactly zero. This is due to imperfections of the hand- built DelFly, in particular a slight misalignment of the tail caused by a twist in the square carbon tube used as fuselage, but this has a negligible effect on the PIV measurements since the misalignment is in the order of a few degrees. 3.2 Flow visualisation This section provides results for the first free-flight flow visualization of the wake of the DelFly. The PIV measurements were performed in a plane oriented perpendicular to the freestream direction, at a distance of approximately 150 mm downstream of the tail, similar to measurements performed with bats of comparable sizes [28]. Results are presented here for the flight condition of a freestream speed of 2.4 m/s, flapping frequency of 12.0 Hz and body angle of 28.7◦. The corresponding reduced frequency, defined as k = πf c/VW (where f is the flapping frequency, c = 80 mm the mean wing chord and VW the freestream velocity) has a value of 1.25. The Reynolds number is 13,000, based on freestream velocity and wing chord. For comparison, the tests were also con- ducted in a tethered setting with the identical DelFly, rigidly fixed according to figure 1, similar to our previous trials [40]. The flight conditions were compa- rable to the free-flight tests (freestream speed of 2.0 m/s, flapping frequency of 12.1 Hz, body angle of 33.8◦ and Reynolds number around 11,000). PIV images were recorded for a duration of approximately one second, at an acquisition frequency of 5 kHz. Given the flapping frequency of 12 Hz, this implies that 12 cycles are captured, with approximately 400 images per cycle, indicating a well-resolved characterisation of the flapping cycle. The results presented in the following are the direct outcome of the measurements, i.e., no additional averaging, smoothing or other form of filtering has been applied that could potentially further improve the quality of the visualisation. The relative position and orientation of the FWMAV with respect to the cen- tre of the measurement region, averaged over the duration of the PIV recording, is displayed in figure 10. While various position set-points were tested during the trials, this relative position allowed to capture the most prominent vortex 20 structures in the wake, originating from the right halve of the wings. The good position stability over the duration of the PIV measurement can be documented by the values of standard deviation from the mean position, which show that apart from a slight drift in the y direction (σy = 26.3 mm) a very good fix was achieved both in the forward (σx = 5.2 mm) and the vertical (σz = 4.5 mm) axes. (a) (b) (c) Figure 10: Relative position and orientation of the FWMAV with respect to the measurement plane (green square), averaged over the duration of the PIV measurement presented in figures 11 and 12. Figure 11 displays a sample time series of four images separated by 0.065 seconds, with the vectors indicating the in-plane velocity components and the colour contours the out-of-plane vorticity. The most prominent feature observed in the visualisations can be associated to the tip vortex of the upper wing in the instroke phase (red, corresponding to counter-clockwise vorticity). A three-dimensional representation of the wake vortex structure is obtained with the convection model, as described in section 2.3, which transforms the temporal information contained in the high-speed velocity field acquisition into a spatial representation by translating the flow field of subsequent images down- stream with the freestream velocity. The result, displaying two flapping cycles, is shown in figure 12. Figure 12a applies to the free-flight condition and figure 12b to the tethered DelFly. The visualisations provide a colour-coding of the helicity (density), which is defined as the scalar product of the velocity and vorticity vectors. Helicity can be used for the detection of vortex cores [59] and non-zero helicity indicates a helical vortex structure with an axial flow. The sign of the helicity allows to distinguish vortex structures with different sense of swirl. 21 Figure 11: Stereo-PIV measurements in the wake of a DelFly in free flight showing a sample sequence of four images, separated by 0.065 seconds, where the rightmost one is captured earliest in time; vectors indicate the in-plane velocity components and the colour contours the out-of-plane vorticity (in 1/s). Free stream velocity 2.4 m/s; flapping frequency is 12.0 Hz. Figure 12: Comparison of three-dimensional wake structure of the DelFly for (a) free-flight and (b) tethered condition; colour coding is for helicity (red: +0.6 m/s2; blue: -0.6 m/s2). 22 The very prominent upper red structure (positive helicity) is the tip vortex formed by the upper wing during the instroke. The less distinct blue structure (negative helicity) is the tip vortex of the bottom wing generated during instroke as well. Structures of the outstroke appear not to be very well captured in this representation, however. Notwithstanding the suboptimal quality of these preliminary results, an im- portant observation is the good qualitative agreement between the free-flight and tethered wake flow structures, and the good repeatability of the two cycles for each case. This supports the conclusion that reliable and meaningful PIV measurement results have been obtained also in the free-flight case. Further processing of the data may allow for a more quantitative and detailed compari- son between the free flight and tethered condition and potentially reveal if there are effects of the tethering to be detected. 4 Conclusions and discussion We presented a methodology, which combined a FWMAV specific control ap- proach for autonomous flight in a wind tunnel with a time-resolved stereoscopic PIV and allowed the first flow visualisation experiments to be carried out with a freely flying flapping-wing robot. The novel FWMAV specific control approach relied on feedback from an on-board IMU and an external motion tracking sys- tem. Applied to the 23 g DelFly FWMAV, an autonomous flight with high accuracy at low speeds (0.8 - 1.2 m/s, maximal root-mean-square error of 28 mm over 1-2 minutes) and good accuracy at high speeds (2 - 2.4 m/s, maximal root-mean-square error of 75 mm over 3-4 minutes) was achieved. Moreover, even higher precision was often achieved for time intervals of several seconds. Thus, the PIV measurements, lasting around one second, could be triggered when the DelFly was at the ideal position, which permitted to use a smaller measurement region and resulted in high resolution flow data. The free-flight PIV measurements were performed at high free stream speeds (2 to 2.4 m/s), where the FWMAV is pitched by 33.8◦ to 28.7◦ and flaps at fre- quencies of 12.1 Hz to 12.0 Hz, which corresponds to Reynolds numbers of 11,000 to 13,000 and reduced frequencies of 1.5 to 1.25. The flow was captured in a planar measurement region oriented perpendicular to the free stream direction and located in the wake approximately 150 mm downstream of the tail. For the initial interpretation of the measurements, the time-series PIV data were transformed into a quasi-three-dimensional representation of the wake structure using a passive convection model. For reference, measurements were also per- formed with a tethered FWMAV at comparable conditions. The first results, presented in the form of helicity isosurfaces, showed a good repeatability among flapping cycles and also qualitative agreement between the free-flight and the tethered cases, suggesting that the free-flight measurements were reliable and meaningful. While the obtained results hold promises for further processing as well as for future experiments with the current setup, the data quality could be further 23 increased by certain improvements of the control approach as well as of the flow visualisation procedure itself. While our control approach was designed and tested primarily for low speeds (0.8 m/s to 1.2 m/s), the first tests revealed that the interpretation of data captured in a plane perpendicular to the free stream direction can be complicated, because the relatively strong induced flow of the flapping wings, aligned with the body that is pitched by ≈ 70◦ to ≈ 50◦, hits the measurement plane at considerable angles. This phenomena will diminish with increasing speed, which allowed to perform meaningful measurements at speeds of 2 m/s to 2.4 m/s. For even better results, the control approach should be revised, as the dynamics of the DelFly become much more challenging at these speeds. However, even with a further increased position accuracy, the stereoscopic PIV method limits the measurements to be conducted in the wake only, as measurements closer to or even around the wings would have to deal with laser reflections on the wings. To enable reliable measurements around the wings, but also meaningful mea- surements at lower speeds, we recommend using a true 3D visualisation method such as tomographic PIV [60] for future experiments. Standard tomo-PIV using conventional seeding is not feasible, however, for the measurement vol- ume size and data acquisition rate required for the present experimental condi- tions. Recent developments have explored the potential of achieving large-scale tomographic measurements by using small (sub-millimetre) neutrally-buoyant helium-filled soap bubbles as tracer particles [61]. Although several studies have indeed proven the feasibility of this approach, we decided not to employ this method for the first trials, because as a relatively new method it still has its own challenges, many of which are related to the soap bubbles used as seeding par- ticles. They are being employed due to their high reflectivity, which is needed when the laser beam of finite power is expanded to illuminate larger volumes. However, the soap bubbles tend to stick to the FWMAV wing foils, which nega- tively affects the wing operation over time. For this reason, the exposure of the wings to the particles needs to be as limited as possible, which needs a specific measurement strategy to be used that minimises this effect. The initial results presented here proved that the developed methodology provides a reliable and repeatable way of obtaining PIV data in free flight and that the data quality is comparable to what is usually achieved in a (traditional) tethered setting. Moreover, this new approach, employing flying robots instead of animals, enables to perform measurements not only in steady state, but also during arbitrary controlled and reproducible manoeuvres. It also allows for further systematic investigations of parameter changes such as wing span, aspect ratio, wing flexibility, etc., something that was not possible before in free flight. Acknowledgements We thank Sarah Gluschitz for making the nice sketch of the experimental setup. 24 References [1] Matthew Keennon, Karl Klingebiel, Henry Won, and Alexander Andriukov. Development of the Nano Hummingbird: A Tailless Flapping Wing Micro Air Vehicle. AIAA paper 2012-0588, pages 1–24, 2012. [2] Kevin Y Ma, Pakpong Chirarattananon, Sawyer B Fuller, and Robert J Wood. Controlled Flight of a Biologically Inspired, Insect-Scale Robot. Science, 340:603–607, 2013. [3] Nina Gaissert, Rainer Mugrauer, Gunter Mugrauer, Agalya Jebens, Kristof Jebens, and Elias Maria Knubben. Inventing a micro aerial vehicle inspired by the mechanics of dragonfly flight. In Lecture Notes in Computer Sci- ence (including subseries Lecture Notes in Artificial Intelligence and Lec- ture Notes in Bioinformatics), volume 8069 LNAI, pages 90–100. Springer Berlin Heidelberg, 2014. doi: 10.1007/978-3-662-43645-5{\ }11. [4] G. C. H. E. de Croon, K M E de Clerq, R Ruijsink, B. D. W. Remes, and C. De Wagter. Design, aerodynamics, and vision-based control of the DelFly. International Journal of Micro Air Vehicles, 1(2):71–97, 2009. doi: 10.1260/175682909789498288. [5] Leif Ristroph, Gunnar Ristroph, Svetlana Morozova, Attila J. Bergou, Song Chang, John Guckenheimer, Z. Jane Wang, and Itai Cohen. Active and passive stabilization of body pitch in insect flight. Journal of the Royal Society, Interface / the Royal Society, 10(85):20130237, 2013. doi: 10. 1098/rsif.2013.0237. [6] Florian T. Muijres, Michael J Elzinga, Johan M Melis, and Michael H. Dickinson. Flies Evade Looming Targets by Executing Rapid Visually Directed Banked Turns. doi: 10.1126/science.1248955. Science, 344(6180):172–177, apr 2014. [7] Katherine M. Sholtis, Ryan M. Shelton, and Tyson L. Hedrick. Field Flight Dynamics of Hummingbirds during Territory Encroachment and Defense. PLOS ONE, 10(6):e0125659, jun 2015. ISSN 1932-6203. doi: 10.1371/ journal.pone.0125659. [8] Richard J. Bomphrey, Toshiyuki Nakata, Per Henningsson, and Huai-Ti Lin. Flight of the dragonflies and damselflies. Philosophical Transactions of the Royal Society of London B: Biological Sciences, 371(1704), 2016. [9] Michael H. Dickinson, Fritz-Olaf Lehmann, and Sanjay P Sane. Wing Rotation and the Aerodynamic Basis of Insect Flight. Science, 284(5422): 1954–1960, 1999. doi: 10.1126/science.284.5422.1954. [10] Sanjay P Sane and Michael H. Dickinson. The aerodynamic effects of wing rotation and a revised quasi-steady model of flapping flight. Journal of Experimental Biology, 205:1087–1096, 2002. 25 [11] Gordon J Berman and Z Jane Wang. Energy-minimizing kinematics in hovering insect flight. Journal of Fluid Mechanics, 582:153–168, 2007. doi: 10.1017/S0022112007006209. [12] S F Armanini, J V Caetano, G C H E de Croon, C C de Visser, and M Mulder. Quasi-steady aerodynamic model of clap-and-fling flapping MAV and validation using free-flight data. Bioinspiration & Biomimet- ics, 11(4):046002, jun 2016. doi: 10.1088/1748-3190/11/4/046002. [13] T Nakata, H Liu, Y Tanaka, N Nishihashi, X Wang, and A Sato. Aerody- namics of a bio-inspired flexible flapping-wing micro air vehicle. Bioinspi- ration & biomimetics, 6(4):045002, dec 2011. doi: 10.1088/1748-3182/6/ 4/045002. [14] Shuanghou Deng, Tianhang Xiao, Mustafa Percin, Bas van Oudheus- den, Hester Bijl, and Bart Remes. Numerical simulation of an X-wing flapping wing MAV by means of a deforming overset grid method. In 22nd AIAA Computational Fluid Dynamics Conference, Reston, Virginia, jun 2015. American Institute of Aeronautics and Astronautics. doi: 10.2514/6.2015-2615. [15] Wee Beng Tay, J.H.S. de Baar, Mustafa Percin, Shuanghou Deng, and Bas W. van Oudheusden. Numerical simulation of a flapping wing MAV based on wing deformation capture analysis. In 34th AIAA Applied Aero- dynamics Conference, Reston, Virginia, jun 2016. American Institute of Aeronautics and Astronautics. doi: 10.2514/6.2016-3552. [16] Shuanghou Deng. Aerodynamics of flapping-wing Micro-Air-Vehicle: An integrated experimental and numerical study. PhD thesis, Delft University of Technology, 2016. [17] Toshiyuki Nakata and Hao Liu. Aerodynamic performance of a hovering hawkmoth with flexible wings: a computational approach. Proceedings of the Royal Society of London B: Biological Sciences, 2011. [18] Adrian L. R. Thomas, Graham K. Taylor, Robert B. Srygley, Robert L. Nudds, and Richard J. Bomphrey. Dragonfly flight: free-flight and tethered flow visualizations reveal a diverse array of unsteady lift-generating mech- anisms, controlled primarily via angle of attack. Journal of Experimental Biology, 207(24), 2004. [19] Charles P Ellington, Coen van den Berg, Alexander P Willmott, and Adrian L R Thomas. Leading-edge vortices in insect flight. Nature, 384:626–630, 1996. doi: 10.1038/384626a0. [20] Alexander P Willmott, Charles P Ellington, and Adrian L R Thomas. Flow visualization and unsteady aerodynamics in the flight of the hawkmoth, Manduca sexta. Philos Trans R Soc Lond B Biol Sci., 352:303–316, 1997. doi: 10.1098/rstb.1997.0022. 26 [21] Masaki Fuchiwaki, Taichi Kuroki, Kazuhiro Tanaka, and Takahide Tababa. Dynamic behavior of the vortex ring formed on a butterfly wing. Exper- iments in Fluids, 54(1):1450, jan 2013. ISSN 0723-4864. doi: 10.1007/ s00348-012-1450-x. [22] Per Henningsson, Dirk Michaelis, Toshiyuki Nakata, Daniel Schanz, Rein- hard Geisler, Andreas Schroder, and Richard J Bomphrey. The complex aerodynamic footprint of desert locusts revealed by large-volume tomo- graphic particle image velocimetry. Journal of the Royal Society Interface, 12(108), 2015. [23] Douglas R Warrick, Bret W Tobalske, and Donald R Powers. Aerodynamics of the Hovering Hummingbird. Nature, 435:1094–1097, 2005. doi: 10.1038/ nature03647. [24] Douglas R Warrick, Bret W Tobalske, and Donald R Powers. Lift produc- tion in the hovering hummingbird. Proceedings of the Royal Society B, 276: 3747–3752, 2009. doi: 10.1098/rspb.2009.1003. [25] Douglas L Altshuler, Marko Princevac, Hansheng Pan, and Jesse Lozano. Wake patterns of the wings and tail of hovering hummingbirds. Experiments in Fluids, 45(5):835–846, 2009. doi: 10.1007/s00348-008-0602-5. [26] Sam Pournazeri, Paolo S. Segre, Marko Princevac, and Douglas L. Alt- shuler. Hummingbirds generate bilateral vortex loops during hovering: ev- idence from flow visualization. Experiments in Fluids, 54(1):1439, jan 2013. doi: 10.1007/s00348-012-1439-5. [27] Richard James Bomphrey, Graham K. Taylor, and Adrian L. R. Thomas. Smoke visualization of free-flying bumblebees indicates indepen- dent leading-edge vortices on each wing pair. Experiments in Fluids, 46(5): 811–821, may 2009. doi: 10.1007/s00348-009-0631-8. [28] Anders Hedenstrom, F. T. Muijres, R. von Busse, L. C. Johansson, Y. Win- ter, and G. R. Spedding. High-speed stereo DPIV measurement of wakes of two bat species flying freely in a wind tunnel. Experiments in Fluids, 46 (5):923–932, may 2009. doi: 10.1007/s00348-009-0634-5. [29] Florian T. Muijres, L Christoffer Johansson, York Winter, and Anders Hedenstrom. Leading edge vortices in lesser long-nosed bats occurring at slow but not fast flight speeds. Bioinspiration & biomimetics, 9(2):025006, may 2014. doi: 10.1088/1748-3182/9/2/025006. [30] L Christoffer Johansson, Sophia Engel, Almut Kelber, Marco Klein Heeren- brink, and Anders Hedenstrom. Multiple leading edge vortices of unex- pected strength in freely flying hawkmoth. Scientific Reports, 3:3264, nov 2013. 27 [31] Victor M Ortega-Jimenez, Nir Sapir, Marta Wolf, Evan A Variano, and Robert Dudley. Into turbulent air: size-dependent effects of von K´arm´an vortex streets on hummingbird flight kinematics and energetics. Proceed- ings. Biological sciences / The Royal Society, 281(1783):20140180, may 2014. doi: 10.1098/rspb.2014.0180. [32] Coen Van Den Berg and Charles P Ellington. The vortex wake of a 'hover- ing' model hawkmoth. Philos Trans R Soc Lond B Biol Sci., 352:317–328, 1997. doi: 10.1098/rstb.1997.0023. [33] Coen Van Den Berg and Charles P Ellington. The three-dimensional leading-edge vortex of a 'hovering' model hawkmoth. Philos Trans R Soc Lond B Biol Sci., 352:329–340, 1997. doi: 10.1098/rstb.1997.0024. [34] James M Birch, William B. Dickson, and Michael H. Dickinson. Force production and flow structure of the leading edge vortex on flapping wings at high and low Reynolds numbers. Journal of Experimental Biology, 207: 1063–1072, 2004. doi: 10.1242/jeb.00848. [35] Fritz-Olaf Lehmann, Sanjay P Sane, and Michael H. Dickinson. The aero- dynamic effects of wing-wing interaction in flapping insect wings. Journal of Experimental Biology, 208:3075–3092, 2005. doi: doi:10.1242/jeb.01744. [36] Van Tien Truong, Jihoon Kim, Min Jun Kim, Hoon Cheol Park, Kwang Joon Yoon, and Doyoung Byun. Flow structures around a flap- ping wing considering ground effect. Experiments in Fluids, 54(7):1575, jul 2013. doi: 10.1007/s00348-013-1575-6. [37] Bo Cheng, Jesse Roll, Yun Liu, Daniel R. Troolin, and Xinyan Deng. Three- dimensional vortex wake structure of flapping wings in hovering flight. Journal of The Royal Society Interface, 11(91), 2013. [38] Yingying Zheng, Yanhua Wu, and Hui Tang. A time-resolved PIV study on the force dynamics of flexible tandem wings in hovering flight. Journal of Fluids and Structures, 62:65–85, 2016. doi: 10.1016/j.jfluidstructs.2015. 12.008. [39] H Ren, Y Wu, and P G Huang. Visualization and characterization of near- wake flow fields of a flapping-wing micro air vehicle using PIV. Journal of Visualization, 16(1):75–83, 2013. doi: 10.1007/s12650-012-0152-z. [40] Mustafa Percin, B. W. van Oudheusden, H. E. Eisma, and B. D. W. Remes. Three-dimensional vortex wake structure of a flapping-wing micro aerial vehicle in forward flight configuration. Experiments in Fluids, 55, 2014. doi: 10.1007/s00348-014-1806-5. [41] Shuanghou Deng and Bas van Oudheusden. Wake structure visualization of a flapping-wing Micro-Air-Vehicle in forward flight. Aerospace Science and Technology, 50:204–211, 2016. doi: 10.1016/j.ast.2016.01.003. 28 [42] Shuanghou Deng, Mustafa Percin, and Bas van Oudheusden. Experimen- tal Investigation of Aerodynamics of Flapping-Wing Micro-Air-Vehicle by Force and Flow-Field Measurements. AIAA Journal, 54(2):588–602, 2016. [43] James Neil Wiken. Analysis of a Quadrotor in Forward Flight. Master's thesis, Massachusetts Institute of Technology, Cambridge, US-MA, 2015. [44] Jan Nowak. Windkanal Freiflugmessungen zur Bestimmung flugmechanis- cher Kenngrossen. Phd thesis, RWTH Aachen University, Aachen, DE, 2010. [45] Christophe De Wagter, Andries Koopmans, Guido C H E de Croon, Bart D W Remes, and Rick Ruijsink. Autonomous Wind Tunnel Free-Flight of a Flapping Wing MAV. In Advances in Aerospace Guidance, Navigation and Control, pages 603–621. Springer, Berlin, DE, 2013. [46] J V Caetano, M Percin, B . W. van Oudheusden, B. D. W. Remes, C. De Wagter, G. C. H. E. de Croon, and C. C. de Visser. Error analysis and assessment of unsteady forces acting on a flapping wing micro air vehi- cle: free flight versus wind-tunnel experimental methods. Bioinspiration & biomimetics, 10(5):056004, oct 2015. doi: 10.1088/1748-3190/10/5/056004. [47] G. C. H. E. de Croon, Mustafa Percin, B. D. W. Remes, Rick Ruijsink, and C. De Wagter. The DelFly - Design, Aerodynamics, and Artificial Intelligence of a Flapping Wing Robot. Springer Netherlands, 2016. doi: 10.1007/978-94-017-9208-0. [48] Kristien M.E. De Clercq, Roeland de Kat, B. D. W. Remes, Bas W. van Oudheusden, and Hester Bijl. Aerodynamic Experiments on DelFly II: Unsteady Lift Enhancement. International Journal of Micro Air Vehicles, 1(4):255–262, 2010. doi: 10.1260/175682909790291465. [49] J.A. Koopmans, S. Tijmons, C. De Wagter, and G. C. H. E. de Croon. Passively Stable Flapping Flight From Hover to Fast Forward Through Shift in Wing Position. International Journal of Micro Air Vehicles, 7(4), jan 2015. [50] B. D. W. Remes, P. Esden-Tempski, F. Van Tienen, E. Smeur, C. De Wagter, and G. C. H. E. de Croon. Lisa-S 2.8g autopilot for GPS-based flight of MAVs. In IMAV 2014: International Micro Air Vehicle Conference and Competition 2014, Delft, The Netherlands, August 12-15, 2014, pages 280–285, aug 2014. [51] Paparazzi UAV, 2016. URL http://wiki.paparazziuav.org/wiki/Main_Page. [52] Florian T Muijres, L Christoffer Johansson, York Winter, and Anders Hedenstrom. Comparative aerodynamic performance of flapping flight in two bat species using time-resolved wake visualization. Journal of the Royal Society, Interface / the Royal Society, 8(63):1418–28, oct 2011. doi: 10.1098/rsif.2011.0015. 29 [53] P. Henningsson, F. T. Muijres, and A. Hedenstrom. Time-resolved vortex wake of a common swift flying over a range of flight speeds. Journal of The Royal Society Interface, 8(59), 2011. [54] Florian T Muijres, Melissa S Bowlin, L Christoffer Johansson, and Anders Hedenstrom. Vortex wake, downwash distribution, aerodynamic perfor- mance and wingbeat kinematics in slow-flying pied flycatchers. Journal of the Royal Society, Interface / the Royal Society, 9(67):292–303, feb 2012. doi: 10.1098/rsif.2011.0238. [55] B. Theys, G. De Vos, and J. De Schutter. A control approach for transi- tioning VTOL UAVs with continuously varying transition angle and con- trolled by differential thrust. In 2016 International Conference on Un- manned Aircraft Systems (ICUAS), pages 118–125. IEEE, jun 2016. doi: 10.1109/ICUAS.2016.7502519. [56] Christophe De Wagter, Dirk Dokter, Guido C H E de Croon, and Bart D W Remes. Multi-lifting-device uav autonomous flight at any transition percentage. In Proceedings of EuroGNC, 2013. [57] Matej Karasek, Andries Jan Koopmans, Sophie F Armanini, Bart D W Remes, and Guido C H E de Croon. Free flight force estimation of a 23.5 g flapping wing MAV using an on-board IMU. In The 2016 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS 2016), Daejeon, Korea, 9-14 October 2016, 2016. [58] Torbjørn Cunis, Matej Karasek, and Guido C H E de Croon. Precision Position Control of the DelFly II Flapping-wing Micro Air Vehicle in a Wind-tunnel. In The International Micro Air Vehicle Conference and Com- petition 2016 (IMAV 2016), Beijing, China, October 17-21, 2016. [59] David Degani, Arnan Seginer, and Yuval Levy. Graphical visualization of vortical flows by means of helicity. AIAA Journal, 28(8):1347–1352, aug 1990. doi: 10.2514/3.25224. [60] F Scarano. Tomographic PIV: principles and practice. Measurement Sci- ence and Technology, 24(1):012001, 2013. doi: 10.1088/0957-0233/24/1/ 012001. [61] Matthias Kuhn, Klaus Ehrenfried, Johannes Bosbach, and Claus Wagner. Large-scale tomographic particle image velocimetry using helium-filled soap bubbles. Experiments in Fluids, 50(4):929–948, apr 2011. doi: 10.1007/ s00348-010-0947-4. 30
1708.01820	3	1708	2018-06-04T13:27:08	Principles for optimal cooperativity in allosteric materials	[ "physics.bio-ph", "cond-mat.mtrl-sci", "cond-mat.soft" ]	Allosteric proteins transmit a mechanical signal induced by binding a ligand. However, understanding the nature of the information transmitted and the architectures optimizing such transmission remains a challenge. Here we show using an {\it in-silico} evolution scheme and theoretical arguments that architectures optimized to be cooperative, which propagate efficiently energy, {qualitatively} differ from previously investigated materials optimized to propagate strain. Although we observe a large diversity of functioning cooperative architectures (including shear, hinge and twist designs), they all obey the same principle {of displaying a {\it mechanism}, i.e. an extended {soft} mode}. We show that its optimal frequency decreases with the spatial extension $L$ of the system as $L^{-d/2}$, where $d$ is the spatial dimension. For these optimal designs, cooperativity decays logarithmically with $L$ for $d=2$ and does not decay for $d=3$. Overall our approach leads to a natural explanation for several observations in allosteric proteins, and { indicates an experimental path to test if allosteric proteins lie close to optimality}.	physics.bio-ph	physics	Principles for optimal cooperativity in allosteric materials L. Yan,1, ∗ R. Ravasio,2, ∗ C. Brito,3 and M. Wyart2, † 1Kavli Institute for Theoretical Physics, University of California, Santa Barbara, CA 93106, USA 2Institute of Physics, ´Ecole Polytechnique F´ed´erale de Lausanne, CH-1015 Lausanne, Switzerland 3Instituto de F´ısica, Universidade Federal do Rio Grande do Sul, CP 15051, 91501-970 Porto Alegre RS, Brazil Allosteric proteins transmit a mechanical signal induced by binding a ligand. However, under- standing the nature of the information transmitted and the architectures optimizing such trans- mission remains a challenge. Here we show using an in-silico evolution scheme and theoretical arguments that architectures optimized to be cooperative, which propagate efficiently energy, qual- itatively differ from previously investigated materials optimized to propagate strain. Although we observe a large diversity of functioning cooperative architectures (including shear, hinge and twist designs), they all obey the same principle of displaying a mechanism, i.e. an extended soft mode. We show that its optimal frequency decreases with the spatial extension L of the system as L−d/2, where d is the spatial dimension. For these optimal designs, cooperativity decays logarithmically with L for d = 2 and does not decay for d = 3. Overall our approach leads to a natural explanation for several observations in allosteric proteins, and indicates an experimental path to test if allosteric proteins lie close to optimality. INTRODUCTION Many proteins are allosteric: binding a ligand at an allosteric site can affect the properties of a distant active site, sometimes located on the other side of the protein [1, 2]. Predicting the existence of such allosteric pathways from protein structure alone would be of great interest [3, 4], since they can be used as targets for drug design [5]. Solving this challenge requires to make progress on both physical and biological questions. First, how can such disordered materials [6] be designed to carry mechanical information specifically over long distances? Are there fundamental limits to what can be achieved? Second, what are allosteric pathways really optimized for? What kind of elastic information do they carry? A physical theory of allostery should address these points. It should also explain the following empirical facts: (i) Some al- losteric proteins [7], including hemoglobin [8, 9], essen- tially function as hinges, while others display a "shear" design where two rigid parts are connected by a weak plane [10]. This classification is however not exhaustive, as in various cases the response to binding a ligand can- not be described in term of a simple shear or hinge mo- tion [11–13]. (ii) The response to binding often corre- sponds mostly to motion along few soft normal modes of the protein [14, 15]. These modes tend to be conserved during evolution [16]. (iii) In some cases the allosteric functional effect at the active site is significant while the physical mean displacement induced by binding the lig- and is small. It has been proposed that for these proteins binding can affect how particles near the active site fluc- tuate around their mean position, while changing little the latter [17–20]. Recently, allostery was investigated using in silico evo- lution schemes where a system evolves to perform a given function [21–25]. Most relevant here are schemes devel- oped to solve inverse elastic problems [23–25], in a spirit similar to topology optimization used in engineering to design functional tools from compliant materials [26–28]. The task studied in [23–25] was to design a material whose response to a specific local strain applied on one of its sides (the allosteric site) leads to a displacement whose geometry is prescribed on the opposite side (the active site). Under broad conditions these algorithms find solutions that achieve such "geometric" tasks essen- tially perfectly. The corresponding architectures turn out to have surprising properties: their response almost van- ishes in the bulk of the material and reappears near the active site [23]. This amplification of the elastic signal is caused by the emergence of a powerful lever, made of a soft elastic region surrounding the active site, where the system is just constrained enough to act as a solid [23, 29]. Although there is great interest in finding whether such architectures exist in nature, an intriguing aspect of this approach is that it does not generate the well-known al- losteric architectures such as the simple shear and hinge designs, in which the response remains of similar magni- tude between the allosteric and the active sites. Here we show that a simple modification of the task, where materials are optimized such that the binding at the allosteric site lowers the binding energy of another ligand at the active site, leads to different design princi- ples. In the context of proteins, this task corresponds to maximizing the cooperativity of binding two ligands, a central feature of various allosteric proteins [1]. We find that there is a zoology of architectures achieving such cooperativity, but they always display a stiff structure (embedded in a softer elastic matrix) with a single very soft extended elastic mode or "mechanism". We lay out the principles behind such designs, and show theoreti- cally that the soft mode frequency should be neither too large nor too small to optimize function: its optimal value decreases with the material size, and scales as L−d/2 in spatial dimension d. We prove that cooperativity then 8 1 0 2 n u J 4 ] h p - o i b . s c i s y h p [ 3 v 0 2 8 1 0 . 8 0 7 1 : v i X r a −1(L/c) for d = 2 and is even independent decays as ln of L in larger spatial dimensions d ≥ 3, where L is the linear extension of the system and c the length scale on which binding takes place. This result is very different from a normal continuous elastic medium where cooper- ativity rapidly decays with distance as L−d. Overall the classification we provide leads to a natural explanation for the key aspects of allostery described in (i,ii,iii). It also shows that a path of large strain values connecting the allosteric and active site induced by binding is not necessary for cooperativity to occur, and it makes fur- ther testable predictions, including the locations where a shear or hinge design would be mostly affected by a mutation and conserved during evolution. 2 average connection ¯σ = Ns/Nc and keep them fixed dur- ing evolution. We find that our results do not depend qualitatively on z as long as z > zc = 2d, the rigidity limit derived by Maxwell [32]. Binding: Binding a ligand exerts forces locally that leads to an imposed local strain. To model this effect at the allosteric site, we choose four adjacent nodes on one side of the system (shown in purple in Fig. 1), and con- sider that binding at that site imposes a displacement δRAl(cid:105) on these nodes, as indicated by purple arrows. (Strictly speaking, this description of binding assumes that the ligands are rigid. However we expect our results to hold true qualitatively as long as the ligands are not significantly softer than the protein itself). Minimizing the elastic energy in the entire system with these con- Al r (cid:105) that can be straints then leads to a response δR(σ) extended (see a formal expression for this response in Supplemental Material Section B and [23]). The corre- sponding energy cost associated with binding writes: FIG. 1: Examples of on-lattice elastic networks. (A) shows a hexagonal lattice (d = 2) with periodic boundary conditions along the horizontal axis (springs crossing the periodic boundary are shown in dashed lines, and are not present when open boundary conditions are used), mimicking a cylindrical geometry. (B) For d = 3, we use a face centered cubic lattice with open boundaries. In all cases, occupied links displaying a spring of stiffness unity are indicated by lines. The stimulus displacement is shown in purple arrows and the target displacement is shown in blue arrows, each are applied on four nodes. All data are presented for L = 20 and z = 5.0 in d = 2 and L = 12 and z = 8.4 in d = 3. METHODS In-silico Evolution Scheme Elastic networks: To model allosteric materials we consider elastic networks, often used to describe pro- teins [14–16]. Specifically, N = Ld nodes are located on a lattice (slightly distorted periodically to avoid straight lines as discussed in Supplemental Material Section A and [30, 31]), and among all Nc links of nearest nodes, a subset of Ns pairs are connected by harmonic springs of stiffness k = 1 , as indicated by lines in Fig. 1. We declare that σα = 1 if a spring is present in the link α and σα = 0 otherwise. Thus the network is entirely described by a connection vector σ(cid:105) made of zeros and ones, whose dimension is the number of links Nc. We define the average coordination number z ≡ 2Ns/N and EAl(σ) = 1 2 (cid:104)δRAl r MδRAl r (cid:105), (1) where M is the stiffness matrix of the network (whose definition is recalled in Supplemental Material Section B) of dimension N d × N d, which depends on the net- work considered. The same procedure is used to model the binding of another ligand at the active site (indicated in blue in Fig. 1), allowing us to define a binding energy EAc(σ). If the two binding events take place simulta- neously, the same procedure leads to the derivation of a joint binding energy EAc,Al(σ). FIG. 2: Illustration of cooperativity. With two binding sites, a protein displays four states. Cooperativity is high if binding a substrate molecule at its active site is difficult when the allosteric site is empty (i.e. EAc is large) whereas it is much simpler when the allosteric site is occupied (i.e. EAc,Al − EAl is small). Cooperativity: We seek to engineer materials in active siteligandallosteric sitesubstrate E=0 E Al E Ac E Ac,Alwhich binding at the allosteric site lowers the binding energy at the active site as much as possible, as illus- trated in Fig. 2. In the absence of the ligand at the al- losteric site, the binding energy at the active site is simply EAc(σ), whereas if present it is EAc,Al(σ)− EAl(σ). We seek to maximize the cooperative energy, simply defined as the difference between these terms: Ecoop = EAc(σ) + EAl(σ) − EAc,Al(σ) ≡ F, (2) which also defines our fitness function. Cooperativity turns out to differ greatly from the geo- metric task in which a displacement imposed at one end of the material must elicit a given displacement at the other end [23–25] (see below and Supplemental Material for a detailed comparison). The architectures associated with the latter task are very asymmetric, in particular they are much softer near the active site than near the allosteric site [23]. By contrast, it is clear from our defini- tion of cooperativity that both active and allosteric sites play a symmetric role. At an intuitive level, the difference can be understood by considering the limit of weak elastic coupling between allosteric and active sites for which one finds Ecoop ≈ (cid:104)F AcδRAl→Ac(cid:105) where F Ac(cid:105) is the exter- nal force field generated by the substrate when it binds to the active site, and δRAl→Ac(cid:105) is the displacement field induced at the active site by binding a ligand at the al- losteric site. Maximizing cooperativity thus requires to have a large and specific response δRAl→Ac(cid:105) (which is essentially what the geometric task accomplishes) and to have a large force scale F Ac(cid:105), which requires the material to be stiff near the active site. This additional constraint makes the cooperative task harder than the geometric one. Evolutionary Dynamics: To generate cooperative architectures, we implement an evolution scheme which selects preferably networks with high fitness. Specifi- cally, we use a Monte-Carlo algorithm where the relo- cation of individual springs is considered, i.e. σ(cid:105) → σ(cid:48)(cid:105) where a randomly chosen vacant link γ becomes occupied σγ = 0 → σ(cid:48) γ = 1 and a randomly occupied link α be- comes empty, σα = 1 → σ(cid:48) α = 0. The new structure is se- )], lected with the probability p = min[1, exp( where 1/Te is inverse evolutionary "temperature" char- acterizing the selection pressure. F (σ(cid:48))−F (σ) Te We find that as the selection pressures increases and Te decreases, there is a rather sudden transition from non- working networks with zero fitness to cooperative ones, as illustrated in inset of Fig.3. The fitness then appears to plateau, and in what follows we choose Te = 10−4 where this plateau is reached. Interestingly, in this plateau re- gion we find that the fitness landscape is glassy: there are many families of solutions that are not dynamically connected on the time scale of our runs, implying the presence of large fitness barriers. The families obtained in a given run are defined by the respective initial con- ditions, and do not display exactly the same fitness as 3 FIG. 3: Evolution of the fitness F vs the number of Monte Carlo steps M CS. Different initial conditions resulted in different architectures, which are analyzed at sufficiently long time to avoid significant transient effects (keeping only the data from the last 3.5 × 104 steps out of the 105 M CS in each run, as delimited by the black vertical line in the plot). The inset shows the fitness F averaged over 25 initial conditions as a function of the evolution temperature Te for the two dimensional network with both open and periodic boundaries. ασj shown in Fig.3. We checked that sequences are much more similar within a family than between different fam- ilies. Indeed, in a single family the mean overlap between α(cid:105) − σ2 is high distinct configurations i and j, q ≡ (cid:104)σi with q ≈ 0.36, while it is small q ≈ 0.03 for different families (•α averages over links, and (cid:104)•(cid:105) averages over configurations). Glassiness also implies that the archi- tectures slowly evolve in time, but less and less so as time goes on. In what follows, we study architectures only in the last third of the run, when transient effects are weaker and fitness is nearly stationary. In total, we generated 25 families in d = 2 and 10 families in d = 3. α Analysis Toolbox In this section we review useful observables character- izing allosteric architectures. Most of them are known in the protein literature, others are novel to the best of our knowledge. Geometry of allosteric response: By computing the structure of proteins crystallized with and without the ligand bound on their allosteric site, one gets access to the internal response of the protein induced by bind- ing, δRAl r (cid:105) in our notations. As recently emphasized in this context [10], a key aspect of this response is its strain, which must be zero in parts of the proteins moving as rigid blocks. The strain thus captures where deformation ↔ (i) can be is actually taking place. The strain tensor 0246810104-0.200.20.40.60.8110-410-310-200.050.10.150.20.252D open2D periodicdirectly computed from any displacement δR(cid:105) = {δRi} where i labels particles or nodes, as shown in Supplemen- tal Material Section C or Ref. [33]. Removing the trace ↔ (i)]1, leads to a local shear tensor where 1 is a d × d identity matrix. It's useful to define scalar observables to visualize the strain, in particular the shear intensity Eshear(i) (not sensitive to compression or dilation) and the bulk intensity Ebulk(i) (sensitive to it) as [10]: ↔ (i) − 1 ↔ γ (i) = d tr[ d(cid:88) d(cid:88) l,m=1 l=1 Eshear(i) = Ebulk(i) = 1 2 1 2 [γlm(i)]2; [ll(i)]2. 4 (7) The extendedness of the vibrational modes is charac- terized by the participation ratio, defined as: (cid:33)−1 (δRω(i) · δRω(i))2 (cid:32) (cid:88) for normalized modes(cid:80) Pω = N i i δRω(i)2 = 1. Translations have a unity participation ratio. By contrast, if a mode only involves the motion of ∼ N0 particles, then Pω ∼ N0/N . Conservation: We quantify the local conservation of the structure by considering the mean occupancy, defined over a period of observation τ : (3) (cid:104)σα(cid:105) ≡ 1 τ σα(t). (8) τ(cid:88) t=1 Rigidity of the structure: For elastic networks, as understood by Maxwell an important aspect of rigidity is the coordination number z(i), counting the local con- nectivity (number of springs) attached to a node i. This notion, sufficient in our model, can be extended to inter- actions relevant in proteins as discussed in [34]. If there is no selection pressure on that link, we expect ¯σ. We thus define the conservation Σ to quantify the deviation from this average [23]: Σα = (cid:104)σα(cid:105) ln (cid:104)σα(cid:105) ¯σ + (1 − (cid:104)σα(cid:105)) ln 1 − (cid:104)σα(cid:105) 1 − ¯σ . (9) Another commonly used observable is the B-factor or Debye-Waller factor [35]. It characterizes the mean square thermal fluctuations of the particle positions. In a harmonic approximation it can be expressed in terms of the vibrational modes (neglecting a temperature- dependent pre-factor): B(i) = 1 ω2 δRω(i) · δRω(i), (4) (cid:88) ω>0 where the ωs and δRω are frequencies and the corre- sponding vibrational modes, that are obtained from the diagonalisation of the stiffness matrix. B-factors however may not pick up the interesting flex- ibility of the structure. For example if a hinge connects two rigid parts, B-factors may be large in the rigid parts too as it is sensitive to rigid motions as well. Here we in- troduce an observable that would reveal the presence of a hinge, as it characterizes the thermal fluctuations of the strain (which is therefore zero by construction for rigid body). We call it the strain B-factor, which for harmonic dynamics follows: SB(i) = 2 ω2 [Eshear,ω(i) + Ebulk,ω(i)], (5) (cid:88) ω>0 where Eshear,ω and Ebulk,ω are the shear and bulk inten- sities for a given mode δRω(i), as defined from Eqs.(3). Spectral analysis: The response to binding can be decomposed into the vibrational modes [16], which form a complete orthogonal basis. We define the overlap: qω = (cid:104)δRAl r δRω(cid:105)2/δRAl r 2, (6) that satisfies(cid:80) ω qω = 1. RESULTS We now document examples of architectures generated by our scheme, focusing on shear, hinge and twist designs. We consider individual families: when average quantities are presented, they always correspond to a time aver- age over the last third of our Monte-Carlo algorithm, as previously described. We then emphasize the features common to all these designs, to be explained in the next section. Shear design: We start by the two-dimensional case where visualization is easier. If periodic boundary con- ditions are considered on the horizontal axis (cylindrical geometry), we find that all 25 architectures correspond to a shear design. This is illustrated in Fig. 4A showing the response to binding: except for a linear path connecting the allosteric and active sites, the motion is essentially that of a rigid body (pure rotations and translations). This is most obvious when plotting the map of the shear intensity Eshear in Fig. 4B, which is essentially zero ex- cepted along that path. Overall, the design is similar to that of the mint box illustrated in Fig. 4C, where strain also localizes on a hyperplane (a line for d = 2 and a plane fr d = 3). At the structural level, we find that the strain path corresponds to a softer region with lower coordination as shown in Fig. 4D and a larger strain B- factor, as illustrated in Fig. 4E. Hinge design. When open boundaries (instead of pe- riodic ones) are used, we find that about 40 to 50 percent of the families lead to hinge architectures, and the rest display a shear design. In the former case, the response exemplified in Fig. 5A can be decomposed into the mo- tion of two rigid bodies connected by a hinge. Again 5 FIG. 4: Shear design: (A) The average cooperative response δRAl arrows. (B) The average shear intensity map Eshear reveals strain localization along a path. (C) A mint box that opens by sliding illustrates the shear mechanism. (D) Map of the average coordination number z. (E) Map of the average strain B-factor SB. (F) Map of the fitness cost of single site mutation normalized by its absolute value ∆F/F. (G) Map of the conservation Σ in the evolution simulation. (H) Overlap qω between the response and the vibrational modes, colored as a function of the participation ratio Pω, showing that a single extended mode dominates the response to binding. induced by binding at the allosteric site is shown in black r this is most apparent in the map of the shear intensity in Fig. 5B, showing that there is little strain excepted for two disconnected regions near the allosteric and active sites. There is thus no connecting path of high strain be- tween these sites. This design is common in our daily life, as illustrated by the clothespin in Fig. 5C. At the struc- tural level, the map of coordination shown in Fig. 5D and that of strain B factor shown in Fig. 5E display a "H" shape with two rather disconnected region being weakly coordinated with a high strain B-factor. Twist design. In three dimensions we find a rich va- riety of architectures, whose structure and response are sometimes hard to describe. Here we present the sim- ple case of a twist architecture, as illustrated in Fig. 6A with the Rubik Cube. To visualize this design, we con- sider the shear intensity in three sections parallel to the x-z plane as illustrated in Fig. 6B. We find that there is little strain except on the central plane connecting the al- losteric (purple) and active (blue) sites shown in Fig. 6C. There is not however a homogeneous shear on that plane: instead, the strain is low at its center and larger near the boundaries. Further evidence for the twist design appears in the allosteric response itself shown in Fig. 6D with the same slicing geometry: the two side planes show reverse rotating motions, whereas the middle plane shows a more complex displacement pattern. Once again, the structural analysis confirms this view: we find that the coordination is large and the strain B-factor is small ex- cept near the boundaries of the central plane, as shown in Figs. 6(E-H). The middle of the central plane thus acts as a well-connected joint around which two quite rigid bodies can rotate. Universal features of cooperative designs: Our in-silico evolution scheme generates different designs, as illustrated with the examples above. However, all these designed architectures follow the same principles, which 6 FIG. 5: Hinge design: (A) The averaged cooperative response δRAl black arrows. (B) Shear intensity Eshear of the response. (C) A clothespin illustrates the hinge mechanism. (D) Map of the average coordination number z. (E) Map of the average strain B-factor SB. (F) Map of the fitness cost of single site mutation normalized by its absolute value ∆F/F. (G) Conservation Σ. (H) Decomposition qω of the response on the vibrational modes ω, colored as a function of the participation ratio Pω. induced by binding at the allosteric site is shown in r we list in the following. These principles are system- atically tested by averaging on the 25 families found in two-dimensions with periodic boundaries in Fig.7. The same analysis holds for other boundary conditions and in three dimensions as well, as documented in Supplemental Material Section E: • The system separates into a rigid and a soft man- ifold, as observed in a class of proteins [36] and in protein models [20]. • The strain associated with the allosteric response is small in the rigid manifold (indicating rigid body or long-wavelength motion), while it is large in the soft manifold. Both properties are apparent in Fig. 7A, showing the two-dimensional density of nodes found with a given strain B-factor (reflecting the local rigidity) and strain intensity (reflecting the strain induced by the allosteric response). This histogram displays a branch of soft nodes, where the strain B-factor is large and positively correlated to the strain intensity. • In all cases, the mutation cost is high precisely in these locations where the system is soft and where the strain intensity is large, as illustrated in Fig. 7B. • Most importantly, the daily-life examples we pro- vided all have a common point: they display a sin- gle mechanism, i.e. a very soft elastic mode. We ob- serve that this is also true in our cooperative archi- tectures: there is always a single soft and extended mode along which most of the response projects to. This fact is already apparent in the decomposi- tion of the response on vibrational modes shown in Figs. 4H, Fig. 5H and Fig. 6K. It is studied system- atically in Fig. 7C and Fig. 7D showing respectively the density of vibrational modes D(ω, Pω) and the overlap q(ω, Pω) as a function of both frequency ω 7 FIG. 6: Twist design: (A) Illustration of a Rubik Cube and its twist mechanism. (B) Two-dimensional sections of the shear strain intensity. The allosteric and active sites are shown in purple and blue respectively. (C) Shear intensity Eshear in the central section. (D) Response δRAl to binding in the same three distinct sections, organized from left to right as in (B). Maps of the average coordination number z on (E) the three sections and (F) the central section. The strain B-factor SB is shown on the three sections (G) and the central one (H). Fitness cost of single site mutation normalized by its absolute value ∆F/F on (I) the three sections and (J) the central one. (K) Decomposition qω of the response on the vibrational modes vs the mode frequency ω, colored as a function of their participation ratios Pω, at two different time points during the run. In three dimensions, most of the spectral decomposition resembles the right panel where several vibrational modes project on the response, although we can always identify time points where a single mode contributes as shown on the left panel. r and participation ratio Pω. Fig. 7C shows a peak of extended (large Pω) modes at low ω, Fig. 7D shows that most of the response projects precisely on these modes. We find that essentially one mode governs the response. This result can also be vi- sualized by classifying modes for each system by decreasing overlap q, and by representing the cu- mulative overlap (the sum of qω for the r modes with the largest overlap) as a function of the rank r, as illustrated in Fig. 7E. In average, the first mode captures more than 90% of the response. It is interesting to note that many properties of mate- rials optimized to be cooperative, whose specific property is to display a single soft elastic mode controlling func- tion, differ from materials studied previously optimized to propagate a given strain - below we will refer to both cases as "cooperative" and "geometric" designs. An ex- tensive comparison is performed in Supplemental Mate- rial Section F and G. Salient differences include that: (a) the magnitude of the response is essentially constant in space in cooperative designs (it decays by five fold or more in geometric designs) (b) the cooperative design is symmetric: binding at the allosteric or at the active site leads to a very similar response (whereas elastic in- formation cannot propagate from the active site to the allosteric site in geometric designs) (c) the cooperative design responds much more specifically than the geomet- ric ones (in the latter case, imposing a strain anywhere in the material typically lead to a strong displacement at the active site) and (d) for geometric designs, the re- sponse does not correspond to a single soft elastic mode, but to a few of them, as already apparent in Fig. 7E. To explain the universal features of cooperative de- signs, and to predict the frequency of the soft extended mode controlling the response, we now investigate the optimality of designs. THEORY Absence of design: We now argue that in a contin- uous elastic medium - where no design is involved - cooperativity decreases very rapidly with the distance L between the allosteric and active sites. Any imposed lo- cal strain can be decomposed into multipole moments (dipole and higher), and the slower decaying response in the far field - sufficiently distant from the source - is dipolar, since higher multipoles decay faster. To model the perturbation induced by ligand binding, we may thus consider without loss of generality two dipoles each of magnitude f c, where f is the applied force and c the dis- tance over which these are exerted. Here we give a simple scaling argument for a medium with elastic modulus G. As mentioned earlier, for L (cid:29) c we have Ecoop ∼ (cid:104)dRF(cid:105) where dR(cid:105) is now the dipolar response induced by the rd−1G , and F(cid:105) the first dipole, of magnitude dR(r) ∼ f c force field of the second dipole. Since F(cid:105) is dipolar its scalar product on dR(cid:105) acts as a derivative taken at r = L, and one obtains Ecoop ∼ f 2c2 GLd ∼ L−d, i.e. a very rapid decay with distance. This result is confirmed numerically for the case of a crystalline network in the Section D of Supplemental Material. Illustration of optimal cooperativity: Shear ar- chitecture. We now show that cooperativity can be greatly improved if the material presents a very soft ex- tended mode. For illustration we consider the geometry of Fig.8 where a cylinder of elastic modulus G is cut on its length L, by a band of width c. This generates a zero mode corresponding to the rotation of a square. If dis- placements at the active or allosteric sites of size δ are imposed as illustrated in Fig.8, they will only couple to that mode (since it costs no energy), and lead to the same response. This statement will be true even if the band of width c is filled up with soft material of elastic modulus 8 FIG. 7: Histogram of network nodes displaying (A) a given strain B-factor SB and shear intensity Eshear (showing that most of the strain induced by the response to binding occurs in regions where the material is soft) and (B) a given shear intensity Eshear and normalized fitness cost −∆F/F (showing that mutations are costly where the response strain is localized). The color bar indicates the relative abundance of the data points. (C) Density of vibrational modes D(ω, Pω) and (D) the overlap q(ω, Pω) as a function of both frequency ω and participation ratio Pω, revealing the presence of a soft extended mode on which most of the response projects to. For (A,B,C,D), the statistics is done over all 25 families of solutions found in the cooperative task in two dimension with a periodic boundary. (E) Cumulative overlap on the first r modes with strongest overlap, where r is denoted the rank. Results are shown both for the cooperative and the geometric tasks, for all dimensions and boundary conditions. Gw, as long as it is small enough (see below). Thus we have EAc ≈ EAl ≈ EAc,Al implying Ecoop ≈ EAl, which can be readily estimated as the amount of elastic energy stored in the soft band, i.e. Ecoop ∼ L c Gwδ2. This results implies that Ecoop = 0 when the material presents a mechanism (i.e. Gw = 0), but increases with Gw. This argument eventually breaks down, however, when it becomes more favorable to deform the rigid ma- terial and to couple to other modes in the system. This takes place when the energy of deforming a continuous moderank100101cumulativeoverlapq00.20.40.60.81Cooperatived=2periodicCooperatived=2openCooperatived=3openGeometricd=2periodicGeometricd=3openEFIG. 8: In a cylindrical geometry, a mechanism - or zero mode - can be constructed by slicing the cylinder, as can be achieved by creating a cut of length L and width c. One then obtains an object with the topology of a square, which now displays an additional zero mode corresponding to a rigid rotation. If the cut is filled up with a soft elastic material, the mode gets a finite frequency. As long as it is small, imposing a local displacement as indicated in the figure at the allosteric or at the active site will be dominated by this mode and will lead to essentially the same response. Thus EAc ≈ EAl ≈ EAc,Al and Ecoop ≈ EAl. 9 medium of modulus G, Econt ∼ Gδ2/ ln(L/c) becomes smaller than the energy associated with the soft mode L c Gwδ2. Comparing these two expressions we get a cross- over for Gw = G∗ w with: w ∼ G∗ cG L ln(L/c) (10) For Gw (cid:29) G∗ w, the role of the soft mode become negligi- ble and the system will respond as a homogeneous elas- tic material (whose cooperativity is small as described above). Thus cooperativity will be maximal for Gw ≈ G∗ w, leading to an optimal cooperativity of order: coop ∼ Gδ2 E∗ ln(L/c) (11) This result is confirmed numerically in the Section D of Supplemental Material. The small energy of the response to binding for large L described by Eq.11 implies the presence of a soft elas- tic mode, which is relevant experimentally. It can be detected in the vibrational spectrum of the protein, and implies large thermal fluctuations. Such fluctuations, in a harmonic approximation, are inversely proportional to the corresponding eigenvalue of the stiffness matrix, of or- der λ∗ ≈ E∗ coop/δR2 where δR2 is the square norm of the allosteric response. For the shear mode considered δR2 ∼ L2δ2 since all particles are moving by a dis- tance of order δ, leading to λ∗ ∼ 1/(L2 ln(L/c)). For the vibrational spectrum such a small eigenvalue will lead to a low frequency ω∗. Assuming for simplicity that all the particles have identical mass leads to: ω∗ ∼(cid:112)λ∗/m ∼ 1 L ln1/2(L/c) (12) FIG. 9: (A) We build a shear architecture using a triangular lattice with a soft band, where the springs have a stiffness kw (cid:28) k = 1 such that the network modulus is proportional to the spring stiffness Gw/G = kw/k. The imposed displacement at the allosteric site is shown in purple arrows, and the associated response in black. Here kw = 0.05, L = 16 and c = L/10. (B) Energy of simultaneous binding EAc,Al and cooperative energy Ecoop versus kw for L = 32. We confirm that Ecoop depends non-monotonically on kw. (C) Overlap between the response and the eigenmodes qω vs mode frequency ω at optimal k∗ w = 0.036 for L = 32 and c = L/10, colored as a function of their participation ratio Pω. which is thus much softer that the lowest-frequency plane wave modes, of frequency 1/L [? ]. It is straightforward to extend these results to three dimensions in the geometry of a shear plane, where we coop ∼ Gcδ2 which does not decay with distance, find E∗ and ω∗ ∼ L−3/2 which is now even much smaller than plane waves modes, thus justifying why the spectrum of our materials show an isolated soft extend mode at low frequency. These results are tested in Fig. 9 for d = 2, which confirms that cooperativity is optimal for a finite frequency of the soft extended mode. Principles of optimal cooperativity: Overall, the common principle emerging from this study is that op- timal cooperativity results from the following antagonist effects. On the one hand, the architectures are such that they nearly present an extended mechanism. Because this mode is much softer than others, an imposed strain strongly couples to it, thus allowing to transfer the elas- tic information over long distances. On the other hand, if this extended soft mode is too soft, the elastic costs activeallostericc δ Lkw10-310-210-1100E10-310-210-1100EAlAcEcoopω10-210-1100overlapqω00.20.40.60.8100.10.20.30.40.5BCAE∝kwassociated with binding become too small, leading to a small cooperative energy. As a result, there is an optimal frequency scale for cooperativity. This idea leads to a natural explanation for the em- pirical facts listed in the introduction. Indeed shear and hinge designs (i) are clear realizations of this principle, which implies the presence of a soft extended modes at low frequency, consistent with observation (ii). We expect that our main result, i.e. the existence of an optimal vibrational frequency for cooperativity, will hold true when non-linearities are taken into account. This prediction can be tested using a combination of molec- ular dynamics (MD) and experiments. MD can be used to measure the effect of point mutations on the thermal fluctuations along the relevant normal mode, and exper- iments can measure the effect of the same mutation on cooperativity. In the spirit of Fig.9.B, we predict that there is an optimal magnitude of fluctuations for cooper- ativity to function properly. It would be very interesting to test if proteins function close to this optimum. Fluctuation-driven cooperativity: Finally, as pointed out in [18, 20], the existence of a soft extended modes of frequency ω∗ leads to the possibility of a co- operative effect with no mean displacement at play, once thermal effects are accounted for (iii). Indeed binding at the active site will hinder motion and increase the soft mode frequency, leading to an entropic cost that can be diminished if binding already took place at the allosteric site. Let us define ωAl, ωAc and ωAc,Al the frequencies of the soft mode after binding at the allosteric site, ac- tive site and both respectively. We can estimate these quantities as ω2Al = ω∗2 + eAl, ω2Ac = ω∗2 + eAc and ω2Ac,Al = ω∗2 + eAl + eAc where eAl (eAc) characterizes the additional energy required for the mode to move when a ligand is bound at the allosteric (active) site. Assum- ing harmonic dynamics, the entropy of a normal mode of frequency ω reads S = kB ln(kBT /ω). Using this expression, one can now estimate the cooperative free energy ∆∆F = −T ∆∆S = kBT ln(ωAc,Alω∗/ωAlωAc) = −kBT ln(1 − eAceAl/(ω∗2 + eAl)(ω∗2 + eAc)) which can indeed be large if ω∗2 is small compared to both eAl and eAc. CONCLUSION AND OUTLOOK We have used in-silico evolution to design materials which are highly cooperative. Strikingly, the architec- tures found differ greatly from materials optimized to propagate a geometrical information over long distances. The latter architectures are based on the emergence of a lever that amplifies the mechanical signal where it is de- sired, which may be relevant in proteins whose task is to trigger large motions when a ligand binds - e.g. to close an ion channel. By contrast, we predict that proteins optimized to be cooperative should display different ar- 10 chitectures, including shear and hinge designs which are well-known in the literature. Intriguingly, we find that there is a great variety of possible functioning architec- tures, especially in the three dimensional case. However, they all function along the same principle: they nearly display an extended mechanism, whose frequency should be neither too large nor too small for optimal function to occurr. Our approach rationalizes several empirical observa- tions on allosteric proteins and it also makes testable predictions. In particular, we predict that a single soft extended mode contributes to function, whose frequency should decrease with protein size. We find that this pre- diction is hard to test stringently from a spectral decom- position of the allosteric response alone, because localized soft modes (typically near the surface of the system) can hybridize with the relevant mode if they lie at similar fre- quencies. As a result, the response appears to project on a few modes (despite the localized modes being irrelevant for function) instead of one. Recent methods have been developed in computer sci- ence to clean-up spectra of localized modes - see e.g. [37] in the field of community detection. An exciting path forward is to adapt these methods to proteins, allowing one to test if a single extended mode indeed contributes to allostery. Ultimately, this suggests a mechanical ap- proach to discover de novo allosteric proteins, as those in which a single extended mode lies at low frequency in the cleaned-up spectrum. Such an analysis would fur- ther predicts where mutations would affect function: we have observed that most damaging mutations hinder the allosteric response, and take place where the extended mode generates high shear. SUPPORTING CITATIONS Reference [38] appears in the Supporting Material. AUTHOR CONTRIBUTIONS L.Y. and M.W. conceived the project and M.W. su- pervised research. L.Y., R.R. and M.W. developed the theory. L.Y., R.R. and C.B. performed the computations and developed the numerical methods. All authors an- alyzed the results and wrote the final manuscript. L.Y. and R.R. contributed equally to this work. ACKNOWLEDGMENTS We thank J.P. Bouchaud, B. Bravi, S. Cocco, T. De Geus, P. De Los Rios, S. Flatt, W. Jie, D. Malinverni, R. Monasson, M. Popovi´c, S. Zamuner, Y. Zheng for discus- sions. L.Y. is supported by the Gordon and Betty Moore Foundation under Grant No. GBMF2919 and in part by the National Science Foundation under Grant No. NSF PHY-1748958. M.W. thanks the Swiss National Science Foundation for support under Grant No. 200021-165509 and the Simons Foundation Grant (#454953 Matthieu Wyart). This material is based upon work performed using computational resources supported by the "Cen- ter for Scientific Computing at UCSB" and NSF Grant CNS-0960316, and by the "High Performance Computing at NYU". ∗ These two authors contributed equally to this work † [email protected] [1] Monod, J., J. Wyman, and J.-P. Changeux, 1965. On the nature of allosteric transitions: a plausible model. Journal of Mol. Biol. 12:88–118. [2] Changeux, J.-P., and S. J. Edelstein, 2005. Allosteric mechanisms of Signal Transduction. Science 308:1424– 1428. [3] Amor, B. R., M. T. Schaub, S. N. Yaliraki, and M. Bara- hona, 2016. Prediction of allosteric sites and mediat- ing interactions through bond-to-bond propensities. Nat. Commun. 7. [4] Halabi, N., O. Rivoire, S. Leibler, and R. Ranganathan, evolutionary units of three- 2009. Protein sectors: dimensional structure. Cell 138:774–786. [5] Nussinov, R., and C.-J. Tsai, 2013. Allostery in disease and in drug discovery. Cell 153:293–305. [6] Liang, J., and K. A. Dill, 2001. Are proteins well-packed? Biophys. J. 81:751–766. [7] Gerstein, M., A. M. Lesk, and C. Chothia, 1994. Struc- tural mechanisms for domain movements in proteins. Biochemistry 33:6739–6749. [8] Perutz, M., 1970. Stereochemistry of Cooperative Ef- fects in Haemoglobin: Haem–Haem Interaction and the Problem of Allostery. Nature 228:726–734. [9] Xu, C., D. Tobi, and I. Bahar, 2003. Allosteric changes in protein structure computed by a simple mechanical model: hemoglobin T-R2 transition. Journal of Mol. Biol. 333:153–168. [10] Mitchell, M. R., T. Tlusty, and S. Leibler, 2016. Strain analysis of protein structures and low dimensionality of mechanical allosteric couplings. Proc. Natl. Acad. Sci. 201609462. [11] Goodey, N. M., and S. J. Benkovic, 2008. Allosteric reg- ulation and catalysis emerge via a common route. Nat Chem Biol 4:474–482. [12] Gandhi, P., Z. Chen, F. Scott Mathews, and E. Di Cera, 2008. Structural identification of the pathway of long- range communication in an allosteric enzyme. Proc. Natl. Acad. Sci. 105:1832–1837. [13] McLaughlin Jr, R. N., F. J. Poelwijk, A. Raman, W. S. Gosal, and R. Ranganathan, 2012. The spatial architec- ture of protein function and adaptation. Nature 491:138– 142. [14] Atilgan, A., S. Durell, R. Jernigan, M. Demirel, O. Ke- skin, and I. Bahar, 2001. Anisotropy of fluctuation dy- namics of proteins with an elastic network model. Bio- phys. J. 80:505–515. 11 [15] De Los Rios, P., F. Cecconi, A. Pretre, G. Dietler, O. Michielin, F. Piazza, and B. Juanico, 2005. Functional dynamics of PDZ binding domains: a normal-mode anal- ysis. Biophys. J. 89:14–21. [16] Zheng, W., B. R. Brooks, and D. Thirumalai, 2006. Low- frequency normal modes that describe allosteric transi- tions in biological nanomachines are robust to sequence variations. Proc. Natl. Acad. Sci. 103:7664–7669. [17] Popovych, N., S. Sun, R. H. Ebright, and C. G. Kalodi- mos, 2006. Dynamically driven protein allostery. Nature Structural & Mol. Biol. 13:831–838. [18] Cooper, A., and D. Dryden, 1984. Allostery without conformational change. Eur. Biophys. J. 11:103–109. [19] Tsai, C.-J., A. del Sol, and R. Nussinov, 2008. Allostery: Absence of a Change in Shape Does Not Imply that Al- lostery Is Not at Play. Journal of Mol. Biol. 378:1–11. [20] McLeish, T. C., T. Rodgers, and M. R. Wilson, 2013. Al- lostery without conformation change: modelling protein dynamics at multiple scales. Phys. Biol. 10:056004. [21] Hemery, M., and O. Rivoire, 2015. Evolution of sparsity and modularity in a model of protein allostery. Phys. Rev. E 91:042704. [22] Tlusty, T., A. Libchaber, and J.-P. Eckmann, 2017. Phys- ical Model of the Genotype-to-Phenotype Map of Pro- teins. Phys. Rev. X 7:021037. [23] Yan, L., R. Ravasio, C. Brito, and M. Wyart, 2017. Ar- chitecture and coevolution of allosteric materials. Proc. Natl. Acad. Sci. 114:2526–2531. [24] Rocks, J. W., N. Pashine, I. Bischofberger, C. P. Goodrich, A. J. Liu, and S. R. Nagel, 2017. Designing allostery-inspired response in mechanical networks. Proc. Natl. Acad. Sci. 114:2520–2525. [25] Flechsig, H., 2017. Design of Elastic Networks with Evolutionary Optimized Long-Range Communication as Mechanical Models of Allosteric Proteins. Biophys. J. 113:558 – 571. [26] Sigmund, O., and K. Maute, 2013. Topology optimiza- tion approaches. Struct. Multidiscip. Optim. 48:1031– 1055. [27] Sigmund, O., 1997. On the Design of Compliant Mecha- nisms Using Topology Optimization. Mechanics of Struc- tures and Machines 25:493–524. [28] Nishiwaki, S., M. I. Frecker, S. Min, and N. Kikuchi, 1998. Topology optimization of compliant mechanisms using the homogenization method. International Journal for Numerical Methods in Engineering 42:535–559. [29] Yan, L., J.-P. Bouchaud, and M. Wyart, 2017. Edge mode amplification in disordered elastic networks. Soft Matter 13:5795–5801. [30] Yan, L., and M. Wyart, 2014. Evolution of covalent net- works under cooling: contrasting the rigidity window and jamming scenarios. Phys. Rev. Lett. 113:215504. [31] Yan, L., and M. Wyart, 2015. Adaptive elastic networks as models of supercooled liquids. Phys. Rev. E 92:022310. [32] Maxwell, J., 1864. On the calculation of the equilibrium and stiffness of frames. Philos. Mag. 27:294–299. [33] Gullett, P., M. Horstemeyer, M. Baskes, and H. Fang, 2007. A deformation gradient tensor and strain ten- sors for atomistic simulations. Modell. Simul. Mater. Sci. Eng. 16:015001. [34] Jacobs, D. J., A. J. Rader, L. A. Kuhn, and M. F. Thorpe, 2001. Protein flexibility predictions using graph theory. Proteins: Struct., Funct., Bioinf. 44:150–165. [35] Yang, L., G. Song, and R. L. Jernigan, 2009. Protein elastic network models and the ranges of cooperativity. Proc. Natl. Acad. Sci. 106:12347–12352. [36] Tama, F., F. X. Gadea, O. Marques, and Y.-H. Sane- jouand, 2000. Building-block approach for determining low-frequency normal modes of macromolecules. Pro- teins: Struct., Funct., Bioinf. 41:1–7. [37] Zhang, P., 2016. Robust Spectral Detection of Global Structures in the Data by Learning a Regularization. In D. D. Lee, M. Sugiyama, U. V. Luxburg, I. Guyon, and R. Garnett, editors, Advances in Neural Information Pro- cessing Systems 29, Curran Associates, Inc., 541–549. [38] Landau, L., and E. Lifshitz, 1986. Theory of Elasticity, volume 7 of Course of Theoretical Physics. Pergamon Press, Oxford, U.K. SUPPLEMENTARY MATERIAL A. Embedding lattice 2D triangular lattice. In our model, we introduce a slight distortion of the lattice to remove long straight lines that occur in a triangular lattice. Such straight lines are singular and lead to unphysical localized floppy modes orthogonal to them. One can remove them by imposing a random displacement on the nodes. Instead, we distort the lines without introducing frozen disorder. We group nodes in lattice by four, labeled as A B C D in Fig. S1. One group forms a cell of our distorted lattice. In each cell, node A stays in place, while nodes B, C, and D move by some distance δ: B along the direction perpendicular to BC, C along the direction perpendicular to CD, and D along the direction perpendicular to DB, as illustrated. We set δ to 0.2, where the straight lines are maximally reduced with this distortion. 3D face-centered cubic (FCC) lattice. We intro- duce a similar distortion to the FCC lattice. Again, we label the lattice nodes into four different types A B C D, as shown in one layer in the z direction. The nodes are labeled in such a way that all 12 nearest neighbors of a node are different from it. For example, the center node in the bottom panel of Fig. S1, labeled as C, is connect- ing to two As and two Bs (in solid lines) in the layer and four Ds with two other As and Bs (in dashed lines) out of the layer. To each layer in z direction, there are two other layers, and in those two layers, D and A, B are located at the same x and y. So we only see half of them (two D, one A and one B) connecting to C in a two di- mensional projection along the z direction in Fig. S1. We move all As along negative y direction, all Bs along pos- itive x direction, all Cs along (− 1√ ), and Ds along ( 1√ ) by δ = 0.2. As shown in Fig. S1, 6 all straight lines are thus perturbed without introducing quenched disorder. ,− 1√ ,− 1√ , 1√ 3 , 1√ 2 6 2 3 12 FIG. S1: Illustration of the distorted triangular lattice (top) and distorted FCC lattice (bottom). B. Linear response of elastic networks Stiffness matrix. Consider a displacement field δ (cid:126)Ri ≡ (cid:126)Ri − (cid:126)Ri0, where (cid:126)Ri0 is the position of the node i in the initial mechanical equilibrium. To the first order in δ (cid:126)Ri, the distance among neighboring nodes, defined as r(cid:104)ij(cid:105) ≡ (cid:126)Ri − (cid:126)Rj between node i and j, changes by δr(cid:104)ij(cid:105) = r(cid:104)ij(cid:105) − r(cid:104)ij(cid:105),0 = S(cid:104)ij(cid:105),lδ (cid:126)Rl + o(δ (cid:126)R2). (S1) l S(cid:104)ij(cid:105),• = n(cid:104)ij(cid:105)((cid:104)i − (cid:104)j), where n(cid:104)ij(cid:105) is the unit vector along link (cid:104)ij(cid:105) from j to i, is the structure matrix. On the other hand, the force on a node is a composition (cid:88) of tensions, (cid:126)Fi = (cid:88) j (cid:88) (cid:104)lm(cid:105) n(cid:104)ij(cid:105)f(cid:104)ij(cid:105) = S(cid:104)lm(cid:105),if(cid:104)lm(cid:105). (S2) For linear springs on the neighboring connections, f(cid:104)ij(cid:105) = k(cid:104)ij(cid:105)δr(cid:104)ij(cid:105), the response force to the displacement is, F(cid:105) = MδR(cid:105), the stiffness (cid:80)(cid:104)lm(cid:105) k(cid:104)lm(cid:105)S(cid:104)lm(cid:105),iS(cid:104)lm(cid:105),j, depends only on connec- where = tion σ(cid:105) and the link directions. The elastic energy corresponds to the displacement field δR(cid:105) of dimension N d is matrix Mi,j (S3) E = (cid:104)FδR(cid:105) = 1 2 1 2 (cid:104)δRMδR(cid:105). (S4) DBACδδDBDABCABDALinear response to an imposed displacement. When we impose a displacement on the subset E of NE nodes, δRE , forces must be applied on these nodes. All other nodes adapt to a new mechanical equilibrium with no net forces on them, and follow a displacement δRr. Thus Eq.(S3) becomes, for this choice of basis: (cid:18) δRE (cid:19) (cid:18) δRE δRr . (cid:126)0 (cid:19) = Q−1M (cid:19) = M (cid:18) (cid:126)F which leads to:(cid:18) (cid:126)F (cid:126)0 with δRr Qij = (cid:19) (cid:26) δij if j ∈ E if j (cid:54)∈ E . −Mij (S5) (S6) (S7) When there are floppy modes in the network, linear equa- tion (S6) may not be solvable. In that case, Q−1 should be understood as the pseudo-inverse so that the net- work does not respond along the floppy directions (cor- responding singular values are zero in Q). Another pos- sibility is to reduce singularity by imposing that each node also interacts with all its next nearest neighbors via weak springs of stiffness kw (cid:28) 1. Both methods lead to qualitatively identical results. For numerical costs, our results were computed using the second approach with wSw. kw = 10−4. So our stiffness matrix M = S t Regarding translations and rotations. When binding a ligand, the translational and rotational degrees of freedom (TR) of the nodes are not determined. If we write the TR degrees of freedom at the imposed nodes as ΨE , a dNE × dT R matrix, which is a set of vectors with dT R = 6 in d = 3 and dT R = 3 in d = 2, any imposed displacement giving the same shape change is then, σSσ +kwS t δRE = δRE 0 + ΨE · (cid:126)c. (S8) where δRE 0 is purely determined by the shape change, 0 · ΨE = (cid:126)0, and (cid:126)c is a parameter vector of dimension δRE dT R to count TR contribution additional to the shape change. We can thus consider a new basis with a dNE by dNE transform matrix U to the original space on imposed nodes so that δRE = U , (S9) (cid:19) (cid:18) (cid:126)δ0 (cid:126)c translations and rotations are isolated from the shape change defined by (cid:126)δ0. In this new basis, the forces (cid:126)F imposed on E obey total force and torque balance, (cid:18) (cid:126)f (cid:19) (cid:126)0 U t (cid:126)F = The linear response problem thus becomes, Q where with  (cid:126)f (cid:126)c δRr (cid:18) U t 0 0 I  = (cid:26) δij (cid:18) U t 0 − Mij 0 I Qij = M = (cid:19) M (cid:18) δRE 0 (cid:19) (cid:126)0 if j ∈ E \ T R otherwise (cid:19) M (cid:18) U 0 0 I (cid:19) . 13 (S11) (S12) (S13) Note that given the separation of the two subspaces the matrix U is of dimension dNE × dNE and the matrix I is d(N − NE ) × d(N − NE ), consistent with M being dN × dN . C. Computing the local strain tensor in a network In a continuous medium, a motion maps a point (cid:126)X in the reference configuration to a new point (cid:126)x in the current configuration, the strain tensor of the motion can thus be computed as, (cid:18) ∂(cid:126)x ∂Xa (cid:19) ab( (cid:126)X) = 1 2 · ∂(cid:126)x ∂Xb − δab , (S14) where a, b labels the spatial dimension. ∆(cid:126)xij = In a discrete medium as networks, the problem is to ↔ Λ = ∂(cid:126)x/∂ (cid:126)X at node i for compute the partial derivative especially non-lattice structures. Ideally, for any neigh- bor j close enough in space, ↔ Λi · ∆ (cid:126)Xij, (S15) where ∆ (cid:126)Xij = (cid:126)Ri0 − (cid:126)Rj0 and ∆(cid:126)xij = (cid:126)Ri − (cid:126)Rj in our ↔ model. We have nb number of such equations for Λi ↔ when nb neighbors are considered. So Λi are usually over-determined when we consider all nearest neighbors (nb = 6 for a 2 × 2 matrix in triangular lattice, and nb = 12 for a 3 × 3 matrix in FCC lattice). Instead of solving Eq.(S15), we define a mean squared error function [33], M SE(i) = (∆(cid:126)xij − ↔ Λi · ∆ (cid:126)Xij)2wj(i), (S16) (cid:88) j . (S10) where we have kept a weight function wj(i) of node j contribution to i in general. Specifically, we set wj(i) = 1 for all nearest neighbors to i on the original embedding nb D. Cooperative energy of two dipoles in a continuous elastic medium Elastic energy of a force monopole. We make the simplifying assumption that the velocity field is di- vergence free (relaxing this assumption will not change the predicted scaling behaviors). The equation for the displacement field when a monopole force (cid:126)f is applied to a constant force over a spherical patch of radius c then follows [38]: ∆(cid:126)u = ∇ · ∇(cid:126)u = − d GΩdcd (cid:126)f , (S19) where G is the shear modulus, Ωd is the solid angle of d dimensional sphere. Defining (cid:126)f = f ey, both force and displacement component are along y direction, we then solve for the divergence of the displacement field using Gauss Theorem, (cid:40) − f ∇uy = GΩdcd rer, GΩd rd−1 er. r ≥ c r < c − f 1 (S20) (cid:33) dr, (cid:90) R c (cid:32)(cid:90) c (cid:90) The total energy of the monopole is approximately, Em = G dd(cid:126)r(∇uy)2 = f 2 GΩd rd+1 c2d + 0 1 rd−1 (S21) where R defines the system size. In 2D, the integral is dominated by the second term, Em = f 2 2πG ln R c . (S22) d(cid:88) i=1 (cid:90) d(cid:88) i=1 (S25) where x+, x−, y+, y− are components in x and y direc- tions contributed by the + monopole and − monopole respectively in the dipole. So the dipole self-energy is, Ed = − 2f 2 Ω2 dG dd(cid:126)r x+,ix−,i, (S26) where the integral is over three regions, within c to the + monopole, within c to the − monopole, and the rest. One can show that the contributions of the first two regions inside monopoles scale as cd+2, while the contribution of the remaining region scales as cd+2, comparable. Outside of the monopoles, x•,i ≈ 1 (cid:90) R c Ed ∼ − f 2 G rd−1dr (cid:40) − f 2 rd xi, so r2d−2 ∼ 1 G ln R a d = 2 G c2−d d > 2 − f 2 . (S27) Cooperative energy of two force dipoles. Similar to the way we computed the dipole self-energy Eq.(S26), the cooperative energy, which is defined as the extra en- ergy from the interaction of two dipoles, can be computed as (cid:90) Ecoop = 2Ed − Etot = − 2f 2 Ω2 dG d(cid:88) dd(cid:126)r i=1 +,i − x0−,i)(xL (x0 +,i − xL−,i). (S28) where xL• are the contributions of the monopole at L. When c/L (cid:28) 1, the contribution outside of both the monopoles and the dipoles dominates the energy, lattice and wj = 0 otherwise. By minimizing the mean squared error with respect to ∆(cid:126)xij∆ (cid:126)Xijwj(i) · ∆ (cid:126)Xij∆ (cid:126)Xijwj(i) (cid:88) ↔ Λi = and j ↔ (i) = 1 2 ↔ Λi, we have (cid:88) (cid:18)↔ (cid:19) Λi − ↔ i · ↔ Λ δ j t −1 , (S17) Elastic energy of a force dipole. To compare with the mechanism discussed in the main text, we hereby compute the cooperative energy of two dipoles of size c separated by L in a homogeneous medium, as illustrated in Fig.8 (main text). Similar to the monopole energy computed above, we could define the dipole energy, (cid:90) . (S18) Ed = G dd(cid:126)r f 2 dG2 Ω2 (x+,i − x−,i)2 = 2Em + Ed, 14 In 3D and above, the integral of the second term con- verges in the large size limit R → ∞, and it has the same scaling as the first term, Ecoop ∼ f 2 G dzρd−2 c2 [ρ2 + z2]d/2[ρ2 + (L − z)2]d/2 ∼ f 2c2 GLd ln L c . (S29) Em = 2df 2 (d2 − 4)ΩdG c2−d. (S23) For given displacement δ applied at the dipoles, So the displacement δ can be achieved by an external force satisfying δ = ∂Em/∂f , Ecoop ∼ G c2δ2 L2 ln L G c2d−2δ2 c Ld d = 2 ln L c d > 2 (S30) δ = f δ = f ; ln R c 4d 1 πG (d2 − 4)ΩdG d = 2 d > 2 c2−d. (S24) showing that Ecoop decays as fast as L−d for two dipoles at a distance L from each other (with weak logarithmic corrections in d = 2). (cid:90) ∞ (cid:90) L dρ 0 c (cid:40) 15 FIG. S2: (A) Cooperative energy computed for a distorted crystal (δ = 0.2) of varying size L with no mechanism. (B) Inverse of the cooperative energy for a crystal with a soft shear band presenting a mechanism, the softness of the band being chosen as the value of kw where the cooperative energy is optimal, see Fig. 9. The two different scalings predicted from continuous elastic media are fitted and shown as solid lines. Numerical verification. In Fig. S2(A) we test our prediction for the cooperativity of a homogenous medium without any design (a distorted crystal with δ = 0.2), and confirm Eq. S30 for d = 2. In Fig. S2(B) we test our prediction for an optimal shear design, and confirm the very weal logarithmic decay of the cooperative energy in two dimensions, as described in Eq.11 in the main text. FIG. S4: Analysis of the cooperative task in two dimensions with open boundaries: histogram of network nodes displaying (A) a given strain B-factor SB and shear intensity Eshear (showing that most of the strain induced by the response to binding occurs in regions where the material is soft), (B) a given shear intensity Eshear and normalized fitness cost −∆F/F (showing that mutations are costly where the response strain is localized) and (C) a given Eshear and conservation Σ (showing that these same locations are highly conserved). (D) Density of vibrational modes D(ω, Pω) and (E) overlap q(ω, Pω) as a function of both frequency ω and participation ratio Pω, revealing the presence of a soft extended mode on which most of the response projects to. E. Principles of cooperative designs: numerical tests F. Geometric task Definition: networks perform the geometric task by minimizing a cost function that measures the deviation of the allosteric response δRAl r (cid:105) from a prescribed shape change located at the active site δRAc(cid:105) [23], i − Ui)2, (cid:115)(cid:88) (S31) E(σ) ≡ minU(cid:105) (δRAl r,i − δRAc i∈Ac where U(cid:105) is a global translation and rotation, which does not change the shape at the active site. Here Ac corre- sponds to four sites defining the active site. We illustrate the method by studying the case d = 2 with periodic boundaries, as well as d = 3 with free boundaries. Our results are averaged over 25 runs with different initial conditions in d = 2 and 10 runs in d = 3. d = 2: This case is documented in Fig. S5, using the observables introduced in the main text. The architec- ture presents two main features. Most importantly, the response is non-monotonic and strongly amplified close to the active site, as shown in Fig. S5B. As demonstrated in [23, 29], this effect is induced by the presence of a marginally connected region with z = 4 (Fig. S5D) which is thus very soft (Fig. S5E). It can be shown to act as a powerful lever [29]. It is also highly conserved and leads to high mutation costs (Fig. S5G,H). FIG. S3: Same analysis as Fig.S4 for d = 3 and open boundaries. In the main text we have listed the principles under- lying the cooperative architectures, and tested them in Fig.7 in two dimensions, with a periodic boundary. The same results hold with open boundaries in d = 2 (Fig.S4) and d = 3 (Fig.S3). AB16 FIG. S5: Two-dimensional geometric task with periodic boundaries. (A) The average response δRAl the allosteric site is shown in black arrows. (B) Map of the average magnitude of response δRr. (C) A fruit picker illustrates the combined mechanisms of edge mode lever and shear. (D) Map of the average coordination number z. (E) Map of the average strain B-factor SB. (F) Map of the average shear intensity Eshear. (G) Map of the fitness cost of single site mutation normalized by the cost of random networks ∆F/F . (H) Map of the average conservation Σ. (I) Decomposition qω of the response on the vibrational modes ω in a specific solution, colored as a function of the participation ratio Pω. induced by binding at r Another aspect of the observed design is the emergence of a shear mode close to the allosteric site, as can be seen from the mean response (Fig. S5A) and from the map of the shear intensity (Fig. S5F). This response is caused by the mergence of a weakly-coordinated band above the allosteric site (Fig. S5D). Overall, the design is thus similar to that of a fruit- picker (Fig. S5C) where a stimulus (violet arrows) leads to a shear (black arrows) that couples to a head (black ar- rows), which acts as a lever and leads to a specific desired response. This design leads to a more complex spectral signature where several modes typically contribute to the response (Fig. S5I), instead of one as for the cooperative designs discussed in the main text. d = 3: The arguably most relevant case corresponds to d = 3 with open boundaries, and is illustrated in Fig. S6. As shown in Fig. S6A, we study the response by focus- ing on two sections: a vertical plane passing through both the allosteric and active sites, and a horizontal plane containing the active site. Once again, we find that the central aspect of the design is the emergence of a weakly-connected region with z ≈ 6 (the isostatic value) surrounding the active site (Fig. S6C), and lead- ing to a very pronounced amplification of the response (Fig. S6B). This lever region is soft (Fig. S6D) and con- served (Fig. S6E,G). In that case, there is no evidence in the shear map of a hinge or shear motion in the material bulk (Fig. S6E). Fig. S6H shows that the lever design alone comes with a rather complex spectral decomposi- tion of the response in which several modes contribute. G. Comparison between cooperative and geometric designs The key difference between the geometric and the co- operative designs is that the former develops a lever, while the latter doesn't. This fact leads to vastly dif- ferent properties of the response to binding. Amplification of the response: The map of the av- erage magnitude of the response shows a non-monotonic behavior between the allosteric and the active sites (Fig. S7A) for geometric designs, not apparent for coop- erative designs (as shown in Fig. S7B,C for both periodic and open boundaries respectively). Symmetry of the response: For cooperative de- signs, the response to binding at the active site is very similar to binding at the allosteric site (Fig. S7E,F), be- cause both type of stimuli mostly couple to the single soft elastic mode in the system. For the geometric de- 17 FIG. S6: Three-dimensional geometric task with open boundaries. (A) Two dimensional sections of the 3D bulk: the vertical plane V and horizontal plane H are shown on left and right respectively in the following panels. (B) Map of the average magnitude of response δRr. (C) Map of the average coordination number z. (D) Map of the average strain B-factor SB. (E) Map of the average shear strain intensity Eshear. (F) Map of the average fitness cost of single site mutation normalized by the cost of random networks ∆F/F . (G) Map of the average conservation Σ. (H) Decomposition qω of the response on the vibrational modes ω in a specific solution, colored as a function of the participation ratio Pω. sign, this is not true at all: stimulating the material in the active site where it is soft has essentially no effect in the rest of the material, as shown in Fig. S7D. Specificity of the response: Finally, in the geomet- ric design a large response at the active site can be trig- gered by binding anywhere in the material, because the lever amplifies any elastic signal it finds, as shown in the two upper panels of Fig. S8. Thus geometric designs are not specific. By contrast, cooperative designs respond much more if the stimulus is triggered at the allosteric site (used to train the material), where the soft extended mode is designed to have a large shear, as shown in the two lower panels of Fig. S8. These results are quantified in Fig.S9. 18 FIG. S7: Magnitude of the response averaged over (A,D) geometric solutions in periodic boundary, (B,E) cooperative solutions in periodic boundary and (C,F) in open boundary. The stimulus strain is imposed at the nodes shown as the purple crosses in (A-C)- precisely where these materials have evolved to respond. By contrast, in (D-F) the strain is imposed at the active site (purple crosses) where the material has not learnt to respond. For the geometric design, the response dies out very rapidly within the material, but for the cooperative design, the response is similar that obtained by stimulating the allosteric site shown above. 19 FIG. S8: Specificity of the (Top two) geometric solutions and (Bottom two) cooperative solutions (with a periodic boundary) toward imposing a stimulus at the surface of the material, at a site shown by the purple crosses which differs from the allosteric site where these materials were trained to respond. Blue crosses indicate the active site. A large response at the active site is always found in the geometric design (due to the presence of a lever) but not for the cooperative design. 20 FIG. S9: Average magnitude of response at the active site to a strain added on the other side of the system at location x (x = 0 corresponds to the position of the allosteric site) for both cooperative and geometric designs. locationx-10-50510MagnitudeδRAc00.20.40.60.811.2Cooperative PeriodicCooperative OpenGeometric Periodic
1006.3937	2	1006	2010-10-12T21:29:42	Nano-scale mechanical probing of supported lipid bilayers with atomic force microscopy	[ "physics.bio-ph", "cond-mat.mes-hall", "cond-mat.mtrl-sci", "cond-mat.soft" ]	We present theory and experiments for the force-distance curve $F(z_0)$ of an atomic force microscope (AFM) tip (radius $R$) indenting a supported fluid bilayer (thickness $2d$). For realistic conditions the force is dominated by the area compressibility modulus $\kappa_A$ of the bilayer, and, to an excellent approximation, given by $F= \pi \kappa_A R z_0^2/(2d-z_0)^2$. The experimental AFM force curves from coexisting liquid ordered and liquid disordered domains in 3-component lipid bilayers are well-described by our model, and provides $\kappa_A$ in agreement with literature values. The liquid ordered phase has a yield like response that we model by hydrogen bond breaking.	physics.bio-ph	physics	Nano-scale mechanical probing of supported lipid bilayers with atomic force microscopy Chinmay Das,1, 2, ∗ Khizar H. Sheikh,1, 3 Peter D. Olmsted,1, † and Simon D. Connell1 1School of Physics and Astronomy, University of Leeds, Leeds LS2 9JT, United Kingdom 2Unilever R &D, Port Sunlight, Wirral, CH63 3JW, United Kingdom 3UCD Conway institute for Biomolecular and Biomedical Research, Dublin, Ireland (Dated: October 25, 2018) We present theory and experiments for the force-distance curve F (z0) of an atomic force mi- croscope (AFM) tip (radius R) indenting a supported fluid bilayer (thickness 2d). For realistic conditions the force is dominated by the area compressibility modulus κA of the bilayer, and, to an excellent approximation, given by F = πκARz2 0/(2d − z0)2. The experimental AFM force curves from coexisting liquid ordered and liquid disordered domains in 3-component lipid bilayers are well- described by our model, and provides κA in agreement with literature values. The liquid ordered phase has a yield-like response that we model as due to the breaking of hydrogen bonds. PACS numbers: 87.16.D-, 87.16.dm, 87.80.Ek, 68.37.Ps I. INTRODUCTION to accurately model the mechanical response. Atomic Force Microscopy (AFM) [1] has become a standard tool for imaging surfaces at high resolution and probing local mechanical properties [2]. Force-distance curves for indentation of AFM tips have been used to characterize the mechanical properties of biological mem- branes [3 -- 5], and the usual approach is to approximate the bilayer as an elastic solid undergoing a Hertzian con- tact [6 -- 8]. However, at physiological conditions most bio- logical membranes are in a fluid bilayer phase [9], whose free energy is described by a bending modulus κ and the area compressibility modulus κA. These are exper- imentally accessible through, for example, micropipette aspiration experiments [10], which give the average value of the elastic moduli over the whole vesicle. However, biological membranes often have different local composi- tions, and thus different local mechanical properties and physiological functions. Despite the growing use of AFM to study lipid bilay- ers, the flexibility of using it to measure local mechanical properties has not been fully exploited. An AFM tip can bend a freely suspended membrane, and compress a supported membrane. In recent work, Steltenkamp et [11] showed how to extract the bending modulus of al. lipid bilayers from AFM force-distance curves for bilay- ers deposited over well defined sized holes (indentation of 'nanodrums'), in which they could safely ignore area com- pression due to the lack of a supported surface. Another issue neglected in previous AFM studies is the double leaflet form of lipid bilayers, which is known to influence the dynamics of fluctuations [12]. Since an AFM tip in- duces an asymmetric response in a supported bilayer [13], the distinction between the two leaflets will be important ∗Electronic address: [email protected] †Electronic address: [email protected] In this paper we consider the force-distance curves ob- tained by indenting an AFM tip into a fluid bilayer sup- ported on a solid substrate. The force-distance curves are calculated from a static analysis of the deformation of the two leaflets and differs from usual Hertzian result of the deformation of elastic bodies. We analyze exper- iments on a dioleoyl-phosphatidylcholine (DOPC) - egg sphingomyelin (SM) - cholesterol (CHOL) phase sepa- rated supported bilayer, which is a model mixture repre- sentative of typical in vivo membranes [14]. For certain composition ratios of the components, this system spon- taneously phase separates into coexisting liquid ordered (Lo, rich in SM and relatively thick because of strong ne- matic order in the acyl tails) and liquid disordered (Ld, rich is DOPC and relatively thin because of the more disordered tails) phases. We show how to determine the area compressibility moduli of the coexisting Lo and Ld phases of a single sample, and find values in agreement with literature values. To the best of our knowledge, this is the first time that the area compressibilities of the two coexisting compositions in fluid bilayers have been ex- tracted directly. This technique should prove invaluable for studying the composition dependence of mechanical properties in lipid bilayers, and can be easily extended to consider more complex interactions between AFM tip and the bilayer. II. THEORY We consider a supported fluid lipid bilayer of thickness 2d probed by an AFM tip of radius R in contact mode, which measures the force as a function of the depth z0 from the unperturbed surface of the layer (see Fig. 1(a)). At typical AFM speeds the viscous forces are negligible. We begin by assuming a hard contact interaction between the tip and the membrane. Electrostatic repulsion from the charged double layers and van der Waals attraction are included later in the paper when comparisons are made with experiments. Since the lipid bilayer is not anchored it remains tension free. We assume that the volume is conserved at the molecular level: as the tip penetrates the bilayer it occupies a volume δV , so that δN = δV /(a0d) lipids are expelled into the surrounding bilayer. Here, a0 is the area per lipid in the absence of the AFM tip. The surface area increases by δA, due to the curved spherical surface of the AFM tip, and the increase in area per head group δa ≡ a − a0 is given by V − δV (cid:19) − a0. N − δN − a0 = a0d(cid:18) A + δA A + δA δa = (1) This increased area induces an elastic cost due to the stretching elasticity of the lipid leaflets. We calculate this not by averaging over the entire spherical cap, but by considering small increases in radius dr, and evaluating δa/a0 at each r (Fig. 1). (a) R R−z 0 r z a 0 d 2 d (b) d r φ r d z(r) z 0 (c) r z 0 dr ds θ FIG. 1: (color online) (a) Schematic geometry of an AFM tip of radius R indenting a fluid bilayer of thickness 2d by an amount z0. The leaflet dividing surface at hb(r) is shown as a solid line. (b) Cylindrical volume elements of depth z(r) at the distance r from the center of the tip. (c) The area 2πr ds of such an element in contact with the lipid. We assume that both leaflets have the same area per lipid a0 and stretching modulus κA/2 for lipid head groups on a flat surface. This should be valid in the absence of specific interaction of the lipid with the sub- strate, although experiments have shown that the sur- face often does have specific interactions [13]. The head groups in the top leaflet are forced to lie on a curved sur- face below the AFM tip. This affects both the area/lipid, and the stretching modulus for the lipids on the top leaflet. We model the local lipid free energy as a sum of a surface energy and a harmonic tail stretching, g(L) ≃ γaL/ cos θ + α/a2 L, where γ is a surface tension, α penal- izes tail stretching, and aL is the projected area for leaflet 2 thickness L. Here, θ(r) = sin−1(r/R) (Fig. 1c) is the tilt angle of the lipid surface. Minimizing g at fixed lipid volume v ≃ LaL leads to an effective stretching modu- lus κA/2 ≃ κA/2 sec2/3 θ and an effective area per lipid 0 = a0 cos1/3 for the top leaflet. at Using the modifications due to the curved surface for the head-groups in the top leaflet, the excess free energy due to the increase in the area per lipid during indenta- tion is G(z0) = κA 4 ZS d2r"sec2/3 θ(r)(cid:18) δa a0(cid:19)2 t a0(cid:19)2 +(cid:18) δa b# , (2) where t and b refers to the top and bottom leaflets and the integration extends over both leaflets. The lower leaflet will generally deform to accommodate the large energy change due to removing too many lipids from the upper leaflet. We let the lower and upper leaflets have thick- nesses hb(r) and ht(r) respectively, with hb(r) + ht(r) = 2d− z0 + R[1− cos θ(r)]. The area changes at each radius r are given by (Fig. 1) at 0(cid:19)2 (cid:18) δa a0(cid:19)2 (cid:18) δa t b ht(r) = (cid:20) d sec1/3 θ(r) hb(r) − 1(cid:19)2 = (cid:18) d 2 , − 1(cid:21) . (3a) (3b) The measure is d2r = r dφdr. The dividing surface hb(r) is determined by minimizing the free energy at each r. For equal stretching moduli in both leaflets, an explicit solution for hb(r) is possible for small tilt angle, hb(r) = 2d − z0 + R (1 − cos θ(r)) 1 + sec1/3 θ . (4) For realistic values z0 ∼ 2nm, d ∼ 3nm and R ∼ 10nm, this approximation introduces less than 0.1% error in hb(r) for the entire range of r. We use this approximation in the rest of the paper to derive analytic expressions for the free energy and force. Using hm(r) from Eq. 4, the free energy is 2G(z0) πκAR2 =Z 1 1− z0 R x(cid:16)1 + x−2/3(cid:17) × (cid:0)1 + x1/3(cid:1) d + R d (1 − x)(cid:3) − 1#2 " x1/3(cid:2)2 − z0 dx, (5) and the force on the AFM tip is given by F = ∂G/∂z0. We use the numerical force derived from Eq. 2 when per- forming fits to the data. For small penetrations z0 the force can be written as F ≃ πκAR 4 "1 + d 3R +(cid:18) d 3R(cid:19)2#(cid:16) z0 d (cid:17)2 + O(z3 0) . . . (6) A surprisingly simple function that fits the entire exper- imental range of forces, correct to within a few percent for R = 3d and much better for larger R, is F = πκAR 4 (cid:18) 2z0 2d − z0(cid:19)2 . (7) The force diverges as z0 approaches 2d because the area/lipid diverges in order to preserve molecular vol- ume. The quadratic free energy (Eq. 2) is no longer valid there. Experimentally this divergence is preempted by pore formation (see below). 0 For comparison, the contact force between two solid (elastic) bodies much larger than the radius of contact (Hertzian contact) is F ∼ z3/2 [15]. More relevant for AFM experiments, the force to indent a finite elastic layer scales as F ∼ R2z3 0/d3 if the layer is bonded to the substrate and F ∼ Rz2 0/d if the layer can slip [7]. The response of fluid bilayers in eqn. 6 scales differently than all of these scenarios, and force is proportional to the area compressibility modulus instead of the Young's modulus. For realistic experimental values the region of validity for this quadratic behavior is limited, as shown in Fig. 2(b). 1.5 0.1 ) R 1 0.05 A κ π ( / F 0.5 0 0 0.2 0.4 0 0 0.5 z0/d 1 FIG. 2: Scaled force F/(πκAR) as a function of tip depth z0/d (solid line) for tip radius R/d = 3, according to Eq. 5. The dashed line shows only the leading quadratic term in z0/d, Eq. 6. Inset: Behaviour at small z0/d. On this scale the approximation of Eq. (7) is indistinguishable from the numerical solution of Eq. (5). III. EXPERIMENTS To test the theory we performed experiments on a sup- ported bilayer comprising DOPC, SM and CHOL at over- all molar ratios DOPC:SM:CHOL=40:40:20. At room temperature this system phase separates into coexisting DOPC-rich liquid disordered and SM-rich liquid ordered domains (Fig. 3). The hydrocarbon tails have large ne- matic order in the Lo phase, leading to a thicker bilayer 3 and higher area compressibility modulus. In contrast, the tails have lower nematic order in the Ld phase with concomitant smaller thickness and lower moduli. DOPC, CHOL (purchased from Sigma) and egg SM (purchased from Avanti) were dissolved in chloroform, dried under a stream of argon for 30 minutes, and then vacuum desic- cated for 30 minutes. The lipid was resuspended in PBS buffer at pH 7.4 to a concentration of 1 mg/ml by vor- texing. To make small single unilamellar vesicles (SUVs), the cloudy lipid suspension was tip sonicated (IKA, U50) at less than 5◦ C for 25 mins (until the solution became clear). The mica (Agar Scientific Ltd.) surface was in- cubated with the SUVs at at 50◦ C and cooled down to room temperature in a incubator over 15 minutes. Af- ter 1h, the sample was gently rinsed with PBS buffer to remove any excess vesicles. Force measurements were performed at 27◦ C in PBS buffer using a Nanoscope IV Mulitmode AFM (Veeco) equipped with a temperature control stage, using can- tilevers (NP, Veeco) with nominal spring constants of 0.12 N/m. Spring constants were measured using the thermal noise method [16] in air, and optical lever sensitivity de- termined against a clean mica surface.The force curves analysed in this paper were all taken from a single force- volume map of the phase separated bilayer shown in Fig- ure. 4, and exported using Nanscope software v5.12r30. Scanning electron microscopy (Camscan series III, FEG-SEM operating at 5 kV with magnification 160k) was used to measure the tip radius. Inset of Fig. 4 shows the tip image with dashed lines along the edges of the four pyramidal faces. The end of the tip can be approx- imated as spherical. The drawn circle (Fig. 4 inset) has a radius of 10 nm. The contrast of the image is poor. Consequently the uncertainty of the exact value of the radius is large. In our analysis we consider the tip radius to be R = 10 ± 5 nm. Fig. 3 shows a tapping mode image of the bilayer along with a one dimensional cross section. There is a ∼ 5 nm thick Ld matrix enclosing ∼ 6 nm thick Lo domains (The heights reported here include the thickness of any wa- ter layer between the bilayer and the mica surface). The composition of the two phases were determined by fol- lowing the tie lines on the ternary phase diagram, which were calculated using Atomic Force Microscopy [17]: Ld : Lo : (DOPC:SM:CHOL) = (68 : 27 : 5) (DOPC:SM:CHOL) = (3 : 71 : 26) (8a) (8b) Phase diagrams on similar ternary mixtures have been calculated using NMR, and the compositions of the liquid-disordered and liquid ordered phases are similar [18]. The uncertainty in the compositions from place- ment of the tie lines is estimated to be less than 2% of the quoted values. For both Lo and Ld phases, force curves from contact mode AFM were used from at least 10 different measurements from different points within different regions ('patches', as in Fig. 3) of the sample. Fig. 4 (symbols) shows the force curve for the DOPC rich bilayer in the liquid disordered phase. Besides the 4 stretching contribution considered so far, the tip experi- ences an attractive force due to van der Waals interaction and a short range repulsive force due to the electric dou- ble layers on the tip surface and the membrane top sur- face. In principle the van der Waals interaction can be calculated from a knowledge of the dielectric constants of the tip, membrane, and the PBS buffer [19]. Simi- larly, the repulsive interaction can be estimated by know- ing the detailed charge distribution and solving Poisson- Boltzmann equation. Phenomenologically, we model the van der Waals attraction as an interaction energy of the form −A/ξ6 between volume elements of the tip and the membrane separated by a distance ξ. Since these forces are short-ranged, we consider the tip as a sphere and for the volume integration over the mem- brane, consider the membrane as infinitely thick. We further assume that the repulsive interaction is strong enough to avoid adsorption. As the tip approaches the bilayer, the bilayer deforms. The extent of the deforma- tion is governed by the minimum of the stretching free energy and the long range interactions (van der Waals and screened Coulomb). We assume that the deforma- tion can be modeled as hard interaction from a tip with radius Rc larger than the physical tip radius R. Although A is poorly known and depends on the detailed dielec- tric properties of the membrane, its precise value only changes Rc and controls the details of the force near contact. For deeper contact the force is overwhelmingly dominated by the stretching modulus κA, so that the force-distance curves yield the same κA, independent of A. The drawn line in Fig. 4 shows the fit from our the- oretical analysis, using a downhill simplex method [20] to minimize the mean square fractional deviation of the prediction from the experimental data over the fitted range. The best fit for the compressibility modulus for R = 10 nm is κA = 0.12 N/m. Because of the uncer- tainty in the tip radius, the range of κA for R between 5 nm and 10 nm is between 0.25 N/m and 0.08 N/m, re- spectively. Our estimate compares well with the litera- ture values κA = 0.13 − 0.6 N/m from osmotic pressure measurements [21] and κA = 0.18 ± 0.04 N/m from mi- cropipette aspiration of GUVs [22] made of pure DOPC. Our model provides an excellent fit until 2d−z0 ≃ 2.5nm, at which point the elastic energy of the deformed bilayer overcomes the cost of forming a hole [23] and the tip abruptly penetrates the full bilayer. (a) Ld Lo 1 nm Ld 6 ) m n ( h t p e d 4 2 (b) 0 FIG. 3: (color online) Phase separated lipid bilayer with liquid ordered and liquid disordered domains. (a) Tapping mode AFM image showing the height profile of the bilayer. (b) One dimensional section along the dashed line in (a). ) N n ( F 8 6 4 2 0 0 2 4 6 2 d - z0 (nm) 8 10 IV. RESPONSE OF LIQUID ORDERED DOMAINS FIG. 4: Force-distance curve for a DOPC rich bilayer in the liquid disordered phase. The data points are from AFM ex- periments and the line is a fit for the theoretical prediction with κA = 0.12 N/m. Inset: SEM image of the tip to measure the tip radius. The dashed lines are along the pyramidal face edges. The circle drawn at the end of the tip has a radius 10 nm. The AFM force curves for the SM rich liquid ordered phase are qualitatively different from those in the coex- isting liquid disorded phase (Fig. 5). The initial deforma- tion (5 nm ≤ 2d − z0 ≤ 6 nm) shows a high modulus con- sistent with the tightly packed character of the Lo phase. Around 2d− z0 ≃ 5 nm the response shows a crossover to 5 3 2 1 0 4 6 8 6 4 2 0 ) N n ( F 8 10 0 2 4 2 d - z0 (nm) 6 8 8 6 4 2 ) N n ( F 0 0 2 4 6 2 d - z0 (nm) FIG. 5: (color online) Force-distance curve from AFM (sym- bols) in the SM rich liquid ordered phase superposed with two separate theoretical fits (lines, using Eq. 5) involving two different κA at small and large tip penetrations z0. Inset: Closeup of the crossover region. a much lower modulus. The symbols in Fig. 5 are from 12 separate force-distance measurements. While the ex- perimental data fall on the same curves away from the crossover region, the transition from stiff to soft behav- ior occurs at different values of z0, which may be due to either the stochastic behaviour of an activated event or flucuations in composition from region to region. We first attempt to model these force curves as due to an effective stretching modulus that differs for small and large penetrations far from the crossover region. Hence we fit the data at small penetration (2d − z0 > 5.3 nm) and large penetration (2d − z0 < 3.5 nm), with effective stretching moduli according to Eq. (5). For R = 10 nm, the small z0 fit gives κA = 1.1 N/m. Recent experi- ments on a bovine brain SM and CHOL equimolar mix- ture found κA = 2.1 ± 0.2 N/m [24]. Since egg SM (16:0 SM) has shorter fatty acid chains than does bovine SM (18:0 SM), and the current composition has compara- tively smaller amounts of CHOL, we expect the mem- brane to be softer (smaller κA), as found. The large penetration (z0) region has a stretching mod- ulus κA = 0.05 N/m, which is much closer to that of the Ld phase shown in Fig. 4 than the unperturbed Lo phase. The AFM tip forces the bilayer immediately below it to decrease in thickness, which thus destroys the strong ne- matic order of the Lo phase and induces a yielding or phase transition of the Lo phase into an Ld phase. It is likely that the composition of this induced Ld phase dif- fers from that of the Ld phase that characterizes equilib- rium coexistence far from the AFM tip (Eq. 8a), because of slow kinetics of composition changes under the AFM FIG. 6: (color online) Force-distance curve from AFM (sym- bols) in the SM rich liquid ordered phase superposed with a microscopically-motivated fit that accounts for a separate en- ergetic contribution from hydrogen-bond breaking (solid line, based on Eq. 9). Also shown are the separate contributions from the van der Waals interaction (dotted line), from the hy- drogen bonds (dot-dashed line) and the area compressibility term (dashed line). tip. Our separate fits to extract κA suffer from narrow available fit window (∼ 0.3 nm) for small z0 and the lack of small force data for the large z0 fit. Also, this proce- dure does not elucidate the reason for two distinct elastic regions separated by a crossover. To understand the qualitatively difference force re- sponses of the Lo and Ld phases, we propose a microscopically-motivated model. SM has both hydro- gen bond donor and acceptor groups and is known to form inter-SM hydrogen bonds [25, 26]. The free energy in the Ld phase, as represented in Eq. 2, is dominated by solvent and tail packing entropies. Hence, to describe the Lo phase we separately include the short range en- ergy of hydrogen bond breaking through a simple Morse potential: U (b) = ED [1 − exp (−(b − b0)/λm)]2, where b is the separation between the donor and acceptor group and b0 is the equilibrium separation. For typical hydro- gen bonds the dissociation energy ED ∼ 2 − 7 KCal/mol and the range λm ∼ 0.02 − 0.07 nm [27, 28]. For small changes in area/lipid and affine deformation the contri- bution to the free energy from distortion of hydrogen bonds is approximately GHB(z0) = eHBZ d2r( ht(r) (cid:20)1 − e hb(r) − 1 + d d − 1 (cid:20)1 − e a0 (cid:17)b(cid:21)2) . λ(cid:16) δa λ(cid:16) δa a0 (cid:17)t(cid:21)2 (9) Here, eHB is inter-lipid hydrogen bond dissociation energy per area and λ ∼ 31/4λm/√a0 for hexagonal arrangement of the lipids. As before, the total free energy, now com- prising contributions from Eqs. (5, 9), is minimized at each r to find the dividing surface between the leaflets and the force is calculated from F = ∂G/∂z0. In the limit of small penetrations this model gives an effective stretching modulus κeff A in the Lo phase of κeff A = κA + 4 eHB λ2 , (10) where κA is thus the stretching modulus of the Ld phase that is left after the Lo phase has been destabilized and there is no remaining hydrogen bond contribution. The fit to the data is shown in Fig. 6. The stochastic nature of the force curves near the rupture point (2d − z0 ≃ 5.7), limits the ability to obtain excellent fits. Our fit gives κA = 0.13 N/m, eHB = 0.006 N/m and λ = 0.1. Assuming an area per lipid a0 ∼ 0.6 nm2, the fitted value λ implies that the range of the Morse potential is λm ∼ 0.06 nm. Simulations show about 0.5 hydrogen bonds per lipid in SM bilayer [25]. Assuming an average hydrogen bond energy of 3.5 KCal/mol, our value for eHB give 0.4 hydrogen bonds broken per lipid. The initial deformation is dominated by the contribution from the hydrogen bonds, and the corresponding force curve leads to an area compressibility modulus κef f A ≃ 2.7 N/m. V. DISCUSSION We have assumed a static force response, despite typ- ical tip velocities vtip ≃ 102nm/s. We can estimate the correction due to finite tip velocity by considering the dissipation from two dimensional viscosity η of the lipid layer. The dissipative force is found to be FD(z0) = ηπz0(2R − z0) 2d2 vtip. (11) The two dimensional shear viscosity for fluid bilayers is expected to be of the order of 10−10N-s/m [29], leading to FD ≃ 10−8 nN, much smaller than the elastic contribu- tions. Hence our static approach is sufficient to describe the AFM force-distance curves on fluid lipid layers. In our calculations we have assumed that the two leaflets have the same area compressibility and preferred area per head group. This may not always be the case, because of surface interactions [13]; for example, sup- ported bilayers often have different melting temperatures than their counterparts in giant unilamellar vesicles. In- corporation of asymmetric membranes into the model is straightforward, although more complex. We have also 6 neglected splay or bending energies. Part of the elastic cost of this is already included in the increased area/lipid against the curved surface, in Eq. 2, but there may also be an additional negligible free energy cost due to the splay of the lipid tails, through the bending modulus of each leaflet. In our analysis, the initial deformation for the Lo phase is described in terms of the stretching of hydrogen bonds. This can be explicitly tested by performing experiments with varying concentrations of SM or using chemicals that disrupt hydrogen bonds. However, this is beyond the scope of the present work. Evidently, local applied pressure can melt the liquid ordered phase into the thinner Ld phase, which is not surprising. We have proposed an explicit microscopic mechanism in terms of breaking hydrogen bonds that are implicated in stabilizing Lo phase. An alternative and more general description could include a Landau theory for the free energy of the Lo-Ld phase transition, with local pressure p added as an external field to destabilize the Lo phase, ∆G ∼ pψ, where ψ is an order parameter proportional to thickness whose value decreases upon a transformation to the Ld phase [30]. The phase trans- formation would then occur first at constant composi- tion, and then one may expect the composition to change slowly as the external force changes the local preference for the different lipid species. The subject of kinetics and composition as a function of applied pressure is interest- ing and important, and we leave this for further work. In summary, we have presented, and validated by ex- periments, a theory for describing the force distance F (z0) relationship for AFM experiments on fluid bilayers, which leads to a remarkably simple expression for F (z0), Eq. (7). This provides a method for finding the area compressibility modulus and the amount of inter-lipid hydrogen bonds of fluid bilayers. The agreement with the existing literature values for the area compressibility is excellent. The main uncertainty in our prediction is due to the uncertainty in the tip radius R. However, the simple linear dependence on R means that relative mea- surements taken with the same tip can be compared very accurately. Acknowledgments The authors thank A. Ferrante, R. Marriot, M. Noro and B. Stidder for useful discussions. This work was sup- ported by Yorkshire Forward through the grant YFRID Award B/302. CD acknowledges SoftComp EU Network of Excellence for financial support. [1] G. Binnig, C. F. Quate, and C. Gerber, Phys. Rev. Lett. Reports 59, 1 (2005). 56, 930 (1986). [3] S. D. Connell and D. A. Smith, Mol. Memb. Biol. 23, 17 [2] H.-J. Butt, B. Cappella, and M. Kappl, Surface Science (2006). 7 [4] K. Vo itchovsky, S. A. Contera, M. Kamihira, A. Watts, 90, 4428 (2006). and J. F. Ryan, Biophys. J. 90, 2075 (2006). [19] J. Israelachvili, Intermolecular and surface forces (Aca- [5] K. D. Costa, A. J. Sim, and F. C. Yin, J. Biomech. Eng. demic Press, New York, 1992), 2nd ed. Trans. ASME 128, 176 (2006). [6] I. N. Sneddon, Internat. J. Engin. Sci. 3, 47 (1965). [7] R. S. Chadwick, SIAM J. Appl. Math. 62, 1520 (2002). [8] Y. T. Cheng, W. Ni, and C.-M. Cheng, Phys. Rev. Lett. [20] W. H. Press, S. A. Teukolsky, W. T. Vetterling, and B. Flannery, Numerical Recipes (Cambridge university press, Cambridge, 1992), 2nd ed. [21] S. Tristram-Nagle, H. I. Petrache, and J. F. Nagle, Bio- 97, 075506 (2006). [9] R. Lipowsky, Nature 349, 475 (1991). [10] E. Evans and W. Rawicz, Phys. Rev. Lett. 64, 2094 (1990). [11] S. Steltenkamp, M. M. M uller, M. Deserno, C. Hennesthal, C. Steinem, and A. Janshoff, Bio- phys. J. 91, 217 (2006). phys. J. 75, 917 (1998). [22] N. Fa, L. Lins, P. Courtoy, Y. Dufrene, P. van Der Smis- sen, R. Brasseur, D. Tyteca, and M.-P. Mingeot-Leclercq, BBA-Biomem 1768, 1830 (2007). [23] H.-J. Butt and V. Franz, Phys. Rev. E 66, 031601 (2002). [24] W. Rawicz, B. A. Smith, T. J. McIntosh, S. A. Simon, and E. Evans, Biophys. J. 94, 4725 (2008). [12] U. Seifert and S. A. Langer, Europhys. Lett. 23, 71 [25] E. Mombelli, R. Morris, W. Taylor, and F. Fraternali, (1993). Biophys. J. 84, 1507 (2003). [13] C. Xing, O. H. S. Ollila, I. Vattulainen, and R. Faller, [26] T. R´og and M. Pasenkiewicz-Gierula, Biophys. J. 91, Soft Matter 5, 3258 (2009). 3756 (2003). [14] S. L. Veatch and S. L. Keller, Phys. Rev. Lett. 94, 148101 [27] Y. Gao, K. V. Devi-Prasad, and E. W. Prohofsky, J. (2005). [15] L. Landau and E. Lifshitz, Theory of elasticity (Butterworth-Heinemann, Oxford, 1986), p. 26, 3rd ed. [16] J. Hutter and J. Bechhoefer, Rev. Sci. Instrum. 64, 3342 Chem. Phys. 80, 6291 (1984). [28] D. Thierry, M. Peyrard, and A. R. Bishop, Phys. Rev. E 47, 684 (1993). [29] M. Sickert and F. Rondelez, Phys. Rev. Lett. 90, 126104 (1993). (2003). [17] S. D. Connell, G. Li, P. D. Olmsted, N. M. Hooper, and [30] S. Komura, H. Shirotori, P. D. Olmsted, and D. Andel- D. A. Smith (2010), to be published. man, Europhys. Lett. 67, 321 (2004). [18] S. L. Veatch, K. Gawrisch, and S. L. Keller, Biophys. J.
1801.01640	1	1801	2018-01-05T05:57:22	Impulse-Response Approach to Elastobaric Model for Proteins	[ "physics.bio-ph" ]	A novel energy landscape model, ELM, for proteins recently explained a collection of incoherent, elastic neutron scattering data from proteins. The ELM of proteins considers the elastic response of the proton and its environment to the energy and momentum exchanged with the neutron. In the ELM, the elastic potential energy is expressed as a sum of a temperature dependent term resulting from equipartition of potential energy among the active degrees of freedom and a wave vector transfer dependent term resulting from the elastic energy stored by the protein during the neutron scattering event. The elastic potential energy involves a new elastobaric coefficient that is proportional to the product of two factors: one factor depends on universal constants and the other on the incident neutron wave vector per degree of freedom. The ELM was tested for dry protein samples with an elastobaric coefficient corresponding to 3 degrees of freedom. A discussion of the data requirements for additional tests of ELM is presented resulting in a call for published data that have not been preprocessed by temperature and wave-vector dependent normalizations.	physics.bio-ph	physics	1 Impulse-Response Approach to Elastobaric Model for Proteins Robert D. Young Department of Physics, Illinois State University Normal, IL 61790-4560 (Dated January 4, 2018) Abstract A novel energy landscape model (ELM) for proteins recently explained a collection of incoherent, elastic neutron scattering data from proteins. The ELM of proteins considers the elastic response of the proton and its environment to the energy and momentum exchanged with the neutron. In the ELM, the elastic potential energy is expressed as a sum of a temperature dependent term resulting from equipartition of potential energy among the active degrees of freedom and a wave-vector transfer dependent term resulting from the elastic energy stored by the protein during the neutron scattering event. The elastic potential energy involves a new elastobaric coefficient that is proportional to the product of two factors – one factor depends on universal constants and the other on the incident neutron wave-vector per degree of freedom. The ELM was tested for dry protein samples with an elastobaric coefficient corresponding to 3 degrees of freedom. A discussion of the data requirements for additional tests of ELM is presented resulting in a call for published data that have not been preprocessed by temperature and wave-vector dependent normalizations. I. Introduction A protein is a folded chain of amino acids with a covalently bonded structure as well as non- bonded interactions that involve hydration waters and bulk solvent. Protein structure and function have been analyzed extensively [1]. Proteins are also dynamic systems that experience continual fluctuations [2]. Fluctuations, in fact, are essential for protein function. Mössbauer absorption spectroscopy, incoherent neutron scattering, dielectric spectroscopy, and other techniques are used to probe protein fluctuations. However, understanding of protein dynamics is not fully developed. Incoherent neutron scattering has developed into a mature experimental tool for probing protein fluctuations [3]. It is customary to use models to extract a measure of spatial fluctuations in the form of the mean-square displacement of the protein hydrogen atom that scatters the neutron. This class of models is called SMM for spatial motion models. Recently shortcomings of the SMM have been outlined in two papers by Frauenfelder and collaborators [4,5]. As a replacement for the SMM, these authors developed a model called ELM for energy landscape model in which the elastic response of a protein plays a central role. The ELM model envisions a quite different role for the dynamic momentum (or equivalently, wave-vector) transfer in incoherent neutron 2 scattering. The ELM passed initial tests by describing elastic incoherent neutron scattering data for several dehydrated proteins [5]. The ELM involves two parameters – one called the elastobaric coefficient  with dimension kelvin-Ångstom (K-Å) and the other called  with dimension kelvin (K). Here an impulse-response approach of the ELM is developed in which elastobaric coefficient  is described by universal constants, the initial neutron wave-vector, and the number of active degrees of freedom, d, of the target proton and its environment. II. Impulse-Response Approach of Elastobaric Model for Neutron Scattering ( )F t The neutron (N) interaction with a proton (P) covalently bound in a molecule results in a time- ( )F t on the proton and its immediate molecular environment [5]. The analytic dependent force form of requires knowledge of the strong interaction between the neutron and proton and, also, the bonded and non-bonded interactions between the proton and its molecular environment. To quantify the elastic response of the protein to the incident neutron without knowing these complicated forces, an impulse-response approach is followed. The total momentum transfer P to the proton and its moiety during the N-P interaction is given by the impulse,   t , and Q is where the time integral extends over the N-P interaction time, P   F t dt  Q     Q  U Q   Q l t   P l , results in potential energy storage   the wave-vector transfer. If F is the average force on a proton and its moiety over the N-P   . The average work done by F , acting over a curvilinear interaction time, then F t displacement in the proton environment where   ov of the incident neutron is not changed by elastic F l  . scattering so is approximated as Qv U Q o k  Using the momentum of the incident neutron, the average speed can be estimated by o ok is the central wave vector of the incident neutron wave packet. The potential energy where stored in the proton environment is then . The potential energy stored then is . The speed l  m v n o  U Q t     ov    U Q    Qv o 2   k m n o Q . (1) 2 Bk T Some, or all, of the transferred energy might be kinetic during the interaction, but the energy is ultimately stored as potential energy as in Eq. (1). The potential energy of the proton and its environment for the active degrees of freedom is assumed for every relevant degree of freedom. For d active degrees of freedom, the potential to be energy . The total potential energy U of the proton and its environment for a specific population of proteins that scatter with momentum transfer Q is then given by  oU before the interaction is given by  U T Q U T (2)  U T  U Q  k T B dk T k T B  Q  Q         2   ,   0 o o 2  k m n B d 2 d 2 where the elastobaric coefficient  is given by       22 m k n B    k o d . (3) 3 Eq. (2) can be rearranged to read T T  * Q  . (4) T has dimension K but is a scaled potential energy. Frauenfelder named the model underlying Eqs. (2) and (3) the elastobaric model [5,6,7,8]. The elastobaric coefficient  is proportional to the initial neutron wave-vector per degree of freedom of the target proton and environment, or equivalently, the initial neutron momentum per degree of freedom. The proportionality factor, , is given in terms of universal constants. Once is determined experimentally, the number of active degrees of freedom of the proton and its environment is known. Eqs. (2) and (3) are the main results for the elastobaric model in the impulse-response approach. As an example, the IN13 neutron spectrometer at Institut Laue-Langevin (ILL) involves scattering with a neutron beam having incident wave-vector  for the neutron beam at IN13. Fig. 1 ok shows the variation of  with changing degrees of freedom for IN13. . According to Eq. (3), 95.7 K Å n Bm k  2.8 Å2 K Å 269.9  2   d  1 2 2  Figure 1. The elastobaric parameter vs degrees of freedom d for 2.8 Å2  .  1 ok The elastic intensity for incoherent neutron scattering on protein samples can be approximated by  S Q T 0, ,    1 2    T       1 T     Q     2 , (5) 4 Eq. (5) is used to fit elastic scattering data for several proteins [5]. For dry protein samples or for samples below about 200 K, the two parameters have values of and describe the data well. See, for example, Figs. 5B and 5C in Reference [5]. The elastic incoherent neutron scattering data for the dry protein samples were taken on the IN13 neutron spectrometer d  , a value for close to the value found in the so that Eq. (3) gives published fits [5]. for dimension 91 K K750 90 K and    -Å -Å 3 III. Variation in Degrees of Freedom Protein samples may have a heterogeneity among the individual proteins of the sample in the elastic response to the incident neutron. For example, the number of degrees of freedom involved in the elastic response may differ among the proteins of the sample. For the proteins treated in Reference [5] the number of degrees of freedom was 3 as shown above. But there are differences in neutron elastic scattering for different proteins [3,5,9-12]. The possibility that differing number of degrees of freedom are activated during the neutron scattering process over an ensemble of proteins in a sample is real. This section explores how to quantify this possibility. Fig. 1 indicates a slowing in the change in  as the number of degrees of freedom increases. For example, if the dimension changes from 2 to 3, then changes from 135 K-Å to 90 K-Å . Data for hydrated samples above 200 K indicate that hydration activates new degrees of freedom in hydrated samples [10,11,12]. So, for example, if the degrees of freedom are increased to 6 at high . However, if the degrees of freedom changes from 5 to 6,  temperature, then changes from 54 K-Å to 45 K-Å . These observations suggest that low temperatures, below 200 K, and dry protein samples would be best to search for the effects of differences in activated degrees of freedom for different proteins. To quantify the discussion above, Fig. 2 displays simulations at 140 K for the elastic intensity  S Q T  in all simulations and the incident wave-vector ok 2.8 Å2 . The simulations for degrees of freedom d = 2, 3, 4 are shown as points at 9 values of wave-vector transfer Q (see caption in Fig. 2). Two additional sets of points are simulated using an assumed sample heterogeneity in which different number of degrees of freedom are affected by the scattering (50-50 % for d = 3, 4 and 25-75 % for d = 2, 3). The parameter in all simulations and the incident wave-vector . The parameter 140 K  1 2.8 Å2 K750 K750 45 K 0,       -Å  1 . ,  ok  5 Figure 2. Simulated values of elastic intensity S(0, Q, 140 K) plotted vs Q for three values of the degrees of freedom (d = 2 red open circles, d = 3 black closed diamonds, d = 4 blue open squares). Fits of Eq. (5) to simulated elastic intensities for the admixtures of 50-50 % d = 3 and 4 (blue solid line) and 25-75 % d = 2 and 3 (red solid line) are shown. The simulated elastobaric parameters are given together with the degrees of freedom. 2.8 Å2  .  1 ok ,  3.4  , simd 2.6 simd sim  sim  79 K-Å 103 K-Å Eq. (5) is then used to fit the simulated points with the result summarized in Fig. 2. The two values of  and d from the fits to the simulated points are (50-50 % for d = 3, 4) and (25-75 % for d = 2, 3). These simulations do not imply fractal dimensions for proteins. The heterogeneity described above is a result of a heterogeneous sample from the standpoint of possible differing number of degrees of freedom being active during neutron scattering in different proteins The results in the preceding paragraphs and Fig. 2 indicate that further progress requires additional elastic incoherent neutron scattering data on proteins data that have not preprocessed by temperature and wave-vector dependent "normalizations". Such normalizations distort the primary data obtained in the experiment [5]. IV. Conclusion Frauenfelder and collaborators have shown that the wave-vector transfer Q in elastic neutron scattering from protein samples results in an inhomogeneity of the target during passage of the neutron [5]. The resulting energy landscape model (ELM) depends crucially on the elastic response of the protein to the perturbation caused by the neutron-proton interaction. The effect is described by the elastobaric coefficient . Here we introduce an impulse-response analysis of the neutron-proton interaction by analyzing the energy and momentum transfer between the neutron 6    22 and the proton and its surroundings. This model shows that . The predicted value of  agrees closely with the fitted value of  in Reference [5]. We also suggest a path to further test the ELM by the publication of primary elastic neutron scattering data that have not been preprocessed by arbitrary temperature and wave-vector dependent "normalizations." Finally, the parameter  requires continued analysis to determine its full meaning [5]. m k n B k d o   Acknowledgement I thank my colleagues Hans Frauenfelder and Paul Fenimore for continued and deep discussions. 7 References 1. G. A. Petsko and D. Ringe, Protein Structure and Function (New Science Press 2004). 2. H. Frauenfelder, The Physics of Proteins. (S. S. Chan and W. S. Chan eds., Springer, 2010). 3. K. Wood, C. Caronna, P. Fouquet, W. Haussler, F. Natali, J. Ollivier, A. Orrechini, M. Plazanet, and G. Zaccai, Chem. Phys. 345, 305 (2008). 4. H. Frauenfelder, P. W. Fenimore, and R. D. Young, Proc. Natl. Acad. Sci. USA 111, 12764 (2014). 5. H. Frauenfelder, R. D. Young, and P. W. Fenimore, Proc. Natl. Acad. Sci. USA 114, 5130 (2017). 6. Y. Wang Y and G. Zocchi G. Europhys Lett. 96, 18003 (2011). 7. J. Hooper. J. Chem. Phys. 132, 014507 (2010). 8. H. Kim, S. A. Hambir, and D. D. Dlott. Phys. Rev. Lett. 83, 5034 (1999). 9. M.T. F. Telling, S. Howells, J. Combet, L. A. Clifton, V. G. Sakai, Chem. Phys. 424, 32 (2013) 10. H. Frauenfelder, G. Chen, J. Berendzen, P. W. Fenimore, H. Jansson, B. H. McMahon, I. R. Stroe, J. Swenson, and R. D. Young, Proc. Natl. Acad. Sci. USA 106, 5129 (2009). 11. M. Fomina, G. Schirò, and A. Cupane, Biophys. Chem. 185, 25 (2014). 12. J. D. Nickels, H. O'Neill, L. Hong, M. Tyagi, G. Ehlers, K. L. Weiss, Q. Zhang, Z. Yi, E. Mamontov, J. C. Smith, and A. P. Sokolov, Biophys. J. 103, 1566 (2012).
1502.03975	2	1502	2015-11-04T20:33:47	Amoeboid motion in confined geometry	[ "physics.bio-ph", "cond-mat.soft", "physics.comp-ph", "physics.flu-dyn", "q-bio.CB" ]	Many eukaryotic cells undergo frequent shape changes (described as amoeboid motion) that enable them to move forward. We investigate the effect of confinement on a minimal model of amoeboid swimmer. Complex pictures emerge: (i) The swimmer's nature (i.e., either pusher or puller) can be modified by confinement, thus suggesting that this is not an intrinsic property of the swimmer. This swimming nature transition stems from intricate internal degrees of freedom of membrane deformation. (ii) The swimming speed might increase with increasing confinement before decreasing again for stronger confinements. (iii) A straight amoeoboid swimmer's trajectory in the channel can become unstable, and ample lateral excursions of the swimmer prevail. This happens for both pusher- and puller-type swimmers. For weak confinement, these excursions are symmetric, while they become asymmetric at stronger confinement, whereby the swimmer is located closer to one of the two walls. In this study, we combine numerical and theoretical analyses.	physics.bio-ph	physics	Amoeboid motion in confined geometry Hao Wu,1, 2, ∗ M. Thi´ebaud,1, 2, ∗ W.-F. Hu,3 A. Farutin,1, 2 S. Rafaı,1, 2, † M.-C. Lai,3 P. Peyla,1, 2 and C. Misbah1, 2 3Department of Applied Mathematics, National Chiao Tung University, 1001 Ta Hsueh Road, Hsinchu 300, Taiwan 1Universit´e Grenoble Alpes, LIPHY, F-38000 Grenoble, France 2CNRS, LIPHY, F-38000 Grenoble, France Many eukaryotic cells undergo frequent shape changes (described as amoeboid motion) that enable them to move forward. We investigate the effect of confinement on a minimal model of amoeboid swimmer. Complex pictures emerge: (i) The swimmer’s nature (i.e., either pusher or puller) can be modified by confinement, thus suggesting that this is not an intrinsic property of the swimmer. This swimming nature transition stems from intricate internal degrees of freedom of membrane deformation. (ii) The swimming speed might increase with increasing confinement before decreasing again for stronger confinements. (iii) A straight amoeoboid swimmer’s trajectory in the channel can become unstable, and ample lateral excursions of the swimmer prevail. This happens for both pusher- and puller-type swimmers. For weak confinement, these excursions are symmetric, while they become asymmetric at stronger confinement, whereby the swimmer is located closer to one of the two walls. In this study, we combine numerical and theoretical analyses. PACS numbers: 47.63.mh, 47.63.Gd, 47.15.G–,47.63.mf Some unicellular micro-organisms move on solid sur- faces or swim in liquids by deforming their body instead of using flagella or cilia–this is known as amoeboid mo- tion. Algae such as Eutreptiella Gymnastica [1], amoe- bae such as dictyostelium discoideum [2, 3], but also leu- cocytes [2, 3] and even cancer cells [4] use this specific way of locomotion. This is a complex movement that recently incited several theoretical studies [5–12] since it is intimately linked to cell migration involved in several diseases. Some experimental results indicate that adhe- sion to a solid substratum is not a prerequisite for cells such as amoebae [2] to produce an amoeboid movement during cell migration and suggest that crawling close to a surface and swimming are similar processes. Recently, it was shown that integrin (a protein involved in adhe- sion process) should no longer be viewed as force trans- ducers during locomotion but as switchable immobilizing anchors that slow down cells in the blood stream before transmigration. Indeed, leukocytes migrate by swimming in the absence of specific adhesive interactions with the extracellular environment [13]. When moving, all micro-organisms are sensitive to their environments. Most microswimmers can follow gra- dients of chemicals (chemotaxis), some microalgae can move toward light sources (phototaxis) [14] or orient themselves in the gravity field (gravitaxis) [15], some other bacteria move along adhesion gradients (hapto- taxis) [16, 17], etc. Spatial confinement is another ma- jor environmental constraint which strongly influences the motion of micro-organisms. As a matter of fact, in the low-Reynolds number world, amoeboid motion generally occurs close to surfaces, in small capillaries or in extracellular matrices of biological tissues. Micro- organisms swim through permeable boundaries, cell walls or micro-vasculature. Therefore, the effect of walls on motile microorganisms has been a topic of increasingly active research [18–27]. It has been calculated a long time ago by Katz [28] and more recently pointed out [21, 23, 25, 27, 29] that swimmers can take advantage of walls to increase their motility. Understanding the be- havior of microswimmers in confinement can also pave the way to novel applications in microfluidic devices where properly shaped microstructures can interfere with swimming bacteria and guide, concentrate, and arrange populations of cells [30]. Living microswimmers show a large variety of swimming strategies [29] as do theoretical models aiming at describing their dynamics in confine- ment. Felderhof [18] has shown that the speed of Taylor-like swimmer increases with confinement. Zhu et al. [21] used the squirmer model to show that (when only tangential surface motion is included) the velocity decreases with confinement and that a pusher crashes into the wall, a puller settles in a straight trajectory, and a neutral swim- mer navigates. When including normal deformation they found an increase of velocity with confinement. Liu et al. [25] analyzed a helical flagellum in a tube and found that except for a small range of tube radii, the swimming speed, when the helix rotation rate is fixed, increases monotonically as the confinement becomes tighter. Ace- moglu et al. [24] adopted a similar model but, besides the flagellum, they included a head and found a decrease of velocity with confinement. Bilbao et al.[22] treated nu- merically a model inspired by nematode locomotion and found that it navigates more efficiently and moves faster due to walls. Ledesma et al.[23] reported on a dipolar swimmer in a rigid or elastic tube and found a speed enhancement due to walls. Here, we investigate, by means of numerical and an- alytical modeling, the effect of confinement on the be- havior of an amoeboid swimmer, which is a deformable object subjected to active forces along its inextensible membrane. Our model swimmer is found to reveal in- teresting features when confined between two walls. (i) We find that straight trajectories might be unstable, in- dependently of the nature of the swimmer (pusher or puller). (ii) For weak confinement, the swimming speed can either increase or decrease depending on the confine- ment strength. For strongly confined regimes, the ve- locity decreases in all cases recalling previous results on different models. (iii) The confined environment is shown to induce a transition from one to another type of swim- mer (i.e. puller or pusher). These behaviors are unique to amoeboid swimming (AS) and point to a nontrivial dynamics owing to the internal degrees of freedom that evolve in response to various constraints. The model. Amoeboid swimming is modeled here by taking a one-dimensional (1D) closed and inextensible membrane, which encloses a two-dimensional (2D) liquid of certain viscosity η and is suspended in another fluid taken to be of the same viscosity, for simplicity. The extra computational complexity of dealing with confined geom- etry restricts our study to 2D, which draws already rich behaviors. The effective radius of the swimmer is R0 = pA0/π, where A0 is the enclosed area. The swimmer has an excess normalized perimeter Γ = L0/(2πR0) − 1 (L0 is the perimeter) with respect to a circular shape (Γ = 0 corresponds to a circle, whereas large Γ signifies a very deflated, and thus amply deformable, swimmer). The strength of confinement is defined as C = 2R0/W , with W the channel width. Bounding walls are parallel to the x direction and y denotes the orthogonal one. A set of active forces is distributed on the membrane that reacts with tension forces to preserve the local ar- clength. The total force density is given by F = Fan − ζcn + ∂ζ ∂s t, (1) where Fan is the active force to be specified below (which we take to point along the normal n for simplicity), ζ is a Lagrange multiplier that enforces local membrane incompressibility, c is the curvature, t is the unit tan- gent vector and s is the arclength. We impose zero total force and torque. In its full generality the active force can be decomposed into a Fourier series Fa(α, t) = Pk=kmax Fk(t)eikα with α = 2πs/L0. We first consider the case kmax = 3, so that we are left with two complex amplitudes F2 and F3. Other configurations of the forces have been explored as well (see below). We consider cyclic strokes represented by F2 = F−2 = −A cos(ωt) and F3 = F−3 = A sin(ωt), where A is the force ampli- tude. k=−kmax The Stokes equations with boundary conditions (force balance condition, continuity of the fluid velocity and membrane incompressibility) are solved using either the boundary integral method (BIM) [31] or the immersed boundary method (IBM) [32]. Besides Γ and C, there is an additional dimensionless 2 FIG. 1: (Color online) Snapshots of an axially moving swim- mer over time (W = 6R0). The dashed profiles show a com- plete period Ts of deformation and then a few shapes are represented over a time of the order of 75Ts. Γ = 0.085. number S = A/(ωη), which is the ratio between the time scale associated with swimming strokes (Ts = 2π/ω) and the time scale of fluid flow due to active force (Tc = η/A). Here we take S = 10.0 (the shape has enough time to re- spond to active forces) and explore the effects of Γ and C. At a large distance from the swimmer, the velocity field is governed by σij = H Firjds. Only the (dimensionless) stresslet Σ = (σxx − σyy)/(η/Ts) enters the velocity field for symmetric swimmers. Σ > 0 defines a pusher and Σ < 0 defines a puller. The force distribution defined above is found to correspond to a pusher in the absence of walls. Below we will see how to monitor a puller or pusher and how the walls change the nature of the swim- mer. Results: Axially moving swimmers. We first consider an axially moving swimmer (AMS) (see Fig. 1). We consider only dimensionless quantities (unless otherwise stated). For example, ¯V = V Tc/R0 will denote the mag- nitude of swimming speed. We find an optimal confine- ment for swimming velocity. Increasing C enhances the speed of the swimmer until an optimal C o, where the speed attains a maximum before it decreases. Around the optimal value C o, low (high) viscous friction between the swimmer and the walls during the forward (recovery) BIM G BIM G BIM G IBM G IBM G IBM G = 0.085 = 0.054 = 0.026 = 0.085 = 0.054 = 0.026 0.006 0.004 V_ 0.002 0 0 0.2 0.4 0.6 0.8 1 1.2 1.4 C FIG. 2: Time-averaged velocity magnitudes (as a function of confinement C) of an axially moving swimmer for different Γ values. 3 Co = 0.085 = 0.054 = 0.026 0.005 0.004 0.003 0.002 0.001 V_ > < 1 0 -1 -2 -3 -4 0.002 0.001 0 -0.001 0 0.2 0.4 0 0.2 0.4 0.6 0.8 1 1.2 1.4 C FIG. 3: Time-averaged hΣi as a function of confinement C showing the transition from pusher to puller. phase of swimming promotes AMS speed. When the con- finement is too strong, large amplitude deformations are frustrated resulting in a loss of speed. The velocity col- lapse at strong confinement was also reported for helical flagellum [24, 25] and is expected to occur for all swim- mer models. Figure 2 shows the swimming velocity mag- nitude for different Γ values. That the wall enhances motility seems to be a quite general fact, as reported in the literature [18, 19, 21–25]. However, we must stress that this is not a systematic tendency. Close inspection shows that at weak confinement velocity first decreases before increasing, as shown in Ref. [33]. Results: Swimmer nature evolution. The value of the dimensionless stresslet Σ depends on the instantaneous swimmer configuration and its sign instructs us on the na- ture of the swimmer. We determine the average stresslet over a navigation cycle. An interesting result is the ef- fect of confinement on the pusher/puller nature of the swimmer. For small C(< 0.5) the swimmer is found to behave as a pusher, while it behaves as a puller for larger C(> 0.5). Figure 3 shows the evolution of hΣi as a func- tion of confinement, where a transition from pusher to 0 0 0.2 0.4 0.8 1 1.2 0.6 C FIG. 5: Time-averaged velocity magnitudes (as a function of confinement C) of different swimmers (Γ = 0.085): migrating along one wall (black diamond dashed line), navigating be- tween two walls (gray circle dotted dashed line) and moving along the channel center (black square solid line). The insets show characteristic trajectories. puller is observed. Results: Instability of the central position. The cen- tral position after a long time is found to be unsta- ble. The swimmer exhibits at small C (weakly confined regime) a zigzag motion undergoing large amplitude ex- cursions from one wall to the other. We refer to this as a navigating swimmer (NS). Figure 4 shows a snapshot, whereas the insets of Fig. 5 display typical trajectories. Despite this complex motion, the velocity in Fig. 5 be- haves with C qualitatively as that of the central swim- mer. The NS trajectory was recently reported [21, 34] in the cases of squirmer and three-bead models and also observed experimentally for paramecium (ciliated motil- / W > Y < 0.08 0.04 0 -0.04 -0.08 pusher puller 0 0.2 0.4 0.6 0.8 1 1.2 C FIG. 4: (Color online) Snapshots of a navigating swimmer over time (W = 6R0). The dashed profiles show a com- plete period Ts of deformation and then a few shapes are represented over a navigation period T of the order of 50Ts. Γ = 0.085. FIG. 6: Average position of the center of mass over a nav- igation period as a function of confinement C. Γ = 0.085. Circles (diamonds) correspond to the symmetric (asymmet- ric) motion of the swimmer. The vertical dotted line is the demarcation line between a pusher and puller. Note that a puller can either navigate or move close to either wall. S G G G s T T / 103 102 10 1 10-2 C-1 C-2 10-1 C 1 FIG. 7: Period of navigation as a function of confinement C. Γ = 0.085. Circles (diamonds) correspond to the symmetric (asymmetric) motion of the swimmer. ity) in a tube [19] pointing to the genericity of naviga- tion. This instability can be explained analytically (see Ref. [33]). A remarkable property is that the navigation mode can be adopted both by the pusher and the puller. This is in contrast with nonamoeboid motion [21], where a pusher is found to crash into the wall whereas the puller settles into a straight trajectory. These last two be- haviors are also recovered by our simulations, provided the stresslet amplitude is large enough (Σ2 >> − ¯V DS, where D is the dimensionless force quadrupole strength; see Ref. [33]). Results: Symmetry-breaking bifurcation. At a critical C ∗ the symmetric excursion of the swimmer becomes un- stable and undergoes a bifurcation characterized by the loss of the central symmetry in favor of an asymmetric excursion in the channel, as shown in the trajectories of Fig. 5 (see the supplemental movies in Ref. [33]). Figure 6 shows the average position in y of the center of mass as a function of confinement: a bifurcation diagram. This bifurcation is very abrupt, albeit it is of supercritical na- ture. Both slightly before and beyond the bifurcation, the swimmer behaves on average as a puller, but still it exhibits two very distinct modes of locomotion: naviga- tion or settling into a quasistraight trajectory (oscillation of the center of mass in this regime is fixed by the amoe- boid cycle). This complexity is triggered by the intricate nature of the amoeboid degrees of freedom. Results: Other force distributions. Including force distributions up to sixth harmonics with various ampli- tudes leaves the overall picture unchanged, pointing to the generic character of AS. The next step has consisted in linking the nature of the swimmer to its dynamics. 4 We have monitored a pusher or puller type of swimmer. If F = 2[sin(ωt) cos(3α) − (β + cos(ωt)) cos(2α)]n, we have a puller; while if F = 2[(−β + sin(ωt)) cos(3α) − cos(ωt) cos(2α)]n, we have a pusher (with β > 0). β monitors the strength of the swimmer nature (weak and strong pusher or puller). We found that for a weak enough stresslet, symmetric and asymmetric navigations prevail both for pullers and pushers. For a strong enough stresslet amplitude (for β > βc ∼ 1), we find that the pusher crashes into the wall, while a puller settles into a straight trajectory. This means that there is a qualita- tive change of behavior triggered by β, on which we shall report on systematic study in the future. Results: Navigation period. The navigation period T exhibits a nontrivial behavior with C (Fig. 7). At small C, the period scales as T ∼ C −1, and as T ∼ C −2 at intermediate confinement, before attaining a plateau at stronger confinement. To dig into the reasons for this complex behavior, we provide here some heuristic argu- ments. In the first regime, the NS swims in a straight and monotonous manner towards the next wall. In that regime, the period is limited by the distance traveled by the swimmer of the order W = 2R/C. This naturally yields the C −1 scaling of Fig. 7 for weak C. In the inter- mediate confinement regime, the magnitude of velocity depends linearly on C, so that the period scales as C −2 (see also Ref. [33]). After the symmetry-breaking occurs, the NS stays close to one of the two walls (inset of Fig. 5), and its center of mass oscillates with the intrinsic stroke period Ts. In this regime, the period is independent of C (diamonds in Fig. 7). Analytical results. We have first performed a linear stability analysis [33]. We find, for small C, that the sta- bility of the swimmer is governed by the stresslet sign: For Σ > 0 (pusher) the straight trajectory is unsta- ble, while it is stable otherwise (puller) . For a neutral swimmer the trajectory is marginally stable (the stabil- ity eigenvalue Ω, for a perturbation of the form y ∼ eΩt, is purely imaginary). We find that in the intermediate C regime the navigation period behaves as C −2. Using a systematic multipole expansion the complex behavior of the velocity as a function of C (at low C) can be ex- plained [33]. Discussion. We believe that the global features re- vealed by our study will persist in 3D although extending our work to 3D simulations will be a challenging task. Besides, in order to better match real cells performing amoeboid swimming (e.g. leukocytes), cytoskeleton dy- namics and its relation to force generation will be impor- tant ingredients to be included in a 3D modeling. C.M., A.F., M.T. and H.W. are grateful to CNES and ESA for financial support. S.R. and P.P. acknowledge financial support from ANR. All the authors acknowledge the French-Taiwanese ORCHID cooperation grant. W.- F. H., M.-C.L. and C.M. thank the MoST for a financial support allowing initiation of this project. ∗ These authors equally contributed to this work † Electronic address: [email protected] [1] J. Throndsen, Norw. J. Bot. 16, 161 (1969). [2] N. P. Barry and M. S. Bretscher, Proc. Natl. Acad. Sci. U.S.A. 107, 11 376 (2010). [3] A. J. Baea and E. Bodenschatz, Proc. Natl. Acad. Sci. U.S.A. 107, E167 (2010). [4] S. Pinner and E. Sahai, J. Micros. 231, 441 (2008). [5] T. Ohta and T. Ohkuma, Phys. Rev. Lett. 102, 154101 (2009). [6] T. Hiraiwa, K. Shitaraa, and T. Ohta, Soft Matter 7, 3083 (2011). [7] A. Shapere and F. Wilczek, Phys. Rev. Lett. 58, 2051 (1987). [8] J. E. Avron, O. Gat, and O. Kenneth, Phys. Rev. Lett. 93, 186001 (2004). [9] F. Alouges, A. Desimone, and L. Heltai, Math. Models Methods Appl. Sci. 21, 361 (2011). [10] J. Loheac, J.-F. Scheid, and M. Tucsnak, Acta Appl. Math. 123, 175 (2013). [11] A. Vilfan, Phys. Rev. Lett. 109, 128105 (2012). [12] A. Farutin, S. Rafai, D. K. Dysthe, A. Duperray, P. Peyla, and C. Misbah, Phys. Rev. Lett. 111, 228102 (2013). [13] T. Lammermann, B. L. Bader, S. J. Monkley, T. Worbs, R. Wedlich-Soldner, K. Hirsch, M. Keller, R. Forster, D. R. Critchley, R. Fassler, et al., Nature (London) 453, 51 (2008). [14] X. Garcia, S. Rafaı, and P. Peyla, Phys. Rev. Lett. 110, 138106 (2013). [15] J. Kessler, Contemp. Phys. 26, 147 (1985). [16] J. B. McCarthy, S. L. Palm, and L. T. Furcht, J. Cell Biol. 97, 772 (1983). 5 [17] I. Cantat and C. Misbah, Phys. Rev. Lett. 83, 235 (1999). [18] B. U. Felderhof, Phys. Fluids 22, 113604 (2010). [19] S. Jana, S. H. Um, and S. Jung, Phys. Fluids 24, 041901 (2012). [20] A. Zottl and H. Stark, Phys. Rev. Lett. 108, 218104 (2012). [21] L. Zhu, E. Lauga, and L. Brandt, J. Fluid Mech. 726, 011701 (2013). [22] A. Bilbao, E. Wajnryb, S. A. Vanapalli, J. Blawzdziewicz, Phys. Fluids 25, 081902 (2013). and [23] R. Ledesma-Aguilar and J. M. Yeomans, Phys. Rev. Lett. 111, 138101 (2013). [24] A. Acemoglu and S. Yesilyurt, Biophys. J. 106, 1537 (2014). [25] B. Liu, K. S. Breuer, and T. R. Powers, Phys. Fluids 26, 011701 (2014). [26] A. Zottl and H. Stark, Phys. Rev. Lett. 112, 118101 (2014). [27] F. Z. Temel and S. Yesilyurt, Bioinspiration Biomimetics 726, 016015 (2015). [28] D. Katz, J . Fluid Mech 64, 33 (2013). [29] E. Lauga and T. Powers, Rep. Prog. Phys. 72, 096601 (2009). [30] M. B. Wan, C. J. O. Reichhardt, Z. Nussinov, and C. Re- ichhardt, Phys. Rev. Lett. 101, 018102 (2008). [31] M. Thi´ebaud and C. Misbah, Phys. Rev. E 88, 062707 (2013). [32] Y. Kim and M.-C. Lai, J. Comp. Phys. 229, 4840 (2010). [33] See Supplemental Material at [URL will be inserted by publisher] for the theoretical analyses of confined swim- ming velocity and the relation between the stability of trajectory and the nature of swimmer shown in the paper as well as the movies for two different swimming modes. [34] A. Najafi, S. S. H. Raad, and R. Yousefi, Phys. Rev. E 88, 045001 (2013).
1801.09231	1	1801	2018-01-28T13:55:15	Monitoring of Wild Pseudomonas Biofilm Strain Conditions Using Statistical Characterisation of Scanning Electron Microscopy Images	[ "physics.bio-ph", "cond-mat.soft", "q-bio.QM", "stat.AP" ]	The present paper proposes a novel method of quantification of the variation in biofilm architecture, in correlation with the alteration of growth conditions that include, variations of substrate and conditioning layer. The polymeric biomaterial serving as substrates are widely used in implants and indwelling medical devices, while the plasma proteins serve as the conditioning layer. The present method uses descriptive statistics of FESEM images of biofilms obtained during a variety of growth conditions. We aim to explore here the texture and fractal analysis techniques, to identify the most discriminatory features which are capable of predicting the difference in biofilm growth conditions. We initially extract some statistical features of biofilm images on bare polymer surfaces, followed by those on the same substrates adsorbed with two different types of plasma proteins, viz. Bovine serum albumin (BSA) and Fibronectin (FN), for two different adsorption times. The present analysis has the potential to act as a futuristic technology for developing a computerized monitoring system in hospitals with automated image analysis and feature extraction, which may be used to predict the growth profile of an emerging biofilm on surgical implants or similar medical applications.	physics.bio-ph	physics	Monitoring of Wild Pseudomonas Biofilm Strain Conditions Using Statistical Characterisation of Scanning Electron Microscopy Images Suparna Dutta Sinha1, Saptarshi Das2, 3, Sujata Tarafdar1, and Tapati Dutta3 1) Condensed Matter Physics Research Centre, Department of Physics, Jadavpur University, Kolkata 700032, India 2) Department of Physics, University of Cambridge, JJ Thomson Avenue, Cambridge CB3 0HE, UK 3) Department of Power Engineering, Jadavpur University, Salt Lake Campus, LB-8, Sector 3, Kolkata-700098, India 4) Physics Department, St. Xavier's College, Kolkata 700016, India Author's Emails: [email protected] (S. Dutta Sinha) [email protected], [email protected] (S. Das) [email protected], [email protected] (S. Tarafdar) [email protected] (T. Dutta) Abstract: The present paper proposes a novel method of quantification of the variation in biofilm architecture, in correlation with the alteration of growth conditions that include, variations of substrate and conditioning layer. The polymeric biomaterial serving as substrates are widely used in implants and indwelling medical devices, while the plasma proteins serve as the conditioning layer. The present method uses descriptive statistics of FESEM images of biofilms obtained during a variety of growth conditions. We aim to explore here the texture and fractal analysis techniques, to identify the most discriminatory features which are capable of predicting the difference in biofilm growth conditions. We initially extract some statistical features of biofilm images on bare polymer surfaces, followed by those on the same substrates adsorbed with two different types of plasma proteins, viz. Bovine serum albumin (BSA) and Fibronectin (FN), for two different adsorption times. The present analysis has the potential to act as a futuristic technology for developing a computerized monitoring system in hospitals with automated image analysis and feature extraction, which may be used to predict the growth profile of an emerging biofilm on surgical implants or similar medical applications. Keywords: Biofilm, SEM image analysis, pattern analysis, fractal dimension, texture features 1. Introduction Biofilms are ubiquitous in nature 1–3, industrial locations 4–6, surgical infections, chronic wounds 7– 9 and other physiological conditions, where bacteria abandon their planktonic status and choose to survive in microbial communities associated with a surface 10–12. In biofilms, the bacteria remain encased within a self-secreted exopolysaccharide matrix 13–15. Biofilm formation has been recognized as the primary virulence factor in peri-operative and post-operative infections as well as in a variety of chronic infections 16,17. Bacteria in biofilms usually present higher resistance to antibiotics 18–20 and higher tolerance to the immune system 21,22 compared to its planktonic counterparts, posing the biggest challenge to public healthcare. Hence understanding the emergence and growth patterns of biofilms through continuous nondestructive monitoring 23,24 is the fundamental requirement for controlling biofilm growth. 1 Biofilm infections within a physiological environment can grossly have either tissue-related or device-related origin. The diversity in the conditions leading to the development of such biofilms, result in the difference in their physical and chemical characteristics - making biofilm monitoring methods extremely application specific. While microbiologically influenced corrosion (MIC) 25,26 represents the main cause of degradation of implanted devices of metallic origin, implants from polymer biomaterials form a robust and versatile class having unparalleled durability, biocompatibility 27, hemocompatibility 28,29, anti-thrombogenicity 30 and resistance to degradation and calcification 31. The present paper focuses on a computerized image monitoring system for device- related biofilms on widely used polymer biomaterials, which will enable a much better understanding of the processes leading to the emergence of biofilms and aid in the quantification of biofilm architecture on different substrates in a variety of growth conditions. 1.1. Fundamentals of Bacterial Growth and Biofilm Formation Bacterial biofilms are three-dimensional sessile structures consisting of bacteria encapsulated within hydrated extracellular polymeric substances (EPS) on a substrate 32. The exopolysaccharide matrix facilitates irreversible attachment between the biofilm cells and the substrate while simultaneously maintaining intercellular interactions, giving rise to a biofilm architecture specific to a particular substrate and definite bacterial strain 33. Medical implants from polymer biomaterials, often serve as the nidus to which bacteria can irreversibly attach via hydrophobic or electrostatic attractions during a surgical insertion,and proliferate in the form of biofilms. Biofilm-producing bacteria usually enter the body during the process of implantation, or exist on the surface of the implant pre-surgery and colonize the implanted device. Advances in infection control strategies include improved operating room, ventilation, sterilization methods, barriers, surgical techniques, and availability of antimicrobial prophylaxis. Despite these activities, surgical site infections (SSI) remain a prime cause of morbidity and mortality among hospitalized patients. This may be partly explained by theemergence of antimicrobial-resistant pathogens 34,35, mostly residing in biofilms, increased numbers of immuno-compromised elderly surgical patients and execution of increased numbers of prosthetic implant operations. Microbial contamination of the surgical site is however a necessary precursor of surgical site infections. Microorganisms may contain or produce toxins and other substances that increase their ability to invade a host, produce damage within the host, or survive on, or in host tissue. Gram-negative bacteria such as Pseudomonas aeruginosa after gaining access to the surgical site, produce a "slime" (mentioned as EPS), which physically shield the bacteria from phagocytes and inhibits the binding or penetration of antimicrobial agents 36. Many sets of stringent preventive measures have been deployed with the aim of preventing infections at the surgery site. However based on reports from National Nosocomial Infections Surveillance (NNIS) system, which monitors trends in nosocomial infections in U.S., acute-care hospitals surgical site complications, mostly accruing from biofilm-related infections, are the third most frequently reported nosocomial infection accounting for 14% to 16% of all nosocomial infections among hospitalized patients 37. Hence to reduce the risk of biofilm related infections, apart from the sets of clinical preventive measures, a systematic but realistic interdisciplinary approach must be applied with the awareness that this risk is influenced by characteristics of the patient, nature of implanted device, mode of surgery and maintenance of sterilization of personnel and hospital. A host of non-destructive methods have been employed to qualitatively and quantitatively study the underlying processes that govern the morphological development of biofilms with respect to different surfaces. These include microscopic techniques, image analysis, spectrochemical methods 24 including Fourier transform-infrared spectroscopy 38, electrochemical 39, calori-metric 40 and piezoelectric 41 approaches. However, rarely any effort has been applied to analyse the architecture of emerging biofilms from a hospital environment that can be of prime utility to healthcare professionals in administering the requisite antibiotics at the earliest opportunity. The computerized monitoring system which forms the topic of the present paper is a significant step towards the development of a futuristic methodology through which a non-invasive online monitoring of the surface of an implanted device may be made possible. The primary aim of the research is to detect the early emergence and growth pattern of biofilm on an implant surface, should there be one developing at the 2 surgery site, through existing image analysis techniques. The present method of monitoring is a unique non-destructive approach to this end. We analyse and quantify biofilm architectures through image analysis of FESEM data, using existing image processing techniques. This technique may be conveniently availed by health professionals in a hospital environment through a computer monitor, only by suitable coupling of a laparoscopic arrangement and FESEM at the surgery site. The microscopic arrangement will produce regular, clear, non-distorted images of the implant surface, with spatial resolution down to almost 1 nanometre, during the post-operative stages of an implant surgery. Early detection of biofilm growth, perceived through image analysis techniques, instead of invasive methods or use of colorimetric assays (which may be toxic to the physiological environment), will enable timely administration of antibiotic prophylaxis, before the infecting microbes turn antibiotic resistant- all eviating the morbidity and mortality rates in biofilm-related implant infections by a considerable amount. 1.2. Brief Survey of the Existing Biofilm Image Analysis Techniques This section, briefly reviews the contributions previously made in existing literature on biofilm quantification using various imaging techniques through different physical or chemical features, as well as previous works on mathematical modelling of biofilm growth. In contemporary research, several methods of biofilm characterisation using statistical techniques has been reported on 2D SEM images 42–48. Yang et al. 49 and Jackson et al. 50 have studied several textural features of microscope images, e.g. entropy, angular second moment, inverse difference moment, as well as morphological features, e.g. fractal dimension, porosity, run length, diffusion distance etc. Several image thresholding methods based on various entropy measures e.g. local entropy, joint entropy, relative entropy, Renyi's entropy have been explored by Yang et al. 51. A software tool for biofilm image segmentation, intensity quantification and spatial arrangement analysis has been reported in Daims et al. 52. New features in 2D microscope images have also been discussed to characterise the biofilm structure e.g. characteristic length by Milferstedt et al. 53, and diameter of cones created on the surface by Perni and Prokopovich 54. There have been similar analysis of biofilm images using 3D confocal laser scanning microscopy (CLSM) by Beyenal et al. 55, using various textural features (energy, entropy, homogeneity), as well as volumetric parameters (run length, aspect ratio, diffusion distance, fractal dimension). Similar confocal image analyses have been carried out by Mueller et al. 56, Truong et al. 57 and Ngo et al. 58, using features like biovolume 59, area to volume ratio, thickness, roughness, horizontal/vertical spreading etc. Some additional features like area distribution, surface to volume ratio have also been studied on confocal images, by Heydorn et al. 60, 61, Sandal et al. 62, Herzberg and Elimelech 63, using a software called Comstat. Image thresholding applied to confocal images has been described by Xavier et al. 64 in order to quantify biovolume and interfacial area. Studies of porosity estimation using confocal images have been performed by Lewandowski 65. Based on biovolume, thickness and roughness measure of confocal images, six different species of bacteria have been distinguished in Bridier et al. 66. Image segmentation of 2D slices of confocal images have been presented by Kyan et al. 67 while image segmentation of the 3D confocal images by Yerly et al. 68. Geostatistical analysis has been carried out on the confocal images by Kim et al. 69. The surface roughness of biofilms have characterized using both height and phase images by Auerbach et al. 70 using atomic force microscope (AFM) images. Phase contrast microscopy based features have been used by Liu et al. 71 to classify bacteria morphotypes e.g. roundness, elongation, compactness, maximum curvature, diameter, width, length, width/length ratio, area etc. Image thresholding and image registration related issues for AFM and CLSM images have been discussed by Webb et al. 72. Bacterial adhesion has been explored in 73 using AFM images through quantification of several surface characteristics e.g. thickness, roughness, advancing and receding contact angle, contact angle hysteresis etc. 3 There have been several studies on biofilm characterisation involving multiple imaging techniques. Surman et al. 74 have compared different microscope imaging techniques e.g. SEM, AFM, CLSM, transmission electron microscopy (TEM), environmental SEM (ESEM), episcopic differential interference contrast microscopy (DIC) with and without fluorescence, Hoffman modulation contrast microscopy (HMC) etc. Similar comparison have been done between SEM and TEM by Sangetha et al. 75and Espinal et al. 76, between SEM and confocal imaging by Villena et al. 77 and Ng et al. 78 and between SEM and variable pressure SEM (VPSEM) Weber et al. 79. A comparison between SEM, CLSM, and phase contrast microscopy has been reported in Norton et al. 80. Bacterial adhesion have also been explored using multiple imaging techniques including SEM, AFM, X-ray photoelectron spectroscopy (XPS), TEM and ultraviolet spectroscopy by Liu et al. 81. Bacteria removal performance has been compared by DeQueiroz and Day 82, using different imaging techniques like SEM, TEM, CSLM, and analysed using Fourier transform infrared (FTIR) spectroscopy. Identical analysis has also been used in other applications like material degradation 83. Comparison of biofilm characterization has been done through combination of SEM, CLSM and light microscopy using gram staining by Kania et al. 84. It is revealed that often combined approaches e.g. SEM and CLSM give better characterisation rather than using a single imaging technique, as reported in 85. A first principle partial differential equation (PDE) based mathematical modelling for bacterial growth has been established in Kreft et al. 86 where the simulated images are quantified using the above mentioned features along with heterogeneity and contrast. A different 2D mathematical model using cellular automata theory has been used to quantify the biofilm inner porosity by Hermanowicz 87. Apart from the grayscale images, the usefulness of various 2D/3D coloured image quantification softwares has been reviewed by Daims and Wagner 88. 1.3. Present Work and Contributions In the present work, we focus on the analysis of field emission scanning electron microscopy (FESEM) image analysis, due to its simplicity and non-destructive recording procedure. We show here that by using a combination of popular image textural features, it is possible to discriminate between several conditions of growth of the biofilm – as presented in the form of three distinct hypotheses. We also report their statistical significance level, using our experimental data. The statistical hypothesis tests are formed to find out whether there are any fundamental difference in the statistical characteristics of the biofilms grown using different substrates and adsorbed protein. Our prime aim is to find out the most significant features to support these hypotheses which are capable of indicating the difference in biofilm growth conditions from an image feature based pattern analysis perspective. 2. Materials and Methods 2.1. Biomaterials and Production of the Conditioning Layer Commercially available clinical grade High density polyethylene (HDPE) and poly-tetra-fluoro- ethylene (PTFE) used in orthopaedic implants and venous catheters respectively, were obtained after fine machine polishing in square configuration (10mm × 10mm) from Plastic Abhiyanta Ltd, India. These samples of biomaterials have been referred to as polymer chips throughout this article. After polishing, the samples were cleaned by 2 min ultrasonication in a 35 kHz ultrasonic bath (Rivotek Instruments, India) and thoroughly rinsed with demineralized water. The water used in all our experiments was of HPLC grade (Lichrosolv) from Merck India. Tris EDTA buffer solution pH 7.4 was obtained from Sigma Aldrich, USA. Bovine serum albumin (Fraction V) and Fibronectin from human plasma (CAS Number 86088-83-7, MDL Number MFCD00131062) lyophilized powder, MW: 45 kDa were obtained from M P Biomedicals, USA. PTFE chips were autoclaved, while HDPE chips being non-autoclavable,were rinsed twice with ethanol, blow dried and preserved in a vacuum desiccator. BSA solution ofconcentration1.5 mg/ml, prepared with buffer solution of pH 7.4 and left 4 for about a week with intermittent mixing to dissolve the BSA completely, was used for adsorption purposes. The reconstituted BSA solution taken in 10 ml glass vials, each containing a single polymer chip of either HDPE or PTFE were kept for 9 hours, and 24 hours. The adsorption times are referred here as exposure time 'τ', where τmax and τmin were 24 hours and 9 hours, respectively. After the stipulated time, the chips were removed from the BSA solutions, rinsed with water, and finally blow dried and preserved in a desiccator ready for growing biofilms. The chips obtained from adsorption experiments possessed different degrees of BSA adsorbed on them, and have been termed as the conditioning layer.These could not be sterilized further as it would lead to the denaturation of the adsorbed protein. The above process of adsorption was repeated for HDPE and PTFE chips with a reconstituted Fibronectin (FN) solution of concentration 1 μg/l. After 24 hours, all chips with conditioning layer of Fibronectin were removed from the Fibronectin solution and dipped thrice in demineralized water. The exposure time τ in Fibronectin is kept as 24 hours. The rest of the chips were left untreated to serve as controls.The concentrations of reconstituted BSA and Fn solutions werekept in proportion with their concentrations inthe plasma. The chips were all treated at 30°C, and τ was kept sufficiently high in all the cases, to obtain a complete surface coverage. After the production of the conditioning layer, the treated chips were thoroughly rinsed in Phosphate Buffer Saline (PBS) in order to remove any non-adsorbing protein molecules, finally leaving only irreversibly adsorbed protein molecules on the polymer surfaces. The untreated chips were rinsed with demineralized water and PBS, and all the experimental chips were finally blow-dried and preserved in a vacuum desiccator. 2.2. Bacterial Strain and Culture Condition A wild type strain of Pseudomonas aeruginosa obtained and isolated from uro-catheters of patients having urinary tract infections (UTIs) at the Department of Urology, Institute of Post Graduate Medical Education & Research, Kolkata, India was used in this study. The strain was previously reported to be a strong biofilm former by 96 well micro-titer plate assay. The goal was to prove experimentally that the pathogenic strain responsible for urinary infections (forming biofilm on silicon rubber) 89 is very much liable to affect orthopaedic implants. Each type of frozen culture was initially inoculated on Tripticase soy agar and incubated at 37°C for 24 hours. Each culture was then transferred via swab to a buffer solution and a suspension equivalent to a Macfarland 0.5 (~1.5 x 108 CFU/ml) was prepared. This suspension was diluted at 1:100 and 1 ml was used to inoculate 100 ml of sterile LB broth (Luria Bertanii Agar (LBB) obtained from Himedia, India).The bacteria grown overnight in LBB at 37°C were diluted in the same medium to an optical density of 0.5 at 600 nm and ready for use as bacterial culture for growing biofilms. 2.3. Biofilm Growth Condition The diluted culture obtained (in section 2.2), was poured over the surface of the treated chips (from section 2.1) placed in the wells of 24 well tissue culture plates (Tarsons, India) and on untreated chips kept in a separate 24 well plate, at 25°C. Each 24 well plate contained a separate set of chips with a definite adsorption time, and each system was closed and sealed without addition or removal of any component with the exception of broth. The sterile LBB was added carefully from time to time to avoid desiccation and incubated at 37°C for 7 days with shaking at 180 rpm. Each set of experiment was performed in triplicate. The plates were sealed and placed on the shaker plate of the biological oxygen demand (BOD) incubator. Care was taken to ensure that each plate was in an upright position during rotation, without tilting, which might affect the growth condition of the biofilms. After the entire 7-day growth period, the polymer chips were aseptically removed and washed thrice with phosphate buffered saline (PBS pH 7.2). This step eliminated all the free floating bacteria and only the sessile forms remained attached to the surface. The chips were then air dried and prepared for FESEM measurements. The chips with attached bacterial cells were covered with 2.5% 5 glutaraldehyde and kept for 3 hours then passed once through the graded series of 25%, 50% and 75% ethanol, and twice through 100% ethanol, for ten minutes each. These are finally transferred to the critical point drier and kept overnight to make them ready for biofilm analysis. To compare the architecture of the biofilms produced by the clinically isolated strain of Pseudomonas aeruginosa on different substrates after 7 days, FESEM measurements were conducted at 5.0 kV-10 kV in a field emission scanning electron microscope (FESEM: Inspect F50, FEI Europe BV, The Netherlands; FP 2031/12, SE Detector R580). For this purpose the dried polymer chips, with and without biofilms, were sputter-coated with a 3-nm thick conductive layer of gold. 2.4. Experimental Conditions and the Research Question Our experiments reveal a difference in biofilm architecture of a specific strain of bacteria, in response to variations in the nature of substrate and conditioning layer. We report here four different combinations of growth condition of biofilms. The main motive is to identify the most significant features that may be used to distinguish between biofilms grown on different substrates, and having dissimilar conditioning layers. We have studied four different cases and compared biofilms grown on bare HDPE/PTFE, with that grown on HDPE/PTFE surfaces adsorbed with BSA/Fibronectin for 9 hours and 24 hours. Three different image analysis techniques have been adopted for statistical feature extraction and hypothesis testing 90 :   Image gray level histogram (first order statistics) – central moments and standardized moments Image texture analysis (second order statistics) based on relative position of the pixels using gray level co-occurrence matrix (GLCM)  Fractal analysis and Haussdorf dimension We hypothesize that, if there is any significant difference between the biofilms determined by differences in – 1) adsorption time of the proteins 2) nature of protein used as conditioning layer and 3) nature of substrate used for growing the biofilms it is possible to discriminate between these conditions, using some textural features of the SEM images. This essentially helps in the statistical characterization of the biofilms when their experimental conditions are not known in a real implant monitoring scenario. The analyses was carried out on images with a fixed magnification factor of 10,000x. Here we make the following three hypotheses and also test the statistical significance levels of each feature in comparing these hypotheses, as follows: 1) Comparison with respect to adsorption time: Biofilms grown on HDPE control (bare surface), HDPE adsorbed with BSA for 9 hours and 24 hours 2) Comparison with respect to conditioning layer: Biofilms grown on HDPE control (bare surface), HDPE BSA 24 hours, HDPE FN 24 hours 3) Comparison with respect to substrate: Biofilms grown on another (PTFE) substrate - PTFE control (bare surface), PTFE BSA 24 hours, PTFE FN 24 hours. We use here, the one way analysis of variance (ANOVA) to test the hypothesis that the population mean in each group are different along with the non-parametric version of ANOVA test 6 and we also report the statistical significance of the best features that are found capable of discriminating between these groups. 3. Description of the Statistical Features of Biofilm Images 3.1. Feature from First Order Statistics of Images The first order statistics of grayscale images refer to the statistical parameters extracted from the one dimensional probability density function of the pixel intensities. The grayscale biofilm image is ip expressed as a normally considered as a 2D matrix of random intensity i with a histogram bin count ratio between the number of pixels with an intensity i and the total number of pixels N. The maximum number of graylevel is given by gN , as per standard grayscale image processing whose value is 256.   ip pixels with graylevel  i N (1) ip , the central moments (mean subtracted) and standardized moments From the histogram count (mean removed, followed by unit standard deviation) can be calculated.The four descriptive statistics (mean, standard deviation, skewness, kurtosis) can now be calculated directly from the normalized image histograms as follows:     I   N  1 g  i  0 ip i ,     I  2    N  1 g  i  0  i  2   p i ,  i  3   p i  3 , (2)       I  3             I  4        i  4   p i  4 . N  1 g   i N 0  1 g  i  0 Here  represents the mathematical expectation operator. Similarly the energy (E) and entropy (H) of the pixel intensity can also be calculated from the histogram as: E   i 2 p i , H    i p i log 2   p i The smoothness texture (SM) 91 is calculated from the variance ( 2 ) as: SM    1 1 1  2    (3) (4) Example of histograms for grayscaled biofilm SEM images on different cases of HDPE substrate have been shown in Figure 1, with 256 discrete intensity levels. It is evident from Figure 1 that, while for the bare HDPE surface the histogram appears continuous, in the other cases few discrete pixel intensities are dominant and the histograms appear sparse. Here all the image histograms have been reported on a normalized scale with respect to the total bin count so that the area under the histograms becomes unity, and hence they could be interpreted as probability density functions (PDFs) in pixel intensity. A similar analysis of the histograms of the biofilms formed on the PTFE surface in Figure 2, shows a more continuous distribution of pixel intensities for all the cases, with a more left-skewed distribution, compared to those in the HDPE case as shown in Figure 1. It is important to note here 7 that the histograms shown in this section do not include all the image dataset but explore a few example images, as representative examples for the feature extraction from grayscale images. Figure 1: (Top panel) Grayscale biofilm imagesfor HDPE, (Bottom panel) Normalised histogram of the pixel intensities. Figure 2: (Top panel) Grayscale biofilm images for PTFE, (bottom panel) Normalised histogram of the pixel intensities. 3.2. Features from Second Order Statistics of Images The second order statistics of grayscale images refer to the spatial correlation analysis on pixel intensities along both the x and y-direction. The image is first converted to Gray Level Co-occurrence Matrix (GLCM), denoted as jp which measures the two dimensional relative frequencies, or the probability of two pixels with intensity i and intensity j occurring in a neighbourhood of distance   92. The GLCM matrix can either be interpreted as co-occurrence of two intensities in the image separated by pixel distance   or, separated by radial distance d and x ,   as ,x ,x  y  y ,i  p i j ,  y 8 , angle θ as  section, the GLCM    or ,x   y  p i j d  . Depending on the discrete graylevels ( , gN ) of the image as described in earlier is derived as a G G matrix, for different combinations of either  ,p i j  ,d  . Different features can now be extracted from the GLCM matrix 93,94, as follows: Here, contrast = variance = inertia =  i i , j  2 j p i ,  j correlation  i   x    j y  y x   p i ,  j   i , j energy = uniformity = angular second moment homogeneity inverse difference moment     p i j ,  2   i , j j ,  p i   1i i j ,  j (5) (6) (7) (8) N g  1 N g  1    and   .   . ij i  0 j  0    p i p x ,    j y are the marginal probability densities, summed over the y and x-direction respectively, i.e.   p i x  N  1 g  j  0  p i j p  , ,   j  y N  1 g  i  0  p i j ,  (9) Parameters   y     denote the mean and standard deviation of the marginal densities , , , x y x along the two axis in (9) respectively. For a homogeneous image there will only be few high jp , so the 2D energy or ASM in equation (7) will be higher. The contrast in amplitude elements in ,i equation (5) measures local intensity variation and favours contributions between elements away from the diagonal of GLCM. Also, because of the inverse relationship in the homogeneity estimate in equation (8), the inhomogeneous areas of the GLCM contribute less, whereas homogenous areas makes it a higher value. The correlation in equation (6) measures the linear dependence between image pixels relative to each other in both the directions. These features could have been extracted from the raw images or after different transformations to see if the discriminatory capability of the features is enhanced under such pre-processing. Here we tested on both the raw grayscale images and its contrast enhanced version via histogram equalization (to get close to a uniform distribution of the pixel intensity) and it was discovered that the raw images are more reliable to extract features because automated preprocessing steps may often introduce a systematic bias in the image data. Therefore, the rest of the analysis in this paper has been shown only with the raw image data with an aim of showing class discrimination and not using any transformed domain image data. 3.3. Fractal Analysis and Hausdorff Dimension For fractal dimension calculation, the grayscale images need to be converted to a binary image for the box-counting based fractal analysis. Setting an optimum threshold has always been a crucial step in fractal analysis since an arbitrarily chosen threshold may spuriously increase or decrease the ratio of black and white pixels and change the pattern of the original continuous grayscale image in its 9 discretised binary version. For example, a threshold at the centre of the graylevel (128), may result in disproportionate white and black pixels, with arbitrary modification of the group mean and variance values, thus distorting the fine granularity of the original grayscale image in its binary version. We here use the Otsu's method 95 for automatic determination of the threshold which minimizes the intra- class variance between the black and white pixels. The box-counting algorithm returns the number of black boxes ( bN ) of size R. For faster  number of numerical computation the box sizes are increased as powers of 2 i.e. pixels. The algorithm terminates when the maximum of size of the image in either x or y direction becomes less than 2P . The negative slope of the box-size (R) vs. box-counting ( bN ) curve in log-log fD ), and their empirical power law plot, is known as the fractal dimension or Hausdorff dimension ( scaling is given by (10). R  0 2 2 ,2 , 2 , 1 , 2P bN fD R (10) In order to find out the fractal dimension fD , the natural logarithm of both box-size ( log R ) and box-counting ( log bN ) were computed first, followed by a least-square regression with a first order polynomial (straight line) fitting 96. The negative slope of this line now indicates the fractal dimension for four different cases of biofilms on the HDPE substrate in Figure 3, using either a bare surface, or with a conditioning layer of BSA adsorbed for 9 hours/24 hours or with a conditioning layer of FN 24 hours. Figure 3: (Top panel) Binary image after thresholding the grayscale image for HDPE substrate, (Bottom panel) Box-counting fractal dimension estimation of the respective cases. A similar exploration on the PTFE substrate has been shown in Figure 4. It is also important to notice that there is minimal loss of basic image characteristics using the optimal image thresholding method to convert the original grayscale SEM image in a binary image for fractal analysis. A different image threshold, if not chosen optimally, could affect the box-counting and hence the fractal dimension estimates. However this has not been explored in the present analysis, as it would be a digression from the main topic of this paper. 10 Figure 4: (Top panel) Binary image after thresholding the grayscale image for PTFE substrate, (Bottom panel) Box-counting fractal dimension estimation of the corresponding binary images. 3.4. Effect of Contrast Enhancement as Preprocessing Figure 5: Effect of contrast enhancement on the image pdf and cdf. The morphological features discussed in the previous sections could have been extracted after some preprocessing of the raw image, e.g. different kind of filtering to improve image clarity. One such method know as contrast enhancement via histogram equalization, is widely used in the image processing literature before feature exraction. An example of this procedure is shown in Figure 5, where the raw grayscale image shows a sigmoid type cumulative distribution function (CDF) of the 11 pixel intensity. The purpose of the contrast enhancement is to make the PDF of the grayscale image close to a uniform distribution, so that the corresponding CDF becomes a ramp like function, with a steady increase in pixel intensity over all the bins. In general the clarity of images can be increased with contrast enhancement and thus may often help in textural feature extraction. In this study, we explore its effect on fractal analysis, as the contrast enhancement may drastically change the proportion of black and white pixels in the binary image after thresholding. Figure 6: Effect of contrast enhancement of the raw image in grayscale to binary conversion. The effect of contrast enhancement on the image thresholding (for fractal analysis) has been shown in Figure 6. It is evident that although with Otsu's method of automatic thresholding, the contrast enhanced grayscale version produces better clarity of the image, some of the fine structures in the biofilm architecture get smeared in the binary version. A simple binary image however preserves this information. It is also found that with the contrast enhanced version, the same set of features produces higher within group variance compared to the raw image based features. Therefore in the remainder of the statistical analysis, only the raw grayscale images and its binary version have been focussed upon. 3.5. Feature Correlation Analysis Due to lack of independent experiments in our present research and difficulty in producing a number of exactly similar sets of bacterial cultures using the wild clinical Pseudomonas aeruginosa strains, the same large images have been resampled with smaller time window, to carry out statistical analysis. This is a viable option for large data where a smaller segment of the data can be assumed to come from different experiments. All the SEM images chosen for the present analysis have a magnification factor 10,000x. For calculating the above mentioned first/second order statistical features and fractal dimension, a smaller blocks of 512×512 pixels have been considered as a fixed scanning window for each of the original images which was slided by 10 pixels along both the x and y directions for calculating resampled statistics from the original SEM images. This method scans 12 through the different local properties of an image and check the consistency of the features using the criteria of smaller within class variance. In each of the groups under study, at least 2 sets of independent SEM images were taken and in some cases 3 sets, for a single experimental condition at the same magnification factor. Since biofilm formation is a stochastic process, it is found to be absolutely nonuniform in stray portions of the substrate while it has a definite pattern on rest of the substrate. The SEM images which were not of good quality and having absolutely non-repetitive pattern were outright discarded. Also the SEM images having varying magnification factor were not included in the present study. Hence finally only those SEM images which passed this criteria of good quality and repetitive patterns were included in the study. In the present study, it was difficult to repeat many experiments with clinical or wild bacterial strains as the same strain may not be available repetitively. The protocol for biofilm formation however remanins the same throughout our experiments for all substrates and all conditioning layers for all strains of the microbes. Hence the experimental results can be considered to be quite robust as also reported previously in Dutta Sinha et al. 89. The calculation of different statistical features were done on each set of data separately (also by resampling each image) and their variability has also been reported in the final results. Figure 7: PTFE control group feature correlation plot. In statistical feature based grouping analysis, it is often found that some of the features may be closely related or correlated. In order to verify it, we here carry out a feature correlation analysis as explored in Figure 7 on the PTFE bare surface, for benchmarking purposes. The features or variables are taken in the following order, before using any ranking for group analysis – fractal dimension (var1), mean (var2), standard deviation (var3), skewness (var4), kurtosis (var5), energy (var6), entropy (var7), smoothness (var8), contrast (var9), correlation (var10), homogeneity (var11), 2D energy (var12), mode (var13). In some cases of the joint distributions, a high (>0.9) value of the correlation coefficient between a pair of features is observed. However a conclusive statement of the correlation analysis needs to be verified on a large dataset. On the PTFE substrate (as shown in Figure 7), the scatter plots show much wider variance and less correlation, indicating new information about the structures of the biofilm SEM images captured by most of the features. A variable correlation 13 analysis helps to identify the redundant information going into the class discrimination problem. In the next subsection these features are first ranked according to their class-separability measures, and then used for hypothesis testing. 3.6. Class Separability Measures and Feature Ranking For binary class problem Fisher's discriminant ratio (FDR) is a widely used measure for ranking features according to their statistical discrimination ability. When dealing with a multi-class discrimination problem, usually the scatter matrix in eqn. (11) is used to quantify the class separability using one/multiple features but the effectiveness of each individual feature needs to be judged at the outset 97. The scatter matrix is given by equation (11), as an extension of FDR for multiclass statistical discrimination problem. J   1 trace S S w   b S w  S b  c   1 i c  i  1 PS i i , P m m m m i 0 0 i i     . Pm i i (11) T  , m 0  c  i  1 Here, {Sm, Sw, Sb} represent mixture scatter matrix, within-class scatter matrix and between-class scatter matrix respectively. Pi is the a priori probability of each class i with respective mean and covariance being {mi, Si}. Also, m0 is known as the global mean vector that considers data points from all the classes. Figure 8: Variation in class separability measure based on scatter matrix with sorted features in the three discrimination problems. The features are sorted using decreasing value of scatter measure J. Considering each feature individually the above covariance matrices reduce to the individual variance. We here rank the features using the univariate scatter score J. The scatter histograms of the top three features have also been explored later on, to show the class separability performance in a 2- dimensional feature space. Before calculating the scatter measures, each feature (f) has been standardized to zero mean and unit variance to avoid any bias due to different ranges of features using 14 f i std   f   i i f  i f  , (12) where   f  denote the mean and variance of ithfeature respectively. ,i f i According to the three distinct hypotheses mentioned in section 2.4, the above 13 statistical features are ranked first according to their decreasing level of significance and shown in Figure 8. It is understandable that for each of these hypothesis all the features will not get a similar priority which is explored in the next section. 3.7. Hypothesis Testing Usually the one way analysis of variance (ANOVA) tests the null hypothesis that the mean values of multiple groups are different with an underlying assumption that all the samples in each group came a Gaussian distribution with a common variance. In more realistic cases, this condition is violated and a non-parametric version of one way ANOVA i.e. the Kruskal-Wallis test is preferred 98. Also the classical F-statistic in ANOVA is replaced by χ2 statistic and the significance level is measured by the p-value. A result is considered to be significant if the p-value is below a certain value. The null hypothesis here for the non-parametric test is that the features from different groups came from the same distribution and their median values are the same, against the alternative hypothesis that their medians are different. A low p-value rejects the null hypothesis and the alternative hypothesis is accepted. Due to multiple comparisons on the same data (although using different features), the significance levels are corrected using Bonferroni's method by dividing the actual p-values (p<0.01) with the number of tests conducted on the same data. 4. Results and Discussions 4.1. Results for Hypothesis 1: Effect of Adsorption Time Variation on HDPE Figure 9: Feature box plot for BSA conditioning layer on HDPE substrate for 9H vs. 24H The ranked features for the hypothesis 1 capturing the effect of protein adsorption time on HDPE substrate has been listed in Table 1, along with the significance levels of the Kruskal-Wallis test on each feature with Bonferroni correction. The features are ranked using the scatter matrix (J) showing a clear separation of the central tendency (mean/median) and small within class variance. The χ2 statistic corresponding to each ANOVA table has been reported showing a gradual decrease in χ2 15 statistic with the scatter matrix (J). The results have been graphically visualised in the form of box- plots for 12 features (except the mode) in Figure 9. In all the boxplots, the median, interquartile range (IQR) and outliers are shown using the red horizontal line, blue box and red cross markers respectively. Here, the fractal dimension for the HDPE BSA are found to have a minimum within class variance and ranging between 1.8-1.85. The higher moments like skewness and kurtosis also shows the deviation from the Gaussian behaviour justifying the choice of nonparametric hypothesis test over the parametric ANOVA test. The steady increase in energy is almost found to have an inverse relation with a steady decrease in entropy when the substrate for biofilm growth changed from HDPE bare surface to HDPE surface adsorbed with BSA for 9 hours and 24 hours. The contrast, homogeneity and 2D energy also show either an increasing or decreasing pattern in these three conditions. Table 1: Sorted features for BSA conditioning layer on HDPE substrate for 9H vs. 24H based on decreasing J Scatter Measure J Feature Description χ2 p-value 8.761 5.311 3.819 3.632 3.288 3.078 1.863 1.835 1.470 1.133 0.3896 0.05793 0.01055 F1 = Homogeneity F2 = Energy F3 = Correlation F4 = 2D Energy F5 = Contrast F6 = Smoothness F7 = Entropy F8 = Standard Deviation F9 = Mean F10 = Mode F11 = Kurtosis F12 = Fractal Dimension F13 = Skewness 1.054×104 7.903×103 9.048×103 8.439×103 1.054×104 8.348×103 7.903×103 8.348×103 8.067×103 7.421×103 4.003×103 3.943×102 3.277×102 0 0 0 0 0 0 0 0 0 0 0 2.417×10-86 6.838×10-72 The results can also be viewed as scatter plots or joint distributions of the most significant set of features. The clear separation of the three groups amongst different combinations of energy, homogeneity and correlation are explored in Figure 10. A kernel density smoothing is used to compute the marginal distributions (projections on either X or Y axes) from the scatter diagrams of the feature pairs of the three groups. Although in some cases the 1D marginal distributions show some degree of overlap between different groups, in a higher dimensional space, i.e. considering feature combinations, instead of simple class discrimination using hypothesis tests reported above will result in a very clear separation between these groups. The gradual separation of the three groups in 1D box plot to 2D and 3D scatter plot indicates that for such class separation problem, feature pairs or groups should definitely be used to quantify and discriminate different biofilm morphologies. 16 (a) (b) 17 (c) (d) Figure 10: Scatter plots and histograms of top three features for hypothesis 1: effect of adsorption time on HDPE (a) F1 vs. F2, (b) F2 vs. F3, (c) F1 vs. F3, (d) F1-F3 18 4.2. Results for Hypothesis 2: Effect of Variation of Conditioning Layer BSA vs. FN A similar statistical analysis has been carried out to test the effect of different proteins (BSA and FN) on the same HDPE substrate, using these features as shown in Table 2. The fractal dimension in this case has been found to be more compact for the HDPE BSA 24 hours case, as shown in Figure 11. Apparently from the scatter diagram of energy and entropy in Figure 12(a), the marginal distributions may appear to be overlapping. However the clear separation between these two groups using another feature 2D energy can be viewed in the Figure 12(b)-(d) in the 2D and 3D feature space. Figure 11: Feature box plot for HDPE substrate with different conditioning layers BSA 24H vs. FN 24H. Table 2: Sorted features for HDPE substrate with different conditioning layers BSA 24H vs. FN 24H based on decreasing J Scatter Measure J Feature Description χ2 p-value 78.25 62.86 15.34 7.182 3.028 2.595 1.982 1.149 1.116 0.8098 0.687 0.2033 0.1272 F1 = Entropy F2 = Energy F3 = 2D Energy F4 = Homogeneity F5 = Contrast F6 = Smoothness F7 = Standard Deviation F8 = Correlation F9 = Mode F10 = Mean F11 = Kurtosis F12 = Fractal Dimension F13 = Skewness 7.903×103 7.903×103 8.439×103 1.054×104 1.054×104 8.348×103 8.348×103 9.048×103 7.421×103 8.067×103 4.003×103 3.943×102 3.277×102 0 0 0 0 0 0 0 0 0 0 0 2.417×10-86 6.838×10-72 19 (a) (b) 20 (c) (d) Figure 12: Scatter plots and histograms of top three features for hypothesis 2: effect of conditioning layer BSA vs. FN on HDPE substrate (a) F1 vs. F2, (b) F2 vs. F3, (c) F1 vs. F3, (d) F1-F3 21 4.3. Results for Hypothesis 3: Testing on another Substrate PTFE Figure 13: Feature box plot for PTFE substrate with different conditioning layers BSA 24H vs. FN 24H. Table 3: Sorted features for PTFE substrate with different conditioning layers BSA 24H vs. FN 24H based on decreasing J Scatter Measure J Feature Description χ2 p-value 0.9764 0.8848 0.7633 0.7345 0.7095 0.6814 0.6542 0.6514 0.6334 0.5828 0.4313 0.1252 0.06988 F1 = Homogeneity F2 = 2D Energy F3 = Fractal Dimension F4 = Contrast F5 = Kurtosis F6 = Skewness F7 = Standard Deviation F8 = Entropy F9 = Energy F10 = Smoothness F11 = Correlation F12 = Mean F13 = Mode 1.054×104 8.439×103 3.943×102 1.054×104 4.003×103 3.277×102 8.348×103 7.903×103 7.903×103 8.348×103 9.048×103 8.067×103 7.421×103 0 0 2.417×10-86 0 0 6.838×10-72 0 0 0 0 0 0 0 As the third hypothesis while testing characteristics of biofilm architectures on the PTFE substrate the fractal dimension is found to be more compact with FN protein. The higher standardized moments – skewness and kurtosis show significant deviation from Gaussianity for all the cases. Also the inverse relationship between energy and entropy can still be inspected in the boxplots in Figure 13. The decrease in χ2 statistic of the ANOVA table is found to be similar to the decreasing scatter matrix in Table 3, as also found before. The 2D scatter plots using the top 3 features in Figure 14 shows many compact but disjoint islands, resulting in some overlapping regions in the 1D marginal distributions, especially using fractal dimension and 2D energy. However the class separation in the 3D feature space is clear in the 3D scatter diagram in Figure 14(d), as different groups form non- overlapping disjoint clouds. 22 (a) (b) 23 (c) Figure 14: Scatter plots and histograms of top three features for hypothesis 3: effect of conditioning layer BSA vs. FN on PTFE substrate (a) F1 vs. F2, (b) F2 vs. F3, (c) F1 vs. F3, (d) F1-F3 (d) 24 4.4. Discussions It is observed from the above reults that different features emerge for different hypothesis or experimental conditions and the variation of the features (extracted from multiple images taken at the same condition) are reported in the box-plots and already been penalised in the feature selection process using Scatter matrix. For any feature, if the variance is high for same experimental condition (coming from different images) or due to resampling (smaller sub-segments of the same image), both would show a low scatter measure as in Figure 8 due to increased variance if either of the case occur. There is less chance that due to low variance (within and between experiments) and high scatter measure, the top 3 features (as shown in this section) would capture non-repeatable patterns. However to have a conclusive answer on this, we plan to report large scale clinical trials with our collaborative hospitals in a future study. However we feel that repeatability through 2-3 independent tests and resampling on each SEM image to create multiples is an established statistical characterisation method which is adopted here as well. Since biofilm structures are complex and heterogeneous, there have been lot of effort to analyse such microstructures, but the exact objective of investigation may be different. While some may determine, the amount of EPS matrix, others may quantify the total number of bacterial cells embedded in biofilm or the effective number of living bacteria in biofilm. Such different targets usually require different approaches. Amonsgt many approaches the colorimetric methods 99,100 are capable of quantifying living bacteria in a biofilm. However the target of such colorimetric assay is quantification of living bacterial cells in a particular biofilm and not the biofilm architecture as a result of microbe-substrate interaction. The present paper describes a novel method of quantification of differences in biofilm architecture on different substrates to gain insight about the contribution of the substrate in modulating the biofilm forming capabilities of a specific microbe. Also, we have chosen to work with a clinical strain, instead of working with reference (laboratory) strains. We have gone forth with the clinical pathogenic strains obtained from the urinary catheters of patients suffering from urinary infections at a public health care unit in Kolkata, India. Our goal was to prove experimentally that a pathogenic strain of bacteria, responsible for urinary infections (forming biofilm on silicon rubber) is very much liable to affect orthopaedic implants, if they get the required access. The present paper shows that the substrate-microbe and conditioning layer affecting the biofilm architecture can be successfully quantified from statistical analysis of SEM images. The study reported here is based on smaller set of clinical strains as a proof of concept study. Large scale analysis of wild clinical strains may be studied in a future work by varying the pathogen species and SEM magnification factor. The objective of this paper is to primarily establish the quantification method by feature extraction, ranking and carrying out hypothesis testing to see separation of different biofilm formation conditions, in the feature space. 5. Conclusions and Scope of Future Work The significant feature sets have been identified for discriminating biofilms of a wild Pseudomonas aeruginosa strain based on biofilm growth conditions using SEM images via image feature extraction and ranking schemes. Three different biofilm growth conditions, including the substrate, conditioning layer protein and absorption time variation which are tested as three distinct hypotheses using the ranked features and the corresponding statistical significance levels have also been reported. Various textural features related to the fractal nature and first/second order statistics of the SEM images in the respective boxplots, kernel density smoothed histograms and 2D/3D joint distributions or scatter plots also qualitatively show the class separability between these image groups. Different set of features have emerged as the most significant ones for comparing different groups including homogeneity, energy, correlation, entropy, 2D energy and fractal dimension etc. However the analysis can be extended with a much larger database and a large pool of features using various other transformed domain features of the SEM images and can be explored in a future research. The 25 present research can be considered as a step forward in that line and briefly reviews the existing contributions on biofilm characterization through SEM images. The long term goal of our research may be summarized as the estimation of substrate-microbe interactions using a clinical strain of bacteria on different substrates. Such quantification approach may serve as an inexpensive procedure for quantifying substrate-microbe interactions and hence determine the degree of bio-incompatibility of different substrates. In addition this method can be beneficial to scientists from different disciplines, working on biofilms, but lacking the training of microbiological methods. We would like to extend our work in future to Gram positive bacteria strains and on other biomaterial surfaces. Future scope of work may also include the use of hybrid characterization techniques through multiple imaging techniques like CSLM or AFM and assimilation of experimental data with computer simulation of biofilm patterns. Acknowledgement SDS acknowledges the funding from the Department of Science and Technology (DST), Govt. of India through the Women's Scientist Scheme – A, project no. LS-466/WOS A/2012-2013. References: (1) McLean, R. J.; Whiteley, M.; Stickler, D. J.; Fuqua, W. C. Evidence of Autoinducer Activity (2) (3) (4) (5) (6) (7) (8) in Naturally Occurring Biofilms. FEMS Microbiology Letters 1997, 154, 259–263. Yang, J.-L.; Li, X.; Liang, X.; Bao, W.-Y.; Shen, H.-D.; Li, J.-L. Effects of Natural Biofilms on Settlement of Plantigrades of the Mussel Mytilus Coruscus. Aquaculture 2014, 424, 228– 233. Mieszkin, S.; Martin-Tanchereau, P.; Callow, M. E.; Callow, J. A. Effect of Bacterial Biofilms Formed on Fouling-Release Coatings from Natural Seawater and Cobetia Marina, on the Adhesion of Two Marine Algae. Biofouling 2012, 28, 953–968. Gutiérrez, D.; Delgado, S.; Vázquez-Sánchez, D.; Martinez, B.; Cabo, M. L.; Rodriguez, A.; Herrera, J. J.; Garcia, P. Incidence of Staphylococcus Aureus and Analysis of Associated Bacterial Communities on Food Industry Surfaces. Applied and environmental microbiology 2012, 78, 8547–8554. Marcato-Romain, C.-E.; Pechaud, Y.; Paul, E.; Girbal-Neuhauser, E.; Dossat-Letisse, V. Removal of Microbial Multi-Species Biofilms from the Paper Industry by Enzymatic Treatments. Biofouling 2012, 28, 305–314. Tang, X.; Flint, S.; Bennett, R.; Brooks, J. The Efficacy of Different Cleaners and Sanitisers in Cleaning Biofilms on UF Membranes Used in the Dairy Industry. Journal of Membrane Science 2010, 352, 71–75. Zimmerli, W.; Moser, C. Pathogenesis and Treatment Concepts of Orthopaedic Biofilm Infections. FEMS Immunology & Medical Microbiology 2012, 65, 158–168. Percival, S. L.; Hill, K. E.; Williams, D. W.; Hooper, S. J.; Thomas, D. W.; Costerton, J. W. A Review of the Scientific Evidence for Biofilms in Wounds. Wound repair and regeneration 2012, 20, 647–657. (9) Wolcott, R.; Rhoads, D.; Bennett, M.; Wolcott, B.; Gogokhia, L.; Costerton, J.; Dowd, S. Chronic Wounds and the Medical Biofilm Paradigm. Journal of wound care 2010, 19, 45– 46. 26 (10) Alhede, M.; Kragh, K. N.; Qvortrup, K.; Allesen-Holm, M.; van Gennip, M.; Christensen, L. D.; Jensen, P. Ostrup; Nielsen, A. K.; Parsek, M.; Wozniak, D.; et al. Phenotypes of Non- Attached Pseudomonas Aeruginosa Aggregates Resemble Surface Attached Biofilm. PloS one 2011, 6, e27943. Fletcher, M.; Savage, D. C. Bacterial Adhesion: Mechanisms and Physiological Significance; Springer Science & Business Media, 2013. Singh, A. V.; Vyas, V.; Patil, R.; Sharma, V.; Scopelliti, P. E.; Bongiorno, G.; Podesta, A.; Lenardi, C.; Gade, W. N.; Milani, P. Quantitative Characterization of the Influence of the Nanoscale Morphology of Nanostructured Surfaces on Bacterial Adhesion and Biofilm Formation. PLoS One 2011, 6, e25029. Flemming, H.-C.; Wingender, J. The Biofilm Matrix. Nature Reviews Microbiology 2010, 8, 623–633. (11) (12) (13) (14) Wingender, J.; Neu, T. R.; Flemming, H.-C. Microbial Extracellular Polymeric Substances: (15) Characterization, Structure and Function; Springer Science & Business Media, 2012. Sheng, G.-P.; Yu, H.-Q.; Li, X.-Y. Extracellular Polymeric Substances (EPS) of Microbial Aggregates in Biological Wastewater Treatment Systems: A Review. Biotechnology advances 2010, 28, 882–894. (16) Mombelli, A.; Décaillet, F. The Characteristics of Biofilms in Peri-Implant Disease. Journal of clinical periodontology 2011, 38, 203–213. (17) Ahmed, R.; Greenlee, J. D.; Traynelis, V. C. Preservation of Spinal Instrumentation after Development of Postoperative Bacterial Infections in Patients Undergoing Spinal Arthrodesis. Journal of spinal disorders & techniques 2012, 25, 299–302. (18) Høiby, N.; Bjarnsholt, T.; Givskov, M.; Molin, S.; Ciofu, O. Antibiotic Resistance of Bacterial Biofilms. International journal of antimicrobial agents 2010, 35, 322–332. Soto, S. M. Role of Efflux Pumps in the Antibiotic Resistance of Bacteria Embedded in a Biofilm. Virulence 2013, 4, 223–229. (19) (20) Mah, T.-F. Biofilm-Specific Antibiotic Resistance. Future microbiology 2012, 7, 1061–1072. (21) (22) Bjarnsholt, T.; Jensen, P. O.; Burmølle, M.; Hentzer, M.; Haagensen, J. A.; Hougen, H. P.; Calum, H.; Madsen, K. G.; Moser, C.; Molin, S.; et al. Pseudomonas Aeruginosa Tolerance to Tobramycin, Hydrogen Peroxide and Polymorphonuclear Leukocytes Is Quorum-Sensing Dependent. Microbiology 2005, 151, 373–383. Lewis, K. Multidrug Tolerance of Biofilms and Persister Cells. In Bacterial Biofilms; Springer, 2008; pp. 107–131. (23) Nivens, D.; Palmer Jr, R.; White, D. Continuous Nondestructive Monitoring of Microbial Biofilms: A Review of Analytical Techniques. Journal of Industrial Microbiology 1995, 15, 263–276. Feng, J.; de la Fuente-Núñez, C.; Trimble, M. J.; Xu, J.; Hancock, R. E.; Lu, X. An in Situ Raman Spectroscopy-Based Microfluidic "lab-on-a-Chip" Platform for Non-Destructive and Continuous Characterization of Pseudomonas Aeruginosa Biofilms. Chemical Communications 2015, 51, 8966–8969. 27 (24) (25) Souza, J. C.; Henriques, M.; Teughels, W.; Ponthiaux, P.; Celis, J.-P.; Rocha, L. A. Wear and Corrosion Interactions on Titanium in Oral Environment: Literature Review. Journal of Bio- and Tribo-Corrosion 2015, 1, 1–13. (26) Koch, T. Microbial Induced Corrosion. Encyclopedia of Lubricants and Lubrication 2014, (27) (28) (29) 1150–1154. Bauer, S.; Schmuki, P.; von der Mark, K.; Park, J. Engineering Biocompatible Implant Surfaces: Part I: Materials and Surfaces. Progress in Materials Science 2013, 58, 261–326. Bakir, M. Haemocompatibility of Titanium and Its Alloys. Journal of biomaterials applications 2012, 27, 3–15. Solouk, A.; Cousins, B. G.; Mirahmadi, F.; Mirzadeh, H.; Nadoushan, M. R. J.; Shokrgozar, M. A.; Seifalian, A. M. Biomimetic Modified Clinical-Grade POSS-PCU Nanocomposite Polymer for Bypass Graft Applications: A Preliminary Assessment of Endothelial Cell Adhesion and Haemocompatibility. Materials Science and Engineering: C 2015, 46, 400– 408. (30) Manabe, K.; Kyung, K.-H.; Shiratori, S. Biocompatible Slippery Fluid-Infused Films Composed of Chitosan and Alginate via Layer-by-Layer Self-Assembly and Their Antithrombogenicity. ACS applied materials & interfaces 2015, 7, 4763–4771. (31) Wang, X.; Zhai, W.; Wu, C.; Ma, B.; Zhang, J.; Zhang, H.; Zhu, Z.; Chang, J. Procyanidins- (32) Crosslinked Aortic Elastin Scaffolds with Distinctive Anti-Calcification and Biological Properties. Acta biomaterialia 2015, 16, 81–93. Rice, S. A.; Wuertz, S.; Kjelleberg, S. Next-Generation Studies of Microbial Biofilm Communities. Microbial Biotechnology 2016, 9, 677–680. (34) (33) Nandakumar, V.; Chittaranjan, S.; Kurian, V. M.; Doble, M. Characteristics of Bacterial Biofilm Associated with Implant Material in Clinical Practice. Polymer journal 2013, 45, 137–152. Sievert, D. M.; Ricks, P.; Edwards, J. R.; Schneider, A.; Patel, J.; Srinivasan, A.; Kallen, A.; Limbago, B.; Fridkin, S. Antimicrobial-Resistant Pathogens Associated with Healthcare- Associated Infections Summary of Data Reported to the National Healthcare Safety Network at the Centers for Disease Control and Prevention, 2009-2010. Infection Control & Hospital Epidemiology 2013, 34, 1–14. Skurnik, D.; Clermont, O.; Guillard, T.; Launay, A.; Danilchanka, O.; Pons, S.; Diancourt, L.; Lebreton, F.; Kadlec, K.; Roux, D.; et al. Emergence of Antimicrobial-Resistant Escherichia Coli of Animal Origin Spreading in Humans. Molecular biology and evolution 2015, msv280. (35) (36) Van Acker, H.; Van Dijck, P.; Coenye, T. Molecular Mechanisms of Antimicrobial Tolerance and Resistance in Bacterial and Fungal Biofilms. Trends in microbiology 2014, 22, 326–333. Talbot, T. R.; Bratzler, D. W.; Carrico, R. M.; Diekema, D. J.; Hayden, M. K.; Huang, S. S.; Yokoe, D. S.; Fishman, N. O. Public Reporting of Health Care-Associated Surveillance Data: Recommendations from the Healthcare Infection Control Practices Advisory Committee. Annals of internal medicine 2013, 159, 631–635. 28 (37) (38) Mei, M.; Chu, C.; Low, K.; Che, C.; Lo, E. Caries Arresting Effect of Silver Diamine Fluoride on Dentine Carious Lesion with S. Mutans and L. Acidophilus Dual-Species Cariogenic Biofilm. Medicina oral, patologia oral y cirugia bucal 2013, 18, e824–31. Renslow, R. S.; Babauta, J. T.; Majors, P. D.; Mehta, H. S.; Ewing, R. J.; Ewing, T.; Mueller, K. T.; Beyenal, H. A Biofilm Microreactor System for Simultaneous Electrochemical and Nuclear Magnetic Resonance Techniques. Water Science and Technology 2014, 69, 966– 973. (39) (40) Gaisford, S. Calorimetry for Monitoring Mixed Bacterial Populations and Biofilms. (41) (42) Biocalorimetry: Foundations and Contemporary Approaches 2016, 345. Tawse-Smith, A.; Atieh, M.; Tompkins, G.; Duncan, W.; Reid, M.; Stirling, C. The Effect of Piezoelectric Ultrasonic Instrumentation on Titanium Discs: A Microscopy and Trace Elemental Analysis in Vitro Study. International journal of dental hygiene 2016, 14, 191– 201. Priester, J. H.; Horst, A. M.; Van De Werfhorst, L. C.; Saleta, J. L.; Mertes, L. A.; Holden, P. A. Enhanced Visualization of Microbial Biofilms by Staining and Environmental Scanning Electron Microscopy. Journal of Microbiological Methods 2007, 68, 577–587. (43) Allen, A.; Semião, A. J.; Habimana, O.; Heffernan, R.; Safari, A.; Casey, E. Nanofiltration (44) (45) (46) (47) (48) and Reverse Osmosis Surface Topographical Heterogeneities: Do They Matter for Initial Bacterial Adhesion? Journal of Membrane Science 2015, 486, 10–20. Lei, J.; Huaiyang, Z.; Xiaotong, P.; Zhonghao, D. The Use of Microscopy Techniques to Analyze Microbial Biofilm of the Bio-Oxidized Chalcopyrite Surface. Minerals Engineering 2009, 22, 37–42. Low, B.; Lee, W.; Seneviratne, C.; Samaranayake, L. P.; Hägg, U. Ultrastructure and Morphology of Biofilms on Thermoplastic Orthodontic Appliances in "fast"and "slow"plaque Formers. The European Journal of Orthodontics 2010, cjq126. Little, B.; Wagner, P.; Ray, R.; Pope, R.; Scheetz, R. Biofilms: An ESEM Evaluation of Artifacts Introduced during SEM Preparation. Journal of Industrial Microbiology 1991, 8, 213–221. Jean, J.-S.; Tsao, C.-W.; Chung, M.-C. Comparative Endoscopic and SEM Analyses and Imaging for Biofilm Growth on Porous Quartz Sand. Biogeochemistry 2004, 70, 427–445. Chung, K. K.; Schumacher, J. F.; Sampson, E. M.; Burne, R. A.; Antonelli, P. J.; Brennan, A. B. Impact of Engineered Surface Microtopography on Biofilm Formation of Staphylococcus Aureus. Biointerphases 2007, 2, 89–94. (49) Yang, X.; Beyenal, H.; Harkin, G.; Lewandowski, Z. Quantifying Biofilm Structure Using (50) Image Analysis. Journal of Microbiological Methods 2000, 39, 109–119. Jackson, G.; Beyenal, H.; Rees, W. M.; Lewandowski, Z. Growing Reproducible Biofilms with Respect to Structure and Viable Cell Counts. Journal of Microbiological Methods 2001, 47, 1–10. (51) Yang, X.; Beyenal, H.; Harkin, G.; Lewandowski, Z. Evaluation of Biofilm Image Thresholding Methods. Water Research 2001, 35, 1149–1158. 29 (52) Daims, H.; Lücker, S.; Wagner, M. Daime, a Novel Image Analysis Program for Microbial Ecology and Biofilm Research. Environmental Microbiology 2006, 8, 200–213. (53) Milferstedt, K.; Pons, M.-N.; Morgenroth, E. Analyzing Characteristic Length Scales in (54) (55) Biofilm Structures. Biotechnology and Bioengineering 2009, 102, 368–379. Perni, S.; Prokopovich, P. Micropatterning with Conical Features Can Control Bacterial Adhesion on Silicone. Soft Matter 2013, 9, 1844–1851. Beyenal, H.; Donovan, C.; Lewandowski, Z.; Harkin, G. Three-Dimensional Biofilm Structure Quantification. Journal of Microbiological Methods 2004, 59, 395–413. (56) Mueller, L. N.; De Brouwer, J. F.; Almeida, J. S.; Stal, L. J.; Xavier, J. B. Analysis of a (57) Marine Phototrophic Biofilm by Confocal Laser Scanning Microscopy Using the New Image Quantification Software PHLIP. BMC Ecology 2006, 6, 1. Truong, V. K.; Lapovok, R.; Estrin, Y. S.; Rundell, S.; Wang, J. Y.; Fluke, C. J.; Crawford, R. J.; Ivanova, E. P. The Influence of Nano-Scale Surface Roughness on Bacterial Adhesion to Ultrafine-Grained Titanium. Biomaterials 2010, 31, 3674–3683. (58) Ngo, Q. D.; Vickery, K.; Deva, A. K. The Effect of Topical Negative Pressure on Wound (59) Biofilms Using an in Vitro Wound Model. Wound Repair and Regeneration 2012, 20, 83–90. Periasamy, S.; Joo, H.-S.; Duong, A. C.; Bach, T.-H. L.; Tan, V. Y.; Chatterjee, S. S.; Cheung, G. Y.; Otto, M. How Staphylococcus Aureus Biofilms Develop Their Characteristic Structure. Proceedings of the National Academy of Sciences 2012, 109, 1281–1286. (60) Heydorn, A.; Nielsen, A. T.; Hentzer, M.; Sternberg, C.; Givskov, M.; Ersbøll, B. K.; Molin, S. Quantification of Biofilm Structures by the Novel Computer Program COMSTAT. Microbiology 2000, 146, 2395–2407. (61) Heydorn, A.; Ersbøll, B. K.; Hentzer, M.; Parsek, M. R.; Givskov, M.; Molin, S. (62) Experimental Reproducibility in Flow-Chamber Biofilms. Microbiology 2000, 146, 2409– 2415. Sandal, I.; Hong, W.; Swords, W. E.; Inzana, T. J. Characterization and Comparison of Biofilm Development by Pathogenic and Commensal Isolates of Histophilus Somni. Journal of Bacteriology 2007, 189, 8179–8185. (63) Herzberg, M.; Elimelech, M. Biofouling of Reverse Osmosis Membranes: Role of Biofilm- Enhanced Osmotic Pressure. Journal of Membrane Science 2007, 295, 11–20. (64) Xavier, J.; Schnell, A.; Wuertz, S.; Palmer, R.; White, D.; Almeida, J. Objective Threshold (65) (66) Selection Procedure (OTS) for Segmentation of Scanning Laser Confocal Microscope Images. Journal of Microbiological Methods 2001, 47, 169–180. Lewandowski, Z. Notes on Biofilm Porosity. Water Research 2000, 34, 2620–2624. Bridier, A.; Dubois-Brissonnet, F.; Boubetra, A.; Thomas, V.; Briandet, R. The Biofilm Architecture of Sixty Opportunistic Pathogens Deciphered Using a High Throughput CLSM Method. Journal of Microbiological Methods 2010, 82, 64–70. (67) Kyan, M.; Guan, L.; Liss, S. Refining Competition in the Self-Organising Tree Map for Unsupervised Biofilm Image Segmentation. Neural Networks 2005, 18, 850–860. 30 (68) Yerly, J.; Hu, Y.; Jones, S. M.; Martinuzzi, R. J. A Two-Step Procedure for Automatic and Accurate Segmentation of Volumetric CLSM Biofilm Images. Journal of Microbiological Methods 2007, 70, 424–433. (69) Kim, T. G.; Yi, T.; Lee, E.-H.; Ryu, H. W.; Cho, K.-S. Characterization of a Methane- Oxidizing Biofilm Using Microarray, and Confocal Microscopy with Image and Geostatic Analyses. Applied Microbiology and Biotechnology 2012, 95, 1051–1059. (70) Auerbach, I. D.; Sorensen, C.; Hansma, H. G.; Holden, P. A. Physical Morphology and (71) Surface Properties of Unsaturated Pseudomonas Putida Biofilms. Journal of Bacteriology 2000, 182, 3809–3815. Liu, J.; Dazzo, F. B.; Glagoleva, O.; Yu, B.; Jain, A. K. CMEIAS: A Computer-Aided System for the Image Analysis of Bacterial Morphotypes in Microbial Communities. Microbial Ecology 2001, 41, 173–194. (72) Webb, D.; Hamilton, M.; Harkin, G.; Lawrence, S.; Camper, A.; Lewandowski, Z. Assessing Technician Effects When Extracting Quantities from Microscope Images. Journal of Microbiological Methods 2003, 53, 97–106. (74) (73) Xu, H.; Murdaugh, A. E.; Chen, W.; Aidala, K. E.; Ferguson, M. A.; Spain, E. M.; Núñez, M. E. Characterizing Pilus-Mediated Adhesion of Biofilm-Forming E. Coli to Chemically Diverse Surfaces Using Atomic Force Microscopy. Langmuir 2013, 29, 3000–3011. Surman, S.; Walker, J.; Goddard, D.; Morton, L.; Keevil, C.; Weaver, W.; Skinner, A.; Hanson, K.; Caldwell, D.; Kurtz, J. Comparison of Microscope Techniques for the Examination of Biofilms. Journal of Microbiological Methods 1996, 25, 57–70. Sangetha, S.; Zuraini, Z.; Suryani, S.; Sasidharan, S. In Situ TEM and SEM Studies on the Antimicrobial Activity and Prevention of Candida Albicans Biofilm by Cassia Spectabilis Extract. Micron 2009, 40, 439–443. Espinal, P.; Marti, S.; Vila, J. Effect of Biofilm Formation on the Survival of Acinetobacter Baumannii on Dry Surfaces. Journal of Hospital Infection 2012, 80, 56–60. (75) (76) (77) Villena, G.; Fujikawa, T.; Tsuyumu, S.; Gutiérrez-Correa, M. Structural Analysis of Biofilms and Pellets of Aspergillus Niger by Confocal Laser Scanning Microscopy and Cryo Scanning Electron Microscopy. Bioresource Technology 2010, 101, 1920–1926. (78) Ng, H. Y.; Tan, T. W.; Ong, S. L. Membrane Fouling of Submerged Membrane Bioreactors: Impact of Mean Cell Residence Time and the Contributing Factors. Environmental Science & Technology 2006, 40, 2706–2713. (79) Weber, K.; Delben, J.; Bromage, T. G.; Duarte, S. Comparison of SEM and VPSEM Imaging Techniques with Respect to Streptococcus Mutans Biofilm Topography. FEMS Microbiology Letters 2014, 350, 175–179. (80) Norton, T.; Thompson, R.; Pope, J.; Veltkamp, C.; Banks, B.; Howard, C.; Hawkins, S. Using Confocal Laser Scanning Microscopy, Scanning Electron Microscopy and Phase Contrast Light Microscopy to Examine Marine Biofilms. Aquatic Microbial Ecology 1998, 16, 199–204. Liu, T.; Yin, B.; He, T.; Guo, N.; Dong, L.; Yin, Y. Complementary Effects of Nanosilver and Superhydrophobic Coatings on the Prevention of Marine Bacterial Adhesion. ACS (81) 31 Applied Materials & Interfaces 2012, 4, 4683–4690. (82) DeQueiroz, G.; Day, D. Antimicrobial Activity and Effectiveness of a Combination of Sodium Hypochlorite and Hydrogen Peroxide in Killing and Removing Pseudomonas Aeruginosa Biofilms from Surfaces. Journal of Applied Microbiology 2007, 103, 794–802. (83) Kaali, P.; Momcilovic, D.; Markstrӧm, A.; Aune, R.; Czel, G.; Karlsson, S. Degradation of Biomedical Polydimethylsiloxanes during Exposure to in Vivo Biofilm Environment Monitored by FE-SEM, ATR-FTIR, and MALDI-TOF MS. Journal of Applied Polymer Science 2010, 115, 802–810. (84) Kania, R. E.; Lamers, G. E.; Vonk, M. J.; Dorpmans, E.; Struik, J.; Tran Huy, P.; Hiemstra, P.; Bloemberg, G. V.; Grote, J. J. Characterization of Mucosal Biofilms on Human Adenoid Tissues. The Laryngoscope 2008, 118, 128–134. Schaudinn, C.; Carr, G.; Gorur, A.; Jaramillo, D.; Costerton, J.; Webster, P. Imaging of Endodontic Biofilms by Combined Microscopy (FISH/cLSM-SEM). Journal of Microscopy 2009, 235, 124–127. (85) (86) Kreft, J.-U.; Picioreanu, C.; Wimpenny, J. W.; van Loosdrecht, M. C. Individual-Based Modelling of Biofilms. Microbiology 2001, 147, 2897–2912. (87) Hermanowicz, S. W. A Simple 2D Biofilm Model Yields a Variety of Morphological Features. Mathematical Biosciences 2001, 169, 1–14. (88) Daims, H.; Wagner, M. Quantification of Uncultured Microorganisms by Fluorescence (89) Microscopy and Digital Image Analysis. Applied Microbiology and Biotechnology 2007, 75, 237–248. Sinha, S. D.; Chatterjee, S.; Maiti, P.; Tarafdar, S.; Moulik, S. Evaluation of the Role of Substrate and Albumin on Pseudomonas Aeruginosa Biofilm Morphology through FESEM and FTIR Studies on Polymeric Biomaterials. Progress in Biomaterials 2017, 1–12. (90) Aggarwal, N.; Agrawal, R. First and Second Order Statistics Features for Classification of Magnetic Resonance Brain Images. Journal of Signal and Information Processing 2012, 3, 146. Prabha, D. S.; Kumar, J. S. Assessment of Banana Fruit Maturity by Image Processing Technique. Journal of food science and technology 2015, 52, 1316–1327. (91) (92) Haralick, R. M.; Shanmugam, K.; Dinstein, I. H. Textural Features for Image Classification. Systems, Man and Cybernetics, IEEE Transactions on 1973, 610–621. (93) Nielsen, B.; Albregtsen, F.; Danielsen, H. E. Statistical Nuclear Texture Analysis in Cancer Research: A Review of Methods and Applications. Critical ReviewsTM in Oncogenesis 2008, 14. (94) Albregtsen, F.; others. Statistical Texture Measures Computed from Gray Level Coocurrence Matrices. Image processing laboratory, department of informatics, university of oslo 2008, 5. (95) Otsu, N. A Thresholding Selection Method from Gray-Level Histogram. IEEE Transactions on Systems, Man and Cybernetics 1979, 9, 62–66. (96) Vicsek, T. Fractal Growth Phenomena; World Scientific, 1992; Vol. 2. 32 (97) (98) Theodoridis, S.; Pikrakis, A.; Koutroumbas, K.; Cavouras, D. Introduction to Pattern Recognition: A Matlab Approach; Academic Press, 2010. Corder, G.; Foreman, D. Nonparametric Statistics for Non-Statisticians: A Step-by-Step Approach; Wiley, 2009. (99) Knezevic, P.; Petrovic, O. A Colorimetric Microtiter Plate Method for Assessment of Phage Effect on Pseudomonas Aeruginosa Biofilm. Journal of Microbiological methods 2008, 74, 114–118. (100) Ruzicka, F.; Hola, V.; Votava, M.; Tejkalova, R. Importance of Biofilm inCandida Parapsilosis and Evaluation of Its Susceptibility to Antifungal Agents by Colorimetric Method. Folia microbiologica 2007, 52, 209. Table of Content for Graphics Figure 1: (Top panel) Grayscale biofilm imagesfor HDPE, (Bottom panel) Normalised histogram of the pixel intensities. Figure 2: (Top panel) Grayscale biofilm images for PTFE, (bottom panel) Normalised histogram of the pixel intensities. Figure 3: (Top panel) Binary image after thresholding the grayscale image for HDPE substrate, (Bottom panel) Box-counting fractal dimension estimation of the respective cases. Figure 4: (Top panel) Binary image after thresholding the grayscale image for PTFE substrate, (Bottom panel) Box-counting fractal dimension estimation of the corresponding binary images. Figure 5: Effect of contrast enhancement on the image pdf and cdf. Figure 6: Effect of contrast enhancement of the raw image in grayscale to binary conversion. Figure 7: PTFE control group feature correlation plot. Figure 8: Variation in class separability measure based on scatter matrix with sorted features in the three discrimination problems. The features are sorted using decreasing value of scatter measure J. Figure 9: Feature box plot for BSA conditioning layer on HDPE substrate for 9H vs. 24H 33 Figure 10: Scatter plots and histograms of top three features for hypothesis 1: effect of adsorption time on HDPE (a) F1 vs. F2, (b) F2 vs. F3, (c) F1 vs. F3, (d) F1-F3 Figure 11: Feature box plot for HDPE substrate with different conditioning layers BSA 24H vs. FN 24H. Figure 12: Scatter plots and histograms of top three features for hypothesis 2: effect of conditioning layer BSA vs. FN on HDPE substrate (a) F1 vs. F2, (b) F2 vs. F3, (c) F1 vs. F3, (d) F1-F3 Figure 13: Feature box plot for PTFE substrate with different conditioning layers BSA 24H vs. FN 24H. Figure 14: Scatter plots and histograms of top three features for hypothesis 3: effect of conditioning layer BSA vs. FN on PTFE substrate (a) F1 vs. F2, (b) F2 vs. F3, (c) F1 vs. F3, (d) F1-F3 34
1110.0983	1	1110	2011-10-05T13:49:35	Self-organizing magnetic beads for biomedical applications	[ "physics.bio-ph", "cs.CE", "physics.flu-dyn" ]	In the field of biomedicine magnetic beads are used for drug delivery and to treat hyperthermia. Here we propose to use self-organized bead structures to isolate circulating tumor cells using lab-on-chip technologies. Typically blood flows past microposts functionalized with antibodies for circulating tumor cells. Creating these microposts with interacting magnetic beads makes it possible to tune the geometry in size, position and shape. We developed a simulation tool that combines micromagnetics and discrete particle dynamics, in order to design micropost arrays made of interacting beads. The simulation takes into account the viscous drag of the blood flow, magnetostatic interactions between the magnetic beads and gradient forces from external aligned magnets. We developed a particle-particle particle-mesh method for effective computation of the magnetic force and torque acting on the particles.	physics.bio-ph	physics	Self-organizing magnetic beads for biomedical applications∗ Gusenbauer Markus†, Kovacs Alexander, Reichel Franz, Exl Lukas, Bance Simon, Ozelt Harald, Schrefl Thomas University of Applied Sciences St. Poelten August 14, 2018 Abstract In the field of biomedicine magnetic beads are used for drug deliv- ery and to treat hyperthermia. Here we propose to use self-organized bead structures to isolate circulating tumor cells using lab-on-chip technologies. Typically blood flows past microposts functionalized with antibodies for circulating tumor cells. Creating these microposts with interacting magnetic beads makes it possible to tune the geom- etry in size, position and shape. We developed a simulation tool that combines micromagnetics and discrete particle dynamics, in order to design micropost arrays made of interacting beads. The simulation takes into account the viscous drag of the blood flow, magnetostatic interactions between the magnetic beads and gradient forces from ex- ternal aligned magnets. We developed a particle-particle particle-mesh method for effective computation of the magnetic force and torque act- ing on the particles. ∗Accepted for publication in the Journal of Magnetism and Magnetic Materials. †Corresponding author, [email protected] 1 1 Introduction 1.1 Circulating Tumor Cell (CTC) CTCs detach from a tumor and can remain in the blood even after the tumor is removed. Their presence increases the chance of net tumors developing. It is important to monitor the number of CTCs in the blood but their low concentration (a few per µl) when compared to normal blood cells (around 5 million per µl) makes this difficult. A new and flexible method is required. Lab-on-chips with fixed arrays are designed to use the properties of the CTCs to filter them. The most common ways are mechanical filters or using antibodies. Membranes with slots smaller than tumor cells fulfill the require- ments for a mechanical filter [1]. Normal blood cells, they have similar sizes, are more deformable than CTCs and can go through. Another possibility is to use cylindrical posts coated with antibodies [2]. If there is a large surface area the possibility for a connection with the tumor cells increases. Recently a CTC-chip based on guided self-assembly of magnetic beads was proposed [3]. Saliba et al. use magnetic traps made by microcontact printing in order to create magnetic chains on a regular grid. They demon- strate cell capture using beads coated with antibodies. In this paper we show that tunable microfluidic chips can be designed by carefully selecting the susceptibility and the applied magnetic fields. Using discrete particle simulations we compute the distance between particle chains in a microflu- idic chip as function of a homogenous magnetic bias field. Such devices might be useful to combine immunomagnetic with mechanical filtering of CTCs [4]. Maimonis et al. show that the capture efficiency can be increased in a mi- crofluidic chip with 2 post arrays of different gap size which enables affinity and size capturing. Our proposed chip technology may offer the possibility to switch between affinity and size capturing through changing the distance between antibody-coated chains by an external field. 1.2 Content of the paper To optimize the design of CTC-chips there are several questions for us. Can we use self-organized magnetic beads to create filter-like structures? What 2 types of particles are applicable? Properties to deal with are susceptibility and diameter of the beads. To answer this questions we developed a simu- lation model with interacting magnetic beads to create particle chains. This is described in section 2 of the paper. In addition to the magnetic part we developed a blood cell model that flows through the channel with the chain barrier. The described methods are used to demonstrate tunability of the chain gaps with an external magnetic field. Results are presented in section 3. 2 Methods 2.1 Magnetic particle dynamics In order to create a lab-on-chip device consisting of self-organizing micro- magnetic beads we want to profit from an external magnetic field and the force of the blood flow. Fig. 1 shows the overview of the proposed microflu- idic chip. Viscous blood flows into a pipe filled with soft-magnetic beads. Magnetic charge sheets with opposite magnetization orientation provide an external magnetic field. Under the influence of the magnetic field a magnetic moment m is created in every particle. With the moments of two nearby beads and the distance r we got a formulation (Eqn. 1) of the interaction force Fi for bead 2 and vice versa for bead 1 [5]. ~F1→2 = 3µ0 4πr5 [( ~m1~r) ~m2 + ( ~m2~r) ~m1 + ( ~m1 ~m2)~r − 5( ~m1~r)( ~m2~r) r2 ~r (1) The gradient force ~Fg (Eqn. 2) on a bead is given by the negative gradient of the energy of the magnetic dipole moment ~m in the field ~B. In three dimensions it is given by ~Fg = ∇( ~m ~B) ~Fg =  Fx Fy  Fz   =  mx∂xBx + my∂xBy + mz∂xBz mx∂xBy + my∂yBy + mz∂yBz  mx∂xBz + my∂zBy + mz∂zBz   3 (2) (3) Figure 1: Formation of particle chains through the interaction of the drag force Fd and magnetic forces. In the picture the magnetic forces are cre- ated by the magnetic field gradient Fg and magnetic interactions Fi. The nonuniform magnetic field is created by 2 magnetic charge sheets. We assume that in the x-y plane the external field is homogeneous due to the distance and size of the charge sheets. For the calculation of the gradient force only the z-field derived in y is important. 0 ~Fg =  mz∂yBz  0   (4) The blood flow creates a drag force [6] on the particles. At a low velocity of 460 µm/s as used at most in CTC-chips [2], the flow is laminar. Turbulent flow would appear at Reynolds number (Eqn. 5) more than unity. Re = 2rρv η 4 (5) During laminar flow the Stokes Law (Eqn. 6) is used to calculate the force Fd on an object in a fluid, with particle radius r, density ρ and viscosity η of the fluid and the relative velocity v of the particle. Krishnamurthy et al. [7] show that Stokes drag is a good approximation for the force acting on the beads. Fd = 6πηrv (6) For the above mentioned calculations of the magnetic particle dynamics we decided to expand the open-source particle simulator Yade [8]. It pro- vides for example gravity force and collision detection using various engines. Additionally we developed magnetic and fluidic engines. In the simulations we integrate the Newtons equation of motion for magnetic particles under the influence of the drag force of the fluid, the gradient force of the external magnetic field and the gradient force of the field created by the particles themselves. For every particle in the pipe we calculate the magnetic field from the permanent magnets analytically [9]. We represent the 2 magnets by charge sheets as commonly done by magnetic simulation [10]. The simulation pa- rameters are listed in table 1. For blood cell dynamics we implemented a cellular engine in Yade. This will be explained in the next section. 2.2 Blood cell dynamics The red blood cell consists of a cytoskeleton covered with a double layer membrane [11]. It has a biconcave form in its native state with a diameter of around 8 µm, a surface of 135 µm2 and a volume of 94 µm3. The struc- ture is built from 33000 hexagons which in reality would contain different proteins such as Spektrin and Aktin. In its loose state there is almost no interaction force between these proteins (Fig. 2a). If there is an external force acting on the cell the cytoskeleton expands and gets stretched (Fig. 2b). The membrane is an almost incompressible layer regarding the plane and shear stresses. Capillaries have a smaller diameter than the red blood 5 Blood velocity viscosity density Particles radius susceptibility magnetic saturation Geometry magnetic charge sheets pipe distance pipe-sheet pole density sheets 460 × 10−6 m/s 0.015 Ns/m2 1.055 g/cm3 10−6 m 0.4 1 T 10 x 10 mm 10 x 0.05 x 0.02 mm 2 mm 1.6 T Table 1: Simulation parameters cells, but because of their elastic properties they make their way using the form of a projectile. Figure 2: Cytoskeleton: a) relaxed, b) stretched Discrete element methods can be used to treat red blood cells numeri- cally. In the following we will show that the behavior of red blood cells can be modeled using a nonlinear mass-spring-system (Fig. 3). The model consists of nearly 600 spherical particles connected by springs arranged orthogonally. The force between two particles i and j (Eqn. 7) depends on the respective spring constants, starting distance l0 and the actual distance d. Yade provides 6 Figure 3: Nonlinear mass-spring model of red blood cell: ks1=500, ks2=400, ks3=300, ks4=200 (Eqn. 7) the damping for all particles in the model. F (xi, xj) = ks ∗ d d ∗ (d − l0) ∗ (d − l0) (7) One possible way of characterizing a red blood cell is using an optical trap for extracting its shear modulus [12]. With an applied force on opposite sides of the cell, due to optical tweezers, the diameter is measured. From the approximated formula (Eqn. 8) the modulus decreases form the starting diameter D0 with a function of the force F and the elasticity µ. D = D0 − F 2πµ (8) To validate this model we computed this cell diameter as function of an applied stretch force (Fig. 4). Different spring constants are used to recreate a similar behavior in the stretching test. They decrease from the middle to the edge of the red blood cell (Fig. 3). The modeled cell is accurate enough to make tests with the chain barrier. This will be demonstrated in the result section. 7 7e-06 ] m [ r e t e m a i d l l e C 6,5e-06 Original blood cell Model 6e-06 0 1e-11 2e-11 3e-11 Stretch force [N] Figure 4: Stretch test of red blood cell, original [12] vs. model 3 Results 3.1 Single particle equilibrium position With a steady blood flow through the microfluidic chip the magnetic parti- cles are flushed out without the presence of an external magnetic field. To find the equilibrium position y0 of a single bead after applying the field we calculate the force balance of the drag force Fd and the gradient force Fg (Eqn. 9). Fig. 6 shows this forces and their sum as function of the position. At the equilibrium state Fd and Fg cancel each other. Fg(y) = Fd =⇒ y0 (9) The magnetic moment ~m(y) of a single particle depends on its volume 8 V , susceptibility χ, which defines the ascending slope of the magnetization curve (Fig. 5), and the magnetic field. This field is the sum of the gradient field Bg(y) and a homogenous bias field Bbias, which we will need to tune the device. ~m(y) = V χ 1 µ0 (Bg(y) + Bbias) (10) We have now 2 possible working points for the softmagnetic particles. In its saturated state (Fig. 5.1) the magnetic moment can't be further in- creased. For better tunability we are using the working point in the initial magnetization curve (Fig. 5.2). This will be shown in the next sections. Figure 5: Operation points in the magnetization curve: 1. magnetic satura- tion, 2. initial magnetization (tunability) The maximum field from the charge sheets is in the center and decreases towards the edge (Fig. 6). Fig. 7 shows the negative drag force and total magnetic force depending on the position of the magnetic particles. The re- sults show that the final position depends on the strength of the uniform bias field and the susceptibility χ of the particles. If there is no crossing point the drag force is to high and the particles get washed out of the pipe (Fig. 7a with a bias field of less than 0.5T ). If one of the simulation parameters changes, e.g. the viscosity of the blood, the drag force changes and therefore the external magnetic field has to 9 change. Otherwise there probably wouldn't be a crossing point, and therefore no equilibrium position. Figure 6: Magnetic charge sheets create a gradient field Bg. A homogenous bias field Bg is added to the existing one for better tunability. Force balance of fluidic drag force Fd and magnetic gradient force Fg leading the particles to the equilibrium position y0. 3.2 Chain formation The simulation starts with softmagnetic beads of radius 1µm and suscepti- bility 0.4. They are randomly located in the fluidic channel (Fig. 8a). Imme- diately after applying the magnetic charge sheets the particles self-organize to chains according to the field lines (Fig. 8b). This chain formation is then shifted to the equilibrium position y0 because of the force balance explained in the section before. Because of chosen fluid velocity of 460 µm/s this lasts 10 0 -1e-10 -2e-10 -3e-10 ] N [ F -4e-10 -5e-10 -6e-10 -7e-10 Fg (+0.00 T) Fg (+0.25 T) Fg (+0.50 T) Fg (+0.75 T) Fg (+1.00 T) Fg (+1.25 T) Fg (+1.50 T) - Fd -8e-10 0,005 0,0075 0,01 0,005 0,01 Position of magnetic bead [m] 0,0075 0,005 0,0075 0,01 a) b) c) Figure 7: Gradient force Fg as function of position with susceptibility a) 0.2 b) 0.4 c) 0.8. Crossing points with the drag force Fd define the equilibrium position y0 of a magnetic particle. around 16s (Fig. 8c). After this procedure the filtering of the blood could start already. Dif- ferent types of CTCs may need different distances between the chains for an optimal filtering. In the second phase of the simulation an additional ho- mogenous bias field is applied on the chip. Depending on the field the chains change their gaps in around 0.13s (Fig. 8e). How it works will be explained in the next section. 11 Figure 8: Timeline of the simulation: Phase 1 - Particle chain creation, Phase 2 - Changing the gap size 3.3 Tunable gap size Eqn. 11 shows the magnetic moment ~m(y) of a single bead with the exis- tance of many particles. When we increase the bias field Bbias the magnetic moment and therefore the magnetic field created in every particle gets more. But only if we are in the working point of the initial magnetization curve (Fig. 5.2). The interaction field Bi is the sum of all other particle fields. This leads to a higher particle interaction force which causes a larger gap. Fig. 9 shows the increasing average gap size between the chains. In order to quantify this effect we computed the average distance between magnetic particle chains as a function of an additional homogenous magnetic field and 12 Figure 9: Average gap size between magnetic particle chains with different susceptibilities and additional external field susceptibility. ~m(y) = V χ 1 µ0 (Bg(y) + Bbias + Bi) (11) With a susceptibility of 0.2 resulting gaps occur only with an external field of more than 0.5T . In this case the force caused by the fluid is greater than the maximum force of the gradient. So the particles get washed out of the pipe. At an intermediate susceptibility of 0.4 the gap size increases linearly with the field from 18µm to 24µm. At high susceptibility the total field will saturate the particles which leads to similar results after 0.5T . 3.4 Particle density For the simulation result it is essential how dense the particles are filled into the pipe. For the particle density ρbead we calculate the number of beads N times the volume of one bead Vbead over the volume of the wrapping box Vbox (Eqn. 12). 13 ρbead = N ∗ Vbead Vbox (12) To evaluate the simulation we compared different starting densities and examined the resulting structures. Up to a density of 0.07 we find good results with not more than 2 bead thick chains. After that the particles get clumpy and can't be used for our application. Fig. 10 demonstrates the different results of bead densities. Figure 10: Endposition of chains with particle density 0.03, 0.07 and 0.11 3.5 Cell chain interaction Very important for the functionality of the filter is the stability of the chains. When a blood cell touches such a chain a certain force is applied on it. And it would destroy the structure if it gets to high. To get a value of the force we started a simulation of our blood cell that flows directly on a particle chain that is fixed in space. Fig. 11 shows the average force per ms over time acting from the blood cell to the chain, especially on one magnetic bead of the chain. The simulation overview is shown in Fig. 12. The magnitude of force on a single magnetic bead in the chain is 1000 times smaller than the drag force of the fluid or the gradient force of the magnetic charge sheets. So it doesn't influence the position and the stability of the chain structure. 14 Figure 11: Force from the blood cell acting on a single magnetic bead of the chain structure. 4 Summary Flexible ways are important to get a high probability of catching cancer cells. Micromagnetic beads are established to create chain barriers on de- mand. Gap sizes are modified as desired for different CTCs. Size-based filtration and the usage of antibodies covered on the beads can take place simultaneously. In this work we demostrated simulation tools for the design of micromagnetic CTC-chips. First step of modelling blood cells was the creation of a nonlinear mass- spring-system. For the success of the project it is essential to create models of all kind of blood cells, healthy ones and circulating tumor cells. More com- plex fluidic behavior is also important for future work. Turbulences could occur due to the selfarrangement of the chains in the magnetic field or the interaction of magnetic beads and blood cells. Results summarized: 15 Figure 12: Red blood cell collides with a single magnetic chain for determing the contact force. • Starting density of micromagnetic beads shouldn't be more then 0.07 for the ratio between volume of beads and wrapping box. • Softmagnetic beads self-organizes to particle chains due to the external aligned magnetic charge sheets within a few µs. • After about 16s the chain arrangement finds its equilibrium position with the force balance of fluidic drag force and magnetic forces. • Applying an additional homogenous field changes the gap size between the chains according to different circulating tumor cells in around 0.13s. • A nonlinear mass spring cell model in the flow creates a force on a magnetic bead in a chain with a magnitude of 1000 smaller than the drag force or the gradient force. So it has no influence on the chain structure. Acknowledgment The authors gratefully acknowledge the financial support of Life Science Krems GmbH, the Research Association of Lower Austria. 16 References [1] B. Lu, T. Xu, S. Zheng, A. Goldkorn, Y. Tai, Parylene membrane slot filter for the capture, analysis and culture of viable circulating tumor cellsdoi:10.1109/MEMSYS.2010.5442361. [2] D. W. Bell, D. Irimia, L. Ulkus, M. R. Smith, E. L. Kwak, S. Digumarthy, A. Muzikansky, P. Ryan, U. J. Balis, R. G. Tompkins, D. A. Haber, M. Toner, S. Nagrath, L. V. Sequist, S. Maheswaran, Isolation of rare circulating tumour cells in cancer patients by microchip technology, Nature 450 (7173) (2007) 1235 -- 1239. doi:10.1038/nature06385. [3] A. Saliba, L. Saias, E. Psychari, N. Minc, D. Simon, F. Bidard, C. Mathiot, J. Pierga, V. Fraisier, J. Salamero, V. Saada, F. Farace, P. Vielh, L. Malaquin, J. Viovy, Microfluidic sorting and multimodal typing of cancer cells in self- assembled magnetic arrays, Proceedings of the National Academy of Sciences 107 (33) (2010) 14524 -- 14529. doi:10.1073/pnas.1001515107. [4] P. J. Maimonis, K. Merdek, K. Dietenhofer, L. Yen, Y. Dong, G. Palmer, Affinity and size capture of circulating tumor cells: A platform for increased sensitivity, Proceedings of the American Association for Cancer Research 2010 (1 Molecular Diagnostics Meeting) (2010) B5. [5] E. P. Furlani, Permanent magnet and electromechanical devices: materials, analysis, and applications, Academic Press, 2001. [6] C. Mikkelsen, M. F. Hansen, H. Bruus, Theoretical comparison of magnetic and hydrodynamic interactions between magnetically tagged particles in mi- crofluidic systems, Journal of Magnetism and Magnetic Materials 293 (1) (2005) 578 -- 583. doi:10.1016/j.jmmm.2005.01.076. [7] S. Krishnamurthy, A. Yadav, P. E. Phelan, R. Calhoun, A. K. Vuppu, A. A. Garcia, M. A. Hayes, Dynamics of rotating paramagnetic parti- cle chains simulated by particle dynamics, stokesian dynamics and lat- tice boltzmann methods, Microfluidics and Nanofluidics 5 (1) (2007) 33 -- 41. doi:10.1007/s10404-007-0214-z. [8] J. Kozicki, F. Donz´e, A new open-source software developed for nu- merical simulations using discrete modeling methods, Computer Meth- ods in Applied Mechanics and Engineering 197 (49-50) (2008) 4429 -- 4443. doi:10.1016/j.cma.2008.05.023. 17 [9] G. Akoun, J. Yonnet, 3D analytical calculation of the forces exerted between two cuboidal magnets, IEEE Transactions on Magnetics 20 (5) (1984) 1962 -- 1964. doi:10.1109/TMAG.1984.1063554. [10] K. Senanan, R. Victora, Theoretical study of nonlinear transition shift in double-layer perpendicular media, Magnetics, IEEE Transactions on 38 (4) (2002) 1664 -- 1669. doi:10.1109/TMAG.2002.1017753. [11] A. Ikai, Einfuhrung in die Nanobiomechanik: Bildgebung und Messung durch Rasterkraftmikroskopie, 1st Edition, Wiley-VCH Verlag GmbH & Co. KGaA, 2010. [12] S. H´enon, G. Lenormand, A. Richert, F. Gallet, A new determination of the shear modulus of the human erythrocyte membrane using optical tweezers (Feb. 1999). doi:16/S0006-3495(99)77279-6. URL http://www.sciencedirect.com/science/article/pii/S0006349599772796 18
1110.3677	1	1110	2011-10-14T14:11:10	Cortical phase transitions, non-equilibrium thermodynamics and the time-dependent Ginzburg-Landau equation	[ "physics.bio-ph", "q-bio.NC", "quant-ph" ]	The formation of amplitude modulated and phase modulated assemblies of neurons is observed in the brain functional activity. The study of the formation of such structures requires that the analysis has to be organized in hierarchical levels, microscopic, mesoscopic, macroscopic, each with its characteristic space-time scales and the various forms of energy, electric, chemical, thermal produced and used by the brain. In this paper, we discuss the microscopic dynamics underlying the mesoscopic and the macroscopic levels and focus our attention on the thermodynamics of the non-equilibrium phase transitions. We obtain the time-dependent Ginzburg-Landau equation for the non-stationary regime and consider the formation of topologically non-trivial structures such as the vortex solution. The power laws observed in functional activities of the brain is also discussed and related to coherent states characterizing the many-body dissipative model of brain.	physics.bio-ph	physics	Cortical phase transitions, non-equilibrium thermodynamics and the time-dependent Ginzburg-Landau equation Walter J. Freeman∗ Department of Molecular and Cell Biology University of California, Berkeley CA 94720-3206 USA Dipartimento di Fisica and Istituto Nazionale di Fisica Nucleare Universit´a di Firenze, I-50022 Sesto Fiorentino (Firenze), Italy Roberto Livi† Masashi Obinata‡ Facolt´a di Scienze and Istituto Nazionale di Fisica Nucleare Universit´a di Salerno, I-84100 Fisciano (Salerno), Italy and Department of Physics, Tsukuba University, Tsukuba, Japan Giuseppe Vitiello§ Facolt´a di Scienze and Istituto Nazionale di Fisica Nucleare Universit´a di Salerno, I-84100 Fisciano (Salerno), Italy The formation of amplitude modulated and phase modulated assemblies of neurons is observed in the brain functional activity. The study of the formation of such structures requires that the analysis has to be organized in hierarchical levels, microscopic, mesoscopic, macroscopic, each with its characteristic space-time scales and the various forms of energy, electric, chemical, thermal produced and used by the brain. In this paper, we discuss the microscopic dynamics underlying the mesoscopic and the macroscopic levels and focus our attention on the thermodynamics of the non-equilibrium phase transitions. We obtain the time-dependent Ginzburg-Landau equation for the non-stationary regime and consider the formation of topologically non-trivial structures such as the vortex solution. The power laws observed in functional activities of the brain is also discussed and related to coherent states characterizing the many-body dissipative model of brain. PACS numbers: 11.10.-z, 87.85.dm, 11.30.Qc I. INTRODUCTION A major effort is under way worldwide to develop superior forms of machine intelligence by applying basic concepts and equations from physics to the interpretation of neurobiological and neuropsychological data. The property that most clearly distinguishes biological intelligence from contemporary machine intelligence is the rich contextualization of information by brains in the construction of knowledge and meaning. Computers and robots operate on information, but they don't know what it means. Books and reprints contain information and display it seriatim in accord with Shannon's theory of information, but knowledge of what the information means is solely in the brains of the writers and readers. We propose that what differentiates knowledge from information are the innumerable linkages among myriad fragments of information, which create the moment-to-moment framework for effective intentional action into the world [1]. We find that meaning can best be described objectively as created and carried by great fields of neural activity, which subjectively we experience as thoughts and perceptions. The contents of such fields are constructed from the fragments of information that are imported by sensory neurons and stored by changes in the synaptic linkages among the cortical neurons. It is the dynamics of the populations in each sensory cortex that subsequently organizes the microscopic fragments into meaningful knowledge by creating macroscopic vector fields of activity that organize hundreds of millions of neurons and trillions of synapses. Our aim in this report is to summarize our observations of cortical fields in humans and other animals engaged in intelligent behaviors, and then to describe the dynamics of cortex using concepts adapted from non-equilibrium thermodynamics [2] and quantum field theory [3, 4]. In particular we focus on adapting the time-dependent Ginzburg-Landau equation so as to describe the construction and ∗ [email protected] - http://sulcus.berkeley.edu † [email protected] ‡ [email protected] § [email protected] - www.sa.infn.it/giuseppe.vitiello/ 2 FIG. 1: Left: The burst of gamma oscillation illustrates the amplitude modulation of the shared carrier wave. Right: AM patterns are compared with and without the conditioned stimuli (CS) present in the inhaled air. The change between trial sets illustrates consolidation: 'off-line' learning requiring participation of the genome. transmission of the wave functions in vector fields that we observe in brain activities and experience as knowledge. A thermodynamic model is needed, because intelligence is extremely energy-intensive. Brains consume free energy in many forms (electric, magnetic, chemical, metabolic) at rates ten-fold greater than any other organ. Salient among these forms is electric current flowing in closed loops that is carried by ions (not electrons) in water. The flux of ions within neurons is essential for their functions: to communicate by transmitting pulses with axons, and to sum synaptic potentials with dendrites [5]. The extracellular flows of loop currents are revealed by ohmic potentials that we record from cortical surfaces (the electrocorticogram, ECoG) and from the scalp (the electroencephalogram, EEG). The cortex is a thin sheet of neurons covering the outer surface of the brain. The average thickness in humans is ∼ 3 mm, and the surface area is ∼ 2, 000 cm2, a ratio of 1 : 105. Whereas the dynamics of microscopic neurons in networks and the scalar potential fields of the EEG/ECoG are described in 3-D, the macroscopic fields of activity from 108 to 109 neurons each supporting 104 synapses [6] topologically operate in 2-D, owing to the long correlation distances compared with the dimensions of neurons. Our standpoint is that cortical dynamics cannot be reduced to the level of neural networks. A macroscopic level of mass action emerges from microscopic dynamics, into which the senses inject their microscopic information and from which macroscopic neural commands are generated. The ECoG and EEG are appropriate to the macroscopic level, because they are summed contributions from interactive masses of neurons. The interactions create dynamic neural assemblies with properties that are undetected with pulses from microelectrodes and inexplicable in terms of neural nets [7]. In the simplest description we conceive cortex as a self-regulating, self-stabilized population of neurons. It modulates and is modulated by other parts of the brain, but it does so on its own terms. Whereas the cortical neural network dynamics involves pulse and wave frequencies, macroscopic population dynamics uses pulse and wave densities, which constitute the state variables in space-time. In our analysis it is important that we distinguish the level of function by properly defining the state variables. To this aim, since in this paper we are mainly interested in the microscopic dynamics out of which macroscopic structures emerge, we consider the many-body dynamics of the vibrational quanta of the electric dipoles of water molecules. Water is the matrix in which neurons, glial cells and the whole net of dendrites and axons are embedded, including in particular the ionic currents that exercise the electric forces on and in nerve membranes that constitute neural activity. The dipole dynamics provides the possibility for long-range correlation, which in turn allows global synaptic communication among the neurons, each with every other as required for the construction of meaning. It should be clearly stated that the water molecules and their dipole oscillations are not the agency of communication among the neurons. As said above, the memory required for macroscopic structure formation is based in the trillions of synapses between neurons. Likewise, in our prior publications we have made it clear that the extracellular electric fields of the EEG and ECoG are epiphenomenal. They are scalar fields of passive dissipation of electric energy as heat, whereas the fields of neural activity are vector fields, which actively derive free energy from chemical energy in every square micron and dissipate that energy in clouds of action potentials and ultimately as heat. Our many-body analysis shows how such a vector field is indeed generated from the basic dynamics of the dipole vibrational quanta [7, 8]. The main sources of our experimental data are high-resolution images of the EEG and ECoG from 8 × 8 planar 3 FIG. 2: A. An example is shown from rabbit ECoG of a cone fitted to the surface given by an 8 × 8 array of values of analytic phase with phase lead at the apex ("explosion"). B. Analytic phase from human ECoG with phase lag at the apex ("implosion"). C. Summary diagram of the variants of phase patterns seen in cinematic displays of the filtered ECoG. Examples can be seen at the URL: http://soma.berkeley.edu arrays of 64 electrodes [9 -- 13]. The 64 signals reveal brief epochs (∼ 0.1 s) of 3 − 5 cycles of narrow band oscillation. The epochs recur 3 − 10 times each second, each time with a different carrier frequency randomly centered in the beta-gamma range. The amplitude of the carrier wave is modulated in a spatial amplitude modulated (AM) pattern (Fig. 1). The set of 64 amplitudes constitutes a 64 × 1 vector that specifies a point in 64-space. Similar patterns accruing from repeated presentation of a conditioned stimulus (CS) form a cluster of points. Differing CSs form multiple clusters, one for each CS that a subject can discriminate. The AM patterns lack invariance with respect to the CS. The same CS in a different context gives a distinctly different cluster. This contextual dependence shows that the AM patterns are dependent on the meaning of the CSs; they are not representations of the CSs but of the knowledge the subject has about the CSs. The AM patterns evolve slowly with repetition over time and the accumulation of new experiences. Each retrieval of a memory changes the memory by adding new context. Each AM pattern is accompanied by a spatial pattern of phase modulation (PM) of the carrier frequency, which has the form of a cone (Fig. 2) (see also Appendix A). The location and sign of the apex vary randomly from each epoch to the next. The gradient of the cone in rad/m is fixed in each epoch and varies between bursts inversely with the carrier frequency in rad/s. The ratio gives the phase velocity in m/s, which is determined by the conduction velocity of pulses on axons running parallel to the cortical surface. We infer that each new AM pattern forms by a state transition, which begins at a site of nucleation on the surface and spreads radially as in the formation of a raindrop. The diameter of the AM pattern is determined by the distance at which phase dispersion by the phase gradient reduces interaction strength below the half-power level. During an epoch the dynamics is stationary in carrier frequency, AM and PM, and the dynamics is near linear. We introduce the Hilbert transform, which gives the high temporal resolution needed to display the transitions between successive images [9, 10]. We also introduce the concept of criticality, which we need to help explain the capacity of cortex to undergo the dramatic transitions that are required to construct and destroy the images enabling perception [2]. Cinematic display of the spatial AM/PM patterns reveals repetitive pulsatile expansions or contractions of the wave functions, and in some instances rotation either clockwise or counterclockwise indicating the stabilization of the AM/PM patterns by formation of vortices, hence vector fields [4]. The point is that the many-body approach allows the formation of spatially delimited domains that constitute the basic facilitating environment wherein the formation occurs of the above described AM and PM structures in the cortex. These domains are also temporally limited in time, in the sense that they have a finite lifetime; they condense and evaporate continually as in a fog, with power-law distributions of duration and diameter. We treat them as the result of a space-time non-homogeneous boson condensation process [14 -- 17] and as neural avalanches by which criticality is maintained [18]. Intermittently we see the long-lasting epochs generated by the long-range correlation of neurons locked in coherent oscillation mode carrying AM patterns relating to perception. The dipole quantum field correlation thus acts to facilitate expansion of the trillions of microscopic synapses modified by learning into the spatially textured macroscopic cognition. The expansion is directed by a mesoscopic Hebbian assembly (Appendix A) into the basin of a macroscopic 4 chaotic attractor that characterizes the AM pattern. The Hebbian assembly provides a burst of transition energy that brains require in order to restrict the expansion to intended, learned sensory input and to exclude environmental noise. From such a perspective, in the present paper we therefore study the thermodynamics of the phase transition processes at the level of the many-body dynamics of the dipole wave quanta with defined topological properties. The AM/PM patterns then can be considered to be the macroscopic manifestations of a boson condensation and may be characterized by its underlying topological properties. The basic symmetry is the (non-Abelian) rotational SU(2) dipole symmetry, and the emerging structure formation is entropic, always incurring loss of information particularly in abstraction and inductive generalization. The topology of symmetry breaking from the search/receiving mode to the constructive/transmitting mode is related with the singularity in the null spike where prior structure is expunged and disorder is initiated, as at center (the core) of a vortex [4]. In Section II we present basic elements of the many-body dissipative model with specific attention to the mechanism of boson condensation in a stationary regime. Further details of the formalism of the dissipative many-body model can be found in Refs. [7, 14, 15]. The non-stationary regime is discussed in Section III where we obtain the time- dependent Ginzburg-Landau equation. The vortex solution is considered in Section IV and the power laws observed in functional activities of the brain are discussed in Section V and related to coherent states characterizing the many- body dissipative model. Finally, Section VI is devoted to conclusions. Further remarks on the relevant aspects of the observed brain activity and some mathematical details are presented in the Appendices. II. THE DISSIPATIVE MANY-BODY MODEL OF BRAIN In the dissipative many-body model of brain [14, 15], one considers the spontaneous breakdown of the rotational symmetry of the dipoles of water molecules. In quantum field theory (QFT) spontaneous breakdown of symmetry produces the formation of ordered patterns through the mechanism of boson condensation. Such ordered patterns, resulting into "in phase" dipole oscillations and often also space ordering are the manifestation of coherent dynamic features, out of which collective macroscopic behaviors of coherent spatial domains appear [19, 20]. The lagrangian describing the dynamics of the system is assumed to be invariant under the SU(2) group of rotations for the molecular φ↓(x, t))T (j =↑,↓, denotes the dipole quantum number and assumes, for dipole vibrational field φ(x, t) = (φ↑(x, t) simplicity, only two values "up" and "down") [21, 22]. On the contrary, the ground state (the "vacuum") of the system cannot be SU(2) invariant since it has to represent the electret state of water. The SU(2) spontaneous breakdown is expressed by the relation h0D(3)(r, t)0i = P(r, t) 6= O, (1) where D(3)(r, t) denotes the dipole density operator in the third direction and P(r, t) is the polarization density. The dipole density operators D(i)(r, t), i = 1, 2, 3, are functionals of the molecular vibrational field φ(x, t). For example, they can be expressed as D(i)(r, t) = φ†(x, t) 1 2 σiφ(x, t), i = 1, 2, 3, with σ the Pauli matrices. The conclusions we will reach, however, are not dependent on the particular expression of D(i)(r, t) in terms of φ(x, t). A general theorem in quantum field theory (QFT) [19, 20, 23 -- 25] shows that spontaneous breakdown of symmetry implies the existence of gapless fields (the Nambu-Goldstone modes or particles (NG)). They appear as collective excitations describing the long-range correlation among the dipoles generating the ordering of the system (electret state). In the dissipative many-body model of brain, a number of functional features of brain can be understood as manifestations of the NG collective modes [14, 15]. The spontaneous breakdown of SU(2) symmetry represented by Eq. (1) means that the vacuum is not invariant under rotations induced by D(1)(r, t) and D(2)(r, t). It is, however, invariant under rotations induced by D(3)(r, t), i.e. under the U(1) rotation group around the third direction: φ → exp(iqλ3(r, t)σ3/2)φ . (2) Here q is the electric charge at the head of the molecular dipole. We assume λ3(r, t) → 0 for t → ∞ and/or r → ∞. Of course, invariance of the lagrangian under local (i.e. with space-time-dependent λ3(r, t)) U(1) rotation group signals the presence (requires the introduction) of the electromagnetic interaction among the dipoles, i.e. the electro- magnetic vector potential Aµ(r, t), which transforms as when the phase transformation around the third axis is performed. In the following, we will use the Coulomb gauge ∇ · A = 0 as gauge condition. In terms of the dipole operators, the local U(1) transformation is Aµ(r, t) → Aµ(r, t) − ∂µλ3(r, t), (3) D(±)(r, t) → exp(∓iqλ3(r, t))D(±), (4) where D(±)(r, t) = D(1)(r, t) ± iD(2)(r, t). 5 (5) Next we observe that actually the ground state of the system does not exhibit invariance under global (i.e. space- time independent λ3) U(1) symmetry; this would require the possibility of changing the phase of the molecular vibrational field φ(r, t) simultaneously at every space-point by the constant amount λ3. Therefore, the global U(1) symmetry is also spontaneously broken. The condition which expresses the symmetry breakdown is the non-vanishing expectation value of D(+)(r, t) in the ground state 0i: h0D(+)(r, t)0i = v(r, t) 6= O, (6) and its complex conjugate relation involving D(−)(r, t). v(r, t) is a complex function and is called the order parameter, which is thus a (classical) vector field. It describes macroscopic collective properties of the system. Its space-time dependence denotes (space-time) non-homogeneities in the ground state. Since the vector potential Aµ is also involved in the dynamics, general theorems of QFT predict some dynamical effects globally named as the Anderson-Higgs- Kibble mechanism [26, 27]. Here, we will not consider such a mechanism (see Refs. [19, 20, 24, 25] for a general discussion). (the NG mode implied by the spontaneous breakdown of SU(2) symmetry). By resorting to the results of Refs. Let bP (r, t) and bP †(r, t) denote the annihilation and the creation operator, respectively, of the dipole-wave quantum [21, 22], we have that the ground state turns out to be eigenvector of the dipole-wave operator bP (r, t): bP (r, t)0i = v(r, t) (2P)1/20i, which shows that the ground state is a coherent state: dipole-wave quanta (the NG boson modes) are coherently condensate in the state 0i [21, 22]. We also have We may also define the polarization density as h0D(−)(r, t)D(+)(r, t)0i = 2P(r, t)h0bP †(r, t)bP (r, t)0i = v(r, t)2, (7) (8) (9) (10) with ρ(r, t) the charge density and δ the (average) dipole length. By writing the NG condensate density as h0D(−)(r, t)D(+)(r, t)0i = ρ(r, t) δ, h0bP †(r, t)bP (r, t)0i = n(r, t), we have ρ(r, t) δ = v(r, t)2 = 2P(r, t) n(r, t). We then write the charge density wave function σ(r, t) as σ(r, t) =pρ(r, t) eiθ(r,t) , with real ρ(r, t) and θ(r, t). We have σ(r, t)2 = ρ(r, t) ≡ N (r, t) in the system ground state. Thus, N (r, t) denotes the charge density condensate. Sometimes σ(r, t) is also called the macroscopic wave function. We observe that the order parameter v(r, t) provides a measure of the NG mode density n(r, t) in the ground state. In terms of σ, the local gauge transformation (4) is σ(r, t) → e−iq λ3(r,t)σ(r, t) , Note that the transformation (4) is induced by the phase transformation θ(r, t) → θ(r, t) − q λ3(r, t), (11) (12) when Eq. (10) is used. One may show [21] that the phase θ(r, t) represents the NG wave dipole field, also called the phason field [20, 28]. As mentioned, the meaning of the macroscopic wave function, or the order parameter v(r, t), resides in the fact that in the presence of the spontaneous breakdown of symmetry, the microscopic quantum components of the system behave in collective or coherent way, namely they undergo in phase motion, so that the system is then globally described by such a common phase of its quantum constituents. This establishes the link between the system macroscopic wave function σ(r, t) and the microscopic one, ψ(r, t) =pa(r, t) exp(iS/), for the quantum components. We assume that ψ(r, t) satisfies the Schrodinger equation. Thus we will set the phase θ(r, t) ≡ S/ . Then for the momentum P = mv we have P ≡ ∇S = ∇θ. In the presence of the A field, the canonical momentum is known to be mv = P − qA (the minimal coupling), thus (cf. Eq. (B4)) m q − v = A − q ∇θ. (13) 6 q We note that the r.h.s. of Eq. (13) can be considered as a gauge transformation with θ(r, t) being the gauge function: A → A′ = A− ∇θ. The requirement that the gauge condition ∇· A = 0 be invariant under gauge transformations, i.e. ∇· A′ = 0, implies that θ(r, t) (and λ3(r, t)) has to be a solution of the equation ∇2θ = 0 (and ∇2λ3(r, t) = 0). In such a case, Eq. (13) gives ∇· v = 0. We remark that in general one may consider the so-called boson transformation (14) θ(r, t) → θ(r, t) − c f (r, t), with c a convenient constant and f (r, t), called boson transformation function, solution of the equation for θ, ∇2f (r, t) = 0, so that Eq. (14) is an invariant transformation for the theory. Such a transformation describes the space-time-dependent (i.e. non-homogeneous) boson transformation and it can be shown [4, 19, 20, 29, 30] that in order to have observable effects f (r, t) has to carry a topological singularity. This theoretical frame thus predicts the appearance of topological structure formation, such as vortices, as further discussed in Section IV (and in Ref. [4]). In the case of the U(1) phase symmetry mentioned above, the stationary function f (x) may carry indeed a vortex singularity given by f (x) = arctan(cid:18) x2 x1(cid:19) . (15) Eq. (15) shows that the phase is undefined on the line r = 0, with r2 = x2 2 (which, as observed in [4], reflects in the observed phase indeterminacy in the process of transition between two AM pattern frames). For an analysis of the vortex properties associated to Eq. (15) see Refs. [29, 30] and the discussion in Section IV. 1 + x2 In the following we study the variation in space and in time of the macroscopic wave function and of the phason field condensate. In QFT these variations denote transformations through physically different vacua, i.e. unitarily inequivalent vacua, or, in other words, they denote "phase transitions". Of course, going through phase transitions, the system moves toward the equilibrium, or stationary regime, which is the one minimizing the free energy density functional F (σ, σ∗, A) with respect to σ∗: ∂F /∂σ∗ = 0. However, during the phase transition processes such a minimization condition is not satisfied, ∂F /∂σ∗ 6= 0. Here we are interested in such dynamical transition processes since the brain is a far from the equilibrium system, indeed. Let us start by considering the Schrodinger equation for σ(r, t): with where the potential energy U is i ∂σ ∂t = Hσ, H ≡ 1 2m (−i∇ − qA)2 + U, U = q ϕ + µ0. (16) (17) (18) Here ϕ ≡ V /q, with V the scalar potential and µ0 the chemical potential. µ0 is included in the Schrodinger's equation for σ in order to account for the variations in the number σ2 = N . In the following our discussion we will closely follow Ref. [31], where the time-dependent Ginzburg-Landau equation for the modulus of the order parameter is obtained and conservation laws as well as dissipative effects are taken into account. The order parameter and the macroscopic wave function σ are not affected by quantum fluctuations (in this sense they are macroscopic fields). However, as said above, we are interested in their variations occurring in the (non- equilibrium) phase transition processes. Therefore, we need to allow variations in the number N , which justify the introduction of the chemical potential µ0 in Eq. (17). We note that a chemical potential term is not included in the Schrodinger's equation for the microscopic wave function ψ(r, t) describing the elementary dipole components, whose number is assumed to be constant during the system time evolution. Use of Eq. (10) into Eq. (16) gives the the equations for the imaginary and the real parts: ∂N ∂t + ∇ · (N v) = 0, (19) µ0 + µ1 + q ϕ + mv2 2 + ∂θ ∂t = 0, 7 (20) respectively. Here and in the following v2 = v· v. Eq. (19) is the continuity equation (to be compared with Eq. (B1)). The additional contribution µ1 to the chemical potential in Eq. (20) is µ1 = − 2 ∇2N 1/2 2m N 1/2 and is due to non-homogeneous density of the condensate. From Eq. (20) one obtains (see Appendix B) with m dv dt = q (E + v × B) − ∇ (µ0 + µ1) , E = − ∂A ∂t − ∇ϕ and B = ∇ × A = − m q (∇ × v). We observe that, provided that ∇ · v = 0 (see the remarks after Eq. (13) and Eq. (D6)), Eq. (19) gives dN dt = ∂N ∂t + v · ∇N = 0, (21) (22) (23) (24) which expresses the charge density conservation in the stationary regime. Since N is related to the order parameter, this means that the long range correlation, namely the ordering, is preserved in time. In the frame of the dissipative many-body model, such a case describes stationary neuronal correlates. However, observations of the functional activity of the brain show that a succession of neuronal correlates ("wave packets"), modulated in amplitudes and in phase, occurs. In its functional activity the brain appears undergoing a continuous stream of phase transitions. We are therefore interested in studying the non-stationary dissipative phase transition processes, where, contrarily to Eq. (24), dN /dt 6= 0. Thus we need to study the non-stationary or time-dependent Ginzburg-Landau equation in order to consider the possibility of dissipative processes. III. TIME-DEPENDENT GINZBURG-LANDAU EQUATION Let us start by considering the Ginzburg-Landau (GL) functional F (σ, σ∗, A) representing the free energy density. For our task we do not need to specify the explicit form of F (σ, σ∗, A). This will depend on the particular dynamical model one adopts for the description of the system under study. In general, it is a non-linear functional of the fields, containing a kinetic energy term (−i∇− qA)2/2m and some potential term. The stationary Ginzburg-Landau (GL) equation is obtained by extremizing the free energy density F (σ, σ∗, A) with respect to σ∗: where, in full generality, we may write: ∂F ∂σ∗ ≡ (cid:20) 1 2m ∂F ∂σ∗ = 0, (−i∇ − qA)2 + α + βσ2(cid:21) σ, (25) (26) with α(T ) and β(T ) = α(T )/N ◦(T ) > 0 acting as a mass term and a positive coupling constant term, respectively. N ◦(T ) = σ2 is the equilibrium condensate density. Their values depend on the temperature T and their explicit expressions depend on the system under consideration and on the adopted phenomenological model (in solid state physics α(T ) is related to the chemical potential µ0, expressing, e.g. in superconductivity [31], the vacuum perme- ability, and β(T ) to the critical (magnetic) fields acting upon the system. Notice that the contribution (α + βσ2)σ in the r.h.s. of Eq. (26) may be considered as derived from the potential V (σ, σ∗) = −(β/2)(σ2 + α/β)2. The (mean value of the) order parameter σ minimizing the potential V (σ, σ∗) is zero (disordered or symmetric ground state) for α ≥ 0, and non-zero for α < 0, with σ2 = −α/β 6= 0 (the ordered or asymmetric ground state)1 1 We also note the strong analogy with the laser phase transition [32], where the potential V (σ, σ∗) is again considered. In the phase transition between the disordered and the ordered (laser) state (the lasering process), the order parameter σ2 changes in time from zero to a non-vanishing value proportional to α, going through the threshold set at α = 0. In the Haken interpretation, α is the pump parameter whose tuning may carry the system in the lasering region; σ denotes the amplitude of the classical electromagnetic mode. 8 Eq. (25) expresses the condition for the stationary (equilibrium) dynamical regime, where the evolution of the macroscopic wave function is controlled by the Schodinger equation (16). Instead, in the case of non-stationary regime, the non-vanishing quantity ∂F /∂σ∗ expresses the rate at which σ(r, t) approaches its stationary value at the minimum of the free energy. It is then customary to consider the generalized time-dependent GL (TDGL) equation: i ∂σ ∂t = Hσ − i γ ∂F ∂σ∗ , (27) where now ∂F /∂σ∗ is not zero, γ is a relaxation parameter and the stationary regime is reached in the limit In such a limit, the equilibrium wave function is σ◦(t) = σ◦ exp(−iǫt/), with Hσ◦ = ǫσ◦. (1/γ)∂F/∂σ∗ → 0. The second term in the r.h.s. of Eq. (27) thus describes the dissipative contribution coming from incoherent relax- ation processes [31]. Using σ = √N exp(iθ), the imaginary and real part of Eq. (27) give (see the Appendix C) the continuity equation: and the relation for the chemical potentials ∂N ∂t + ∇ · (N v) = − 2G τGL N respectively, where µ0 + µ1 + µ2 + qϕ + mv2 2 + ∂θ ∂t = 0, µ2 = − 2γN ∇ · (N v) ≡ − ξ2 GL τGL m q ∇ · J N , (28) (29) (30) is related to the relaxation parameter γ, is of dissipative nature and proportional to non-homogeneities of the con- densate. We have used the definition J ≡ qN v. In Eq. (28) G is given by G = ξ2 GL 2m 2 (cid:18) mv2 2 + µ1(cid:19) + N N ◦ − 1 . (31) and τGL and ξ2 the gradient operator on the Eq. (29) we obtain GL are the GL relaxation time and coherent length, respectively (see Eqs. (C1)). By operating with = q (E + v × B) − ∇(µ0 + µ1 + µ2). Eqs. (28) and (32) have to be compared with Eqs. (19) and (22), respectively. m dv dt By using Eq. (31) and ∇ · v = 0, the continuity equation (28) can be rewritten as dN dt ≡(cid:18) ∂ ∂t + v · ∇(cid:19) N = −ΓRN 6= 0 where we have set ΓR ≡ 1 τR = 2G τGL (32) (33) (34) We thus see that the relaxation term Rdiss ≡ ΓRN accounts for the rate of change of the condensate dN /dt. From the expression of G, Eq. (31), we see that we may write ΓR = Γ1 + Γ2, where Γ1 = ξ2 GL τGL 2m 2 mv2 = 2DGL(cid:16) mv (cid:17)2 , (35) with the diffusion coefficient DGL ≡ ξ2 GL/τGL, and Γ2 related with non-homogeneities (µ1 and N 6= N ◦). Γ2 is related with dissipative processes whose life-time is usually longer than the GL relaxation time τGL. In general, the condition τR = τGL/2G ≫ τGL, requiring, in order to hold, small values of G, ensures fast formation of quasi-local equilibrium in the condensate, controlled by τGL, compared with longer decay time of the condensate density N . At a critical temperature Tc, a fast transition to the equilibrium condensate regime occurs; one may then expect G → 0, so that τGL ≪ τR = τGL/2G, as T → Tc. Eq. (33) gives the general TDGL equation for the normalized wave function χ = σ/σ◦ ≡ (N/N ◦)1/2: 1 τGL (cid:20)(1 − χ2) − ξ2 DGL∇2χ − dχ dt = − GL(cid:16) mv (cid:17)2(cid:21) χ , 9 (36) where dχ/dt = ∂χ/∂t + v · ∇χ. quasi-local equilibrium, we have In the approximation of ΓR ≈ Γ1 ≡ 2/τE, and in the case of fast formation of In such a limit, for τGL ≪ τE, Eq. (36) becomes τGL τE = ξ2 GL(cid:16) mv (cid:17)2 ≪ 1. which further reduces to DGL∇2χ − dχ dt = −(1 − χ2) χ τGL , ξ2 GL ∇2χ + χ − χ3 = 0 (37) (38) (39) when the stationary condition dχ/dt = 0 is met, i.e. ∂χ/∂t = 0 and v · ∇χ = 0. Eq. (39) reproduces indeed the stationary GL equation (C6) when the disequality (37) holds. IV. THE VORTEX SOLUTION Our discussion in the previous Sections concerning the stationary regime and the departure from it occurring in the phase transition processes is centered on the behavior of the quantities τGL(T ), ξ2 GL(T ) and ηGL(T ) defined in Eqs. (C1). These quantities, in turn, depend on the behavior of α(T ), which, as we have observed in Section III, has the meaning of the mass term in the equations for the macroscopic wave function (the order parameter) σ, or the wave function χ normalized to σ◦, namely χ = σ/σ◦ ≡ (N/N ◦)1/2 satisfying the general TDGL equation (36) (equivalent to Eq. (33)). We might consider non-instantaneous phase transition processes where a time-dependent temperature might still be defined. In these processes, assume that the transition starts at the critical temperature TC and the stable configuration is reached at the so-called "Ginzburg temperature" TG, with TG < TC, after a certain interval of time during which the system is said to be in the critical regime [33]. In practice, the model here adopted aims at describing brain functioning as a repeated passage through a critical temperature state towards which the brain naturally relaxes after having reached a lower-temperature stable configuration, that corresponds to the response to some external stimulus or any other kind of perturbation. The crucial point is that this "critical dynamics" is always performed as a transition between different thermodynamic equilibrium states (at TC and at TG): irrespectively of the nonequilirium processes involved into this transition, this implies that the general thermodynamic description derived from the fluctuation theorem [34] applies to this model of critical dynamics of the brain. The rate of entropy production that is inherent in the process described by the TDGL equation can be accordingly related to the amount of heat dissipated by the brain to reach new equilibrium configurations below the critical temperature TC. We want to model time dependence of α(T ) during such a critical regime evolution. Here we remark that the departure from the stationary regime (at TC ), representing the system ground state at a given time in the system evolution2, namely the start of the critical regime, is driven by fluctuations which may be spontaneous (typically occurring in a quantum vacuum) and which can trigger a phase transition, if the requisite transition energy is provided by some endogenous or external stimulus. These ground state fluctuations turn into temperature fluctuations since in the dissipative model the ground state is in fact a thermal state [7, 14, 15]. At the end of the critical regime (at TG) the systems arrives at a "new" ground state configuration and the "phase transition" is thus completed. As a matter of fact, the brain system undergoes a continuous sequence of phase transitions in a path through the (infinitely) many ground states whose existence is allowed by QFT [7, 14, 15]. This is a delicate point, where the dissipative many-body model turns out to be very helpful in depicting what a traditional non-equilibrium thermodynamical approach cannot explain. It is well known that mammalian brains operate at a steady state constant temperature that is homeostatically regulated. Then it is crucial reconcile the 2 See Refs. [7, 14, 15] for the discussion of the existence of infinitely many vacua (ground states) representing different, physically inequivalent, microscopic configurations of the system. 10 temperature invariance of brains with the temperature fluctuations intrinsic to "far-from-equilibrium" activity seen in laboratory experiments as well as in everyday life. (As noted in Section I, brains consume free energy at rates ten-fold greater than any other organ caused by the heat exchanges at a local molecular level due to the ceaseless electrochemical and metabolic reactions). The resolution of the apparent contradiction between homeostasis and the far from the equilibrium brain activity is implicit in the dissipative many-body model, which suggests that a phase transition entails a brief, local fluctuation of temperature during the transition that punctuates the pseudo -- equilibrium steady state. A peak is followed by reversal (a so-called biphasic transient), leading the system to a state that differs from the one before, as observed and predicted by the model. (For brevity we do not report more on such a hysteresis-like property in brain dynamics; see e.g. [7, 35 -- 38]). In fact, such a transient fluctuation in temperature has been demonstrated for nerve and modified muscle otherwise held in thermal steady state in experiments with the squid giant axon and siphon [39]. It shows that an action potential is accompanied by brief cooling of the nerve and synapse when the sodium ions expand into the intracellular compartment and the potassium ions expand outwardly. The cooling is soon overshadowed by the heat released by the burst of metabolism that pumps the ions back to their compartments and restores the energy reserve. The thermal reservoir of the water provides the buffering and averaging, such that the dissipative system can undergo substantial local fluctuations in temperature, which are masked in the global heat production. Thus we see another crucial role of water by continually holding temperature constant in the average, although locally (in space and in time) fluctuating. Our reference to the dependence of the system on temperature has to be understood as its dependence on such localized transients. It is interesting to note that in completely physically different contexts (such as inneuroscience, condensed matter physics, particle physics and cosmology) intensive theoretical and experimental research has shown that extended objects with non trivial topology (as for instance vortices) appear during the critical transition processes (see e.g. [16, 17, 40, 41]) and persist for varying time intervals thereafter. In the present Section we are indeed going to discuss the vortex solution to the TDGL equation. Due to the nonlinearity of the TDGL, numerical simulations have been associated often to the theoretical analysis (see [33] and Refs. quoted therein). Let us write Eq. (36) as 1 DGL dχ dt = ∇2χ + GL (cid:20)(1 − χ2) − ξ2 1 ξ2 GL(cid:16) mv (cid:17)2(cid:21) χ , (40) 1 dχ dt = γ dχ which admits as a solution the (quasi-)static vortex solution for dt ≈ 0 (in such a case, indeed, it is recognized to be the vortex equation) [19, 35]. The condensate produces the classical vortex envelope σ(r). Since σ0 is temperature dependent, the vortex envelope changes with T . The phase transition and the vortex formation appear thus as the effect of localized (non-homogeneous) boson condensation. DGL Actually, the phase transition begins with an abrupt, adiabatic, localized decrease of the order parameter (a measure of the analytic power of the background activity) to near zero, denoted as the null spike, resulting locally in loss of order and coherence, hence onset of symmetry. Concomitant to this, the spatial variance of the analytic phase increases and a discontinuity in space and in time of the analytic phase appears. The start of the phase transition is thus adiabatic and it appears to be instantaneous at the time scale of the beta and gamma cycle durations. The extreme spatiotemporal localization of the null spikes indicates that they are associated to singularities. These constitute the core out of which, in the non-instantaneous phase transition process, the vortices start to form. The singularities coincide with the apex of the phase cones which start to develop and spread over the system. We can thus say that the null spikes mediate or precipitate the phase transition toward the formation of a new AM pattern. The non-homogeneous boson condensation at the origin of the formation of vortices and ordered localized patterns requires energy (ordered states are lower in energy and separated by an energy gap from the symmetric state). Consequently, apart the adiabatic instantaneous start, the phase transition itself appears to begin with cooling and is non-instantaneous. The cooling appears as a manifestation of the process of symmetry breakdown eventually leading to the emergence of new extended AM and PM patterns. It is thus different from the very localized cooling that may occur due to localized (e.g. on the axon) metabolic activity. It is interesting to observe that these features predicted by the model find a correspondence in the features observed at the level of neuronal assemblies. It appears that the neural transition requires three conditions as precursors. The first is macroscopic arousal with a high level of background 1/f activity (see our remarks in the following Section). Second is activation of a Hebbian nerve cell assembly, which amplifies, generalizes and abstracts a sensory signal into a burst of high intensity but microscopic firing of pulses. Third is the initiation of a narrow-band oscillation in that burst. These three changes require an increase in dissipation of free energy as heat and therefore in temperature. But the phase transition itself begins with cooling as predicted by the dissipative model. core size is of the order ξ2 From the TDGL equation (see also Ref. [4] and [35]) one may also derive that, in the first approximation the vortex GL ∝ α−1. This means that the vortex core is increasing with temperature increase, as T 11 approaches TC from below, the vortex envelope disappearing at TC. Rising temperature above TC leads to symmetry restoration (unstructured ground state). Vice-versa, going from above back to TC the unstructured background activity (fully symmetric) with vanishing or very low analytic amplitude exhibits, at TC, undefined analytic phase, namely the singularity (null spike) as it is at the center line of the vortex core. As T is lowered below TC, the critical regime starts, vortices appear, whose core shrinks as temperature further decreases. As mentioned in Section II the singular NG boson (the phason) condensation is essential for obtaining a nontrivial topological charge. In Refs. [29, 30] it is shown that the space-time-dependent order parameter can be gauged away by an appropriate gauge transformation when the boson condensation function is regular. When the boson condensation function presents singularities corresponding to regions occupied by the normal state (i.e. without condensation) the boson condensation is manifested in macroscopic structures. The vortex, e.g., is singular on the z-axis, at r = 0. In this case, the topological charge is characterized by the integer winding numbers n 6= 0. Associated to the vortex there is the quantized flux φ = 2π n/q [4, 19, 35]. When the l.h.s of Eq. (40) is not neglected, one can show that in order to have a solution behaving as σ ∝ exp−(ω/γ)t, the condition (41) √k2), with k the wave number and has to hold during the critical regime between TC and TG for each k-mode (k ≡ m2(t) the time-dependent "effective mass". In our discussion then we may follow the arguments presented in Ref. [4], where we have considered the formation of vortices during the critical regime and the appearance of null spikes (there we have considered the harmonic limit approximation consisting in neglecting the nonlinear term in the TDGL equation (36) [16, 17, 33]). k2 ≥ m2(t) , Suppose the critical regime starts and ends at the times t = 0 and t = τ , respectively. For a given k, Eq. (41) holds up to a time τk such that m2(t) is larger than k2 for t > τk. The corresponding k-mode can propagate in a span of time 0 ≤ t ≤ τk, which determines the dimensions to which the domains can expand and the "effective causal horizon" [40, 41] will be inside the system (possible formation of more than a domain) or outside (single domain formation) according to whether the time occurring to the k-mode to reach the boundaries of the system is longer or shorter than the allowed propagation time, respectively. In order to determine the value of τk, one must assign the explicit form of m2(t). Its time-dependence may be fixed in such a way to allow vortex formation [17]. One possible modeling of m2(t) is then: m2(t) = m2 0 e2h(t) . (42) Let ξ denote the correlation length corresponding to the k-mode propagation and L ∝ m−1 propagation time is implicitly obtained from Eqs. (41) and (42) as: 0 . Then the correlation h(τk) = ln(cid:18) k m0(cid:19) ∝ ln(cid:18) L ξ(cid:19) . (43) We see that L acts as an intrinsic infrared cut-off. Small values of k are indeed excluded by the non-zero minimum value m0 of m. Consequently, long wave-lengths are not allowed. This means that only domains of finite size can be obtained, which is what happens in the brain activity. The linear size of the domains over which the correlation length extends at the end of the critical regime is of the order of λk ∝ m−1(τk). It remains to choose the explicit analytic expression for h(t) in Eq. (42). The choice we may adopt for h(t) is [17]: h(t) = ± at bt2 + c , (44) where a, b, c are (positive) parameters chosen so to guarantee dimensionless h(t). Let λ be an arbitrary constant and define the ratios c/aλ ≡ τQ, aλ/b ≡ τ0. We have h(τQ) = h(τ0) and the time derivative of h(t) (and thus of m2(t)) is zero at t = τ = ±√τQτ0, which thus plays the role of the equilibrium time scale. We have h(t) = ± 1 1 λτQ 1 + t2 τ 2 t ≈ ± Γ 2 t , (45) for t2/τ 2 ≈ 1 and Γ ≡ 1/λτQ. appearing during the critical regime is given by [17, 41]: In the linear approximation one finds that the number of vortices ndef possibly ndef ∝ m2(τ ) ≈ m2 0 τ /λτQ . (46) 12 Since the size of the vortex core is given by (m(t))−1, Eqs. (42) and (45) show that such a size evolves in time as e∓Γ t, t < τ (t < τk for the k-mode). Thus, as observed in Ref. [4], we have both, converging (imploding) and diverging (exploding) regimes, which is in agreement with the phase cone behaviors observed in laboratory. The "normal" (disordered) state is confined to the vortex core, then in the case of enlargement of the vortex core (exploding regime) disorder (local correlations) prevail. Instead, the shrinking of such a region (imploding regime) may signal that ordering (long range correlation) is prevailing (the vortex is "squeezed out"). This is in agreement with laboratory observations which show that explosion or implosion is obtained if the local connections or the long axon connections predominate, respectively [42]. Observations also show that many phase cones exhibit little or no rotation but repetitive outward or inward pulsations with each cycle. When rotational gradients (vortices) are observed, the singularity is associated to the vortex core singularity. All four types of these observed spatiotemporal phase gradients (See Fig. 2) are thus predicted by the model. Finally, we note that conventional neurodynamics may explain the negative gradient (e.g. in terms of a pacemaker), but not the positive gradient. Moreover, in the conventional framework there is no explanation of why both gradients, the positive and the negative one, occur, one or the other at random. Furthermore, we offer as a prediction the expectation that the locations of the initiating null spike, the apex of the following phase cone, and the center of rotation or expansion-contraction of the vortices will be closely related though not necessarily coincident. At present the available experimental techniques indicate that this occurs, but they lack sufficient resolution and precision to provide experimental proof. V. POWER LAWS AND COHERENT STATES In this Section we consider one more of the predictions of the many-body dissipative model which is confirmed in laboratory observations. According to the model, the system ground state and the phase transition processes are characterized by the boson transformations, such as the one of the NG field described by Eq. (14) (see also Eq. (12)). Boson transformations are the ones by which (Glauber) coherent states are constructed [19, 43]. Coherence is therefore a distinctive feature of the basic dynamics of the system and one may expect that it manifests itself at a mesoscopic and macroscopic level since coherent states are known to be the quantum states most near to classical (macroscopic) states. Elsewhere [3, 4, 7] it has been stressed that one of the merits of the dissipative many-body model is that "classicality" emerges naturally out of the basic quantum dynamics, not as an artificial classical limit imposed by hand. The laboratory observations are in fact at a classical macroscopic level, as it should be (it has to be stressed that in our analysis neurons, glia cell and other biological structures are classical objects; the quantum variables are the dipole vibrational modes). However, some of the system features are macroscopic quantum features, in the sense that they cannot be explained without recurs to the underlying quantum dynamics. Among such features, there are the power-laws observed in the brain functional activity, on which we focus our attention in this Section. In the frame of the entire analytical function formalism, power-laws have been shown to be related to coherent states [44, 45]. Let us very briefly summarize such a result. Power-laws denote scale free, self-similar phenomena of fractal nature characterized by a straight line graph in log-log coordinates, with the fractal dimension d corresponding to the slope of the line (the tangent y/x of the straight line angular coefficient). Self-similarity is a characterizing fractal property [46, 47] expressed by (q γ)n = 1, for any integer n, (47) where for convenience (see below) the number q is parameterized as q = 1/f d and γ is the parameter characterizing the specific considered fractal, whose dimension d is then given by d = − Logγ Logf . (48) This equation clearly describes a straight line in the log-log coordinates (x = Logf , y = Logγ). Here we denote log10 by the symbol Log. We now remark that (q γ)n, up to the normalization factor 1/√n!, is the nth element of the basis of the entire analytical functions defined in the space of the complex numbers z = q γ. In Eq. (47) the number q γ is considered to belong to the restriction to the real axis of the z-complex plane. The connection with coherent states is then established by observing that they are in fact constructed in the Fock-Bargmann representation (FBR) by use of the entire analytical functions [19, 43]. More precisely, the coherent states for such a z = q γ variable are defined to be "q-deformed" coherent states of coherence strength γ, or, also, squeezed coherent states, q being the deformation or squeezing parameter. We will not insist on such technical points here and the reader may consult Refs. [44, 45] for details. The point is that, as a result, self-similar processes of fractal dimension d can be associated to q-deformed quantum coherent states, namely they appear as macroscopic systems generated by local deformations of coherent quantum dynamics, they are macroscopic quantum processes induced by quantum boson transformations. 13 FIG. 3: A. Multichannel recording from a high-density array fixed on the auditory cortex of a rabbit at rest. B. The power spectral densities in the working ECoG were computed for all 64 signals and averaged. The power-law trend lines (1/f 0 and 1/f 2.6) were drawn by hand to emphasize the multiple peaks of power in the theta and beta-gamma ranges above the line, which were missing in the resting ECoG. From Figs. A1.01 and A1.02 in [12] Let us now turn to the laboratory observations (see e.g. [9 -- 12, 48]) and to the graphs of the power laws depicted in the lo-log coordinates in Figure 3 and 4. On the basis of what said above, it is evident that we face macroscopic quantum processes characterized by a coherent dynamics. Coherence is in fact the property to which the analysis in terms of neuronal correlates also leads. This can be seen as follows. In laboratory observations [9 -- 12, 48] multichannel recording yields a collection of state variables. Weakly coherent signals are obtained in ECoG with electrodes which are sufficiently far apart. The cortex may be considered to be modular, and each module gives a signal. Coherent signals are instead obtained when the electrodes are closely spaced. In such a case we may talk of variables defining a state space. They are all from the same module, and their variations in time define a trajectory through the modular state space. Many data have been collected [9 -- 12] on ECoG spatial imaging coming from small, high-density electrode arrays fixed on the surfaces of the olfactory, visual, auditory, or somatic modules (typically 8× 8 electrodes, spaced 0.79 mm apart, giving a 5.6 × 5.6 mm square aperture onto the ECoG). Figure 3.A shows the unfiltered background ECoG, obtained in such a way, which is featureless in appearance. Spectral analysis of the ECoG shows broad distribution of the frequency components in Figure 3.B, where the temporal power spectral density (P SDT ) in log-log coordinates is power-law, 1/f α, in segments. Below an inflection in the theta-alpha range the P SDT is flat, α = 0. Above, the log10 power decreases linearly with increasing log10 frequency in the beta-gamma range (12.5 − 80 Hz) with the exponent α between 2 and 3. One can show that in slow wave sleep the exponent averages near 3 [49]; in seizures it averages near 4. Above 75 Hz the slope either increases (α = 4, [50]), or it flattens to α = 0. On the basis of what said above, such values of the slope α provide corresponding values of the fractal dimension d ≡ α, with deformation parameter q ≡ 1/f α and coherent strength γ corresponding to the power spectral density. In Figure 3.B, each peak of power above the 1/f trend line in the beta-gamma range (arrows in Figure 3.B) reflects a brief epoch of narrow band oscillation. These multiple peaks indicate a departure from the scale free regime (the straight line) and therefore the presence of structures emergent from the background activity. Figure 4.A, B shows the post stimulus time histograms (PSTH) of the microscopic pulses from a representative neuron in an excitatory population due to low and high intensity electric shocks. The averaging gives the impulse response of the macroscopic population. It presents a rapid rise and an exponential decay to the background firing rate. When the shock strength is reduced to threshold, the decay rate approaches zero. In dynamical terms the asymptotic convergence of activity to the background after perturbation is evidence that a point attractor governs the cortical background activity at unity gain (pp. 285-305 in [48]). An attractor by definition represents a stable (ground) state. The power law in Figure 4.C, D signals the scale free ("spatial similarity") property of such ground states. The spatial similarities reveal the long spatial distances across which synaptic interactions can sustain coherence, namely the high-density coordination of neuron firing by synaptic interaction, in agreement with the dissipative model prediction. In the frame of our discussion of theoretical and observational results, the coherent mutual excitation among pyramidal cells thus appears as an especially attractive conclusion, also considering that they comprise about 80% of cortical neurons, and they provide 90% of the synapses of cortical origin to each other. 14 FIG. 4: A, B. The symbols ∆ show the post stimulus time histogram (PSTH) of a representative neuron in response to a weak electric shock, fitted with the solution to a 4th order linear differential equation [48]. The prolonged discharge without inhibitory overshoot is reverberation that is due to mutual excitation. C, D. The decay rate determines the inflection frequency of the power-law P SDT , and the rise rate determines the exponent, α. The predicted range of the trend lines is 2 < α in 1/f α. We propose that the segments give the canonical form of resting cortical P SDT . From Figs. 5, 7 in [49] VI. CONCLUSIONS The brain is an open system submitted to continuous exchange of energy and information with the environment. Sensory inputs drive the brain out of its background activity and large amplitude modulated patterns appear [9 -- 12] which are the expression of long range neuronal correlation. Moreover the brain manifests a form of conditional stability, criticality, denoting a state of readiness to transit from one state to another, like the one of a subject who is expecting one of several CS, by which to choose one of several courses of action. In the expectant state of searching, a sensory cortex can be viewed as having a set of attractors, one for each expected CS, and each with its basin around an attractor in an attractor landscape [51]. The CS determines the choice by directing the state trajectory into a particular basin. Convergence to the attractor directs the cortex to form the AM pattern that is selected by the CS. The cortical assemblies of neurons can complete widespread phase transitions in very few milliseconds, regardless of their correlation lengths and carrier frequencies. This property can help explain the rapidity of the flow of mental images. The distances across which coherent oscillations are maintained (the correlation lengths) are far larger than the modal diameters of the dendrites of the participating neurons and can include multiple cortical areas, even the entire scalp, which can help to explain the multisensory integration that is required for Gestalt formation. Moreover, spectral analysis of the ECoG shows the occurrence of 1/f α power-law form of the temporal spectral density in log-log coordinates which indicates that the activity is scale-free. The above properties are the macroscopic manifestation of the mesoscopic and microscopic dynamics. The formation of AM and PM assemblies of neurons have been associated, in previous works [3, 7], with the spontaneous breakdown of symmetry occurring at the level of the many-body system components and their coexistence and temporal succession to the dissipative character of the dynamics. The large distances across which coherent oscillations are observed to occur in the brain is accounted by the long range of the correlation, extending even to the whole system volume, predicted as a consequence of the spontaneous breakdown of symmetry in QFT. In this paper, continuing our study of the many-body dynamics underlying these behaviors, we have focused our attention on the non-equilibrium non- 15 instantaneous phase transitions occurring at the level of the many-body dynamics. In the stationary regime (Section II), as customary, one considers the extremizing condition of the system free energy ∂F ∂σ∗ = 0 whose solutions describe the ground state of the system. In the brain activity, however, the brain is continuously moved out from its ground state activity entering a non-stationary dynamical regime. We have thus considered (Section III) a non-vanishing value for the quantity ∂F ∂σ∗ , which now expresses the rate at which the system approaches to the stationary regime. In such a way we have obtained the time-dependent Ginzburg-Landau equation. During such a non-equilibrium phase transition, named the critical Ginzburg-Landau regime, the system dynamics turns out to be characterized by topologically non-trivial structures (vortices), which in the formalism of the dissipative model of brain are described by non-homogeneous boson condensation processes. The size and the life-time of these topological structures have been discussed (Section IV) and their behavior commented upon in relation to the temperature changes in the domain of the critical regime, from the critical temperature TC at which the transition starts to the so-called Ginzburg temperature TG, with TG < TC. The number of vortices appearing in the phase transition process has been estimated, in agreement with a previous analysis [4]. In such a critical phase transition the system moves towards a new stationary ground state, which thus describes an attractor for the system non-linear evolution. Since our frame is the one of QFT, which posses infinitely many unitarily inequivalent ground states, the critical system transition to the stationary regime is not at all described by a trivially determined trajectory in the state space of the system. Elsewhere [3, 7, 36 -- 38] we have shown that such a trajectory is indeed a deterministic chaotic trajectory in an attractor landscape, the initial conditions being determined by the CS specificity. Finally, the observation of the 1/f α power-law of the temporal spectral density in the lo-log coordinate plot, which exhibits a scale free activity in the brain ground state activity, has been described, in agreement to previous studies of the relation between fractals and coherent states [44, 45], as the macroscopic manifestation of the underlying coherent many-body dynamics, namely in terms of the scale-free property of coherent states. In such a derivation, the properties of the entire analytical functions have been used. Appendix A The ECoG and EEG present nearly random but bounded amplitude and spectral distributions, often power-law (1/f ). The oscillations in the beta-gamma range occur in brief epochs with spatial patterns of amplitude and phase that are statistically constant. In those stationary epochs, the two major operations of normal cortical dynamics of dendritic integration of waves and axonal transmission of pulses are executed in near-linear domains [48]. Superposition justifies our use of the tools of linear analysis. The interactions among excitatory neurons that produce the random 1/f oscillations in the ECoG and EEG background activity at rest has been modeled with a single positive feedback loop [49], and the interactions among excitatory and inhibitory neurons that produce the beta-gamma oscillations (12 − 80 Hz) with a negative feedback loop [48, 52]. Multiple AM/PM patterns coexist in different frequency bands, overlapping in space and time. Whether they interact and if so how strongly has not been determined; they appear to have something like the imperturbability of solitons. The stability of the patterns embedded in and arising from the background turbulence leads us to view cortical dynamics as bistable. In the behavioral state of expectancy the cortex is in a receiving mode that is characterized by disorganized, random pulse firing and chaotic wave activity. We adopt the AM pattern feature vector as an order parameter. For such a state it is zero, and say that the cortical populations are in a gas-like phase. In the behavioral state of perceiving the cortex is in a transmitting mode of strong interaction, which imposes order on the firing of pulses and synchronization on the carrier wave. The order parameter is non-vanishing positive. The change resembles the phase transition from a gas-like phase to a liquid-like phase of condensation. Usually on close examination of the analytic phase we can find a discontinuity in the time series at which the new AM/PM patterns begin [2] suggesting a Type I phase transition. After ∼ 3 − 5 cycles of the carrier the patterns more slowly dissipate, suggesting a Type II phase transition. The transitions only occur when a CS is presented and the subject is in an expectant state. We believe that the phase transition requires a non-vanishing activation energy, strong enough in order to minimize the risk of false alarms and to conserve energy, and that this energy is provided by a Hebbian assembly, which is an intracortical mesoscopic network that has been formed by prior reinforcement learning each discriminable CS. Its functions are to amplify, abstract and generalize each weak sensory input by reverberation (mutual excitation) to activate the entire assembly by any part. The Hebbian assembly also greatly facilitates narrow band oscillation in the beta-gamma range [48]. The bandwidth is not zero; the narrow distribution about the center carrier frequency causes beats, which appear as reductions in amplitude that recur at intervals proportional to the bandwidth regardless of the center frequency [53]. During a beat the power in the designated frequency band can briefly go to zero [54] in one or more null spikes [4]. We propose that the null spike manifests a space-time singularity at which the phase transition begins. Its location may weakly correspond to those of the apex of the phase cone and the center of vortex rotation. 16 Evidence for the vector fields is derived from recording and interpreting the highly textured spatial patterns of oscillations provided by the images of scalar fields of potential from the EEG and ECoG. The neural mechanisms of vector field formation are modeled with large networks of differential equations that we solve by piece-wise linear approximation (Chapter Six in [48] and by implementation in VLSI [55]. The solutions show the existence of a point attractor that homeostatically regulates the background activity that maintains cortex in a stable state of criticality. This is the receiving phase [49], in which the oscillatory impulse response has en exponential envelope with an exponent, a < 1. At criticality a = 0, endowing cortex with exquisite sensitivity to selected input, form scale-free, power-law distributions of its parameters and state variables, and establish remarkably long correlation distances that can include the entire cortex of each hemisphere. In the transmitting state a > 0, and the cortex transits to a limit cycle attractor in a form of subcritical Hopf bifurcation. It is the amplified output of the Hebbian assembly that selects the carrier frequency, and it is the null spike that drives the cortex inexorably to the limit cycle attractor. Let ψ(r, t) =pa(r, t) exp(iS/) denote the wave function for the system quantum components, i.e. the microscopic wave function. It satisfies the Schrodinger equation from which the continuity equation Appendix B ∂a ∂t + ∇ · (a v) = 0 can be derived. As well known, by defining v = 1 m ∇S and Q = − 2 ∇2a1/2 2m a1/2 , one obtains the hydrodynamic equation of motion (the Madelung picture) ∂v ∂t + (v · ∇)v = − 1 m ∇(V + Q), (B1) (B2) (B3) where V is the potential appearing in the Schrodinger equation and Q is the so-called quantum potential. In the presence of a vector potential A the (hydrodynamic) velocity v is given by v = 1 m (∇S − qA). (B4) We see that the chemical potential µ1 in Eq. (21) is equivalent to the quantum potential Q in Eq. (B2). We consider now the derivation of Eq. (22). We apply the operator ∇ to both members of Eq. (20) and use the identities: ∇ mv2 2 = m(v · ∇)v + mv × (∇ × v) , dv dt ≡ ∂v ∂t + (v · ∇)v = ∂v ∂t + ∇ v2 2 − v × (∇ × v) . where v2 = v · v. The rotor of Eq. (B4) gives ∇ × v = − q m ∇ × A = − q m B. Combining these relations we get Eq. (22). By adopting the notation of Ref. [31], we put Appendix C τGL(T ) ≡ γ(T ) α(T ) , ξ2 GL(T ) ≡ 2 2mα(T ) , ηGL(T ) ≡ β(T ) α(T ) ≡ 1 N ◦(T ) , (B5) (B6) (B7) (C1) and using Eq. (26) and Eq. (27) takes the form: Use of the relations then leads us to Eqs. (28) and (29). R ≡ ξ2 GL(cid:16)∇ + q i A(cid:17)2 − ηGLσ2 + 1, i ∂σ ∂t = Hσ + i R τGL σ. v · ∇ρ1/2 ρ1/2 = ∇ · N 1/2v N 1/2 , ∇ · v = 0 . 17 (C2) (C3) (C4) The (stationary) case τGL → 0 (i.e. γ → 0), can be studied by considering Eq. (27). Multiplying both members times γ, we see that the limit γ → 0 leads to the stationary GL equation (25). On the other hand, by multiplying both members of Eq. (C3) times τGL, the limit τGL → 0 leads to ρ (cid:19)2 ∇ · J (cid:17)2(cid:21) χ − χ3 = 0, Rσ ≡ −(cid:18)G − iξ2 ∇2χ +(cid:20)1 − ξ2 The real part of Eq. (C5) gives G σ = 0, i.e. GL(cid:16) mv σ = 0 ξ2 GL (C5) (C6) m GL where χ = σ/σ◦ ≡ (N/N ◦)1/2. The vanishing of the imaginary part of Eq. (C5) implies which means that in the stationary case non-homogeneities in N only arise in the plane perpendicular to v. ∇ · J = 0, i.e. v · ∇N = 0, (C7) (D1) (D2) We can define µ ≡ µ0 + µ1 + µ2 and introduce the total chemical potential (cf. Eqs. (29) and (30)) µΣ = µ + mv2 2 ≡(cid:18)µ0 − 2 2m ξ2 GL τGL m q N (cid:19) + ∇ · J mv2 . 2 The gauge invariant potential Φ is defined as Φ ≡ qϕ + ∂θ ∂t Appendix D ∇2N 1/2 N 1/2 − = −(cid:18)µ + mv2 2 (cid:19) ≡ −µΣ. and we see that µΣ is always opposite in phase to Φ. We may also introduce ζ ≡ µ + qϕ and the electrochemical potential ζΣ given by We see that it depends on the rate of change of the NG mode (the phason mode θ). Finally, the gauge invariant vector and scalar potentials are given by ζΣ ≡ ζ + mv2 2 ≡ µΣ + qϕ = − ∂θ ∂t . (D3) Ag = − respectively. The corresponding electric field is mv q and ϕg = Φ q , E = − ∂Ag ∂t − ∇ϕg (D4) (D5) The Coulomb gauge condition ∇ · Ag = 0 then implies that ∇ · v = 0 (D6) and Eq. (B7) is also reobtained. Due to Eq. (D6), the velocity normal to any boundary thus vanishes, n · v = 0, and at boundaries any current such that n · v 6= 0 converts into a current with n · v = 0 and the condensate density N has nonuniform distribution. References 18 [1] J. Barham, Biosystems 38, 235 (1996) [2] W. J. Freeman, Neural Networks 21, 257 (2008) http://repositories.cdlib.org/postprints/2781 [3] W. J. Freeman and G. Vitiello, J. Phys. A: Math. Theor. 41, 304042 (2008); http://Select.iop.org; q-bio.NC/0701053v1 [4] W. J. Freeman and G. Vitiello, Int. J. Mod. Phys. B 24, 3269 (2010) http://dx.doi.org/10.1142/S0217979210056025 [5] W. J. Freeman, How Brains Make Up Their Minds, Columbia UP, New York 2001 [6] V. Braitenberg and A. Schuz, Cortex: Statistics and Geometry of Neuronal Connectivity, 2nd edn. Springer-Verlag, Berlin 1998 [7] W. J. Freeman and G. Vitiello, Phys. of Life Reviews 3, 93 (2006); http://dx.doi.org/10.1016/j.plrev.2006.02.001, http://repositories.cdlib.org/postprints/1515, q-bio.OT/0511037 [8] W.J. Freeman and G. Vitiello, J. Physics Conf Series 174, 012011 (2009). http://www.iop.org/EJ/toc/1742-6596/174/1. [9] W. J. Freeman, Clin. Neurophysiol. 115, 2077 (2004); http://repositories.cdlib.org/postprints/1006 [10] W. J. Freeman, Clin. Neurophysiol. 115, 2089 (2004); http://repositories.cdlib.org/postprints/1486 [11] W. J. Freeman, Clin. Neurophysiol. 116 (5), 1118 (2005); http://repositories.cdlib.org/postprints/2134, http://authors.elsevier.com/sd/article/S1388245705000064 [12] W. J. Freeman, Clin. Neurophysiol. 117 (3), 572 (2006); http://repositories.cdlib.org/postprints/1480/, http://dx.doi.org/10.1016/j.clinph.2005.10.025 [13] W. J. Freeman, R. Kozma, B. Bollob´as, O. Riordan, Chapter 7. Scale-free cortical planar network, in: Handbook of Large-Scale Random Networks. Series: Bolyai Mathematical Studies, Vol. 18, B. Bollob´as, R. Kozma, D. Miklos (Eds.), Springer, New York 2009, pp. 277 [14] G. Vitiello, Int. J. Mod. Phys. B 9, 973 (1995). [15] G. Vitiello, My Double Unveiled, John Benjamins, Amsterdam, 2001. [16] E. Alfinito, O. Romei and G. Vitiello, Mod. Phys. Lett. B 16, 93 (2002). [17] E. Alfinito and G. Vitiello, Phys. Rev. B 65, 054105 (2002). [18] M. G. Kitzbichler, M. L. Smith, S. R. Christensen, E. Bullmore, PLoS Comput Biol 5 (3), e1000314 (2009). doi:10.1371/journal.pcbi.1000314 [19] M. Blasone, P. Jizba and G. Vitiello, Quantum Field Theory and its macroscopic manifestations, Imperial College Press, London, 2011 [20] H. Umezawa, Advanced field theory: Micro, macro, and thermal physics, AIP, New York 1993. [21] E. Del Giudice, S. Doglia, M. Milani and G. Vitiello, Nucl. Phys. B 251 (FS 13), 375 (1985). [22] E. Del Giudice, S. Doglia, M. Milani and G. Vitiello, Nucl. Phys. B 275 (FS 17), 185 (1986). [23] J. Goldstone, Nuovo Cimento 19, 154 (1961). J. Goldstone, A. Salam and S. Weinberg, Phys. Rev. 127, 965 (1962). [24] C. Itzykson and J. Zuber, Quantum field theory, McGraw-Hill, New York, 1980. [25] H. Umezawa, H. Matsumoto and M. Tachiki, Thermo Field Dynamics and condensed states, North-Holland, Amsterdam, The Netherland 1982. [26] P.W. Anderson, Basic Notions of Condensed Matter Physics, Benjamin, Menlo Park, 1984. [27] P. Higgs, Phys. Rev. 145, 1156 (1966). T.W.B. Kibble, Phys. Rev. 155, 1554 (1967). [28] L. Leplae and H. Umezawa, Nuovo Cim. 44, 410 (1996) [29] H. Matsumoto, N. J. Papastamatiou, H. Umezawa and G. Vitiello, Nucl. Phys. B 97, 61 (1975). [30] H. Matsumoto, N. J. Papastamatiou, H. Umezawa, Nucl. Phys. B 97, 90 (1975). [31] A.A. Barybin, Advances in Condensed Matter Physics, 425328 (2011) [32] H. Haken, in Proc.Int. School of Physics E.Fermi, Nonlinear spectroscopy, ed. N. Bloembergen, North-Holland, Amsterdam 1977, p.350 H. Haken, Laser theory, Springer-Verlag, Berlin 1984 [33] L.M.A. Bettencourt, N.A. Antunes and W.H. Zurek, Phys. Rev. D62, 065005 (2000). N.A. Antunes, L.M.A. Bettencourt and W.H. Zurek, Phys. Rev. Lett. 82, 2824 (2000). [34] As a general references see G. E. Crooks, J. Stat. Phys. 90, 1481 (1998). C. Jarzynski, Phys. Rev. Lett. 78, 2690 (1997). C. Jarzynski, Phys. Rev. E 56, 5018 (1997). [35] R. Manka and G. Vitiello, Annals of Phys. 199, 61 (1990) [36] G. Vitiello, Int. J. Modern Physics B 18, 785 (2004). [37] E. Pessa and G. Vitiello, Mind and Matter 1, 59 (2003). 19 [38] E. Pessa and G. Vitiello, Int. J. Modern Physics B 18, 841 (2004). [39] B. C. Abbot, Journal of General Physiology 43, 119 (1960) [40] T.W.B. Kibble, in Topological defects and the non-equilibrium dynamics of symmetry breaking phase transitions, eds. Y.M. Bunkov and H. Godfrin, NATO Science Series C 549, Kluwer Acad., Dordrecht, 2000, p. 7. G.E. Volovik, in Topological defects and the non-equilibrium dynamics of symmetry breaking phase transitions, eds. Y.M. Bunkov and H. Godfrin, NATO Science Series C 549, Kluwer Acad., Dordrecht, 2000, p. 353. [41] W.H. Zurek, Phys. Rep. 276, 177 (1997) and Refs. therein quoted. [42] R. Kozma, Unpublished data (2005). [43] A.M. Perelomov, Generalized Coherent States and their Applications, Springer, Berlin, 1986. [44] G. Vitiello, New Mathematics and Natural Computing 5, 245 (2009). [45] G. Vitiello, Fractals and the Fock-Bargmann representation of coherent states, in Quantum Interaction. Third International Symposium (QI-2009), Saarbruecken, Germany, Eds. P. Bruza, D. Sofge, et al.. Lecture Notes in Artificial Intelligence, Edited by R.Goebel, J. Siekmann, W.Wahlster, Springer-Verlag Berlin Heidelberg 2009, (pp. 6-16). [46] H.O. Peitgen, H. Jurgens and D. Saupe, Chaos and fractals. New frontiers of Science, Springer-Verlag, Berlin 1986 [47] P. Bak and M Creutz, in Fractals in Science, eds. A. Bunde and S. Havlin, Springer -Verlag, Berlin 1995. [48] W. J. Freeman, Mass Action in the Nervous System, Academic Press, New York 1975, 2004. http://sulcus.berkeley.edu/MANSWWW/MANSWWW.html [49] W.J. Freeman and J. Zhai, Cogn. Neurodyn. 3 (1), 97 (2009). http://repositories.cdlib.org/postprints/3374, http://dx.doi.org/10.1007/s11571-008-9064-y [50] L.M. Miller and C.E. Schreiner J Neurosci 20, 7011 (2000) [51] C.A. Skarda, W.J. Freeman Brain and Behavioral Science 10, 161(1987) [52] W. J. Freeman and H. Erwin, Freeman K-set. Scholarpedia 3(2), 3238 (2008). http://www.scholarpedia.org/article/Freeman K-set [53] S.O. Rice, Mathematical Analysis of Random Noise and Appendixes. Technical Publications Monograph B-1589. Bell Telephone Labs Inc., New York 1950. [54] W.J. Freeman, Cognitive Neurodynamics 3 (1), 105(2009). http://repositories.cdlib.org/postprints/3387. [55] J.C. Principe, V.G. Tavares, J.G. Harris and W. J. Freeman, Design and implementation of a biologically realistic olfactory cortex in analog VLSI. Proceedings IEEE 89, 1030 (2001)
1507.00649	1	1507	2015-07-02T16:25:22	Frequency-Dependent Selection at Rough Expanding Fronts	[ "physics.bio-ph", "cond-mat.stat-mech", "q-bio.PE" ]	Microbial colonies are experimental model systems for studying the colonization of new territory by biological species through range expansion. We study a generalization of the two-species Eden model, which incorporates local frequency-dependent selection, in order to analyze how social interactions between two species influence surface roughness of growing microbial colonies. The model includes several classical scenarios from game theory. We then concentrate on an expanding public goods game, where either cooperators or defectors take over the front depending on the system parameters. We analyze in detail the critical behavior of the nonequilibrium phase transition between global cooperation and defection and thereby identify a new universality class of phase transitions dealing with absorbing states. At the transition, the number of boundaries separating sectors decays with a novel power law in time and their superdiffusive motion crosses over from Eden scaling to a nearly ballistic regime. In parallel, the width of the front initially obeys Eden roughening and, at later times, passes over to selective roughening.	physics.bio-ph	physics	Frequency-Dependent Selection at Rough Expanding Fronts Jan-Timm Kuhr1, ∗ and Holger Stark1 1Institut fur Theoretische Physik, Technische Universitat Berlin, Hardenbergstrasse 36, 10623 Berlin, Germany (Dated: November 5, 2018) Microbial colonies are experimental model systems for studying the colonization of new territory by biological species through range expansion. We study a generalization of the two-species Eden model, which incorporates local frequency-dependent selection, in order to analyze how social inter- actions between two species influence surface roughness of growing microbial colonies. The model includes several classical scenarios from game theory. We then concentrate on an expanding public goods game, where either cooperators or defectors take over the front depending on the system pa- rameters. We analyze in detail the critical behavior of the nonequilibrium phase transition between global cooperation and defection and thereby identify a new universality class of phase transitions dealing with absorbing states. At the transition, the number of boundaries separating sectors decays with a novel power law in time and their superdiffusive motion crosses over from Eden scaling to a nearly ballistic regime. In parallel, the width of the front initially obeys Eden roughening and, at later times, passes over to selective roughening. PACS numbers: 02.50.Le, 68.35.Ct, 68.35.Rh, 87.18.Hf, 87.23.-n I. INTRODUCTION Living species are usually confined to their territory, a spatial region defined by geographical borders, climate, or other environmental constraints. Uninhabited regions are colonized through range expansion, where individuals reproduce and disperse at the front of their territory [1]. This process is seen in biological invasions [2], as a result of shifting climate zones [3 -- 5], during colonizations in our own species' history [6 -- 8], tumor growth [1, 9, 10], and biofilm growth [11 -- 13]. Evidently, expansions occur on very different spatial (micrometers to 107 meters) and temporal (hours to millennia) scales. In this article we aim to characterize range expansion under the influence of short-range "social interactions" of individuals at the front. Such interactions are present if success in reproduction depends on the presence of nearby individuals of the own and/or other species. Here, we set up a model for the expanding front based on evo- lutionary game theory [14 -- 16] and investigate its rough- ening dynamics for two interacting species. Besides ex- ploring an interesting non-equilibrium growth process, we hope to contribute to interpreting experiments on range expansion in multi-species colonies of simple organisms. In experiments, microbial growth is excellently suited to study range expansion and other processes in pop- ulation dynamics and evolution such as spatial spread of infections and adaptation to an environment (see for example Ref. [17]). Microbes reproduce fast, their en- vironment and genotype can be controlled, and exper- imental conditions are easily reproducible. Grown in a Petri dish, the spatial patterns of single-species microbial colonies have long been a rich field of study [18 -- 22]. The observed patterns crucially depend on motility, availabil- ity of nutrients, and the growth medium, to name but a ∗ [email protected] few. However, even under conditions of negligible motil- ity and abundant nutrients a colony's front is rough and has interesting statistical properties [23 -- 25]. Multi-species colonies are composed of more than one species and show additional intriguing features, even if the species are identical except for a marker [26]. During reproduction they keep their marker but compete with other species for space at the front. Thereby, sectors of single species form, which are separated by boundaries. Their statistical and dynamic properties are determined by the evolving roughness of the expanding front [27, 28]. Usually, when two or more species of microbes live in a common environment, they influence each other dur- ing reproduction. In particular, reproductive success of any species, also called its fitness, depends on the pop- ulation sizes of all the species. This constitutes "social interactions" between the species commonly referred to as frequency-dependent selection. Research in the field has initiated a wealth of fascinating experiments [10, 29 -- 35] either in well-mixed liquid culture without any spatial order [31, 36] or in the Petri dish, where the populations are spatially structured [37, 38]. Many of the experimen- tal observations can be discussed within the framework of evolutionary game theory [14 -- 16]. For example, light has been shed on a long-standing theoretical question in evolution [14, 15, 39, 40]: Why do individuals cooperate if non-cooperators can exploit them? Literature empha- sizes the importance of a population to be structured in groups [15, 41 -- 45], for example, by spatial distance. While within a single group cooperators are always in- ferior to non-cooperating defectors, if the latter interact only with their neighborhood, distant large groups of co- operators will ultimately outcompete defectors. Some models also stress the central role of demographic fluctu- ations and of populations growing in size [46, 47]. Both, experiments and theory, explain the advantage of coop- erators during colony growth by their ability to locally advance faster [38, 48, 49], vividly termed "survival of 5 1 0 2 l u J 2 ] h p - o i b . s c i s y h p [ 1 v 9 4 6 0 0 . 7 0 5 1 : v i X r a the fastest" [38]. Cooperation between between nearby cells is often me- diated by some biochemical compound (a public good ) which the microbes release into the extracellular envi- ronment. This compound then promotes reproduction of neighboring cells. In general, a released substance may act beneficial or detrimental to other individuals, also depending on their species, and implies some cost to the producer. Examples include secretion of digestive inver- tase to break down sucrose [36, 50 -- 52], siderophores to scavenge iron from the environment [31, 53, 54], poly- mers which support biofilms [41, 55, 56], release of tox- ins [57, 58] (sometimes through lysis [59, 60]), surfactants which facilitate swarming [61], and the exchange of amino acids [37]. This plethora of biochemical compounds, released by cells and affecting nearby cells, implies a wealth of spe- cific features, which certainly are not covered by a sin- gle model. However, since the released biomolecules usually mediate short-range interactions between indi- viduals, properties on large scales should be indepen- dent of microscopic details. Hence, we formulate a sim- ple model which captures the essence of an interaction while ignoring complicated details. The classical Eden model [62, 63], a simple growth process on a lattice, has been used successfully to mimic growing cell colonies.. It generates a cluster (the colony), the surface of which ex- hibits scaling properties also found for expanding fronts of microbial colonies [64, 65]. Extended to two identi- cally growing but still distinguishable species, it gener- ates sectors occupied by a single species only [28, 66]. Indeed, this behavior is found for two-species micro- bial colonies [26]. Moreover, boundaries between sectors move superdiffusively as in the experiments. In this article we explore a generalization of the two- species Eden model, which incorporates local frequency- dependent selection. We thereby aim to analyze how social interactions influence surface roughness of grow- ing microbial colonies. We set up an expanding public goods game, where either cooperators or defectors take over the front depending on the system parameters [14 -- 16]. Right at the transition the front displays critical behavior, which we analyze in detail. In particular, we establish that our model belongs to a new universality class of phase transitions dealing with absorbing states. At the transition, the number of boundaries separating sectors decays with a novel power law in time and their superdiffusive motion crosses over from Eden scaling to a nearly ballistic regime. In parallel, the width of the front initially obeys Eden roughening and, at later times, passes over to what we call selective roughening. The remainder of this article is organized as follows. To analyze multi-species microbial colony growth, we in- troduce the Eden model with frequency-dependent selec- tion in Sec. II and analyze its phenomenology in Sec. III. We then concentrate on the expanding public goods game with its social dilemma in Sec. IV and analyze the critical behavior at the transition between long-term cooperation 2 FIG. 1. Two-species Eden model with frequency-dependent selection on a hexagonal lattice. The bacterial colony grows from the bottom line (lattice sites with narrow black edge) of length L, where individuals of species 1 (blue) and species 2 (red) occupy the lattice sites. The colony expands into the empty, infinitely-extended half-space. Individuals capable of reproduction (indicated by bold edges), have at least one empty lattice site as a nearest neighbor. In a reproduction event, one of these empty neighboring sites (i, j) is chosen with equal probability and the reproducing individual changes the corresponding state si,j to its own state 1 or 2. Each individual has its own reproduction rate bi,j given in Eq. (1). For example, the reproduction rate of the individual at site 2 + 3T + 2P. Along the (3, 3) (bold black edges) is b3,3 = b0 transverse direction periodic boundary conditions apply. and long-term defection by applying statistical analysis. Finally, we discuss and summarize our findings in Sec. V. II. EDEN MODEL WITH FREQUENCY-DEPENDENT SELECTION In this work we employ a lattice model (see Fig. 1) to analyze range expansion at rough fronts under the in- fluence of frequency-dependent selection. We set up a cellular automaton on a two-dimensional hexagonal1 lat- tice of transverse extension L and an infinite longitudinal extension. Periodic boundary conditions are applied in the transverse direction. The state {s} of the system at time t is specified by the state variables si,j of lattice sites (i, j). Consider a system with two species (extension to more species is straightforward). For any time t, any site (i, j) is either empty (si,j = 0) or occupied by an indi- vidual of either species 1 (si,j = 1) or 2 (si,j = 2). All individuals which have at least one free nearest neighbor site can reproduce. To perform a reproduction step, we choose one of these fecund individuals with a probability proportional to its reproduction rate (see below) and a 1 On a square lattice it is impossible to enclose a cluster A within a cluster B, which only contains nearest neighbor sites of cluster A. On a hexagonal lattice this is possible. species 2species 1transverse direction ( )longitudinal direction ( )new individual of the same species is placed with equal probability on one of the free neighboring sites. 3 In contrast to the Eden model [62] and some of its two- species generalizations [28, 67], reproduction rates in our model depend on the states of the nearest-neighbor sites. Let n1 and n2 denote the number of nearest neighbors of species 1 and 2, respectively, then the reproduction rate of an individual at lattice site (i, j) is if si,j has no free neighbors, (1) 0 bi,j = 1 + n1R + n2S if si,j = 1, b0 2 + n1T + n2P if si,j = 2. b0 1 and b0 Here, b0 2 are the respective contributions to the reproduction rates of species 1 and 2, which are indepen- dent of the states of their nearest neighbors. Frequency- dependent selection is introduced through the parameters R, S, T, and P. production rate of the population is btot :=(cid:80) With the reproduction rates bi,j we implement a ran- dom sequential update of the system using a simplified version of the Gillespie algorithm [68]. The overall re- i,j bi,j and an individual at site (i, j) is selected to reproduce with probability bi,j/btot. We then choose one of the empty nearest-neighbor sites of the reproducing individual at random and place there a new individual of the same species. This implies that there are no mutations. Since the mean time until the next reproduction event is b−1 tot, we update time by t → t + b−1 tot after each reproduction event. We assume that individuals do not die and that they are immobile. Therefore, any site with si,j (cid:54)= 0 remains in its specific state indefinitely. As initial condi- tion we occupy all sites of an initial line randomly, but in equal parts, with species 1 and 2, if not stated otherwise. The formulated model generalizes version C of the Eden model, introduced by Jullien and Botet [69], to a two-species system. We already applied a similar model to range expansion without frequency-dependent selec- tion but included the possibility of mutations [67]. If R, S, T, and P are zero, our model reduces to that of Saito and Muller-Krumbhaar [28], however they used a square lattice. Since diffusion is not included in the model, con- figurations and patterns behind the front are frozen. This corresponds to observations in microbial experiments on range expansion [26, 60]. In game theory the parameters R, S, T, and P from Eq. (1) define the payoff matrix of a two-strategy game [14 -- 16]. Different scenarios, some well known in game theory, are implemented if we set these parameters accordingly. III. PHENOMENOLOGY We now describe some generic examples of our model for growing microbial colonies (see Fig. 2), which emerge for typical parameter settings, and discuss their charac- teristic features. We then concentrate on so-called so- FIG. 2. Growth patterns of our model for different param- eters, which correspond to typical settings. Time is always t = 15 and lattice size is L = 200. Cooperators are depicted in blue and defectors in red. If not stated otherwise b0 2 = 1, R = P = S = T = 0, and the initial ratio of both species is 1:1. (a) neutral growth, (b) selective advantage for species 2 (b0 2 = 1.5), which occupies 10% of initial sites, (c) coordina- tion game (R = P = 1), (d) snowdrift game (S = T = 1), and (e) public goods game (R = 0.1,T = 1.1). 2 = b0 cial dilemmas, where one species (defectors) exploits the other one (cooperators). Due to the spatial extent of our system, cooperators are able to outcompete defectors in a defined parameter region. This is in contrast to a single group, where all members interact with each other and, therefore, cooperators are always inferior to defectors. In the simplest case selection is frequency-independent, R = S = T = P = 0, and both species reproduce with the same rate, b1 0 [see Fig. 2(a)]. For this "neutral" setting we observe roughening of the front typical for the Eden model [62, 63]. Simultaneously, sectors composed of a single species merge and thereby coarsen [28, 66]. This inherent process happens according to the following sce- nario. If the tips of two advancing boundary lines meet, they annihilateand the enclosed sector loses contact to the front. Consequently, the number of boundaries and sectors can only decrease. Sectors repeatedly coarsen as they merge in these events. When all boundaries have 0 = b2 vanished, the front "has fixed" to a single species keeps on expanding. In finite systems, L < ∞, fixation to a single species always occurs since in our stochastic model there is always a finite rate at which boundaries annihi- late. Hence, two absorbing states exist. Eventually, the expanding front will fix either to species 1 or species 2. In Fig. 2(b), species 2 has a larger reproduction rate, b2 0 > b1 0, and therefore a constant selective advantage. As the front expands faster at locations where it is com- posed of species 2, the roughness of the front increases. Indentations of the front usually are caused by sectors of species 1, whereas species 2 creates bulges. Furthermore, boundaries are biased such that sectors of species 2 widen while sectors of species 1 shrink laterally. Hence, sectors of species 2 merge and coarsen quickly. Eventually, the expanding front will fix to species 2, which has almost happened in Fig. 2(b). If the reproduction rates depend on the state of near- est neighbors (frequency-dependent selection), new pat- terns arise. In this article we are mainly interested in frequency-dependent selection and therefore set b0 1 = b0 2 = 1 from here on. We now discuss some interesting cases, see Fig. 2(c) -- (d), which correspond to well-known settings in game theory [14 -- 16]. In coordination games (R > T and P > S), see Fig. 2(c), the front expands slower near boundaries than in the centers of sectors. Therefore, indentations in the front are typically found where boundaries currently are or recently have been. After sectors have coarsened for some time, most individuals are located inside sectors. Therefore, most of them only have neighbors of their own kind, which raises the average reproduction rate and the overall front advances faster. In snowdrift games (R < T and P < S), see Fig. 2(d), the front expands faster near boundaries. They annihi- late less frequent as compared to Fig. 2(c) since narrow sectors grow faster. Boundaries are also strongly twisted and associated with bulges of the front. In this article we are mostly interested in social dilem- mas where one species (called cooperators) raises the reproduction rate of all neighbors regardless of their species. The increased reproduction rate is called a pub- lic good in game theory, since it is of benefit to all nearby individuals, but it also costs resources. In contrast, the other species (called defectors) takes advantage of the public good for its own reproduction, but does not con- tribute to the reproduction of its neighbors in the same way. Defectors save resources for their own reproduction and therefore have an advantage. In this scenario de- fectors do not at all contribute to reproduction of their neighbors and, therefore, we set S = P = 0 and just vary R and T. According to Eq. (1), these two parame- ters increase the respective reproduction rates of coopera- tors and defectors if they have cooperating neighbors. In Fig. 2(e) we present a setting, where species 2 (defectors) rapidly takes over large parts of the front. Defectors ben- efit from the initially large number of boundaries, where they take advantage of nearby cooperators, and conquer 4 FIG. 3. Schematics of possible scenarios in an expanding public goods game. Depending on the parameters R, S, T, and P of Eq. (1), cooperators (C, blue) and defectors (D, red) advance with different speeds, as indicated by arrows. (a) Cooperators outrun the trailing defectors. From an advanced position along the front cooperators can then expand laterally and take over the front. (b) A thin layer of defectors keeps up with the cooperators' sector and, eventually, completely covers the cooperators. most of the front. Only cooperators, living in sufficiently large sectors, can keep up with the front during this early period and may then take over the front, depending on the parameter values. In a situation like this, it is not a priori clear if the front eventually fixes either to cooperators or to defec- tors. Depending on the values of R and T, cooperators can either outrun defectors and, from their advanced po- sition at the front, overgrow their competitors laterally, see Fig. 3(a). Or, defectors cover cooperators with a thin layer and thereby take over the front, see Fig. 3(b). Close to the transition between both scenarios, the front dis- plays increasing roughness since both species are able to take over while their fronts grow with different speeds. To characterize this transition quantitatively, we performed extensive simulations and applied methods from surface roughening [63, 70 -- 72] and the theory of phase transi- tions dealing with absorbing states [72, 73]. IV. EXPANDING PUBLIC GOODS GAME: CRITICAL BEHAVIOR In this section we quantitatively analyze the transi- tion between long-term cooperation and long-term defec- tion for an expanding public goods game. As the transi- tion is approached, several observables show critical scal- ing [72, 73]. Following our earlier work [67], we perform finite-size scaling to localize the transition. Furthermore, we determine critical exponents and thereby establish a new universality class for the transition between the two adsorbing states. In the vicinity of the transition we also study the dynamics of the sector boundaries including the decline of their mean number during coarsening and their superdiffusive motion as well as the roughening of the expanding front. CDD(b)CDD(a)5 FIG. 4. Probability Pfix of the front to fix to cooperators plotted versus T for several system sizes L at R = 0.1. Er- ror bars give the standard error of the mean for each data point. Lines are fits of the data points to the Fermi function, 1/ exp(c(T − T1/2) + 1), where T1/2 and c are fit parameters. Inset: The transition point T1/2(L) relative to the fitted crit- ical value Tc ≈ 1.58 (black crosses) follows a power law in L → ∞: Tc − T1/2(L) = (A/L)1/4.2 (black line). A. Finite-size scaling and phase diagram As boundaries merge, sectors coarsen and the system progresses towards one of the two absorbing states. At finite system sizes L this is a stochastic process and both adsorbing states are reached with a certain probability. However, in the thermodynamic limit, L → ∞, the mag- nitude of fluctuations relative to the mean value goes to zero and one of the absorbing states is reached with cer- tainty. We now use the method of finite-size scaling to determine the transition point between both states [73]. In Figure 4 we present the probability Pfix that the front fixes to cooperators and plot it versus T for several lattice sizes L at R = 0.1. We distinguish two regimes: one where cooperators dominate (Pfix(T , L) > 1 2 ) and one where defectors take over. We locate the transition 2 . As L → ∞, Pfix point at T1/2(L) by Pfix(T1/2(L), L) = 1 converges to a step function, since in infinite systems the absorbing states are reached with certainty. The step is positioned at Tc := limL→∞ T1/2(L). From the theory of critical scaling applied to absorbing states, we expect that close to the critical point Tc the states of the lattice sites are correlated on the transverse distance ξ⊥. Ap- proaching Tc, ξ⊥ diverges as ∆−ν⊥, where ∆ := Tc − T is the distance to the critical point and ν⊥ is a critical ex- ponent. For finite systems, an absorbing state is reached if L ≈ ξ⊥ ∼ ∆−ν⊥ . (2) The transition occurs at T1/2(L) and rearranging Eq. (2), FIG. 5. Phase diagram of the two-species public goods game with range expansion. Parameter regimes, where the expand- ing front fixes either to cooperators or defectors in large sys- tems with L → ∞, are indicated by blue and red shade, re- spectively. Red crosses are from simulations with system size L = 1000, where defectors always outcompeted cooperators (Pfix = 0) while blue dots identify events where cooperators survived (Pfix > 0). The black diamonds indicate critical points Tc for L → ∞ determined from finite-size scaling (see Eq. (3) and Fig. 4). Note that in finite systems defectors have an advantage in a larger parameter region. we obtain T1/2(L) ≈ Tc − (A/L)1/ν⊥ . (3) The characteristic length A is related to the microscopic length scale, which here is the lattice constant, and de- tails of our model. It is not important to the following analysis. The inset of Fig. 4 shows the best fit of our data to Eq. (3), which yields the critical exponent ν⊥ ≈ 4.2 and the critical point Tc ≈ 1.58 at R = 0.1. The above procedure can be repeated for different val- ues of R to map out the phase diagram (see Fig. 5). One realizes that the benefit of cooperators from their own species, R, has a much more pronounced influ- ence on the final state than the defectors' benefit from cooperating neighbors, T. This makes sense, since at large times t (cid:29) 1 the front contains large single-species sectors. Hence, the number of sector boundaries Nb, where defectors can benefit from cooperators, is small: Nb (cid:28) L. Therefore, almost all cooperators have coop- erating neighbors, while only a few defectors have this advantage. This is an example of "preferential assort- ment", where the benefit of cooperation is almost entirely available to other cooperators [15, 31, 32, 43 -- 45, 74, 75]. So, for a wide range of parameter combinations coopera- tors can indeed outcompete defectors. However, for large enough T the dynamics at the boundaries still determines the final state of the front and defectors outcompete co- operators. 00.20.40.60.8100.20.40.60.811.21.41.6PfixT500020001000500200100502010L-0.8-0.6-0.4-0.200.211.522.533.54log (Tc - T1/2(L))log Lt-0.2400.20.40.60.810246810TRDefectors winCooperators win6 FIG. 6. Mean time to fixation tfix as a function of defector benefit T for various system sizes L at R = 0.1. The inset depicts the rescaled fixation time, where ∆ = Tc − T. All data collapse on a single master curve for critical exponents ν(cid:107) = 3.5 ± 0.1 and ν⊥ = 4.2 ± 0.1 and critical point Tc = 1.59 ± 0.03. Accordingly, tfix grows with L and the position of its maximum approaches Tc for L → ∞. B. Critical exponents of the phase transition In the previous section IV A we already encountered the critical exponent ν⊥. We now continue to determine further critical exponents of the phase transition. These exponents are universal. They only depend on the dimen- sion of the system, the number of components of the order parameter, and symmetries of the model [72, 73]. They are independent of microscopic details and do not vary along a phase transition line. Our system has properties similar to "compact directed percolation" (CDP) [76]. This is a stochastic process with a flat front, which also has two distinct absorbing states. Using this similarity, we proceed by determining critical exponents, which are known for CDP [73], and compare both models. At the critical transition, T = Tc, none of the two species has an advantage. Heterogeneous fronts, com- posed of more than one sector, exist for long times be- fore the front fixes to one of the absorbing single-species states. This can be quantified by the mean time to fix- ation, tfix, presented in Fig. 6. The data show that the fixation time has a maximum, the position and value of which grow with system size. Along the longitudinal direction, in which the front propagates, states are correlated on the longitudinal dis- tance ξ(cid:107). As before, we expect it to scale like ∆−ν(cid:107) close to the transition. Since the front propagates with a mean velocity, ξ(cid:107) is proportional to a correlation time. Close to Tc this time becomes very long, which is known as critical slowing down. Substituting Eq. (2) into ξ(cid:107) ∼ ∆−ν(cid:107) , we FIG. 7. Survival probability PC of cooperators starting from a single site plotted versus time for several values of T. Other parameters are L = 1000 and R = 0.1. For T > Tc, PC decreases exponentially, while for T < Tc the front fixes to cooperators with a non-zero probability. At the transition point Tc, the survival probability decays in time with a power law with exponent β(cid:48)/ν(cid:107) = 2.2 ± 0.1. Inset: The number of cooperator sites at the front, NC , decays exponentially in time for T > Tc and is non-monotonic for T < Tc. At the transition NC decreases with a power law with critical exponent Θ = 1.3 ± 0.1. find the scaling relation ξ(cid:107) ∼ Lν(cid:107)/ν⊥ . (4) We expect the mean fixation time to be proportional to the correlation time ∼ ξ(cid:107). Therefore, in the inset of Fig. 6 we plot tfix rescaled by Lν(cid:107)/ν⊥ versus ∆ rescaled by L−1/ν⊥ . All curves of the main plot collapse on a single master curve for Tc = 1.59 ± 0.03, and critical exponents ν⊥ = 4.2 ± 0.1 and ν(cid:107) = 3.5 ± 0.1. The values of Tc and ν⊥ are in good agreement with our fit to Eq. (3). So, fixation of the front is determined by the characteristic time τfix ∼ ξ(cid:107) ∼ L0.83±0.05 . (5) Two more critical exponents right at the transition are related to the survival probability of one species or state, which initially occupies a single site while all the other sites are occupied by the other state. We choose a single cooperator site in a line of defectors and determine the probability PC(t) that after time t there are still cooper- ators at the front and also calculate the average number of cooperator sites at the front, NC(t). In Fig. 7 we plot both quantities versus time for different defector bene- fit T . At the transition situated between T = 1.55 and 1.6, we find that both PC and NC (see inset) decay with power laws in time: PC(t) ∼ t−β(cid:48)/ν(cid:107) and NC(t) ∼ t−Θ. The respective best fits yield β(cid:48)/ν(cid:107) = 2.2 ± 0.1 and Θ = 1.3 ± 0.1. Using our result for ν(cid:107), we find β(cid:48) = 7.7 ± 0.6. 140012001000800600400200000.20.40.60.811.21.41.6Ttfix500020001000500200L00.20.40.60.81024680-1-2-3-4-5-6-7-10123log PClog tt-2.2T=2.3T=1.9T=1.7T=1.6T=1.55T=1.5T=1.3T=0.9R=0.1t-1.3log NClog t10-1-2-3-4-1012TABLE I. Critical exponents for the phase transition to the absorbing states (either long-term global defection or long- term global cooperation) for the expanding public goods game with frequency dependent selection. ν⊥ 4.2 ± 0.1 ν(cid:107) 3.5 ± 0.1 β 0 β(cid:48) 7.7 ± 0.6 Θ 1.3 ± 0.1 In general, in phase transitions to absorbing states the critical exponent β governs the stationary density of "ac- tive sites", when approaching the transition [73]. In our case, the "active" sites can either be cooperators or defec- tors. Since the stationary state is either an all cooperator or an all defector front, the density ∼ Tc − T β jumps from 0 to 1 and, hence, β is 0. Our results for all the critical exponents are summarized in Table I. The set of critical exponents determines the univer- sality class of a phase transition. To our knowledge no other non-equilibrium transition has been found, which shares the same set of exponents. Hence, the transition between long-term cooperation and long-term defection in our expanding public goods game with frequency de- pendent selection constitutes a new universality class. C. Dynamics of boundaries In this section we investigate the dynamics of the boundaries which separates sectors of cooperators and defectors from each other. In rough fronts the local front orientation is tilted against the main growth direction. When the front grows further, this tilt directs the move- ment of boundaries [26, 28]. Ultimately, when two of them meet, they annihilate. In Fig. 8 we plot their mean number Nb versus time for several values of the defector benefit T . Right at the transition (dashed black line), Nb shows a power law decay. We now discuss the different regimes in Fig. 8. For neutral systems, where frequency-dependent selec- tion is absent (T = R = 0), any inclination of the front is created by stochastic surface or Eden roughening [63, 70 -- 72]. The surface undulations obey KPZ-scaling [77] and thereby drive the decay of Nb [26, 28]. Boundaries move superdiffusively along the front with a mean-square dis- placement proportional to t4/3 [27]. On average, they annihilate after having traveled the mean distance L/Nb between the boundaries, for which they need the time ∼ (L/Nb)3/2. Hence, boundaries annihilate with a rate proportional to N 3/2 b , which implies Nb ∼ −N 3/2 b Nb . So, the number of boundaries decreases as Nb(t) ∼ t−2/3 , as already observed by Saito and Muller-Krumbhaar [28]. This power law is excellently reproduced by our simu- (6) (7) 7 FIG. 8. Mean number of sector boundaries Nb plotted as a function of time for several values of T. Other parameters are L = 1000 and R = 0.1. Solid black line: Nb(t) ∼ t−2/3, as predicted in Ref. [28] for neutral growth (T = R = 0). At large times the simulated curve deviates from the power law due to finite system size. Dashed black line: Close to the critical point T = Tc ≈ 1.6, we find a power-law decay Nb(t) ∼ t−2.5. lations in the case of neutral growth, R = T = 0, as illustrated by the solid black line in Fig. 8. For R (cid:54)= 0 and T (cid:54)= 0, species reproduce with different rates. Hence, the fronts of two neighboring sectors (occu- pied by different species) advance with different speeds. This influences the tilt of the front orientation, in ad- dition to stochastic roughening in neutral systems, and thereby the movement of the separating boundary. Thus, we do not expect Eq. (6) to be valid. Indeed, Fig. 8 reveals different regimes for the mean number of boundaries Nb. For T > Tc, Nb decays expo- nentially in time in line with the exponential decay of the survival probability PC in Fig. 7 and similar to the case of selective advantage in Ref. [28]. For T < Tc boundaries annihilate less frequently. Narrow defector sectors persist in the front dominated by cooperators since almost all in- dividuals in the defector sectors have cooperating neigh- bors. This results in the upward curvature of the curves in Fig. 8. However, the defector sectors cannot expand laterally and ultimately loose contact to the front due to random fluctuations, and Nb declines exponentially. At Tc the number of boundaries decreases with a power law Nb ∼ t−χ, where a new exponent χ ≈ 2.50±0.05 appears. This power law implies that the number of boundaries de- clines from the initial value Nb(t = 0) ∼ L to the order of 1 in the coarsening time τcoarse ∼ L1/χ ≈ L0.40±0.01 . (8) Comparing with Eq. (5) reveals 1/χ < ν(cid:107)/ν⊥. This suggests that for large systems fixing the front to one species takes much longer than coarsening to a few sec- -101234-8-20-4-6log tlog(Nb/Nb(t=0))t-2.5t-2/3R=T=0T=2T=1.7T=1.6T=1.55T=1.5T=1.4T=0.9R=0.18 FIG. 9. Time evolution of two boundaries with initial dis- tance L/2 in a system close to criticality (T = 1.6) and at R = 0.1. Standard deviation σ of the transverse distance plot- ted versus time t for several system sizes L. The boundaries move superdiffusively. Initially, σ ∼ tη with η = 0.71 ± 0.05, and consistent with Eden scaling (t2/3, bold black line). At later times a crossover to σ ∼ tη(cid:48) with η(cid:48) = 0.9 ± 0.1 occurs (dashed black line). FIG. 10. The width w of the front plotted versus time for several system sizes L close to the critical point at T = Tc ≈ 1.6. Inset: After rescaling time t by τ× ∼ L1/χ and width w by w× ∼ Lγ/χ, the curves collapse onto a single master curve. The solid black line indicates Eden roughening with w ∼ t1/3, whereas the dashed black line shows selective roughening with w ∼ t1.3 for t (cid:29) τ×. Note that the saturation of w at large times is due to the finite system size. tors. Hence, the few remaining boundaries move differ- ently compared to early times since they have to annihi- late to fix the front to a single species. To check if this is the case, we employ the initial con- dition, where the front is composed of only two sectors of size L/2 each, separated by two boundaries. We quantify the boundaries' random motion by monitoring the tem- poral evolution of the standard deviation for the trans- verse distance (cid:96)⊥, σ(t) :=(cid:112)(cid:104)[(cid:96)⊥(t) − (cid:104)(cid:96)⊥(t)(cid:105)]2(cid:105) . We subtract the mean distance (cid:104)(cid:96)⊥(t)(cid:105) to take care of any transient drift, when the front relaxes from its initially flat to the rough shape, and an expected small drift if T is not exactly Tc. From Fig. 9 we see that, for early times, σ grows like a power law, σ ∼ tη, with η = 0.71 ± 0.05. This is consistent with meandering boundaries induced by Eden roughening, σ ∼ t2/3 [26 -- 28]. We expect such a behavior since the roughness of the front has not fully developed yet. For large systems and later times we find a crossover to σ ∼ tη(cid:48) with η(cid:48) = 0.9 ± 0.1. This confirms our ear- lier statement that at late times the few boundaries re- maining after coarsening move differently. Indeed, they show an even stronger superdiffusive motion than Eden scaling, which is associated with the long-lived and pro- nounced surface undulations in systems with frequency- dependent selection. (9) Here, h(i, t) is the longitudinal position of the front at its transverse site i and ¯h(t) is the mean position L(cid:88) i=1 ¯h(t) := 1 L h(i, t) . (11) D. Surface roughening of the expanding front We now discuss the surface roughness or undulations of the expanding front and compare our results to the clas- sical Eden model. We measure the roughness of a front, which has not yet fixed to one species, by calculating its width w for system size L and at time t: (cid:18) 1 L(cid:88) (cid:2)h(i, t) − ¯h(t)(cid:3)2(cid:19)1/2 . (10) w(L, t) := L i=1 Figure 10 plots the width w(L, t) close to the critical point at T = Tc ≈ 1.6. Initially, the roughness of the front grows like in the original Eden model [62]: w ∼ tγ, where the growth exponent γ = 1/3 belongs to the KPZ- universality class [63]. At intermediate times a new regime sets in, where w ∼ tγ(cid:48) increases with enhanced growth exponent γ(cid:48) ≈ 1.3±0.1. This marks the transition from Eden roughening to "selective roughening". Here, the typical shape of the front is determined by advancing cooperator sectors and trailing defector sectors and ulti- mately drives the accelerated increase in the front's width w. The crossover to this regime happens at time τ×, which increases with system size L as Fig. 10 shows. This 0123400.511.522.53log tlog σt2/3t0.910000500020001000500200L00.511.5200.511.522.5log(t)log(w)t1/350002000100050020010050Lt1/3t1.3log(w/L/)-0.500.511.52-1-0.500.511.52log(t/L1/)makes sense since we expect selective roughening to dom- inate over Eden roughening when the lateral extension of sectors is comparable to L, i.e., τ× ∼ τcoarse ∼ L1/χ. Due to Eden roughening the width at the crossover is w× ∼ τ γ× ∼ Lγ/χ ≈ L0.13. Indeed, rescaling width and time with w× and τ×, respectively, collapses all data in Fig. 10 onto a single master curve, as the inset demon- strates. To conclude, surface roughening close to critical- ity occurs in two regimes. Until crossover time τ×, one observes Eden roughening, whereas for times larger τ× selective roughening occurs until the front fixes to one species. The dynamics of the width of the front is sum- marized by L(γ−γ(cid:48))/χ t (cid:28) τ× ∼ L1/χ , t (cid:29) τ× . (12) tγ tγ(cid:48) (cid:40) w(L, t) ∼ V. SUMMARY AND CONCLUSION In this work we studied a generalized Eden model, where two species compete with each other at the rough expanding front. Individuals of the two species influence each other by frequency-dependent selection, which acts between nearest neighbors. We analyzed the evolution- ary dynamics at the expanding front, where single-species sectors form and coarsen. Ultimately, the front fixes to one species, which we identify with an absorbing state of our model. In its general form the model can implement several scenarios including selective advantage, and also well- known game theoretical settings like the snow drift game or the coordination game. Each of them creates distinct patterns, which should be analyzed in detail in future work. For the prominent example of a public goods game, we find that cooperators prevail in a wide param- eter regime, as expected for a spatial version of a social dilemma [15, 43 -- 45]. For other parameter values defec- tors take over the front, as usual. We identify the transition between long-term coopera- tion and long-term defection as a nonequilibrium critical phase transition between two absorbing states. The set of critical exponents (see Table I), which we determined by analyzing critical and finite size scaling, shows that the phase transition belongs to a new universality class. We attribute this result to the fact that the front in our model is rough and not flat as in usual absorbing states. Close to the critical transition the front's roughness ex- hibits a crossover in time from slow Eden roughening to fast selective roughening. Strong roughening has also been observed at phase transitions in a related model by Lavrentovich and Nelson [78]. Interestingly, the critical exponents determined in our present work violate the so-called generalized hyperscal- ing relation [79] dν⊥ = ν(cid:107)Θ + β + β(cid:48) , (13) 9 where the number of transverse dimensions d is 1. This relation holds for most universality classes of phase tran- sitions to absorbing states [73]. It is not obvious why our model does not obey the scaling relation. The roughness of the front correlates with superdiffu- sive motion of the boundaries separating sectors. Two factors contribute to the movement of the boundaries on long length scales. On the one hand, the direct competi- tion between the species on either side of the boundaries pushes them towards the sector composed of the more slowly reproducing species. On the other hand, bound- aries follow the local tilt of the front. In the public goods game cooperator sectors are advanced, while defector sec- tors lag behind. Near the critical transition, defectors outcompete their direct cooperating neighbors but the front is tilted towards sectors filled by defectors, so the two factors move the boundaries in opposite directions. At the phase transition both effects cancel and the front fixes with equal probability to either species. The strong roughening correlates with superdiffusive motion of the boundaries with nearly ballistic scaling. Accordingly, whether a species takes over the expand- ing front is determined by two contributions: its repro- duction rate relative to its competitor and its position relative to the average front position. The influence of different reproduction rates of neighboring species can directly be compared and is summarized in the phrase "survival of the fittest". The position at the front deter- mines the available space for progeny, which then have the opportunity to expand sidewards. This is illustrated by the phrase "survival of the fastest" [38]. In our model the number of sectors only decreases. It does not include experiments with mutually beneficial interactions between different species, which do not gen- erate sectors [37, 80, 81]. In future extensions of our model this may be remedied by including motility of in- dividuals [82, 83], by allowing reproduction to more dis- tant lattice sites [78], or by increasing the maximal num- ber of individuals per lattice site from one. Moreover, it is worthwhile to consider interactions ranging beyond nearest neighbors, since biomolecules, released by indi- vidual microorganisms, may diffuse in the extracellular medium [84 -- 86]. In the public goods game scenario this would stabilize narrow sectors of defectors so that they do not lose contact to the front. In general, range expansion of multiple species will de- velop enhanced roughness at the growing front. As we demonstrated here, the corresponding models have new and interesting statistical properties. From a biological point of view, roughness is important. It affects the ter- ritories that different species occupy and thereby their evolutionary success through the strong random motion of sector boundaries. This may also be relevant for range expansion in a real environment and not just in a test tube. To better understand the properties and conse- quences of rough expanding fronts, further theoretical work is needed. At the same time further experiments should look for the fingerprint of roughness in microbial colony growth. ACKNOWLEDGMENTS We thank the research training group GRK 1558 funded by Deutsche Forschungsgemeinschaft for finan- cial support. We further thank Erwin Frey, Maria Eckl, Florian Gartner, and Raphaela Gessele for discussion and collaboration on a related model. 10 [1] J. D. Murray, Mathematical Biology I. An Introduction, [31] A. S. Griffin, S. A. West, and A. Buckling, Nature 430, 3rd ed., Vol. 1 (Springer, 2007). 1024 (2004). [2] R. W. Griffiths, D. W. Schloesser, J. H. Leach, and W. P. Kovalak, Can. J. Fish. Aquat. Sci. 48, 1381 (1991). [3] S. R. Loarie, P. B. Duffy, H. Hamilton, G. P. Asner, C. B. Field, and D. D. Ackerly, Nature 462, 1052 (2009). [32] J. U. Kreft, Biofilms 1, 265 (2004). [33] S. A. West, S. P. Diggle, A. Buckling, A. Gardner, and A. S. Griffin, Annu. Rev. Ecol. Evol. Syst. 38, 53 (2007). [34] M. E. Hibbing, C. Fuqua, M. R. Parsek, and S. B. Pe- [4] I. C. Chen, J. K. Hill, R. Ohlemuller, D. B. Roy, and terson, Nat. Rev. Micro. 8, 15 (2010). C. D. Thomas, Science 333, 1024 (2011). [35] S. Mitri and K. Richard Foster, Annu. Rev. Genet. 47, [5] E. Peacock et al., PLoS ONE 10, e112021 (2015). [6] C. Stringer, Nature 423, 692 (2003). [7] H. Liu, F. Prugnolle, A. Manica, and F. Balloux, Am. 247 (2013). [36] J. Gore, H. Youk, and A. van Oudenaarden, Nature 459, 253 (2009). J. Hum. Genet. 79, 230 (2006). [37] M. J. Muller, B. I. Neugeboren, D. R. Nelson, and A. W. [8] C. Moreau, C. Bherer, H. Vezina, M. Jomphe, D. Labuda, Murray, Proc. Natl. Acad. Sci. USA 111, 1037 (2014). and L. Excoffier, Science 334, 1148 (2011). [38] J. D. Van Dyken, M. J. Muller, K. M. Mack, and M. M. [9] A. Br´u, S. Albertos, J. Luis Subiza, J. L. Garc´ıa-Asenjo, Desai, Curr Biol 23, 919 (2013). and I. Br´u, Biophys. J. 85, 2948 (2003). [10] J. B. Xavier, Mol Syst Biol 7, 483 (2011). [11] L. Hall-Stoodley, J. W. Costerton, and P. Stoodley, Nat. Rev. Micro. 2, 95 (2004). [39] C. Darwin, On the Origin of Species by Means of Natural Selection, or the Preservation of Favoured Races in the Struggle for Life (John Murray, 1859). [40] R. Axelrod, The Evolution of Cooperation, Revised Edi- [12] C. D. Nadell, J. B. Xavier, and K. R. Foster, FEMS tion (Basic Books, 2009). Microbiol. Rev. 33, 206 (2009). [13] S. Mitri, J. B. Xavier, and K. R. Foster, Proc. Natl. [41] P. B. Rainey and K. Rainey, Nature 425, 72 (2003). [42] J. S. Chuang, O. Rivoire, and S. Leibler, Science 323, Acad. Sci. USA 108, 10839 (2011). 272 (2009). [14] J. Hofbauer and K. Sigmund, Evolutionary Games and Population Dynamics, 1st ed. (Cambridge University Press, 1998). [43] M. A. Nowak, S. Bonhoeffer, and R. M. May, Proc. Natl. Acad. Sci. USA 91, 4877 (1994). [44] H. Ohtsuki, C. Hauert, E. Lieberman, and M. A. Nowak, [15] G. Szab´o and G. F´ath, Phys. Rep. 446, 97 (2007). [16] E. Frey, Physica A 389, 4265 (2009). [17] A. Buckling, R. C. Maclean, M. A. Brockhurst, Nature 441, 502 (2006). [45] F. Fu, M. A. Nowak, and C. Hauert, J. Theor. Biol. 266, and 358 (2010). N. Colegrave, Nature 457, 824 (2009). [46] A. Melbinger, J. Cremer, and E. Frey, Phys. Rev. Lett. [18] J. A. Shapiro, BioEssays 17, 597 (1995). [19] E. B. Jacob, Y. Aharonov, and Y. Shapira, Biofilms 1, 239 (1999). 105, 178101 (2010). [47] J. Cremer, A. Melbinger, and E. Frey, Sci. Rep. 2 (2012). [48] J. B. Xavier and K. R. Foster, Proc. Natl. Acad. Sci. USA [20] I. Golding, I. Cohen, and E. Ben-Jacob, Europhys. Lett. 104, 876 (2007). 48, 587 (1999). [21] M. Matsushita, J. Wakita, H. Itoh, K. Watanabe, T. Arai, T. Matsuyama, H. Sakaguchi, and M. Mimura, Physica A 274, 190 (1999). [22] T. Matsuyama and M. Matsushita, Forma 16, 307 (2001). [23] T. Vicsek, M. Cserzo, and V. Horvath, Physica A 167, 315 (1990). [49] C. D. Nadell, K. R. Foster, and J. B. Xavier, PLoS Comput. Biol. 6, e1000716 (2010). [50] D. Greig and M. Travisano, Proc. Biol. Sci. 271 (Suppl 3), S25 (2004). [51] J. H Koschwanez, K. R Foster, and A. W Murray, PLoS Biol. 9, e1001122 (2011). [52] M. Sen Datta, K. S. Korolev, I. Cvijovic, C. Dudley, and [24] J.-i. Wakita, H. Itoh, T. Matsuyama, and M. Matsushita, J. Gore, Proc. Natl. Acad. Sci. USA 110, 7354 (2013). J. Phys. Soc. Jpn. 66, 67 (1997). [53] F. Pattus and M. A. Abdallah, J. Chin. Chem. Soc. 47, [25] M. Huergo, M. Pasquale, A. Bolz´an, A. Arvia, and 1 (2000). P. Gonz´alez, Phys. Rev. E 82, 031903 (2010). [54] C. Ratledge and L. G. Dover, Annu. Rev. Microbiol. 54, [26] O. Hallatschek, P. Hersen, S. Ramanathan, and D. R. 881 (2000). Nelson, Proc. Natl. Acad. Sci. USA 104, 19926 (2007). [55] C. D. Nadell and B. L. Bassler, Proc. Natl. Acad. Sci. [27] B. Derrida and R. Dickman, J. Phys. A 24, L191 (1991). [28] Y. Saito and H. Muller-Krumbhaar, Phys. Rev. Lett. 74, 4325 (1995). [29] B. J. Crespi, Trends Ecol. Evol. 16, 178 (2001). [30] G. J. Velicer, Trends Microbiol. 11, 330 (2002). USA 108, 14181 (2011). [56] J. van Gestel, F. J. Weissing, O. P. Kuipers, and A. T. Kov´acs, ISME J. 8, 2069 (2014). [57] M. A. Riley and D. M. Gordon, Trends Microbiol. 7, 129 (1999). 11 [58] D. M. Cornforth and K. R. Foster, Nat. Rev. Micro. 11, 285 (2013). [59] V. T. Lee and O. Schneewind, Genes Dev. 15, 1725 (2001). [60] M. F. Weber, G. Poxleitner, E. Hebisch, E. Frey, and M. Opitz, J. R. Soc. Interface 11, 20140172 (2014). equilibrium phase transitions: Absorbing phase transi- tions, 1st ed., Vol. 1 (Springer, 2008). [74] W. D. Hamilton, J. Theor. Biol. 7, 1 (1964). [75] W. D. Hamilton, J. Theor. Biol. 7, 17 (1964). [76] J. W. Essam, J. Phys. A 22, 4927 (1989). [77] M. Kardar, G. Parisi, and Y.-C. Zhang, Phys. Rev. Lett. [61] J. B. Xavier, W. Kim, and K. R. Foster, Mol. Microbiol. 56, 889 (1986). 79, 166 (2011). [78] M. O. Lavrentovich and D. R. Nelson, Phys. Rev. Lett. [62] M. Eden, Proc of the Fourth Berkeley Symposium on 112, 138102 (2014). Mathematical Statistics and Probability 4, 223 (1960). [79] J. F. F. Mendes, R. Dickman, M. Henkel, and M. C. [63] A.-L. Barab´asi and H. E. Stanley, Fractal concepts in sur- face growth, 1st ed. (Cambridge University Press, 1995). [64] M. Plischke and Z. R´acz, Phys. Rev. A 32, 3825 (1985). [65] R. Jullien and R. Botet, J. Phys. A 18, 2279 (1985). [66] A. Ali and S. Grosskinsky, Adv. Complex Syst. 13, 349 Marques, J. Phys. A 27, 3019 (1994). [80] B. Momeni, K. A. Brileya, M. W. Fields, and W. Shou, eLife 2, e00230 (2013). [81] A. T. Kov´acs, Front. Microbiol. 5 (2014). [82] T. Reichenbach, M. Mobilia, and E. Frey, Nature 448, (2010). 1046 (2007). [67] J.-T. Kuhr, M. Leisner, and E. Frey, New J. Phys. 13, [83] A. Gelimson, J. Cremer, and E. Frey, Phys. Rev. E 87, 113013 (2011). (2013). [68] D. T. Gillespie, J. Comput. Phys. 22, 403 (1976). [69] R. Jullien and R. Botet, Phys. Rev. Lett. 54, 2055 (1985). [70] J. Krug and H. Spohn, in Solids far from Equilibrium, edited by C. Godr`eche (Cambridge University Press, 1992). [84] T. Julou, T. Mora, L. Guillon, V. Croquette, I. J. Schalk, D. Bensimon, and N. Desprat, Proc. Natl. Acad. Sci. USA 110, 12577 (2013). [85] B. Allen, J. Gore, M. A. Nowak, and C. T. Bergstrom, eLife 2, e01169 (2013). [71] T. Halpin-Healy and Y.-C. Zhang, Phys. Rep. 254, 215 [86] R. Menon and K. S. Korolev, Phys. Rev. Lett. 114, (1995). [72] G. ´Odor, Rev. Mod. Phys. 76, 663 (2004). [73] M. Henkel, H. Hinrichsen, and S. Lubeck, Non- 168102 (2015).
1707.00492	2	1707	2018-03-22T09:52:44	Transition and formation of the torque pattern of undulatory locomotion in resistive force dominated media	[ "physics.bio-ph" ]	In undulatory locomotion, torques along the body are required to overcome external forces from the environment and bend the body. Using a resistive force theory model, we find that the torque from external resistive forces has a traveling wave pattern with decreasing speed as the wave number (the number of wavelengths on the locomotor's body) increases from 0.5 to 1.8, and the torque transitions into a two-wave-like pattern and complex patterns as the wave number increases to values of 2 and greater values. Using phasor diagrams analysis, we reveal the formation and transitions of the pattern as the consequences of the integration and cancellation of force phasors.	physics.bio-ph	physics	Transition and formation of the torque pattern of undulatory locomotion in resistive force dominated media Tingyu Ming and Yang Ding∗ Beijing Computational Science Research Center Haidian District, Beijing 100193, China (Dated: March 28, 2018) In undulatory locomotion, torques along the body are required to overcome external forces from the environment and bend the body. These torques are usually generated by muscles in animals and closely related to muscle activations. In previous studies, researchers observed a single traveling wave pattern of the torque or muscle activation, but the formation of the torque pattern is still not well understood. To elucidate the formation of the torque pattern required by external resis- tive forces and the transition as kinematic parameters vary, we use simplistic resistive force theory models of self-propelled, steady undulatory locomotors and examine the spatio-temporal variation of the internal torque. We find that the internal torque has a traveling wave pattern with a de- creasing speed normalized by the curvature speed as the wave number (the number of wavelengths on the locomotor's body) increases from 0.5 to 1.8. As the wave number increases to 2 and greater values, the torque transitions into a two-wave-like pattern and complex patterns. Using phasor diagram analysis, we reveal that the formation and transitions of the pattern are consequences of the integration and cancellation of force phasors. INTRODUCTION TABLE I. Wave numbers observed in nature Organism Spermatozoon Nematode Snake Eel Scup Sandfish Undulatory locomotion is a common way for animals to move in various environments (e.g. spermatozoa in water [1], sandfish in sand [2], snakes on land [3, 4], and fish in water [5], for reviews, see [6–8] ) and a popular mode of locomotion for bio-inspired robots [9–11]. This type of lo- comotion consists of bending the body or some portion of the body to form a traveling wave in the direction oppo- site to the motion direction to generate propulsion. For organisms in environments dominated by resistive forces, such as spermatozoa swimming at low Reynolds num- bers (Re) and snakes slithering on the ground, how the propulsive forces from the environments are generated is quite well understood [7, 12]. To bend the body and generate propulsion, internal torques (bending moments) are required to overcome both the restoring forces and damping forces of the body and the external forces from the resistance of surrounding media. For macroscopic animals such as eels and snakes, the internal torques are generated by muscle forces act- ing on the body. Therefore, in previous theoretical and computational studies, spatio-temporal torque patterns were used to explain and predict muscle activation pat- terns [13–15]. Ongoing interdisciplinary research over the past several decades has provided a general overview of the torque and muscle activation: they both exhibit trav- eling wave patterns from head to tail. However, the waves of the muscle activation and the torque travel faster than the wave of the curvature, which is a phenomenon known as neuromechanical phase lags [5, 16]. Consequently, muscles activate after they begin to shorten in the an- terior part of the body, and muscles begin to activate before they begin to shorten in the posterior part of the body. Wave number a Source 1.25-1.4 [1, 18] 0.55-1.31 1.60 (on ground), 3.5 (in sand) 1.7 0.65 [19, 20] [5] [3, 21] [5] 1.0 [2] a When only amplitude and wavelength are given in the reference, we assume that the motion is sinusoidal and approximated the wave number as ξ = L/ R λ the undulation amplitude, and λ is the wavelength. 0 p1 + B2 sin2 xdx, where L is the body length, B is By imposing kinematics and considering contributions from the resistance of the environment and the passive body properties of fish (e.g. saithe and lamprey), qualita- tive agreements between predicted torque patterns and muscle activation patterns have been achieved [13–15]. For the relatively simple case of the sandfish lizard swim- ming in sand, where resistive forces dominate and the body is nearly uniform, a quantitative agreement has been obtained using resistive force theory (RFT) [17]. However, how the torque pattern is formed and whether the pattern is always a traveling wave are still open ques- tions. Another approach for studying the mechanics of undulatory locomotion is to start with the internal forces/torques and observe the kinematics as a result of the couplings between the internal drives, passive body properties and external environments. Two closely re- lated kinematic parameters are the wavelength and the wave number (the number of wavelengths on the locomo- tor's body), which, in reality, vary in different species and for the same species in different environments (Tab. I). For example, the wavelength decreases as the viscosity increases in nematodes and spermatozoa [19, 22]. By im- posing a neuron activation pattern, muscle forces, or a relationship between internal shear force and curvature, previous studies showed the trend of decreasing wave- length in spermatozoa and fish swimming when the rel- ative strength of the external resistance to the internal driving forces/torques is reduced [23–25]. However, how the variations in internal torque, kinematics and other components interact with each other is still not well un- derstood. In robots using undulatory gaits, torques are generally generated directly by motors (e.g., [26]), although new actuation mechanisms are emerging [27–29]. Torque is also a convenient way for detecting unexpected forces and avoiding damage to robots [30]. A deep understanding of the features of torques such as their magnitudes, power output, and phase relationships with curvature in various configurations and environments is useful for designing driving systems [31, 32]. Here, we consider steady forward undulatory loco- motion in resistive-forces-dominated media with simple kinematics and body shape. We show the basic torque pattern and its transitions to new patterns as the wave number increases. Further analysis reveals the formation of the torque pattern and the underlying mechanism of the transitions. MODEL We consider an undulatory locomotor bending its slim and uniform body as a traveling serpentine wave in a plane (Fig. 1). We use body length as unit length and one undulation period Tp as unit time. The curvature is prescribed as κ = Aξ sin[2π(ξs + t)], where s ∈ [0 1] is the arc length measured from the tail, A controls the undulation amplitude relative to the wavelength, ξ is the wave number, and t is the time. The speed of the cur- vature wave becomes vκ = 1/ξ. A is set to 7.54, which gives an amplitude-to-wavelength ratio (≈0.24) that is close to experimentally observed ratios [1–3]. For every time instant, we use a body frame in which the tail end is at the origin and pointing toward the x− axis. The tangent angle of a segment at s to the x+ axis can be computed by integrating the curvature along the body: θ(s) = R s 0 κdl. The position of the segment can be com- puted as r(s) = (x, y) = (R s 0 sin(θ)dl). By taking the time derivative, the velocity vb of a segment relative to the tail end can be computed. Assuming that the tail end is moving at velocity vtail and rotating at angular velocity ω, the velocity at the body position s in the lab frame becomes v = vb + vtail + ωez × r. To determine the motion of the body and the distri- bution of the force on the body, we use an RFT model similar to those in [3, 17]. In the RFT model, the body is divided into infinitesimal segments. Assuming that the force (F(s)) experienced by one segment is inde- pendent of other segments, the force can be calculated based on the geometry, orientation, and velocity of the 0 cos(θ)dl, R s 2 locomotion direction t=0 v v θ F F v F y x Τ (+) FIG. 1. Diagram of the model. The black curve represents the body, the magenta arrows represent velocities, and the green arrows represent forces from the medium. The black dot is a representative point on the body at which the internal torque (T ) is calculated from the forces in the dashed box. The sign and arrow indicate the direction of the torque. ξ = 1.5, A = 7.54, and t = 0. segment. The perpendicular and parallel components of the force on a segment can be written as F⊥(v⊥, vk) and Fk(v⊥, vk), respectively, where v⊥ and vk are the perpen- dicular and parallel components of the segment velocity v. We first consider the simplest case, in which the head drag is negligible and the forces are from viscous drag: F⊥ = C⊥v⊥ and Fk = Ckvk, where C⊥ = 2 and Ck = 1 are the drag coefficients for a very thin cylinder [33]. The total external torque on the body can be computed as Ftotal = R 1 r(l) × F(l)dl. We assume that inertia is negligible, which is a good approx- imation for micro-swimmers in fluids, crawlers on land, and swimmers in granular materials [1–3]. Under this assumption, the resultant net force Ftotal and the net torque Ttotal related to the tail (reference) frame are both zero, from which vtail and ω can be determined. The motion and force distributions on the body are shown in Supplementary Video S1, and the MATLAB scripts of the computation are provided in the Supplementary Materials. F(l)dl and Ttotal = ez R 1 0 0 To compute the internal torque (T ) required to over- come the external forces at a point on the body, we ana- lyze the torque balance on the anterior side of the body at that point and simply find that T (s) = −T e(s) = − R 1 s [r(l) − r(s)] × F(l)dl (see Fig. 1 for an example), where T e is the total external torque from the anterior side of the body. Integrating over the posterior side of the body gives the same results. To compute the wave speed of the torque, we use the fitting function √2hT (s)i sin[2π(s/λT +t)+φT ], where hT (s)i is the stan- dard deviation of the torque at s, λT is the wavelength of the torque wave along the body, and φT is a fitting parameter for the phase. The hT (s)i term is used to cap- ture the variation in torque amplitude and the prefactor √2 comes from the ratio between the maximum and the standard deviation of the sine function. The fitting pa- rameters λT and φT are obtained from the best fitting of the torque. The speed of the torque wave is defined as vT = λT /Tp = λT and the speed ratio of the torque wave to the curvature wave is vT /vκ = λT ξ. RESULTS To focus on the torque pattern, we normalize the torque by its maximum value for each wave number ξ. We find that the torque exhibits a traveling wave pattern for ξ < 1.8 (Fig. 2a-d). In this regime, the amplitude of the torque is smaller near the ends and greater in the middle. As in previous studies, the torque wave travels faster than the curvature wave, and different phase lags between the curvature and the torque along the body are observed. When ξ approaches 2, the magnitude of the torque in the middle of the body suddenly decreases, and a pat- tern of two apparently separated traveling waves forms (Fig. 2e); each wave is similar to a wave with ξ = 1 and takes half of the body. For ξ > 2, the two waves merge as the wave direction near the middle of the body be- comes the opposite direction of the curvature wave. The torque pattern is no longer one or two traveling waves (Fig. 2f). Up to at least ξ = 11, similar transitions occur by adding one additional traveling wave pattern near the middle of the body when ξ reaches integer numbers. See Supplementary Video S2 for torque patterns with smaller ξ increments. The speed of the torque wave normalized to the speed of the curvature decreases from 5.2 to 1.3 as the wave number increases from 0.5 to 1.8 (Fig. 3a). This result is consistent with the results of previous studies, namely, the muscle activation is nearly synchronized for short wavelengths [5]. The fit of a single traveling wave is poor when ξ > 1.8; therefore, the wave speed is not defined and shown in Fig. 3a. The energy output per cycle required to overcome the external force at each point on the body is computed by integrating the power over a cycle, i.e., W = R 1 0 T κdt. The decrease in the phase difference between T and κ and the decrease in the amplitude of T at the middle as ξ increases result in a more uniform distribution of the energy output over the body (see the blue line in 3b). When ξ = 2, the instantaneous power and energy output of the middle segment are zero. As ξ further increases to 2.3, the power of the middle segment becomes negative, which means that the energy generated by other parts of the body is transferred to this part. A few variation are tested to evaluate the influences of the external forces and kinematics on the torque pat- tern and transitions. First, two additional types of re- sistive force laws obtained in previous experiments are considered: forces for granular media are described by F⊥ = Cn sin[arctan(γ sin(φ))] and Fk = [Cf cos(φ) + Cl(1− sin(φ))], where Cn = 5.57, Cl = −1.74, Cf = 2.30, γ = 1.93, and φ = arctan(v⊥/vk) [34]. These force laws 3 are empirical fitting functions for an aluminum cylin- der dragged with different orientations in 3mm glass beads. For anisotropic frictional forces, F⊥ = µtv⊥/v and Fk = [µf H(vk) + µb(1 − H(vk))]vk/v, where µf = 0.3, µb = 1.3µf and µt = 1.8µf are the friction coeffi- cients in the forward, backward, and normal directions, respectively [3]. H(x) is the Heaviside step function. Since we focus on the torque pattern and normalize the torque by its maximum value, the absolute magnitudes of the forces are irrelevant here. These force laws and co- efficients are obtained by measuring the frictional forces while unconscious snakes slide on clothes at different ori- entations. The torque pattern is qualitatively the same when the force laws are replaced by those for granular and frictional environments, and only subtle differences are observed (Fig. 4a & b). We also compute the force dis- tribution using Lighthill's elongated body theory (EBT) where only the lateral inertial forces from the fluid are considered (the derivation is provided in the Supplemen- tary Information) [35]. The inertial forces considered in the EBT give a torque pattern that is similar to that from the RFT, albeit with a phase shift of +π/2 (1/4 period) (Fig. 4d). To study the effect of undulation amplitude, we in- crease the amplitude to A = 12.57, at which the segments nearly overlap. Surprisingly, we find that the torque pat- tern is insensitive to amplitude (see Fig. 4d and Supple- mentary Video S9). When the amplitude increases lin- early toward the tail, similar transitions occur, albeit at a greater wave number (Fig. 4e). Since interactions be- tween segments through the medium are neglected, the results from the amplitude variations only include the geometric effects. An example case with head drag is also examined (Fig. 4f). Based on an experiment on bull spermatozoa, the head is approximated as a sphere with isotropic drag, and the drag coefficient is chosen such that the head drag is 58% of the body drag when the body is straight and moving perpendicular to its axis [36], i.e. Fh = 0.58C⊥v(1). The resulting torque is significantly larger for the anterior half of the body and the transition of torque to the two-wave pattern occurs at a smaller wave number. To elucidate the mechanism underlying the transition of the torque, we analyze the phases of the torque at the middle point of the body and at a point infinitely close to the head as examples (Fig. 5). In a previous study [17], the torque at the middle point was roughly decomposed into three parts to explain the neuromechan- ical phase lag. Here we use a more quantitative tool– phasor diagram–to visualize and analyze the relation- ships among the phases of curvature, force and torque. Further simplification is needed prior to the analysis. Since the torque pattern is not sensitive to amplitude, a small amplitude (A = 0.6) is used such that the loco- motor is nearly a straight line on the x axis undulating in place. In this case, only the lateral displacement (y) 4 FIG. 2. The internal torque as a function of body position and time for different wave numbers. The magnitude of the torque is normalized by the maximum torque and represented by color. The solid and dashed lines indicate the maximum and minimum curvatures, respectively. The red dotted box in (e) indicates the torque pattern that appears similar to the pattern in (b). The green and blue arrows indicate the local waves traveling posteriorly and anteriorly, respectively. A = 7.54 (a) 5 v T v κ 3 1 0.5 1 (b) W 0 ξ = 0.5 ξ = 1.0 ξ = 1.7 ξ = 1.8 ξ = 2.0 ξ = 2.3 1 1.5 ξ 2 0 Tail <−− s −−> Head 1 FIG. 3. The speed of the internal torque wave normalized by the speed of the curvature wave as a function of the wave number (a) and the energy per cycle required as a function of the body position for different wave numbers (b). and lateral forces (Fy ≈ F⊥) need to be considered and the longitudinal forces (≈ Fk) are negligible. Then the equation for computing the torque at position s can be simplified as T (s) = − R 1 s (l − s)Fydl. Prior to the analy- sis, we also note that the spatio-temporal patterns of the lateral force are affected by the requirements of force and torque balances. For ξ = 0.5, a wave number less than 1, the phase difference between the lateral forces at the head and at the middle point is greater than π/2, which is the phase difference of the curvatures at these two points (Fig. 5). This result can be understood by considering the balance of the lateral forces and the torque: zero to- tal lateral force requires both negative forces and positive forces to be present at any time; the zero torque condition further requires that the negative forces be distributed on both sides when the force in the middle is positive (a sim- ilar argument was made by Gray [37]). Nonetheless, the phase differences of the forces between the middle point and the end points increase with increasing wave number. In phasor diagrams, a variable is represented by a pha- sor (vector), whose projection on the horizontal axis is the instantaneous value of the variable and the rota- tion of the phasor corresponds to the time evolution. Since T and T e only have a sign difference, we exam- ine the phasors of the external torque T e and first focus on the phasor of T e at the middle point of the body (T e mid = T e(s = 0.5) in Fig. 5b). The force contribution to T e at the middle point (i.e. (l − 0.5)Fy for l > 0.5) is discretized and visualized using phasors in Figure 5b. The integrative nature of the torque at the middle point makes the phase of T e between the phases of the forces. Interestingly, the torque T e at the middle point is pre- cisely either out of phase (for ξ < 2) or in phase (for 2 < ξ < 2.3) with the force at the middle point. This alignment can be understood by considering the symme- try and torque balance when the force at the middle is zero (this case is similar to the one shown in Fig. 1 with the fore-aft forces ignored): the lateral displacement is 1 Frictional (a) ξ= 2.00 Granular (b) ξ= 2.00 Inertial (EBT) (c) ξ= 2.00 0.8 e m T i 0.4 0 0 A=12.57 (d) 1 0.4 ξ= 2.00 0.8 1 0 A (e) 0.4 ξ= 2.30 0.8 1 0 W. head (f) 0.4 0.8 1 ξ= 1.80 0.8 e m T i 0.4 0 0 0.4 0.8 1 0 Tail <−− s −−> Head Tail <−− s −−> Head Tail <−− s −−> Head 0.4 0.8 1 0 0.4 0.8 1 FIG. 4. Torque pattern variation for different force laws, kine- matics and geometry. Torque pattern with (a) frictional, (b) granular (b), and (c) inertial force laws. A = 7.54 in (a-c).(d) Large amplitude A = 12.57. (e) Increasing amplitude toward the tail. A = 7.54(1 − s). (f) With a large head. Insets in (d- f) are schematic diagrams of the corresponding models. The head is not drawn to scale. See Supplementary Videos S3-8 for the respective torque patterns as a function of ξ for (a-f). symmetric about the middle point while the lateral ve- locity and force distributions are antisymmetric. The antisymmetric force distributions generate torque about the middle point with the same sign but the total torque on the body must be zero. Therefore, the torque from each half of the body (i.e. T e(0.5)) at this time instant must be zero. When ξ approaches 2, one full wavelength appears on each side of the body, the torque contribu- tions cancel out, and T e mid becomes zero. As ξ continues to increase, the T e mid becomes in phase with the local force Fmid. This situation corresponds to a breakdown in the torque pattern of a single traveling wave and a reversal of the local torque wave at the middle. mid and T e To understand the speed variation of the torque wave (Fig. 3a), we compare the phase differences of the torque, force and curvature at the middle point and at the head (s = 1). At a point infinitely close to the head, the torque T e head is simply in phase with the local force at the head (Fhead). Therefore, the phase difference be- tween the torques at the middle point and at the head (the angle between T e head in Fig. 5b) is consider- ably smaller than the phase difference between the forces (the angle between Fmid and Fhead in Fig. 5b) and the phase difference between curvatures (the angle between κmid and κhead in Fig. 5b). As ξ increases from 0.5 to 1.8, the phase of T e head in- creases at the same rate of Fhead. Therefore, the phase difference of torque increases from a smaller number (ap- proximately 0.25π for ξ = 0.5) to nearly π. Since such an increase is greater in proportion compared to the increase in curvature, which is from 0.5π to 1.8π, the speed of the torque wave relative to the curvature wave decreases. mid remains the same, but T e 5 1 0 −1 As shown in the above analysis, the torque pattern is primarily determined by the phase distribution of the forces modulated by distance. This picture can also help explain the observed torque variations (Fig. 4). When the force laws are changed to granular or frictional ones, the phase distributions of the forces on the body are sim- ilar; therefore, similar torque patterns are observed. The phase of the reactive force is proportional to the time derivative of the velocity and is hence ahead of the phase of the resistive force by π/2; therefore, the phase of the torque is shifted by the same amount. When the cur- vature amplitude increases toward the tail, the motion and forces on the head are relatively small; thus, the ef- fective phase range from the head to tail is smaller than the nominal one indicated by ξ, and the two-wave tran- sition is delayed (greater ξ). For the case with a head, because the head force is nearly out of phase with the total torque from other parts when the two-wave transi- tion occurs (see the red and magenta arrows in Fig. 5b, ξ = 1.8), the enhancement of the head drag causes the cancellation and reversal of the torque to occur earlier (smaller ξ). DISCUSSION In the resistive force theory for viscous fluids, the as- sumption that the force on one segment is independent of the movement of other segments might introduce signifi- cant errors in the forces [33]. Such error can be alleviated by using slender body theory [38]. In slender body the- ory, the body of the swimmer is assumed to be slender, and the ratio between the radius of the body and body length a/L is much smaller than 1. Singularity solutions of point forces and dipoles are arranged along the body centerline ,and the velocity at a point is computed as the superposition of the singularity solutions to include the effect of the interaction between segments (see [33] for the details of the explanation and implementation). Here, we use a biologically relevant body shape 1/L = 1/30 [39] and the same kinematic parameters in RFT. We found that the transition of the torque pattern from slender body theory is qualitatively the same as those from re- sistive force theory but the transition to the two-wave pattern occurs at smaller ξ (≈1.8) (Fig. 6a & b). The wave speed ratio from SBT is also slightly smaller than the result from RFT (e.g., 3.0 vs. 3.3 at ξ = 1). Ex- amination of the force distribution for a small amplitude reveals one mechanism for the early transition: forces at the head and tail are larger because at the ends, the seg- ments experience greater drag force as less segments are nearby to "help" induce the flow. Similar to the case with a head (Fig. 4f), the head and tail forces are nearly out of phase with the total torque from other parts and the enhancement of the drags and the ends causes the cancellation and reversal of the torque to occur earlier. 6 FIG. 5. Composition of the torque at the middle and phasor diagrams. (a) The lateral force Fy distribution as a function of body position (s ≈ x) and time. (b) Phasor diagrams. The blue arrows represent the contribution of the force on the anterior part of the body to the torque at the middle of the body when the force at the middle is maximum. The integration regions are marked by the thick blue lines in (a). T e head represent the torque at the middle and the head, respectively, and the corresponding forces are marked by green and red dots in (a). The black arrows represent the curvature phasors. The lengths of the force, torque, and curvature phasors are drawn to reflect only their relative magnitudes for the same value of ξ. mid and T e = 1.00 = 1.80 (b) 1 (a) 0.8 i e m T 0.4 (c) 1 Fy 0 -1 0 0 1 Tail <-- Position --> Head 0.4 0.8 0 0.4 Tail <-- Position --> Head 0.8 1 0 1 Tail <-- Position --> Head 0.8 0.4 SBT RFT 1 (a) Elastic 0.8 e m T i 0.4 0 1 0.8 (c) Viscous = 1.00 = 2.00 (b) (d) FIG. 6. Torque pattern computed using slender body theory for viscous fluids. (a) & (b) are the torque patterns for differ- ent wave numbers. See Supplementary Video S10. A = 7.54. (c) Comparison of the lateral force along the body using resis- tive force theory and slender body theory. t = 0, and A = 0.6. The resistance to the bending from the body can also contribute to the torque, therefore, we further discuss the effects of elastic and viscous forces in the body in a general way. We assume that the elastic force requires an additional torque Ts = Ceκ and that the viscous force requires an additional torque Tv = Cv κ. Since when these torque dominates, the torque pattern just coincides with the curvature pattern, a traveling wave, we consider the case in which these torques are signif- icant but smaller compared to the torque from exter- nal forces. Therefore, the coefficients Ce and Cv are chosen such that maximal values of these torques are 20% of the maximal values of the torque from exter- nal forces, i.e. T = T e/ max(T e) + 0.2 κ/ max( κ) and T = T e/ max(T e) + 0.2κ/ max(κ). For ξ < 1.8, the in- e m T i 0.4 0 0 1 Tail <-- Position --> Head 0.8 0.4 0 0.4 0.8 1 Tail <-- Position --> Head FIG. 7. Torque pattern variation when the elastic or viscous body forces are included. The four subfigures represent com- bination of two kinds of forces and two the wave numbers. See Supplementary Videos S11-12 for the respective torque patterns as a function of ξ for (a,b) and (c,d). clusion of elastic force causes the wave speed of the torque in the middle part to decrease (Fig. 7a). For example, vT /vκ = 2.9 for ξ = 1. The inclusion of viscous force causes the wave speed of the torque in the middle part to decrease. For example, vT /vκ = 1.8 for ξ = 1. In this study, we also adopted a highly simplified lo- comotion gait, but organisms adopt gaits different from a single-mode sinusoidal curvature wave during turning and other maneuvers [40, 41]. The torque pattern and neural control required for these maneuvers may be quite different and warrant further study. As shown in our variation study and previous studies, inertia, body elasticity, interactions between body parts, and complex body geometry may all affect the torque and muscle activation patterns. Therefore, the predicted torque from our simple model probably cannot match the muscle activation of a particular organism in detail. However, the torque predicted by our model is certainly an important part of the total torque that needs to be overcome by many organisms. Our results predict that muscle activation is no longer a traveling wave when the dominant forces that the ani- mal must overcome are external resistive forces and the wave number is greater than two (e.g., the snake in [21]). However, to our knowledge, muscle activation and neu- ral control in animals with wave numbers greater than 2 have not been studied. From another perspective, our results predict that muscle activation of a traveling wave cannot produce a uniform bending wave for more than two wavelengths if external forces significantly contribute to the torque. For robotic systems, our results show that the distributions of torque magnitude and energy out- put along the body can be adjusted by varying the wave number; this information may guide the design of driving systems and the use of passive materials. In summary, our study provides a general picture of the torque pattern from resistive forces in undulatory lo- comotion, including new and complex patterns that have not previously been observed. By introducing the pha- sor diagram for undulatory locomotion, we show that the torque pattern can be understood from the integration of distance-modulated force phasors and that the rapid transitions occurring near integer numbers are the result of the cancellation of the force phasors. The phasor dia- gram method may be a useful tool to further investigate the interplay between torque, passive body forces, body shape, and external forces in undulatory locomotion. 7 [2] R. D. Maladen, Y. Ding, C. Li, and D. I. Goldman, Science 325, 314 (2009). [3] D. Hu, J. Nirody, T. Scott, and M. Shelley, Proceedings of the National Academy of Sciences 106, 10081 (2009). [4] Z. Guo and L. Mahadevan, Proceedings of the National Academy of Sciences 105, 3179 (2008). [5] C. Wardle, J. Videler, and J. Altringham, Journal of Experimental Biology 198, 1629 (1995). [6] E. Lauga and T. Powers, Reports on Progress in Physics 72, 096601 (2009). [7] N. Cohen and J. Boyle, Contemporary Physics 51, 103 (2010). [8] R. M. Alexander, Principles of Animal Locomotion (Princeton University Press, Princeton, USA, 2003). [9] M. Tesch, K. Lipkin, I. Brown, R. Hatton, A. Peck, J. Rembisz, and H. Choset, Advanced Robotics 23, 1131 (2009). [10] A. Crespi and A. Ijspeert, IEEE Transactions on Robotics 24, 75 (2008). [11] R. D. Maladen, Y. Ding, D. P. I. banhowar, The International Journal of Robotics Research 30, 793 (2011). and B. Um- Goldman, [12] D. I. Goldman and D. L. Hu, American Scientist 98, 314 (2010). [13] J.-Y. Cheng and R. Blickhan, Journal of Theoretical Bi- ology 168, 337 (1994). [14] J. Cheng, T. Pedley, and J. Altringham, Philosophical Transactions of the Royal Society of London. Series B: Biological Sciences 353, 981 (1998). [15] F. Hess and J. Videler, Journal of Experimental Biology 109, 229 (1984). [16] V. J. Butler, R. Branicky, E. Yemini, J. F. Liewald, A. Gottschalk, R. A. Kerr, D. B. Chklovskii, and W. R. Schafer, Journal of the Royal Society Interface 12, 20140963 (2015). [17] Y. Ding, S. S. Sharpe, K. Wiesenfeld, and D. I. Goldman, Proceedings of the National Academy of Sciences 110, 10123 (2013). [18] C. Brokaw, Journal of Experimental Biology 43, 155 (1965). [19] C. Fang-Yen, M. Wyart, J. Xie, R. Kawai, T. Kodger, S. Chen, Q. Wen, and A. Samuel, Proceedings of the National Academy of Sciences 107, 20323 (2010). [20] S. Berri, J. H. Boyle, M. Tassieri, I. A. Hope, and N. Co- hen, HFSP journal 3, 186 (2009). [21] S. S. Sharpe, S. A. Koehler, R. M. Kuckuk, M. Serrano, P. A. Vela, J. Mendelson, and D. I. Goldman, Journal of Experimental Biology 218, 440 (2015). [22] C. Brokaw, Journal of Experimental Biology 45, 113 (1966). [23] R. Johnson and C. Brokaw, Biophysical Journal 25, 113 ACKNOWLEDGMENTS (1979). Funding for Y.D. and T.Y.M. was provided by NSFC grant No. 11672029, NSAF-NSFC grant No. U1530401, and the Recruitment Program of Global Young Experts. ∗ [email protected] [1] J. Gray and G. Hancock, Journal of Experimental Biol- ogy 32, 802 (1955). [24] E. D. Tytell, C.-Y. Hsu, T. L. Williams, A. H. Cohen, and L. J. Fauci, Proceedings of the National Academy of Sciences (2010). [25] T. McMillen, T. Williams, and P. Holmes, PLoS Com- putational Biology 4, e1000157 (2008). [26] H. Choset, J. Luntz, E. Shammas, T. Rached, D. Hull, and C. Dent, in Proceedings of SPIE, Vol. 3990 (Interna- tional Society for Optics and Photonics, 2000) p. 148. [27] Q. Yan, L. Wang, B. Liu, J. Yang, and S. Zhang, Journal of Bionic Engineering 9, 156 (2012). [28] W.-S. Chu, K.-T. Lee, S.-H. Song, M.-W. Han, J.-Y. Lee, H.-S. Kim, M.-S. Kim, Y.-J. Park, K.-J. Cho, and S.-H. 8 Ahn, International Journal of Precision Engineering and Manufacturing 13, 1281 (2012). [29] B. K. Nguyen, J. H. Boyle, A. A. Dehghani-Sanij, and N. Cohen, in International Conference on Robotics and Biomimetics (IEEE, 2009) pp. 765–770. [30] A. De Luca and R. Mattone, in Proceedings of the 2005 IEEE International Conference on Robotics and Automa- tion (IEEE, 2005) pp. 999–1004. [31] P. Liljeback, K. Y. Pettersen, Ø. Stavdahl, and J. T. Gravdahl, Robotics and Autonomous Systems 60, 29 (2012). [32] C. Wright, A. Johnson, A. Peck, Z. McCord, A. Naakt- geboren, P. Gianfortoni, M. Gonzalez-Rivero, R. Hatton, and H. Choset, in International Conference on Intelligent Robots and Systems (IEEE, 2007) pp. 2609–2614. [33] B. Rodenborn, C.-H. Chen, H. L. Swinney, B. Liu, and H. Zhang, Proceedings of the National Academy of Sci- ences 110, E338 (2013). [34] R. D. Maladen, Y. Ding, and D. P. B. Umban- I. Goldman, A. Kamor, howar, Journal of The Royal Society Interface 8, 1332 (2011). [35] M. Lighthill, Journal of Fluid Mechanics 9, 305 (1960). [36] B. Friedrich, I. Riedel-Kruse, J. Howard, and F. Julicher, Journal of Experimental Biology 213, 1226 (2010). [37] J. Gray, Journal of Experimental Biology 23, 101 (1946). [38] J. Lighthill, SIAM review 18, 161 (1976). [39] B. T. Moore, J. M. Jordan, and L. R. Baugh, Plos One 8, e57142 (2013). [40] V. Padmanabhan, Z. S. Khan, D. E. Solomon, A. Arm- and strong, K. P. Rumbaugh, S. A. Vanapalli, J. Blawzdziewicz, PloS one 7, e40121 (2012). [41] G. Saggiorato, L. Alvarez, Kaupp, G. Gompper, arXiv:1703.07705 (2017). J. F. Jikeli, U. B. and J. Elgeti, arXiv preprint
1903.05712	1	1903	2019-03-13T21:06:48	Membrane-mediated interactions	[ "physics.bio-ph", "cond-mat.soft" ]	Interactions mediated by the cell membrane between inclusions, such as membrane proteins or antimicrobial peptides, play important roles in their biological activity. They also constitute a fascinating challenge for physicists, since they test the boundaries of our understanding of self-assembled lipid membranes, which are remarkable examples of two-dimensional complex fluids. Inclusions can couple to various degrees of freedom of the membrane, resulting in different types of interactions. In this chapter, we review the membrane-mediated interactions that arise from direct constraints imposed by inclusions on the shape of the membrane. These effects are generic and do not depend on specific chemical interactions. Hence, they can be studied using coarse-grained soft matter descriptions. We deal with long-range membrane-mediated interactions due to the constraints imposed by inclusions on membrane curvature and on its fluctuations. We also discuss the shorter-range interactions that arise from the constraints on membrane thickness imposed by inclusions presenting a hydrophobic mismatch with the membrane.	physics.bio-ph	physics	Membrane-mediated interactions Anne-Florence Bitbol, Doru Constantin and Jean-Baptiste Fournier Abstract Interactions mediated by the cell membrane between inclusions, such as membrane proteins or antimicrobial peptides, play important roles in their biolog- ical activity. They also constitute a fascinating challenge for physicists, since they test the boundaries of our understanding of self-assembled lipid membranes, which are remarkable examples of two-dimensional complex fluids. Inclusions can couple to various degrees of freedom of the membrane, resulting in different types of in- teractions. In this chapter, we review the membrane-mediated interactions that arise from direct constraints imposed by inclusions on the shape of the membrane. These effects are generic and do not depend on specific chemical interactions. Hence, they can be studied using coarse-grained soft matter descriptions. We deal with long- range membrane-mediated interactions due to the constraints imposed by inclusions on membrane curvature and on its fluctuations. We also discuss the shorter-range interactions that arise from the constraints on membrane thickness imposed by in- clusions presenting a hydrophobic mismatch with the membrane. Anne-Florence Bitbol Sorbonne Universit´e, CNRS, Laboratoire Jean Perrin (UMR 8237), Paris, France (pre- vious and Department of Physics, Princeton University, Princeton, NJ, USA), e-mail: anne-florence.bitbol@ sorbonne-universite.fr address: Lewis-Sigler Institute for Integrative Genomics Doru Constantin Laboratoire de Physique des Solides, CNRS, Univ. Paris-Sud, Universit´e Paris-Saclay, Orsay, France, e-mail: [email protected] Jean-Baptiste Fournier "Mati`ere Laboratoire (MSC), UMR 7057 CNRS, Univer- sit´e Paris 7 Diderot, Paris Cedex 13, France, e-mail: jean-baptiste.fournier@ univ-paris-diderot.fr et Syst`emes Complexes" 1 2 1 Introduction Anne-Florence Bitbol, Doru Constantin and Jean-Baptiste Fournier Although membrane proteins were traditionally described as free to diffuse in the cell membrane [1], it was soon acknowledged that the lipid bilayer can influence their organization and thus have an impact on many aspects of their activity [2]. Hence, interactions between proteins and the host membrane, as well as the resulting protein-protein interactions, have become fundamental topics in biophysics. Membrane inclusions such as proteins can couple to various degrees of freedom of the membrane (curvature, thickness, composition, tilt, etc.), thus giving rise to several types of membrane-mediated interactions. It is noteworthy that these in- teractions are often non-specific, i.e. they do not involve the formation of chemi- cal bonds between the various components. Thus, understanding these interactions calls for a description of the membrane as a self-assembled system whose proper- ties are collectively determined, and not merely given by the chemical properties of the molecules involved [3]. Over the last few decades, it has become clear that the concepts developed in soft matter physics to describe self-organized systems are extremely useful in this context, and that coarse-grained effective models such as the Helfrich model of membrane elasticity [4] can yield valuable insight. In this chapter, we review the membrane-mediated interactions between inclu- sions such as membrane proteins that arise from direct constraints imposed by these inclusions on the shape of the membrane. Our point of view is mostly theoretical, in agreement with the history of this research field, but we also discuss the numerical and experimental results that are available. For clarity, we treat separately the effects that result from the coupling of the inclusions with membrane curvature and those that arise from their coupling with membrane thickness. Note however that a given inclusion can couple to both of these degrees of freedom. The first case, presented in Section 2, leads to interactions with a much larger range than the characteristic size of the inclusions, which will be referred to as "long-range interactions". Such effects can be described starting from the coarse-grained Helfrich model [4]. The second case, discussed in Section 3, yields a much shorter-range interaction, and requires more detailed effective models of the membrane. Other types of membrane-mediated interactions, arising from other underlying membrane degrees of freedom such as lipid composition and tilt, will not be dis- cussed in detail. Besides, important applications such as the crystallization of mem- brane proteins and the interaction between constituents of such crystals, are outside of the scope of this chapter. 2 Long-range membrane-mediated interactions Inclusions such as proteins are generally more rigid than the membrane. Therefore, they effectively impose constraints on the shape of the membrane, especially on its curvature, which plays a crucial part in membrane elasticity. These constraints in turn yield long-range membrane-mediated interactions between inclusions. Membrane-mediated interactions 3 We will review the first theoretical predictions of these interactions, before mov- ing on to further results in the analytically tractable regime of distant inclusions embedded in almost-flat membranes, including anisotropy, multi-body effects, and dynamics. Extensions to other geometries will then be discussed, including the com- pelling but tricky regime of large deformations, where numerical simulations pro- vide useful insight. Finally, we will examine the available experimental results. 2.1 First predictions 2.1.1 Seminal paper The existence of long-range membrane-mediated forces between inclusions in lipid membranes was first predicted in Ref. [5]. The curvature elasticity of the membrane was described by the tensionless Helfrich Hamiltonian [4]. For an up-down sym- metric membrane, it reads: H = dA (c1 + c2)2 + ¯κ c1c2 , (1) (cid:90) (cid:104)κ 2 (cid:105) (cid:111) (cid:90) (cid:110)κ 2 (cid:2)∇2h(rrr)(cid:3)2 where κ is the bending rigidity of the membrane and ¯κ is its Gaussian bending rigidity, while c1 and c2 denote the local principal curvatures of the membrane, and A its area. This elastic energy penalizes curvature. For small deformations of the membrane around a planar shape, Eq. 1 can be approximated by H[h] = drrr + ¯κ det[∂i∂ jh(rrr)] , (2) where h(rrr) is the height of the membrane at position rrr = (x,y) ∈ R2 with respect to a reference plane, and (i, j) ∈ {x,y}2. The Hamiltonian in Eq. 2 is massless and features a translation symmetry (h → h + C where C is independent of position) that is broken in a ground-state configuration, yielding Goldstone modes. The as- sociated long-range correlations give rise to long-range membrane-mediated inter- actions. Neglecting the effect of the membrane tension σ, as in Eqs. 1 and 2, is legitimate below the length scale(cid:112)κ/σ. Note that the simplified Hamiltonian in Eq. 2 is quadratic in the field h, i.e. the field theory is Gaussian. In Ref. [5], inclusions are characterized by bending rigidities different from those of the membrane bulk. A zone with slightly different rigidities can represent a phase- separated lipid domain, while a very rigid zone can represent a protein. Both regimes (perturbative and strong-coupling) are discussed, in the geometry of two identical circular domains of radius a at large separation d (cid:29) a (see Fig. 1). An interaction potential proportional to 1/d4 is obtained in both regimes. In the perturbative regime, the interaction depends on the perturbations of κ and ¯κ in the inclusions and on the value of κ in the membrane, as well as on kBT . 4 Anne-Florence Bitbol, Doru Constantin and Jean-Baptiste Fournier Fig. 1: Ground-state shape of a membrane containing two rigid disk-shaped inclu- sions that impose the contact angles α1 and α2, obtained by solving the Euler- Lagrange equation (see Ref. [6]). The membrane shape is described by its height h with respect to the plane z = 0. The radius of the inclusions is denoted by a, and the center-to-center distance by d. Besides, a low-temperature interaction is obtained for rigid inclusions that im- pose a contact angle with the membrane, e.g. cone-shaped inclusions [5, 7]: U1(d) = 4πκ(α2 1 + α2 2 ) a4 d4 , (3) where α1 and α2 are the contact angles imposed by inclusion 1 and inclusion 2 (see Fig. 1). This interaction is obtained by calculating the membrane shape that mini- mizes the membrane curvature energy in Eq. 2 in the presence of the inclusions. It arises from the ground-state membrane deformation due to the inclusions, and van- ishes for up-down symmetric inclusions. It is repulsive. Note that this interaction does not depend on the Gaussian bending rigidity of the membrane [7], in contrast with the perturbative case [5]. Indeed, the Gaussian curvature energy term only de- pends on the topology of the membrane and on boundary conditions [7]. Hence, in most subsequent studies of the membrane-mediated forces between rigid membrane inclusions, the Gaussian curvature term in Eq. 2 is discarded. Another interaction, which is attractive and originates from the thermal fluctua- tions of the membrane shape, was predicted as well between rigid inclusions [5, 8]: U2(d) = −6kBT a4 d4 . (4) Importantly, this fluctuation-induced interaction is independent of elastic constants and of contact angles. It exists even for up-down symmetric inclusions (imposing α1 = 0 and α2 = 0) that do not deform the ground-state membrane shape. Multipole expansions valid for a (cid:28) d were used to calculate these interactions for rigid inclusions. Details on these expansions are presented in Refs. [8, 6]. Only the leading-order terms in a/d were obtained in Ref. [5]. This method was recently pushed further, yielding higher-order terms in a/d [6]. Membrane-mediated interactions 2.1.2 Point-like approach 5 Ref. [9] extended the study of of Ref. [5]. Membrane elasticity was described by Eq. 2 as in Ref. [5], but different membrane-inclusion couplings were considered. Rigid inclusions were treated through a coupling Hamiltonian favoring a relative orientation of their main axis and of the normal of the membrane. The membrane- mediated interaction was calculated in the limit of very small inclusions, where the ultraviolet cutoff of the theory Λ appears. The radius a of the inclusions was related to Λ through Λ = 2/a [9], yielding agreement with the results of [5]: the total interaction energy obtained is the sum of U1 and U2 (Eqs. 3 and 4). This opened the way to direct point-like descriptions of membrane inclusions. In Ref. [10], a perturbative approach was taken, where the coupling with the membrane and the inclusions was assumed to be linear or quadratic in the local mean curvature at the point location of the inclusion. In Ref. [11], the insertion energy of a protein in the membrane was approximated by a term proportional to the Gaussian curva- ture of the membrane at the insertion point. Then, in Refs. [12, 13], inclusions were modeled as more general local constraints on the membrane curvature tensor. Con- sidering inclusions as point-like is justified in the case of membrane proteins, since their typical radius is comparable to membrane thickness, which is neglected when the membrane is considered as a surface, as in Eq. 2. This description simplifies the calculation of membrane-mediated interactions, by eliminating the need for a multipole expansion. In practice, one writes the partition function of the membrane described by the elastic energy in Eq. 2 (discarding Gaussian curvature), modeling inclusions as point curvature constraints [12, 13]. For one inclusion imposing a lo- cal isotropic curvature c in rrr0, these constraints read ∂ 2 y h(rrr0) = c and ∂x∂yh(rrr0) = 0. Then, the part of the free energy that depends on the distance d be- tween the inclusions is the sum of U1 and U2 (Eqs. 3 and 4), where the effective radius a of the point-like inclusions appears through the cutoff Λ = 2/a, and the effective contact angle is α = ac. x h(rrr0) = ∂ 2 Refs. [14, 15] formalized the connection between the original description of in- clusions as rigid objects [5], and the more convenient point-like description. The effective field theory formalism developed in Refs. [14, 15] for membranes (see also Ref. [16] for fluid interfaces, and Ref. [17] for a review) considers inclusions as point-like particles, and captures their structure and the boundary conditions they impose via localized coupling terms. In practice, a series of generic scalar localized terms consistent with the symmetries of the system is added to the curvature energy describing the bare membrane. Each term in the series is polynomial in the deriva- tives of the membrane height h, taken at the point position of the inclusion. The coefficients of each term of the series are then obtained by matching observables, such as the ground-state membrane shape responding to an imposed background, between the full model with extended inclusions and the effective field theory [15]. These Wilson coefficients are analogous to charges, polarizabilities etc. of the in- clusions and describe the interplay between the membrane and the inclusions, by encoding the long-range effects of short-range coupling [17]. Membrane-mediated interactions can be obtained from this effective field theory. It gives back the lead- 6 Anne-Florence Bitbol, Doru Constantin and Jean-Baptiste Fournier ing terms in a/d obtained previously, with a generalization to inclusions with dif- ferent radii, and yields higher-order corrections [14, 15]. This general and powerful method could be extended to complex inclusions with specific Wilson coefficients, and also enables general derivation of scaling laws through power counting. How- ever, one should bear in mind that its existing application to rigid disk-shaped in- clusions a priori yields results specific to this particular model of the inclusions. In particular, the discrepancy obtained with previous point-like approaches on cer- tain higher-order terms [15] should be regarded as a different result obtained for a different model, since previous point-like approaches did not aim to fully mimic rigid disk-shaped inclusions. Note that higher-order terms were recently calculated in the framework of extended disks [6], showing agreement with [15] and pushing the expansion further. 2.1.3 Two types of interactions The long-range membrane-mediated interaction between rigid inclusions comprises two leading-order terms that both depend on the fourth power of a/d (Eqs. 3 and 4) [5]. Subsequent works [9, 12, 13, 14, 15] demonstrated that the total interac- tion is the sum of these two terms, one coming from the ground-state deformation of the membrane by the inclusions (Eq. 3) and the second one arising from entropic effects (Eq. 4). However, it should be noted that the separation of these two terms is mostly of formal interest, since the ground-state shape, which is obtained by mini- mizing the Hamiltonian of the system, may not be of much practical relevance. In practice, one may be able to measure experimentally the average shape of a mem- brane, but in general it would not coincide with the ground-state one, except in the regime of small deformations. In this regime, which has been the focus of most the- oretical work, the membrane Hamiltonian is quadratic (Eq. 2): then, the separation of the two terms makes sense. Let us now discuss each of these two terms. The first term, U1 (Eq. 3), arises from the interplay of the ground-state deforma- tions of the membrane due to the presence of each of the inclusions, and it was first obtained in Ref. [5] by taking the (fictitious) zero-temperature limit. It also corre- sponds to the membrane-mediated interaction within a mean-field approximation. The second term, U2 (Eq. 4), is a fluctuation-induced or entropic effect, which exists even if both inclusions impose vanishing contact angles. Remarkably, in the case of rigid inclusions, the only energy scale involved is kBT : this interac- tion is universal. It arises from the constraints imposed by the inclusions on the thermal fluctuations of the shape of the membrane, which is a field with long- range correlations. It is analogous to the Casimir force in quantum electrodynam- ics (see e.g. [9, 18, 8, 12, 13]), which arises from the constraints imposed by non-charged objects (e.g. metal plates) on the quantum fluctuations of the electro- magnetic field [19, 20]. This fluctuation-induced interaction is thus often termed "Casimir" or "Casimir-like". In Ref. [21], the fluctuation-induced force between membrane inclusions was recovered from the entropy loss associated to the sup- pression of fluctuation modes, thus reinforcing the formal analogy with the Casimir Membrane-mediated interactions 7 force. Fluctuation-induced forces analogous to the Casimir force exist in several other soft matter systems, where thermal fluctuations play an important part [22, 23]. They were first discussed by Fisher and de Gennes in the context of critical binary mixtures [24]. This "critical Casimir" force has been measured experimentally be- tween a colloid and a surface immersed in a critical binary mixture [25]. Interest- ingly, such critical Casimir forces have been predicted to exist in membranes close to a critical point in lipid composition, and that they are very long-range, with power laws up to (a/d)1/4 [26]. Their sign depends on the boundary conditions imposed by the inclusions [26], as in the three-dimensional critical case [23]. Let us now compare the magnitude of these two types of interactions. For two identical inclusions imposing the same contact angle α, the interactions in Eq. 3 and 4 have the same modulus if (cid:114) 3 (5) Using the typical value κ ≈ 25kBT gives α ≈ 6◦: for larger contact angles, the mean-field repulsion dominates over the fluctuation-induced attraction. 4π α = kBT κ . 2.2 Further developments on distant inclusions embedded in almost-flat membranes 2.2.1 In-plane anisotropy Until now, we discussed the simple case of two inclusions with isotropic (i.e. disk- shaped) in-plane cross-section, which was the first case investigated [5]. However, real membrane inclusions, such as proteins, have various shapes. Fig. 2 shows a schematic of the different cases at stake: those in panels a and b were discussed above, and those in panels c and d will be discussed here. In Ref. [9], the case of anisotropic cross-sections was treated through a cou- pling between membrane curvature and symmetric traceless tensor order parame- ters constructed from the main direction of the inclusion cross-section, integrated over the surface of the inclusion cross-section. The interaction energies obtained are anisotropic, and depend on d as 1/d4 for up-down symmetric inclusions that interact only through the fluctuation-induced interaction (see Fig. 2c), just as in the case of isotropic cross-sections. However, inclusions that break the up-down symmetry of the membrane feature an anisotropic interaction with a stronger 1/d2 power law. Its angle dependence is cos(2(θ1 + θ2)), where θi is the angle between the main in-plane axis of inclusion i and the line joining the two inclusion centers (Fig. 2c). This orientation dependence is that of a quadrupole-quadrupole interac- tion [27, 28], and the interaction energy is minimized whenever θ1 + θ2 = 0 (or equivalently θ1 + θ2 = π). This interaction is attractive for a wide range of relative 8 Anne-Florence Bitbol, Doru Constantin and Jean-Baptiste Fournier Fig. 2: Schematic representation of the different cases for inclusions with separation d much larger than their characteristic size a, embedded in a membrane with small deformations around the flat shape. In each case, a view from above and a longitu- dinal cut are presented. Thermal fluctuations of the shape of the membrane are only represented in the bottom right cut of panel a. orientations, while the analogous interaction between inclusions with an isotropic cross-section is always repulsive (see Eq. 3). The in-plane anisotropic case of rigid up-down symmetric rods imposing vanish- ing contact angles to the membrane on their edges was treated in Refs. [27, 8]. Only the fluctuation-induced interaction is then at play (as in Fig. 2c). In this study, thin rods were considered in the limit of vanishing width, and in the "distant" regime where their length L is much smaller than their separation d. The opposite case d (cid:28) L will be discussed in Sec. 2.4.2. The power law obtained is in 1/d4, as in the case of isotropic cross sections (Eq. 4), and the only energetic scale involved in this fluctuation-induced force is kBT . The angular dependence of the interaction is cos2[2(θ1 + θ2)], yielding energy minima for θ1 + θ2 = 0 and π/2. Anisotropic cross-sections were revisited within the point-like approach in Refs. [12, 29]. In this model, inclusions couple to the membrane by locally imposing a generic curvature tensor, with eigenvalues (principal curvatures) denoted by K +J and K−J. The interaction between two such identical inclusions then reads, to leading order in a/d [12, 29]: (cid:8)2J2 cos(2(θ1 + θ2)) + JK [cos(2θ1) + cos(2θ2)](cid:9) , (6) U3(d) = −8πκ a4 d2 where θi are angles between the line joining the inclusion centers and their axis of smallest principal curvature (see Fig. 2d). This term vanishes for isotropic inclu- sions (J = 0), consistently with Refs. [5, 9]. Furthermore, in the fully anisotropic case K = 0, corresponding to a saddle, the power law and the angular dependence both agree with the up-down symmetry breaking and anisotropic cross-section case of Ref. [9]. Eq. 6 shows that in the generic case where J and K are nonzero, the angular degeneracy of the lowest-energy state is lifted, and (assuming without loss of generality that K and J have the same sign) the inclusions tend to align their axis Membrane-mediated interactions 9 of smallest principal curvature along the line joining their centers. Their interac- tion is then attractive [12]. This interaction (Eq. 6) was recovered in Ref. [14] (with different angle notations), and generalized to inclusions with different radii. Subleading terms in 1/d4 were also calculated in Refs. [12] and [14], featuring different results (as for the subleading terms in the isotropic case). One should keep in mind that the models at stake are different, since Ref. [12] considers fully point- like inclusions while Ref. [14] models disk-shaped ones with finite radius through the effective field theory. While the agreement of these models on the leading-order term is a nice sign of robustness, there is no reason to expect an exact agreement at all orders. Ref. [12] also investigated the fluctuation-induced interaction, but its leading- order term was found not to be modified with respect to the isotropic case (Eq. 4). This is at variance with the anisotropy obtained in Refs. [27, 8] for the flat rods, but one should keep in mind that the point-like saddles do not correspond to the limit of the distant flat rods. 2.2.2 Multi-body effects and aggregation A crucial and biologically relevant question is how long-range membrane-mediated interactions drive the collective behavior of inclusions, in particular aggregation. One would be tempted to start by summing the pairwise potentials discussed above, but these long-range membrane-mediated interactions are not pairwise additive. Non-pairwise additivity is a general feature of fluctuation-induced interactions. For instance, the existence of a three-body effect in the van der Waals -- London interac- tion was demonstrated in Ref. [30]. The interaction due to the ground-state mem- brane deformation is not additive either. Indeed, if one considers inclusions that impose boundary conditions to the membrane on their edges, a shape minimizing the energy in the presence of one inclusions will generically not satisfy the boundary conditions imposed by the other one, yielding non-additivity [17]. Three-body and four-body long-range membrane-mediated interactions were first calculated within a perturbative height-displacement model, breaking up-down symmetry but retaining in-plane anisotropy, in Ref. [9]. The distance dependence of the three-body term involves terms in 1/(d2 23) where di j is the distance between particles i and j. These interactions were also investigated in Ref. [10], in a different perturbative approach, considering in particular inclusions that favor a given aver- age curvature, and then in Ref. [11] in a point-like framework, but this particular calculation was recently shown to miss some contributions [14]. 12d2 In Ref. [12], the multi-body interactions and the aggregation of point-like inclu- sions locally imposing a curvature tensor were investigated. This generic model can include both up-down symmetry breaking and in-plane anisotropy depending on the curvature tensor imposed. The leading three-body interaction was found to involve terms in 1/(d2 23), as in Ref. [9], and to vanish for inclusions imposing a zero cur- vature tensor [12]. Monte-Carlo simulations including the full multi-body interac- tions were performed, allowing to study the phase diagram of the system (see Fig. 3). 12d2 10 Anne-Florence Bitbol, Doru Constantin and Jean-Baptiste Fournier Polymer-like linear aggregates were obtained for sufficient values of K and J, as pre- dicted from the leading pairwise term (Eq. 6). A gas phase was found for small J, consistent with the fact that for isotropic inclusions (J = 0) that break the up-down symmetry (K (cid:54)= 0), the leading pairwise interaction is repulsive (Eq. 3). Finally, for small K and large J, aggregates were obtained, some of which had an "egg-carton" structure. This is made possible by the angular degeneracy of the lowest-energy state for K = 0 in the leading pairwise term (Eq. 6). Multi-body interactions were shown to be quantitatively important, but the effect of the fluctuation-induced interaction (Eq. 4) was found to be negligible [12]. The analytical calculation of multi-body effects was performed in this framework in Ref. [29], where the "egg-carton" ag- gregates were also further studied and related to experimentally-observed structures. Fig. 3: Typical equilibrium aggregates obtained from Monte-Carlo simulation of 20 identical point-like anisotropic curvature-inducing inclusions. Each panel represents a different set of (J,K) values. Reproduced from Ref. [12]. Coarse-grained molecular-dynamics simulations of the highly anisotropic curvature- inducing N-BAR domain proteins adhering on membranes have demonstrated linear aggregation of these proteins on the membrane. This is a first self-assembly step, Membrane-mediated interactions 11 which then yields the formation of meshes enabling budding [31]. This is qualita- tively in good agreement with the predictions of Ref. [12]. The influence of the long-range elastic repulsion between isotropic inclusions that break the up-down symmetry of the membrane on their aggregation was also discussed in Ref. [32], but within a less specific framework including other types of interactions. In this work, this repulsive interaction (Eq. 3) plays the role of an energetic barrier to aggregation. In Ref. [33], the collective behavior of inclusions locally penalizing local cur- vature (either only mean curvature or also Gaussian curvature) was studied us- ing a mean-field theory for the inclusion concentration and Monte-Carlo simula- tions. Since the inclusions considered retain both up-down symmetry and in-plane isotropy, the only membrane-mediated interaction at play is an attractive fluctuation- induced one similar to that in Eq. 4. Direct interactions were also included. Aggre- gation was found to occur even for vanishing direct interactions, provided that the rigidity of the inclusions was sufficient [33]. Hence, fluctuation-induced interactions may be relevant for aggregation, at least in the absence of other, stronger, interac- tions. Note that Eq. (4) shows that the amplitude of fluctuation-induced interactions is quite small. For instance, d = 4a yields U2 ≈ 0.02kBT (all the results discussed so far are strictly relevant only for d (cid:29) a). In Ref. [14], the general effective field theory framework was used in the case of in-plane isotropic inclusions. The leading-order and next-order three-body in- teraction terms due to the ground-state membrane deformation between up-down symmetry-breaking inclusions were obtained, as well as the leading three-body and four-body fluctuation-induced interactions. 2.2.3 Membrane tension membrane tension. This is appropriate for length scales below(cid:112)κ/σ. As σ is in relevant to go beyond(cid:112)κ/σ. For small deformations around a planar shape, the Until now, we have focused on the regime where bending rigidity dominates over the range 10−6 − 10−8 N/m for floppy membranes, while κ (cid:39) 10−19 J, this length scale is then of order 1 µm. However, membrane tensions can span several orders of magnitude [34] depending on external conditions (e.g. osmotic pressure), so it is quadratic Hamiltonian of a membrane including tension reads (cid:90) (cid:110)κ 2 (cid:2)∇2h(rrr)(cid:3)2 [∇∇∇h(rrr)]2(cid:111) + σ 2 H[h] = drrr , (7) where notations are the same as in Eq. 2, and where the Gaussian curvature term has been discarded. Note that, in a self-assembled membrane not submitted to external actions, each lipid adopts an equilibrium area. Hence, a membrane has no intrinsic surface tension (contrary to a liquid-gas interface), and stretching the membrane has an energy cost quadratic in the area variation. However, one usually considers a patch of membrane in contact with a reservoir made up by the rest of the membrane, 12 Anne-Florence Bitbol, Doru Constantin and Jean-Baptiste Fournier so the tension term in Eq. 7 can be interpreted as arising from the chemical potential of this reservoir. For length scales much larger than(cid:112)κ/σ, tension dominates and Eq. 7 can be simplified into (cid:90) H[h] = σ 2 drrr [∇∇∇h(rrr)]2 . (8) This case applies to a tense membrane at large scales, but also to a liquid interface (neglecting gravity). From a formal point of view, techniques similar to those em- ployed in the bending-dominated case can be used, since the Hamiltonian is also quadratic with a single term. Let us first focus on inclusions that do not break the up-down symmetry of the membrane. In Refs. [27, 8], the fluctuation-induced interaction between two distant up-down symmetric rigid thin rods embedded in such a surface was calculated. It was found to be similar to the analogous bending-dominated case (see above), with the same 1/d4 power law, but with a different angular dependence. Refs. [35, 36] considered the tension-dominated case of ellipsoidal colloids trapped at a fluid interface. In the case where the colloid height fluctuations are included but their contact line with the fluid is pinned, long-range fluctuation- induced interactions were obtained. This case is analogous to that of rigid in-plane anisotropic membrane inclusions preserving the up-down symmetry. Interestingly, the power law obtained was found to depend on whether or not in-plane orienta- tional fluctuations of the colloids were allowed. If they are not allowed, the result of Refs. [27, 8] with the 1/d4 power law is recovered in the limit of full anisotropy. If they are allowed, a weaker anisotropic interaction with 1/d8 power law is ob- tained [36]. This strong dependence of the power law of fluctuation-induced forces on boundary conditions was confirmed in Ref. [16] through the effective field the- ory method, in the specific case of in-plane isotropic (disk-shaped) rigid inclu- sions [14, 17]. In the case of membranes, the physical case should allow orienta- tional fluctuations of the inclusions, and hence the 1/d8 power law should be con- sidered. It is attractive and reads: U4(d) = −9kBT a8 d8 . (9) Hence, we expect a crossover between a 1/d4 power law (Eq. 4) and a 1/d8 power law (Eq. 9) as the tension becomes more important. In Ref. [37], a scattering-matrix approach analogous to the one developed for the study of Casimir forces [38, 39, 40] was developed, and applied to the full Hamiltonian in Eq. 7 including both tension and bending. The focus was on disk-shaped elastic inclusions preserving the up-down symmetry, and on their fluctuation-induced interaction. The results obtained in the case of rigid inclusions were consistent with Eq. 4 in the bending-rigidity -- dominated regime, and with Eq. 9 in the tension-dominated regime. Moreover, the crossover between these two regimes was studied numerically. The method developed in Ref. [37] can potentially deal with more general cases, involving multiple complex inclusions. It appears to Membrane-mediated interactions 13 be complementary to the effective field theory method of Refs. [14, 17], and was more straightforward in the transition regime where both tension and bending are relevant [37]. Let us now focus on the interaction due to the ground-state deformation of the membrane. Ref. [41] studied the case of conical inclusions breaking up-down sym- metry but retaining in-plane isotropy, and considered the full Hamiltonian in Eq. 7. They showed that for non-vanishing tension, this interaction has a sign that de- pends on the relative orientation of the cones with respect to the membrane plane (i.e. on the signs of the angles they impose), contrary to the vanishing-tension case (see Eq. 3). Furthermore, at long distances between inclusions, the interaction is exponentially cut off with a decay length(cid:112)κ/σ (it involves Bessel K functions). This property was confirmed in Ref. [42]. Hence, at long distances, the fluctuation- induced force in Eq. 9 should dominate over the force due to the ground-state de- formation. Conversely, in the case of colloids or inclusions with anisotropic cross- sections, Refs. [28] and [16] demonstrated the existence of a long-range interaction due to the ground-state deformation of the membrane. The leading term of this in- teraction is anisotropic and decays as 1/d4. In Ref. [43], the effect of tension on the aggregation of the highly anisotropic curvature-inducing N-BAR domain proteins adhering on membranes was investi- gated through coarse-grained molecular-dynamics simulations. Increasing tension was shown to weaken the tendency of these proteins to linear aggregation, in agree- ment with the predicted weakening of the ground-state membrane-mediated inter- action. 2.2.4 Summary of the interaction laws Table 1 presents a summary of the power laws of the leading-order term of the membrane-mediated interactions in the various situations discussed until now. 2.2.5 External forces and torques Until now, we have only discussed cases where inclusions couple to the membrane shape through its curvature, either explicitly or implicitly (e.g. through rigidity). This is the relevant case in the absence of external forces or torques. External forces can yield local constraints directly on the height of the membrane, e.g. quadratic ones in the case of local trapping or linear ones in the case of local pulling [10]. More specifically, inclusions may experience direct mechanical constraints if they are attached to the cytoskeleton, and torques in the presence of electrical fields be- cause of their dipole moments [13]. In these cases, one expects membrane-mediated interactions to be enhanced, because the ground-state deformations will generically be stronger than in the case where inclusions can freely reorient to minimize them, and because the constraints imposed on fluctuations will be stronger too. 14 Anne-Florence Bitbol, Doru Constantin and Jean-Baptiste Fournier Dominant term in the Hamiltonian Geometry in Eq. 7 Fluctuation-induced interaction Interaction due to the ground-state deformation -- Vanishes if up-down symmetric Disks 1/d4 [5, 9, 12] 1/d4 [5, 9, 12] Bending rigidity κ Disks +anisotropy Distant rods 1/d4 [9, 12] 1/d4 [27, 8] 1/d2 [9, 12] Disks 1/d8 [36, 37] Exponentially suppressed Tension σ Disks +anisotropy Distant rods 1/d8 [36, 37] 1/d4 [27, 8] 1/d4 [28, 16] Table 1: Summary of the power laws obtained for the leading-order terms of the two types of membrane-mediated interactions, as a function of the separation d be- tween the inclusions, in the regime of small deformations of a flat membrane and distant inclusions. Different inclusion geometries are considered. In the case labeled "disks+anisotropy", the anisotropy can be either in the inclusion shape (e.g. ellip- soidal [9]) or in the constraint it imposes (e.g. an anisotropic local curvature [12]). The case of inclusions subject to external torques was studied in Ref. [13], for point-like inclusions setting a curvature tensor, in in-plane isotropic case. Both ex- ternal fields strong enough to effectively pin the orientations of the inclusions, and finite external fields that set a preferred orientation, were considered. In both cases, membrane-mediated forces are strongly enhanced, even more in the strong-field case. A logarithmic fluctuation-induced interaction was obtained, as well as an in- teraction due to the ground-state deformation which either scales as 1/d2 if the preferred orientations are the same for both inclusions, or logarithmically if they are different. Interestingly, these interactions depend on the relative orientation of the preferred curvatures set by the inclusions, while in the torque-free case, the interac- tion only depends on their absolute values (see Eq. 3) [13]. In Ref. [35], colloids at a fluid interface were considered, with different types of boundary conditions. In the case were the position of the colloids is considered to be frozen (both in height and in orientation), strong logarithmic fluctuation-induced interactions are obtained. 2.2.6 Fluctuations of the interactions Until now, we have discussed the average values at thermal equilibrium of membrane- mediated forces. Thermal fluctuations already play an important part since they are the physical origin of fluctuation-induced forces. But membrane-mediated forces themselves fluctuate as the shape of the membrane fluctuates. The fluctuations of Membrane-mediated interactions 15 these forces have been studied in Ref. [44], using the stress tensor of the mem- brane [45, 46]. This approach is inspired from those used previously for the fluc- tuations of Casimir forces [47], and of Casimir-like forces between parallel plates imposing Dirichlet boundary conditions on a thermally fluctuating scalar field [48]. The case of two point-like membrane inclusions that locally impose a curvature tensor was studied in Ref. [44], for in-plane isotropic inclusions but including the up-down symmetry breaking case. Integrating the stress tensor on a contour sur- rounding one of the two inclusions allowed to calculate the force exerted on an inclusion by the rest of the system, in any shape of the membrane [49]. The average of the force obtained gives back the known results Eqs. 3 and 4 that were obtained from the free energy in previous works. The variance of the force was also calcu- lated, showing that the membrane-mediated force is dominated by its fluctuations. The distance dependence of the fluctuations, present in the sub-leading term of the variance, was also discussed. Interestingly, it shares a common physical origin with the fluctuation-induced (Casimir-like) force [44]. 2.3 Dynamics Fundamental interactions, e.g., electrostatic ones, are usually considered as instan- taneous, in the sense that they propagate at a velocity much higher than that of the particles experiencing them. This is not the case for membrane-mediated interac- tions, as the spreading of membrane deformations involves slow dissipative phe- nomena. The dynamics of membrane-mediated interactions is a promising subject for future research. Studying out-of-equilibrium membrane-mediated interactions intrinsically requires taking into account the dynamics of the membrane. Taking care both of the motion of the membrane and of that of the inclusions is very diffi- cult. Hence, the first theoretical study in this direction to our knowledge, Ref. [50], considered two immobile inclusions that simultaneously change conformation, i.e., that simultaneously create a source of deformation, and therefore trigger a time- dependent interaction as the membrane deformation spreads dissipatively. In Ref. [50], inclusions were modeled as simple point-like sources of mean curvature that are triggered simultaneously at t = 0. One could imagine cylindri- cal integral proteins such as ion channels transforming into conical ones upon re- ceiving a chemical signal. The time-dependent Hamiltonian of these inclusions is Hinc(t) = θ (t)∑i Bi∇2h(rrri), with θ (t) the Heaviside step function, Bi the curving strength, and rrri the position of inclusion i. The dynamical reaction of the membrane to such a perturbation was studied in Ref. [50]. As shown in the pioneering works of Refs. [51, 52], the dominant dissipation mechanism at short length scales is the friction between the two monolayers of the membrane. The corresponding dissipated power per unit area is b(vvv+−vvv−)2, where vvv± are the velocities of the two lipid monolayers (the monolayers are denoted by + and −) and b ≈ 109 Js/m4 is the intermonolayer friction coefficient. In addi- tion, the membrane is subjected to viscous forces from the bulk solvent, of viscosity 16 Anne-Florence Bitbol, Doru Constantin and Jean-Baptiste Fournier η ≈ 10−3 Js/m3, and each monolayer behaves as a compressible fluid with elastic 2 k(ρ± ± e∇2h)2. In this expression, ρ± are the monolayer relative energy density 1 excess densities (normalized by their equilibrium density), measured on the mem- brane mid-surface, e ≈ 1 nm is the distance between this surface and the neutral surface of the monolayers (where density and curvature effects are decoupled) and k ≈ 0.1 J/m2. For most practical purposes, the two-dimensional viscosities of the monolayers can be neglected [53]. Taking into account all these effects, Ref. [50] showed that the relaxation dy- namics of a Fourier mode {h(q,t),ρ±(q,t)} in the membrane with two identical triggered inclusions is given, to linear order, by a set of two first-order dynamical equations: ∂ (ρ + − ρ−) = −kq2(ρ + − ρ−) + 2keq4h, (10) 2b 4ηq ∂t ∂ h ∂t = −(σq2 + κq4)h + keq2(ρ + − ρ−) + F(qqq,t), (11) where F(qqq,t) is the Fourier transform of −δ Hinc/δ h(rrr,t), σ is the membrane ten- sion, and κ = κ + 2ke2 the bending rigidity at frozen lipid density [51]. Solving these linear differential equations for time evolution and integrating over the Fourier modes qqq yields the time-dependent membrane deformation produced by one or more inclusions. Then the force f (t) exerted by one inclusion on the other can be obtained by integrating the membrane stress tensor [45, 46, 54] around one inclusion. Fig. 4: (a) Force f (t) normalized by B2/(κe3) exchanged by two inclusions sepa- rated by d versus time t normalized by 4ηe3/κ × 103. The parameters are d = 20e, σ = 10−3κ/e2, ke2/κ = 1 and be2/η = 1000. (b) Dependence of the equilibrium force, feq, and of the maximum of the dynamical force, fm, as a function of d nor- malized by e. Two striking behaviors were observed in Ref. [50] (see Fig. 4): (i) the force f (t) reaches a maximum fm and then decreases to the equilibrium force feq. (ii) While feq decreases exponentially with the separation d between the inclusions, the maximum Membrane-mediated interactions 17 force fm decreases as a power-law ∼ d−3 until it reaches feq. Hence fm is long- ranged. Although these results were obtained with a simplified Hamiltonian for the inclusions, it is likely that the general trends observed will also apply to more realis- tic cases. It should be straightforward to extend the model of Ref. [50] to inclusions that trigger at different times, but considering the movement of the inclusions at the same time as the movement of the membrane would be more challenging. 2.4 Other geometries Until now, we focused on the case of inclusions with separation d larger than their characteristic size, embedded in a membrane with small deformations around the flat shape. This is the case that has attracted the most attention in the literature, because of its relevance for proteins embedded in the membrane, and because of its technical tractability. We now move on to other geometries. 2.4.1 Spherical vesicle Ref. [55] focused on the membrane-mediated interaction arising from the ground- state deformation between two disk-shaped inclusions embedded in the membrane of a spherical vesicle, and imposing contact angles. The case of the spherical vesicle is practically relevant both in biology and in in-vitro experiments. The energy of the membrane was considered to be dominated by bending rigidity, which requires the length scales at play (in particular the vesicle radius) to be small with respect to(cid:112)κ/σ. The covariant Helfrich Hamiltonian (Eq. 1 with no Gaussian curvature term) was adapted to small deformations with respect to a sphere. The interaction was evaluated thanks to an expansion of the energy-minimizing profile of the membrane, and it was found to be strongly enhanced with respect to the flat-geometry interaction (Eq. 3) at length scales where the spherical shape of the vesicle is relevant. At sufficient angular separation, the effective power law of the interaction is ∼ 1/d1/3 [55]. This sheds light on the strong impact of the underlying geometry of the membrane on membrane-mediated forces. Qualitatively, in a flat membrane, the interaction is weaker because the curvature energy in Eq. 1 can be minimized quite well between the inclusions (with an almost perfect saddle that has very little curvature energy), which is not possible in the spherical geometry. Similarly, in the case of external torques (Sec. 2.2.5), the imposed orientations did not allow for this low-energy saddle, thus enhancing the interaction. 2.4.2 Close parallel rods We already discussed the case of rigid rods of length L, at a distance d (cid:29) L [27, 8], which is close to the point-like case. The opposite regime d (cid:28) L is also relevant bi- 18 Anne-Florence Bitbol, Doru Constantin and Jean-Baptiste Fournier ologically, since it can model semi-flexible polymers adsorbed on the membrane. In Ref. [56], the effect of the reduction of the membrane fluctuations by the presence of a semiflexible (wormlike) polymer was discussed. An effective nematic interac- tion was found between different segments of the polymer, and it was shown that this interaction can yield an orientational ordering transition. Let us first consider rods that do not break the up-down symmetry of the mem- brane. The case of such stiff parallel rods in the limit d (cid:28) L (see Fig. 5a) embedded in a membrane with energy dominated either by bending rigidity (Eq. 2) or by ten- sion (Eq. 8) was studied in Ref. [18]. A constant scale-free Casimir-like interaction per unit length is then expected [57], and indeed the Casimir-like interaction poten- tial is then proportional to −kBT L/d [18]. This interaction is much stronger than the one between point-like objects (Eq. 4), because the constraints imposed on fluc- tuation modes are much stronger in the geometry of parallel close rods. Ref. [18] further showed that such rods tend to bend toward one another below a certain criti- cal distance, and that their interaction is screened by out-of-plane fluctuations if the rigidity of the polymer is finite. Fig. 5: Rods embedded in membranes. (a) Geometry: two parallel rods of length L at separation d (cid:28) L. (b) and (c) Two examples of rod types. All rigid rods impose a vanishing curvature along them: ∂y∂yh = 0 on the rod. (b) Rod that allows curving ("c") and twisting ("t") across it. (c) Rod that does not allow curving or twisting across it: it imposes ∂x∂xh = 0 and ∂x∂yh = 0 as well as ∂y∂yh = 0 (see Ref. [58]). This situation was further studied in Ref. [58]. Rods were modeled as constraints imposed on the membrane curvature along a straight line, allowing to define four types of rods, according to whether the membrane can twist along the rod and/or curve across it (see Fig. 5b-c for two examples of these rod types). The numerical prefactors of the potential in L/d were obtained for interactions between the differ- ent types of rods, and they were all found to be attractive, provided that the rods are rigid, i.e. that they impose ∂y∂yh = 0 along them, with the notations of Fig. 5. However, repulsion was obtained between objects imposing completely antagonis- tic conditions (i.e. a rigid rod only imposing ∂y∂yh = 0 along it, see Fig. 5b, and a non-rigid "ribbon" only imposing ∂x∂xh = 0 along it), which is reminiscent of results obtained in critical binary mixtures [23]. In addition, the interaction energy Membrane-mediated interactions 19 was studied numerically versus d/L, thanks to a discretization scheme [59], show- ing the transition between the asymptotic behaviors at large d/L [27] and at small d/L [58] were recovered. Finally, the bending and coming into contact of the rods due to the fluctuation-induced interaction was discussed: it was predicted to occur below a certain value of d [58]. The L (cid:29) d geometry gives insight into what happens between two generic in- clusions that are very close to one another, through the proximity force approxi- mation [60]. This approximation was used in the case of disk-shaped inclusions in Refs. [37, 58], showing that the fluctuation-induced interaction potential then scales as 1/d1/2. In Ref. [61], the interaction due to the ground-state deformation between parallel rigid cylinders adsorbed on a membrane and interacting with it through an adhesion energy was studied. The membrane was assumed to be in the regime of small de- formations, but both tension and bending were accounted for (see Eq. 7), and the geometry where d (cid:28) L was considered. The interaction due to the ground-state de- formation was calculated explicitly in this effectively one-dimensional case. It was found to be repulsive for a pair of cylinders adhering to the same side of the mem- brane, and attractive for cylinders adhering to opposite sides (and hence imposing an opposite curvature). This is at variance with the point-like case, where the in- teraction only depends on the modulus of the curvatures imposed (see Eq. 3). The dependence in d is in tanh(d/(cid:112)κ/σ ) in the first case, and in coth(d/(cid:112)κ/σ ) in the second one [61]. 2.4.3 Large deformation regime All cases discussed until now focused on small deformations. Then, the Hamilto- nian of the membrane is quadratic, and the field theory is Gaussian. This provides tractability, both to solve the Euler-Lagrange equations that give the ground-state shape, which are then linear, and to compute thermodynamical quantities such as the free energy. Here, we will discuss the biologically relevant but much trickier regime of large deformations. In Ref. [62, 63], the covariant membrane stress and torque tensors associated to the full Helfrich Hamiltonian [45] were used to determine formal expressions of the forces between objects adsorbed on fluid membranes that are due to the ground- state deformation of the membrane. These expressions are valid without assuming small deformations, but the ground-state shape needs to be determined in order to obtain a more explicit expression. This is not an easy task in the large-deformation regime. Equilibrium shapes in the large deformation regime were further investi- gated in Ref. [64], allowing to plot the force between cylinders, in the case of a fixed adhesion area between them and the membrane. The direction of the force and its asymptotic exponential decay at large d/(cid:112)κ/σ were found to remain the same as in the small-deformation regime [61]. This situation was also investigated numer- ically in Ref. [65] in the case of cylinders interacting with the membrane through an adhesion energy, yielding phase diagrams of the system. 20 Anne-Florence Bitbol, Doru Constantin and Jean-Baptiste Fournier In Ref. [66], the entropic contribution to the membrane-mediated interaction be- tween two long cylinders adsorbed on the same side of a membrane was studied in the regime of large deformations, in the case of a fixed adhesion area between the cylinders and the membrane. The free energy of the system was calculated by assuming Gaussian fluctuations around the ground-state shape. Interestingly, this entropic contribution enhances the ground-state repulsion between the two cylin- ders [66], while the fluctuation-induced interaction between identical rods in the small-deformation regime is attractive [18, 58]. This is presumably a non-trivial effect coming from the non-linearities at play in the large deformations. It would be interesting to go beyond the approximation of Gaussian fluctuations around the ground-state shape. Solving the shape/Euler-Lagrange equation for membranes beyond the domain of small deformations is technically very hard for most geometries, and incorpo- rating fluctuations too, but numerical simulations can provide further insight. The coarse-grained molecular-dynamics membrane simulations without explicit solvent description of Ref. [67] showed that the elastic interaction between two isotropic curvature-inducing membrane inclusions (quasi-spherical caps) can become attrac- tive at short separations, provided that the inclusions induce a strong enough cur- vature. Recall that the interaction due to the ground-state deformation, which is dominant with respect to the fluctuation-induced one for large enough curvatures imposed by inclusions, is always repulsive in the regime of small deformations (see Eq. 3). This hints at highly non-trivial effects of the large-deformation regime. The attractive membrane-mediated interaction was found to be able to yield aggregation of the caps and vesiculation of the membrane [67] (see Fig. 6). The case of curved phase-separated lipid domains was explored in Ref. [68] through coarse-grained molecular-dynamics simulations. The interaction between domains was found to be attractive, but the angles imposed by the domains were smaller than those yielding attraction in Ref. [67]. Fig. 6: Successive snapshots of a coarse-grained simulation of a membrane with several curvature-inducing inclusions. A process of vesiculation is induced by the elastic interaction between inclusions, which becomes attractive at short separations. Reproduced from Ref. [67]. Membrane-mediated interactions 21 A numerical minimization via Surface Evolver of the Helfrich Hamiltonian Eq. 1 for a membrane with two in-plane isotropic curvature-inducing inclusions was pre- sented in Ref. [69], and forces were calculated by studying infinitesimal displace- ments. A change of sign of the membrane-mediated interaction due to the ground- state deformation of the membrane was obtained, consistently with Ref. [67]. The repulsive interaction, agreeing quantitatively with Eq. 3 at large d/a and for small deformations, turned attractive for d/a of order one, provided that the curvature im- posed by the inclusions (and hence the membrane deformation) was large enough. The separation d is defined as the center-to-center distance projected on a reference plane, while a is the real radius of the inclusions, so that in the large deformation regime where inclusions are very tilted, it is possible to have d < 2a. Attraction occurs in this regime, which is inaccessible to the small-deformation approach. Re- cently, Ref. [70] studied anisotropic protein scaffolds, modeling e.g. BAR proteins, in the large-deformation regime, through similar numerical minimization methods: strongly anisotropic attractive interactions were obtained. Ref. [71] presented a Monte-Carlo simulation of spherical nanoparticles ad- sorbed on a spherical vesicle modeled as a triangulated surface. Aggregation of the nanoparticles and inward tubulation of the vesicle were observed, implying strong attractive interactions. Note however that adhesion might have a strong impact on these structures [72]. A similar coarse-grained description of a membrane vesicle was used in Ref. [73] to investigate the collective effects of anisotropic curvature- inducing inclusions, modeling e.g. BAR proteins. Vesicles were strongly deformed by the numerous inclusions, with sheet-like shapes or tubulation depending on inclu- sion concentration, and aggregation and nematic ordering of these inclusions were observed. 2.5 Experimental studies While membrane-mediated interactions have been the object of significant theoreti- cal and numerical attention, quantitative experimental tests of the theoretical predic- tions remain scarce to this day. A very active research area in biophysics deals with the morphological changes of the cell (invagination [74], vesiculation [75], etc.) un- der the action of various proteins (see [76] for a recent review). However, many other ingredients than membranes and inclusions are at play in these biological sys- tems, for instance the cytoskeleton, out-of-equilibrium events, etc., which makes it hard to isolate membrane-mediated interactions. Biomimetic lipid membranes such as giant unilamellar vesicles [77] are a good model system to study such effects. In principle, the inclusions could be real proteins, but these molecules have complex shapes, which makes it difficult to test predictions of models developed for simple geometries. Many studies have focused on the simpler and more easily controlled system of colloids adhering to membranes (see Ref. [72] for a review), and some have investigated interactions between phase-separated membrane domains [78]. However, even in these simpler cases, membrane-mediated interactions may involve 22 Anne-Florence Bitbol, Doru Constantin and Jean-Baptiste Fournier other effects, such as adhesion of the colloids, variability of contact angles imposed by domains, etc. An experimental study of the aggregation of spherical colloidal particles adhering to biomimetic lipid membranes was presented in Ref. [79]. The observed aggrega- tion of two particles was deemed consistent with a short-range (e.g. exponential) attractive force, and no signature of a longer-range force was obtained. Note that theoretical studies predict a mostly repulsive membrane-mediated force in this ge- ometry, except at very high deformations and small distances. Surprisingly, triplets were observed to form chains, and a linear ring-like aggregate was observed around the waist of a vesicle. Linear chain-like arrangements were also obtained in sim- ulations of a very similar situation in Refs. [80, 81], for certain sizes of particles and adhesion regimes. Ref. [81] used a scaling argument to show that this was not due to membrane-mediated interactions, but to the adhesion of the particles to the membrane, as a linear aggregate yields a higher adhesion area than a compact one. Such a phenomenon would thus not arise in the case of inclusions [72]. Apart from proteins and colloids, another source of membrane deformation is the presence of phase separated (liquid-ordered/liquid disordered) domains, which can be partially budded. Contrast between the domains is obtained in fluorescence microscopy by adding a dye which partitions into one phase [82] or by selec- tively labeling one lipid species [83]. Selective deuteration can also be used to induce contrast in small-angle nuclear scattering [84]. In Ref. [78], the stability of partially budded domains was interpreted as a signature of repulsive interac- tions, since flat ones rapidly fused. The strength of this interaction was evaluated by measuring the distribution of inter-domain distance, and then by evaluating the effective spring constant of the confining potential. It was found to be consistent with the membrane-mediated interaction arising from the ground-state deformation of a tension-free membrane in the small-deformation and large-separation regime (Eq. 3). In Ref. [83], a good agreement was obtained between the observed in-plane distribution of the domains and the predictions of the elastic theory in the presence of tension [41] (see Figure 7). Fig. 7: Shape of the dimpled domains (left), interacting domains on the surface of the same vesicle (center) and repulsive interaction potential, with a fit to theoretical predictions from Ref. [41] (right). Adapted from Figures 3 and 4 of reference [83]. Membrane-mediated interactions 3 Short-range membrane-mediated interactions 23 In Part 1, we dealt with long-range membrane-mediated interactions between inclu- sions, which arise from the curvature constraints imposed by rigid inclusions. There exist several other ways in which inclusions can couple to the surrounding mem- brane and thus interact with other inclusions through the membrane, but these ef- fects are generally short-ranged. The study of these interactions was in fact initiated before that of their long-range counterparts [5]. Membrane proteins were shown ex- perimentally to tend to immobilize neighboring lipids [85]. A membrane-mediated attraction between proteins was predicted to arise due to this local ordering [86], and to decay exponentially above the correlation length of the membrane order pa- rameter [87]. Proteins can locally perturb the thickness of the membrane due to this local ordering, but they may also couple preferentially to one component of a lipid mixture [88]. Here, we are going to focus on the coupling of proteins to membrane thickness. Intrinsic membrane proteins can have a hydrophobic mismatch with the membrane: their hydrophobic thickness is slightly different from that of the unperturbed mem- brane. Hydrophobic mismatch is ubiquitous, and has important biological conse- quences, since the activity of many membrane proteins has been shown to depend on membrane thickness [89]. As proteins are more rigid than membranes, the mem- brane generically deforms in the vicinity of the protein, in order to match its thick- ness and avoid exposing part of the hydrophobic chains of lipids to water. This local deformation of the membrane thickness yields a membrane-mediated interaction between two such proteins. Membrane thickness deformations are not included in the traditional Helfrich description of the membrane [4]. Describing them is tricky since they occur on the nanometer scale, which corresponds to the limit of validity of usual continuum theories where only long-scale terms are kept. Let us focus on these models before moving on to the actual interactions. 3.1 Models for local membrane thickness deformations 3.1.1 Early models The idea that the membrane hydrophobic thickness must locally match that of an intrinsic protein was first used in theoretical descriptions of lipid-protein interac- tions that focused on the thermodynamic phase behavior of the lipid-protein system and on protein aggregation. In Ref. [90], a thermodynamic model called the "mat- tress model" was proposed in order to describe the phase diagrams of lipid bilayers containing proteins with a hydrophobic mismatch. More detailed theoretical investigations of local membrane thickness deforma- tions and of resulting membrane-mediated interactions were motivated by exper- imental results on the antimicrobial peptide gramicidin. In lipid membranes, two 24 Anne-Florence Bitbol, Doru Constantin and Jean-Baptiste Fournier gramicidin monomers, one on each side of the bilayer, can associate to form a dimer, which acts as an ion channel. While isolated monomers do not deform the mem- brane, the dimeric channel generically possesses a hydrophobic mismatch with the membrane [91]. Conductivity measurements yield the formation rate and lifetime of the channel, which are directly influenced by membrane properties [92, 93, 94]. Hence, gramicidin constitutes a very convenient experimental system to probe the effects of local membrane thickness deformations. The first attempt to explain the dependence of gramicidin channel lifetime on the membrane thickness was provided by Ref. [93]. It is based on the idea that the rel- evant membrane energy variation upon dimer breaking is mostly due to membrane tension, which pulls apart the monomers in a membrane with hydrophobic thick- ness larger than that of the dimer. The resulting estimate of the gap between the two monomers in the transition state is δ (cid:39) 1.8 nm [93]. However, this is far larger than the separation required for the breaking of the hydrogen bonds that stabilize the dimer [91], which is of order 1 A. Hence, this first model was not complete. 3.1.2 Huang's model The first full continuum model describing membrane thickness deformations was proposed in Ref. [95]. The Hamiltonian per unit area of the membrane was written by analogy with a smectic A liquid crystal, in which the elongated molecules orga- nize in layers with the molecules oriented along the layers' normal. These two sys- tems present the same symmetries. The most important energetic terms in smectic A liquid crystals correspond to compression of the layers, and to splay distortion, i.e. curvature orthogonal to the layers [96]. In addition, the contribution of the "surface tension" of the membrane was included [95]. Restricting to symmetric deformations of the two monolayers, the effective Hamiltonian H of the membrane reads [95] (cid:90) H = dxdy (cid:20) Ka 2d2 0 (cid:0)∇2u(cid:1)2(cid:21) u2 + γ 4 (∇∇∇u)2 + κ 8 . (12) In this expression, u denotes the thickness excess of the membrane relative to its equilibrium thickness d0 (see Fig. 8), Ka is the stretching modulus of the mem- brane, d0 its equilibrium thickness, γ its "surface tension", and κ an elastic constant associated to splay. Finally, x and y denote Cartesian coordinates on the mid-plane of the membrane. Ref. [95] assimilated γ to the tension of a Plateau border and κ to the Helfrich bending modulus, which may be questioned (see below). The corresponding typi- cal values allowed to neglect the contribution of the "tension" term. By minimizing the resulting membrane Hamiltonian, analytical expressions were obtained for the membrane deformation profiles close to a mismatched protein such as the grami- cidin channel, obtaining a decay length of a few nanometers. This model yields a satisfactory agreement with the experimental results of Ref. [93]. Membrane-mediated interactions 25 Fig. 8: Cut of a bilayer membrane (yellow) containing a protein with a hydrophobic mismatch, represented as a square (orange). The equilibrium thickness of the bilayer is d0, while the actual thickness is denoted by d0 + u. 3.1.3 Models based on the work of Dan, Pincus and Safran Refs. [97, 98] proposed another construction of the membrane Hamiltonian associ- ated to thickness deformations. The energy per lipid molecule in each monolayer of the membrane was written for small deformations as a generic second-order expan- sion in the variation of area per lipid and in the local "curvature" of the monolayer thickness (different from the curvature of the shape of the membrane involved in the Helfrich model, which disregards thickness). Incompressibility of the lipids was used to relate the monolayer thickness and and the area per lipid. Using the same notations as in Eq. 12, and restricting again to up-down symmetric deformations of the membrane, the membrane Hamiltonian of Ref. [98] reads: (cid:0)c0 − c(cid:48) 0Σ0 (cid:1)u∇2u + κ 8 (cid:0)∇2u(cid:1)2(cid:21) , (13) (cid:90) H = dxdy (cid:20) Ka 2d2 0 u2 + κ c0 2 ∇2u + κ 2d0 where c0 is the spontaneous curvature of a monolayer, while c(cid:48) with respect to the area per molecule, and Σ0 the equilibrium area per lipid. 0 denotes its derivative The main difference between this model and that of Ref. [95] is that the effect of monolayer spontaneous curvature is included in Eq. 13. It was shown in Refs. [97, 98] that this ingredient can yield oscillations in the membrane deformation profile, and in the resulting interaction potential between two mismatched proteins. Note that no "tension" term is included in Eq. 13, but the "tension" term in Eq. 12 was neglected in all the calculations of Ref. [95] too. The model of Refs. [97, 98] was generalized in Refs. [99, 100], where results of coarse-grained molecular-dynamics simulations for mismatched proteins in lipid membranes were also presented. The deformations of the average shape of the membrane (i.e., those usually described by the Helfrich model), and the small- scale protrusions were accounted for, as well as the symmetric thickness deforma- tions [99, 100]. The effect of Gaussian curvature was also included in Ref. [100], and Ref. [101] added the effect of tilt. The model of Refs. [97, 98, 99, 100] was further generalized in Ref. [102], where an additional term, proportional to the squared gradient of thickness, was included in the initial expression of the energy per lipid molecule in each monolayer of the membrane. Physically, this term should involve a microscopic interfacial tension 26 Anne-Florence Bitbol, Doru Constantin and Jean-Baptiste Fournier contribution, associated to variations of the area per lipid. Note that this is different from the Plateau border tension discussed and discarded in Ref. [95], as in a Plateau border, molecules can move along the surface and exchange with the bulk, yielding a smaller tension. Macroscopic membrane tension was also incorporated explicitly in Ref. [102], through a chemical potential µ set by the rest of the membrane on the patch considered: σ = −2µ/Σ0 then plays the part of an externally applied ten- sion. The Helfrich Hamiltonian with tension Eq. 7 was recovered from this model for average height deformations. In the case where the average shape of the mem- brane is flat, and integrating out anti-symmetric thickness deformations to focus on symmetric ones, the Hamiltonian reads: (cid:90) H = dxdy (cid:20) σ d0 u + Ka 2d2 0 u2 + K(cid:48) a 2 (∇∇∇u)2 + K(cid:48)(cid:48) a 2 (∇2u)2 , (14) (cid:21) plus omitted boundary terms (see Ref. [102]), with (c0 − c(cid:48) 0Σ0) + k(cid:48) a , (15) a = −κ0 K(cid:48) d0 κ0 K(cid:48)(cid:48) a = , 4 (16) and the same notations as in Eqs. 12 and 13, and where the new contribution k(cid:48) a with respect to Eq. 13 arises from the term proportional to the squared gradient of the thickness u. (The definition of u in Ref. [102] is slightly different from that of Refs. [97, 98, 99, 100], but it does not affect the present discussion.) The predictions of the model of Ref. [102] were compared with numerical pro- files of membrane thickness close to a mismatched protein [99, 100, 103], and with experimental data regarding gramicidin lifetime [93] and formation rate [94]. This analysis yielded consistent results for the term stemming from the gradient of the area per molecule, and its order of magnitude was found to be of order of the con- tribution of the interfacial tension between water and the hydrophobic part of the membrane. In addition, the presence of this new term allowed to explain for the first time a systematic dependence in previous numerical data. 3.1.4 Isotropic cross-section The first models of short-range interactions between transmembrane proteins as- sumed that the proteins are coupled to a local order parameter describing the inter- nal state of the membrane, either the conformational/chain-packing properties of the lipids, or the bilayer thickness u [86, 87]. Both are equivalent for a fully incompress- ible membrane hydrophobic core. In Refs. [88, 104], a generic Landau -- Ginzburg expansion of the free energy density in terms of u and its first gradient was used to investigate the energy of a hexagonal lattice of embedded proteins imposing a value u0 of the order parameter, i.e., a fixed hydrophobic mismatch, on their edge. Approximating the Wigner-Seitz cell of the lattice by a circle, which yields cylindri- Membrane-mediated interactions 27 cal symmetry, the authors derived a monotonically attractive short-range interaction caused by the overlap of the membrane regions deformed by the inclusions. As discussed in Sec. 3.1, several models based on the thickness order parame- ter u have been developed. They have been used to study membrane-mediated in- teractions. These models essentially introduced terms involving the second-order derivative of u, based on the (recently questioned [102]) expectation that the term proportional to (∇∇∇u)2 was negligible. In particular, Ref. [95] introduced a term pro- portional to (∇2u)2, by analogy with the splay term for smectic liquid crystals. Later, Ref. [97] introduced additional terms, linear in ∇2u and in u∇2u, which arise from the spontaneous curvature of the monolayers and its dependence on the area per lipid. This initiated a series of works [97, 105, 98] aiming to estimate the elastic energy of a hexagonal lattice of proteins with hydrophobic mismatch. These works showed that the interaction potential can be non-monotonous, with short-distance repulsion and a minimum energy at finite separation. These effects can arise from the spontaneous curvature term, but also from a fixed contact angle between the membrane hydrophobic-hydrophilic interface and the inclusion, thereafter referred to as "slope". The associated multi-body effects were investigated in Ref. [106] through a Monte-Carlo simulation of inclusions fixing both the membrane thick- ness and its slope, in a membrane described by the elastic energy in Eq. 12. This study also demonstrated the interest of the structure factor to test the models. An- other term involving second-order derivatives of the thickness profile u, proportional to its Gaussian curvature, was included in Ref. [100], improving the agreement with coarse-grained molecular-dynamics numerical simulations. Note that oscillations in the interaction potential were observed in the coarse-grained molecular-dynamics simulations of Ref. [107]. The term proportional to (∇∇∇u)2 in the elastic energy density was originally dis- carded on the grounds that it originates from a negligible microscopic surface ten- sion assimilated to that of a Plateau border [95]. However, it was recently shown by us to also originate form gradients of lipid density, and therefore to contribute significantly to the elastic Hamiltonian [102]. Note in addition that the term in u∇2u introduced in Refs. [97, 98] contributes to the (∇∇∇u)2 term once integrated by parts. In the end, these models converge towards the most general quadratic expansion in terms of u and its first and second-order derivatives [102, 108]. In standard statis- tical field theory, it is justified to neglect higher-order gradients, because the focus is on large-scale physics and the coarse-graining length is much larger than the range of the microscopic interactions [109]. However, here, such arguments do not hold since the distortions around proteins relax on a length comparable with the bilayer thickness. Therefore, in practice, one should rather rely on comparison with exper- iments and simulations to determine how many terms to include in the expansion. Our current understanding is that all linear and quadratic terms involving derivatives of u up to second order should contribute, and that the best strategy is to try to fit the parameters of the elastic Hamiltonian and of the protein -- membrane coupling using experimental or numerical data [108]. The focus of this chapter is on membrane-mediated interactions arising from di- rect constraints on the membrane shape (mean shape and thickness). Hence, we will 28 Anne-Florence Bitbol, Doru Constantin and Jean-Baptiste Fournier not discuss in detail the role of the underlying lipid tilt degree of freedom [101] in membrane-mediated interactions [110, 111, 112, 113, 114, 115]. However, tilt cer- tainly plays a part in these interactions. For instance, proteins with no hydrophobic mismatch but with a hour-glass shape [110, 111] may induce a membrane deforma- tion due to the boundary conditions they impose on lipid tilt. A legitimate question, though, is how necessary it is to include this degree of freedom. Statistical physics allows to integrate out virtually any degree of freedom [109]. The resulting effec- tive elasticity for the remaining degrees of freedom takes into account the underlying distortion energy of the removed ones. For instance, integrating out the tilt degree of freedom in the presence of an hour-glass shaped inclusion would produce an effec- tive boundary energy depending on the inclusion thickness and on its angle with the membrane. What is not clear is how many orders in the derivatives of u, both in the bulk and in the boundary energy, one would have to introduce in order to properly account for the removed degrees of freedom. Future works in this direction could be interesting. 3.1.5 Anisotropic cross-section While most theoretical studies of short-range membrane-mediated interactions have considered cylinder-shaped inclusions, actual membrane proteins have various shapes. As in the case of long-range interactions, in-plane anisotropy may result in direc- tional membrane-mediated interactions, which may impact the formation of multi- protein complexes. In Ref. [116], an analytical method was developed to study membrane-mediated interactions between in-plane anisotropic mismatched inclusions. The effective Hamiltonian H associated to membrane thickness deformations was expressed as (cid:90) (cid:26) Ka 2d2 0 H = dxdy u2 + γ (cid:20) u d0 (cid:21) (cid:0)∇2u(cid:1)2(cid:27) + (∇∇∇u)2 8 + κ 8 , (17) where we have used the notations defined in Eq. 12. This model is based on that of Ref. [95] (see Eq. 12), but includes an additional "tension" term in u/d0. Such a term is also included in Ref. [102] (see Eq. 14), but without the assumption that its prefactor is related to that of the squared thickness gradient term. This assumption should be viewed as a simplifying hypothesis, given the contribution of monolayer curvature to the squared thickness gradient term [97, 98, 99, 100] and the difference between externally applied tension and interfacial tension [102] (see Sec. 3.1.3). In Ref. [116], the solution of the Euler-Lagrange equation associated with Eq. 17 in the case of a single cylinder-shaped inclusion was expressed using Fourier-Bessel series. Then, using an ansatz introduced in Ref. [41] in the context of long-range membrane-mediated interactions, the ground-state shape of the membrane in the presence of two inclusions was written as a sum of two such series. The coefficients of the successive terms of these series can be chosen in order to match the boundary conditions imposed by both inclusions, using expansions in a/d < 1. Membrane-mediated interactions 29 This method was extended to weakly anisotropic inclusions, modeling mechanosen- sitive channels of large conductance (MscL) in Ref. [116]. The in-plane cross- section of these pentameric proteins was described as a circle perturbed by a small- amplitude sinusoidal, with fifth-oder symmetry. Boundary conditions along the edge of these proteins were expressed perturbatively in the amplitude of the sinusoidal, allowing to use the method described above. The resulting anisotropic membrane- mediated interaction features an energy barrier to dimerization, and demonstrates that the tip-on orientation is more favorable than the face-on one, except at very short distances. Gating of the MscL channel was also studied in Ref. [116], by modeling open and closed channels as having different diameters and hydrophobic thicknesses [117]. The impact of having different oligomeric states of MscL on these interactions and on gating by tension (see Ref. [118]) was studied in Ref. [119]. The method developed in Ref. [116] was used in Ref. [120] to study the effect of membrane-mediated interactions on the self-assembly and architecture of bacterial chemoreceptor lattices. Chemotaxis enables bacteria to perform directed motion in gradients of chemicals. The chemoreceptors that bind to these chemicals are trans- membrane proteins that organize into large honeycomb lattices of trimers of dimers at the poles of bacteria [121]. In Ref. [120], it was shown that membrane-mediated interactions between chemoreceptor trimers of dimers, modeled as inclusions with three-fold symmetry, correctly predict the structure of the arrays observed in exper- iments. Indeed, at short distances, the face-on relative orientation of the trimers is favored by these anisotropic interactions. In addition, the collective structure of the honeycomb lattice, studied approximately through the pairwise nearest-neighbor in- teractions, was shown to be more favorable than other types of aggregates at realistic densities of proteins. Gateway states to this lattice were also predicted, and it was shown that membrane-mediated interactions may contribute to the cooperativity of chemotactic signaling. 3.2 Numerical studies at the microscopic scale Continuum models account for the microscopic degrees of freedom (i.e. the po- sitions and conformations of all molecules involved) in a coarse-grained way, via effective terms in the elastic energy and the associated prefactors. However, even in the absence of a mesoscopic deformation due to hydrophobic mismatch, the pres- ence of an inclusion constrains the configurations accessible to the lipid chains that surround it [86, 122, 123, 112]. Further insight can thus be gained by treating such microscopic degrees of freedom explicitly, in particular those describing the con- formation of the lipid chains. Recent advances in numerical simulations have made such approaches possible. Here, we give a brief overview of such studies. Note that numerical studies focusing on larger-scale features were mentioned earlier. Refs. [123, 124] used the lateral density-density response function of the alkyl chains, obtained by molecular dynamics simulations of lipid bilayers, to determine the interaction between "smooth" (no anchoring) hard cylinders inserted into the bi- 30 Anne-Florence Bitbol, Doru Constantin and Jean-Baptiste Fournier layer. Three values were considered for the cylinder radius. For the largest one (9 A, comparable to that of the gramicidin pore, for instance), the long-range interaction is repulsive for all the lipids studied (DMPC, DPPC, POPC and DOPC), with an additional short-range attraction for DMPC. This study does not discuss how the interaction might vary with the concentration of inclusions. Other studies followed suit [125, 126, 127, 128]. A complete description should in principle combine the effects of hydrophobic mismatch and of these changes in chain order [129, 114]. Such a complete model is currently lacking, due to the theoretical difficulties but also due to the dearth of experimental data that could be used to test and validate it. As in the case of lipid tilt (see Section 3.1.4), one can wonder how integrating out these underlying degrees of freedom would affect an effective model written in terms of u, what effective boundary conditions non-mismatched inclusions would then impose, and whether such a model would be sufficient. 3.3 Experimental studies It has proven very difficult to directly measure the interactions between membrane inclusions. 3.3.1 Electron microscopy First among such attempts were freeze-fracture electron microscopy (FFEM) studies [130, 131, 132, 133] that analyzed the spatial distribution of inclusions to determine their radial distribution function g(r). The data was then described using liquid state theories [134, 135, 136] in terms of a hard-core model with an additional interaction, either repulsive or attractive depending on the system. These pioneering results were not followed by more systematic investigations, probably due to the intrinsic difficulty of the technique. It is also very difficult to check whether the distribution function observed in the sample after freezing still corresponds to that at thermal equilibrium. 3.3.2 Atomic force microscopy It has been known for a long time that atomic force microscopy (AFM) can resolve lateral structures down to the nanometer scale [137], but data acquisition used to be relatively slow. This changed with the introduction of high-speed AFM [138], which allows taking "snapshots" of the system and determining the radial distribu- tion function. The latter gives access to [OK?] the interaction potential between in- clusions, as illustrated by Ref. [139] for ATP-synthase c-rings in purple membranes (see Fig. 9). Membrane-mediated interactions 31 Fig. 9: Interaction between ATP synthase c-rings. (A) Histogram of the center-to- center distance of c-rings. (B) Membrane-mediated two-protein interaction energy landscape. Reprinted from Figure 2 of reference [139]. 3.3.3 Small-angle scattering A promising way of studying membrane-mediated interactions is through small- angle radiation (X-ray or neutrons) scattering from oriented samples, as demon- strated by Refs. [140, 141, 142]. This non-invasive technique is very well adapted to measurements of membrane-mediated interactions since the wavelength used is of the same order of magnitude as the typical length scales over which one must probe the system (nanometers). One can thus measure the structure factor of the two-dimensional system formed by the inclusions in the membrane and obtain the interaction potential between them. This strategy was recently used to study alamethicin pores in DMPC membranes [143], inorganic particles contained in bilayers of a synthetic surfactant [144, 145] and gramicidin pores in several types of membranes [146]. 4 Conclusion Membrane-mediated interactions between inclusions constitute a very rich topic. Their study gives insight into the behavior of complex two-dimensional biolog- ical membranes. In particular, these interactions may have important impacts on membrane protein aggregation, and on the formation of specific biologically func- tional assemblies. Interestingly, inclusions can also serve as membrane probes, since membrane-mediated interactions are in part determined by the properties of the host membrane. The field of long-range membrane-mediated interactions has been dominated by theory, yielding interesting theoretical developments such as the fluctuation-induced interaction, the general effective field theory and scattering approaches, and the questions currently raised by the dynamics of these interactions. Some experimen- 32 Anne-Florence Bitbol, Doru Constantin and Jean-Baptiste Fournier Fig. 10: Interaction potential U(r) of BuSn12 particles within DDAO bilayers. The lower curve is the interaction potential of the particles in ethanol. The solid vertical line marks the hard core interaction with radius 4.5 A. Reprinted from Figure 3 of reference [144]. tal and numerical studies have enriched this field, and we hope for further progress allowing for more quantitative comparison with theory. The study of short-range membrane thickness deformations was motivated by quantitative experiments on gramicidin. Work on these deformations and on the as- sociated membrane-mediated interactions has led to several developments of the theoretical description of membrane elasticity at the nanoscale. Importantly, the small-scale deformations involved are at the limit of the domain of validity of stan- dard coarse-grained continuum theories, making comparison to precise experimen- tal and numerical data even more crucially important. An interesting fundamental feature of membrane-mediated interactions is the ex- istence of many-body effects, arising from the interplay of the deformations caused by each of the inclusions. It would thus be particularly interesting to vary the con- centration of inclusions in experiments. Acknowledgments A.-F. B. acknowledges support from the Human Frontier Science Program. D. C. and J.-B. F. acknowledge support by ANR Grant MEMINT (2012-BS04-0023). 2Rr - 2RMembrane-mediated interactions References 33 1. S. J. Singer and G. L. Nicolson. The fluid mosaic model of the structure of cell membranes. Science, 175:720 -- 31, 1972. 2. E. Sackmann. Physical basis for trigger processes and membrane structures. In D. Chapman, editor, Biological Membranes, volume 5, pages 105 -- 143. Academic Press, London, 1984. 3. M. Ø Jensen and O. G. Mouritsen. Lipids do influence protein function -- the hydrophobic matching hypothesis revisited. Biochim. Biophys. Acta-Biomembranes, 1666:205 -- 226, 2004. 4. W. Helfrich. Elastic properties of lipid bilayers -- Theory and possible experiments. Zeitschrift fur Naturforschung C -- Journal of Biosciences, 28:693 -- 703, 1973. 5. M. Goulian, R. Bruinsma, and P. Pincus. Long-range forces in heterogeneous fluid mem- branes. EPL, 22(2):145 -- 150, 1993. Erratum: Europhysics Letters 23(2):155, 1993. Com- ment: Ref. [7]. A factor 2 in the fluctuation-induced force was corrected in Ref. [8]. 6. J.-B. Fournier and P. Galatola. Higher-order power series expansion of the elastic interaction between conical membrane inclusions. Eur. Phys. J. E, 38:86, Aug 2015. 7. J.-B. Fournier and P. G. Dommersnes. Comment on "Long-range forces in heterogeneous fluid membranes". EPL, 39(6):681 -- 682, 1997. 8. R. Golestanian, M. Goulian, and M. Kardar. Fluctuation-induced interactions between rods on a membrane. Phys. Rev. E, 54(6):6725 -- 6734, 1996. 9. J. M. Park and T. C. Lubensky. Interactions between membrane inclusions on fluctuating membranes. J. Phys. I, 6(9):1217 -- 1235, 1996. 10. R. R. Netz. Inclusions in fluctuating membranes: exact results. Journal de Physique I -- France, 7:833 -- 852, 1997. 11. K. S. Kim, J. Neu, and G. Oster. Curvature-mediated interactions between membrane pro- teins. Biophys. J., 75(5):2274 -- 2291, Nov 1998. 12. P. G. Dommersnes and J.-B. Fournier. N-body study of anisotropic membrane inclusions: Membrane mediated interactions and ordered aggregation. Eur. Phys. J. B, 12(1):9 -- 12, 1999. 13. P. G. Dommersnes and J.-B. Fournier. Casimir and mean-field interactions between mem- brane inclusions subject to external torques. EPL, 46(2):256 -- 261, 1999. 14. C. Yolcu, I. Z. Rothstein, and M. Deserno. Effective field theory approach to Casimir inter- actions on soft matter surfaces. EPL, 96:20003, 2011. 15. C. Yolcu and M. Deserno. Membrane-mediated interactions between rigid inclusions: an effective field theory. Phys. Rev. E, 86:031906, 2012. 16. C. Yolcu, I. Z. Rothstein, and M. Deserno. Effective field theory approach to fluctuation- induced forces between colloids at an interface. Phys. Rev. E, 85:011140, 2012. 17. C. Yolcu and M. Deserno. The Effective Field Theory approach towards membrane-mediated interactions between particles. Adv. Colloid Interface Sci., 208:89 -- 109, 2014. 18. R. Golestanian. Reduced persistence length and fluctuation-induced interactions of directed semiflexible polymers on fluctuating surfaces. EPL, 36:557, 1996. 19. H. B. G. Casimir. On the attraction between two perfectly conducting plates. Proc. K. Ned. 20. P. W. Milonni. The Quantum Vacuum -- An Introduction to Quantum Electrodynamics. Aca- Akad. Wet., 51(7):793 -- 796, 1948. demic Press, 1994. 21. W. Helfrich and T.R. Weikl. Two direct methods to calculate fluctuation forces between rigid objects embedded in fluid membranes. Eur. Phys. J. E, 5:423 -- 439, 2001. 22. M. Kardar and R. Golestanian. The "friction" of vacuum, and other fluctuation-induced forces. Rev. Mod. Phys., 71(4):1233 -- 1245, 1999. 23. A. Gambassi. The Casimir effect: from quantum to critical fluctuations. JPCS, 161:012037, 2009. 24. M. E. Fisher and P. G. de Gennes. Wall phenomena in a critical binary mixture. C. R. S´eances Acad. Sci. - Ser. B, 287(8):207 -- 209, 1978. 25. C. Hertlein, L. Helden, A. Gambassi, S. Dietrich, and C. Bechinger. Direct measurement of critical Casimir forces. Nature, 451(7175):172 -- 175, 2008. 34 Anne-Florence Bitbol, Doru Constantin and Jean-Baptiste Fournier 26. B. B. Machta, S. L. Veatch, and J. P. Sethna. Critical casimir forces in cellular membranes. Phys. Rev. Lett., 109:138101, Sep 2012. 27. R. Golestanian, M. Goulian, and M. Kardar. Fluctuation-induced interactions between rods on membranes and interfaces. EPL, 33(3):241 -- 245, 1996. 28. D. Stamou, C. Duschl, and D. Johannsmann. Long-range attraction between colloidal spheres at the air-water interface: The consequence of an irregular meniscus. Phys. Rev. E, 62(4, B):5263 -- 5272, Oct 2000. 29. P. G. Dommersnes and J.-B. Fournier. The many-body problem for anisotropic membrane in- clusions and the self-assembly of "saddle" defects into an "egg carton". Biophys. J., 83:2898 -- 2905, 2002. 30. B. M. Axilrod and E. Teller. Interaction of the van der Waals type between three atoms. J. Chem. Phys., 11(6):299 -- 300, 1943. 31. M. Simunovic, A. Srivastava, and G. A. Voth. Linear aggregation of proteins on the mem- brane as a prelude to membrane remodeling. Proc. Natl. Acad. Sci. USA, 110(51):20396 -- 20401, Dec 2013. 32. N. Destainville. Cluster phases of membrane proteins. Phys. Rev. E, 77:011905, 1998. 33. T. R. Weikl. Fluctuation-induced aggregation of rigid membrane inclusions. EPL, 54:547 -- 553, 2001. 34. W. Rawicz, K. C. Olbrich, T. McIntosh, D. Needham, and E. Evans. Effect of chain length and unsaturation on elasticity of lipid bilayers. Biophys. J., 79:328 -- 339, 2000. 35. H. Lehle and M. Oettel. Importance of boundary conditions for fluctuation-induced forces between colloids at interfaces. Phys. Rev. E, 75:011602, 2007. 36. E. Noruzifar and M. Oettel. Anisotropies in thermal Casimir interactions: Ellipsoidal colloids trapped at a fluid interface. Phys. Rev. E, 79(5):051401, 2009. 37. H.-K. Lin, R. Zandi, U. Mohideen, and L. P Pryadko. Fluctuation-induced forces between inclusions in a fluid membrane under tension. Phys. Rev. Lett., 107:228104, 2011. 38. C. Genet, A. Lambrecht, and S. Reynaud. Casimir force and the quantum theory of lossy optical cavities. Phys. Rev. A, 67(4): 043811, Apr 2003. 39. T. Emig, N. Graham, R. L. Jaffe, and M. Kardar. Casimir forces between compact objects: The scalar case. Phys. Rev. D, 77(2):025005, Jan 2008. 40. S. J. Rahi, T. Emig, N. Graham, R. L. Jaffe, and Kardar M. Scattering theory approach to electrodynamic Casimir forces. Phys. Rev. D, 80:085021, 2009. 41. T. R. Weikl, M. M. Kozlov, and W. Helfrich. Interaction of conical membrane inclusions: Effect of lateral tension. Phys. Rev. E, 57:6988 -- 6995, 1998. 42. AR Evans, MS Turner, and P Sens. Interactions between proteins bound to biomembranes. Phys. Rev. E, 67:041907, Apr 2003. 43. M. Simunovic and G. A. Voth. Membrane tension controls the assembly of curvature- generating proteins. Nat. Commun., 6:7219, May 2015. 44. A.-F. Bitbol, P. G. Dommersnes, and J.-B. Fournier. Fluctuations of the Casimir-like force between two membrane inclusions. Phys. Rev. E, 81:050903(R), 2010. 45. R. Capovilla and J. Guven. Stresses in lipid membranes. J. Phys. A, 35(30):6233 -- 6247, 2002. 888, 2007. 46. J.-B. Fournier. On the stress and torque tensors in fluid membranes. Soft Matter, 3(7):883 -- 47. G. Barton. On the fluctuations of the Casimir force. J. Phys. A, 24(5):991 -- 1005, 1991. 48. D. Bartolo, A. Ajdari, J.-B. Fournier, and R. Golestanian. Fluctuations of fluctuation-induced Casimir-like forces. Phys. Rev. Lett., 89(23):230601, 2002. 49. A.-F. Bitbol and J.-B. Fournier. Forces exerted by a correlated fluid on embedded inclusions. Phys. Rev. E, 83:061107, 2011. 50. J.-B. Fournier. Dynamics of the force exchanged between membrane inclusions. Phys. Rev. Lett., 112:128101, Mar 2014. Erratum: Phys. Rev. Lett., 114:219901, 2015. 51. U. Seifert and S. A. Langer. Viscous modes of fluid bilayer membranes. EPL, 23:71 -- 76, 52. E. Evans and A. Yeung. Hidden dynamics in rapid changes of bilayer shape. Chem. Phys. 1993. Lipids, 73:39 -- 56, 1994. 54. A.-F. Bitbol, L. Peliti, and J.-B. Fournier. Membrane stress tensor in the presence of lipid density and composition inhomogeneities. Eur. Phys. J. E, 34:53, 2011. 55. P. G. Dommersnes, J.-B. Fournier, and Galatola P. Long-range elastic forces between mem- brane inclusions in spherical vesicles. EPL, 42:233 -- 238, 1998. 56. R. Podgornik. Orientational ordering of polymers on a fluctuating flexible surface. Phys. Rev. E, 52:5170 -- 5177, 1995. 57. H. Li and M. Kardar. Fluctuation-induced forces between manifolds immersed in correlated fluids. Phys. Rev. A, 46(10):6490 -- 6500, 1992. 58. A.-F. Bitbol, K. Sin Ronia, and J.-B. Fournier. Universal amplitudes of the Casimir-like interactions between four types of rods in fluid membranes. EPL, 96:40013, 2011. 59. K. Sin Ronia and J.-B. Fournier. Universality in the point discretization method for calculat- ing Casimir interactions with classical Gaussian fields. EPL, 100(3), Nov 2012. 60. B. V. Derjaguin. Analysis of friction and adhesion -- IV. The theory of the adhesion of small particles. Kolloid Zeitschrift, 69:155, 1934. 61. T. R. Weikl. Indirect interactions of membrane-adsorbed cylinders. Eur. Phys. J. E, 12:265 -- 62. M. M. Muller, M. Deserno, and J. Guven. Interface-mediated interactions between particles: A geometrical approach. Phys. Rev. E, 72:061407, 2005. 63. M. M. Muller, M. Deserno, and J. Guven. Geometry of surface-mediated interactions. EPL, 273, 2003. 69:482 -- 488, 2005. Membrane-mediated interactions 35 53. J.-B. Fournier. On the hydrodynamics of bilayer membranes. Int. J. Nonlinear Mech., 75:67 -- 76, Oct 2015. 64. M. M. Muller and M. Deserno. Balancing torques in membrane-mediated interactions: Exact results and numerical illustrations. Phys. Rev. E, 76:011921, 2007. 65. Sergey Mkrtchyan, Christopher Ing, and Jeff Z. Y. Chen. Adhesion of cylindrical colloids to the surface of a membrane. Phys. Rev. E, 81:011904, Jan 2010. 66. P. Gosselin, H. Mohrbach, and M. M. Muller. Interface-mediated interactions: Entropic forces of curved membranes. Phys. Rev. E, 83:051921, 2011. 67. B. J. Reynwar, G. Illya, V. A. Harmandaris, M. M. Mueller, K. Kremer, and M. Deserno. Ag- gregation and vesiculation of membrane proteins by curvature-mediated interactions. Nature, 447:461 -- 464, 2007. 68. H. Yuan, C. Huang, and S. Zhang. Membrane-Mediated Inter-Domain Interactions. Bio- NanoSci., 1:97 -- 102, Jun 2011. 69. B. J. Reynwar and M. Deserno. Membrane-mediated interactions between circular particles in the strongly curved regime. Soft Matter, 7:8567, 2011. 70. Y. Schweitzer and M. M. Kozlov. Membrane-Mediated Interaction between Strongly Anisotropic Protein Scaffolds. PLOS Comp. Biol., 11(2), Feb 2015. 71. A. H. Bahrami, R. Lipowsky, and T. R. Weikl. Tubulation and aggregation of spherical nanoparticles adsorbed on vesicles. Phys. Rev. Lett., 109(18):188102, Nov 2012. 72. A. Sari´c and A. Cacciuto. Self-assembly of nanoparticles adsorbed on fluid and elastic mem- branes. Soft Matter, 9:6677 -- 6695, 2013. 73. N. Ramakrishnan, P. B. Sunil Kumar, and J. H. Ipsen. Membrane-mediated aggregation of curvature-inducing nematogens and membrane tubulation. Biophys. J., 104(5):1018 -- 1028, Mar 2013. 74. Winfried Romer, Ludwig Berland, Val´erie Chambon, Katharina Gaus, Barbara Windschiegl, Dani`ele Tenza, Mohamed R. E. Aly, Vincent Fraisier, Jean-Claude Florent, David Perrais, Christophe Lamaze, Grac¸c¸a Raposo, Claudia Steinem, Pierre Sens, Patricia Bassereau, and Ludger Johannes. Shiga toxin induces tubular membrane invaginations for its uptake into cells. Nature, 450:670 -- 675, 2007. 75. Benedict J. Reynwar, Gregoria Illya, Vagelis A. Harmandaris, Martin M.Mueller, Kurt Kre- mer, and Markus Deserno. Aggregation and vesiculation of membrane proteins by curvature- mediated interactions. Nature, 447:461 -- 464, 2007. 76. H. T. McMahon and J. L. Gallop. Membrane curvature and mechanisms of dynamic cell membrane remodelling. Nature, 438:590 -- 596, 2005. 36 Anne-Florence Bitbol, Doru Constantin and Jean-Baptiste Fournier 77. M. I. Angelova and D. S. Dimitrov. Liposome electroformation. Faraday Discuss. Chem. Soc., 81:303 -- 311, 1986. 78. S. Semrau, T. Idema, T. Schmidt, and C. Storm. Membrane-mediated interactions measured using membrane domains. Biophys. J., 96:4906 -- 4915, 2009. 79. I. Koltover, J. O. Radler, and C. R. Safinya. Membrane mediated attraction and ordered aggregation of colloidal particles bound to giant phospholipid vesicles. Phys. Rev. Lett., 82:1991 -- 1994, 1999. 80. T. Yue and X. Zhang. Cooperative Effect in Receptor-Mediated Endocytosis of Multiple Nanoparticles. ACS Nano, 6(4):3196 -- 3205, Apr 2012. 81. A. Saric and A. Cacciuto. Fluid Membranes Can Drive Linear Aggregation of Adsorbed Spherical Nanoparticles. Phys. Rev. Lett., 108(11):118101, Mar 2012. 82. Sharon Rozovsky, Yoshihisa Kaizuka, and Jay T. Groves. Formation and spatio-temporal evolution of periodic structures in lipid bilayers. Journal of the American Chemical Society, 127(1):36 -- 37, 2005. 83. Tristan S. Ursell, William S. Klug, and Rob Phillips. Morphology and interaction between lipid domains. Proceedings of the National Academy of Sciences, 106(32):13301 -- 13306, 2009. 84. Frederick A. Heberle, Robin S. Petruzielo, Jianjun Pan, Paul Drazba, Norbert Kucerka, Robert F. Standaert, Gerald W. Feigenson, and John Katsaras. Bilayer Thickness Mismatch Controls Domain Size in Model Membranes. Journal of the American Chemical Society, 135(18):6853 -- 6859, 2013. 85. P. C. Jost, O. H. Griffiths, R. A. Capaldi, and G. Vanderkooi. Evidence for boundary lipid in membranes. Proc. Natl. Acad. Sci. USA, 70:480 -- 484, 1973. 86. S. Marcelja. Lipid-mediated protein interaction in membranes. Biochim. Biophys. Acta: Biomembr., 455:1 -- 7, 1976. 87. H. Schroder. Aggregation of proteins in membranes. An example of fluctuation induced interactions in liquid crystals. J. Chem. Phys., 67:1617, 1977. 88. J. C. Owicki, M. W. Springgate, and H. McConnell. Theoretical study of protein-lipid inter- actions in bilayer membranes. Proc. Natl. Acad. Sci. USA, 75:1616 -- 1619, 1978. 89. J. A. Killian. Hydrophobic mismatch between proteins and lipids in membranes. Biochim. Biophys. Acta: Biomembr., 1376:401 -- 416, 1998. 90. O. G. Mouritsen and M. Bloom. Mattress model of lipid-protein interactions in membranes. Biophys. J., 46:141 -- 153, 1984. 91. D. A. Kelkar and A. Chattopadhyay. The gramicidin ion channel: A model membrane pro- tein. Biochim. Biophys. Acta: Biomembr., 1768:2011 -- 2025, 2007. 92. H. A. Kolb and E. Bamberg. Influence of membrane thickness and ion concentration on the properties of the gramicidin A channel: autocorrelation, spectral power density, relaxation and single-channel studies. Biochim. Biophys. Acta: Biomembr., 464:127 -- 141, 1977. 93. J. R. Elliott, D. Needham, J. P. Dilger, and D. A. Haydon. The effects of bilayer thickness and tension on gramicidin single-channel lifetime. Biochim. Biophys. Acta: Biomembr., 735:95 -- 103, 1983. 94. M. Goulian, O. N. Mesquita, D. K. Fygenson, C. Nielsen, O. S. Andersen, and A. Libchaber. Gramicidin channel kinetics under tension. Biophys. J., 74:328 -- 337, 1998. 95. H. W. Huang. Deformation free energy of bilayer membrane and its effect on gramicidin channel lifetime. Biophys. J., 50:1061 -- 1070, 1986. 96. P. G. de Gennes. The Physics of Liquid Crystals. Clarendon Press, Oxford, 1974. 97. N. Dan, P. Pincus, and S. A. Safran. Membrane-induced interactions between inclusions. Langmuir, 9:2768 -- 2771, 1993. 98. H. Aranda-Espinoza, A. Berman, N. Dan, P. Pincus, and S. Safran. Interaction between inclusions embedded in membranes. Biophys. J., 71:648, 1996. 99. G. Brannigan and F. L. H. Brown. A consistent model for thermal fluctuations and protein- induced deformations in lipid bilayers. Biophys. J., 90(5):1501 -- 1520, 2006. 100. G. Brannigan and F. L. H. Brown. Contributions of gaussian curvature and nonconstant lipid volume to protein deformation of lipid bilayers. Biophys. J., 92(3):864 -- 876, 2007. Membrane-mediated interactions 37 101. M. C. Watson, E. S. Penev, P. M. Welch, and F. L. H. Brown. Thermal fluctuations in shape, thickness, and molecular orientation in lipid bilayers. J. Chem. Phys., 135:244701, 2011. 102. A.-F. Bitbol, D. Constantin, and J.-B. Fournier. Bilayer elasticity at the nanoscale: the need for new terms. PLoS ONE, 7(11):e48306, 2012. 103. B. West, F. L. H. Brown, and F. Schmid. Membrane-protein interactions in a generic coarse- grained model for lipid bilayers. Biophys. J., 96(1):101 -- 115, 2009. 104. J. C. Owicki and H. McConnell. Theory of protein-lipid and protein-protein interactions in bilayer membranes. Proc. Natl. Acad. Sci. USA, 76:4750 -- 4754, 1979. 105. N. Dan, A. Berman, P. Pincus, and S. A. Safran. Membrane-induced interactions between inclusions. J. Phys. II France, 4:1713 -- 1725, 1994. 106. T. A. Harroun, W. T. Heller, T. M. Weiss, L. Yang, and H. W. Huang. Theoretical analysis of hydrophobic matching and membrane-mediated interactions in lipid bilayers containing gramicidin. Biophys. J., 76(6):3176 -- 3185, Jun 1999. 107. J. Neder, B. West, P. Nielaba, and F. Schmid. Membrane-mediated Protein-protein Interac- tion: A Monte Carlo Study. Curr. Nanosci., 7(5):656 -- 666, Oct 2011. 108. F. Bories, D. Constantin, J.-B. Fournier, and P. Galatola. To be published. 109. P. M. Chaikin and T. C. Lubensky. Principles of condensed matter physics. Cambridge University Press, 1995. 110. J.-B. Fournier. Coupling between membrane tilt-difference and dilation: A new "ripple" instability and multiple crystalline inclusions phases. EPL, 43(6):725 -- 730, 1998. 111. S. May and A. Ben-Shaul. Molecular theory of lipid-protein interaction and the Lα-H transition. Biophys. J., 76:751 -- 767, 1999. 112. S. May and A. Ben-Shaul. A molecular model for lipid-mediated interaction between pro- teins in membranes. Phys. Chem. Chem. Phys., 2:4494 -- 4502, 2000. 113. S May. Membrane perturbations induced by integral proteins: Role of conformational re- strictions of the lipid chains. Langmuir, 18:6356 -- 6364, Aug 2002. 114. Klemen Bohinc, Veronika Kralj-Iglic, and Silvio May. Interaction between two cylindrical inclusions in a symmetric lipid bilayer. J. Chem. Phys., 119:7435 -- 7444, 2003. 115. Y. Kozlovsky, J. Zimmerberg, and M. M. Kozlov. Orientation and interaction of oblique cylindrical inclusions embedded in a lipid monolayer: A theoretical model for viral fusion peptides. Biophys. J., 87:999 -- 1012, Aug 2004. 116. C. A. Haselwandter and R. Phillips. Directional interactions and cooperativity between mechanosensitive membrane proteins. EPL, 101(6):68002, 2013. 117. T. Ursell, K. C. Huang, E. Peterson, and R. Phillips. Cooperative gating and spatial organiza- tion of membrane proteins through elastic interactions. PLoS Comput. Biol., 3(5):803 -- 812, May 2007. 118. R. Phillips, T. Ursell, P. Wiggins, and P. Sens. Emerging roles for lipids in shaping membrane-protein function. Nature, 459(7245):379 -- 385, May 2009. 119. Osman Kahraman, William S. Klug, and Christoph A. Haselwandter. Signatures of protein structure in the cooperative gating of mechanosensitive ion channels. EPL, 107(4), Aug 2014. 120. C. A. Haselwandter and N. S. Wingreen. The role of membrane-mediated interactions in the assembly and architecture of chemoreceptor lattices. PLoS Comput. Biol., 10(12):e1003932, Dec 2014. 121. A. Briegel, X. Li, A. M. Bilwes, K. T. Hughes, G. J. Jensen, and B. R. Crane. Bacterial chemoreceptor arrays are hexagonally packed trimers of receptor dimers networked by rings of kinase and coupling proteins. Proc. Natl. Acad. Sci. USA, 109(10):3766 -- 3771, Mar 2012. 122. T. Sintes and A. Baumgartner. Protein attraction in membranes induced by lipid fluctuations. Biophys. J., 73:2251 -- 2259, 1997. 123. Patrick Lague, Martin J. Zuckermann, and Benoıt Roux. Lipid-mediated interactions be- tween intrinsic membrane proteins: A theoretical study based on integral equations. Biophys. J., 79:2867 -- 2879, 2000. 124. Patrick Lague, Martin J. Zuckermann, and Benoıt Roux. Lipid-mediated interactions be- tween intrinsic membrane proteins: Dependence on protein size and lipid composition. Bio- phys. J., 81:276 -- 284, 2001. 38 Anne-Florence Bitbol, Doru Constantin and Jean-Baptiste Fournier 125. Lorant Janosi, Anupam Prakash, and Manolis Doxastakis. Lipid-Modulated Sequence- Specific Association of Glycophorin A in Membranes. Biophysical Journal, 99(1):284 -- 292, 2010. 126. Richard A. Kik, Frans A. M. Leermakers, and J. Mieke Kleijn. Molecular modeling of pro- teinlike inclusions in lipid bilayers: Lipid-mediated interactions. Physical Review E, 81(2), 2010. 127. Jejoong Yoo and Qiang Cui. Membrane-mediated protein-protein interactions and connec- tion to elastic models: A coarse-grained simulation analysis of gramicidin A association. Biophysical journal, 104:128 -- 138, January 2013. 128. Thomas A. Dunton, Joseph E. Goose, David J. Gavaghan, Mark S. P. Sansom, and James M. Osborne. The Free Energy Landscape of Dimerization of a Membrane Protein, NanC. PLoS Computational Biology, 10(1):e1003417, 2014. 129. S. Marcelja. Toward a realistic theory of the interaction of membrane inclusions. Biophys. J., 76:593 -- 594, 1999. 130. B. A. Lewis and D. M. Engelman. Bacteriorhodopsin remains dispersed in fluid phospholipid bilayers over a wide range of bilayer thicknesses. J. Mol. Biol., 166:203 -- 210, 1983. 131. Y. S. Chen and W. L. Hubbell. Temperature- and light-dependent structural changes in rhodopsin-lipid membranes. Exp. Eye Res, 17:517 -- 532, 1973. 132. R. James and D. Branton. Lipid- and temperature-dependent structural changes in Achole- plasma laidlawii cell membranes. Biochim. Biophys. Acta, 323:378 -- 390, 1973. 133. James R. Abney, Jochen Braun, and John C. Owicki. Lateral interactions among membrane proteins: Implications for the organization of gap junctions. Biophys. J., 52:441 -- 454, 1987. 134. L. T. Pearson, B. A. Lewis, D. M. Engelman, and S. I. Chan. Pair distribution functions of bacteriorhodopsin and rhodopsin in model bilayers. Biophys. J., 43:167 -- 174, 1983. 135. L. T. Pearson, J. Edelman, and S. I. Chan. Statistical mechanics of lipid membranes, protein correlation functions and lipid ordering. Biophys. J., 45:863 -- 871, 1984. 136. Jochen Braun, James R. Abney, and John C. Owicki. Lateral interactions among membrane proteins: Valid estimates based on freeze-fracture electron microscopy. Biophys. J., 52:427 -- 439, 1987. 137. F. Oesterhelt, D. Oesterhelt, M. Pfeiffer, A. Engel, H. E. Gaub, and D. J. Muller. Unfolding pathways of individual bacteriorhodopsins. Science, 288:143 -- 146, 2000. 138. Toshio Ando, Noriyuki Kodera, Eisuke Takai, Daisuke Maruyama, Kiwamu Saito, and Aki- toshi Toda. A high-speed atomic force microscope for studying biological macromolecules. Proceedings of the National Academy of Sciences, 98(22):12468 -- 12472, 2001. 139. Ignacio Casuso, Pierre Sens, Felix Rico, and Simon Scheuring. Experimental evidence for membrane-mediated protein-protein interaction. Biophysical Journal, 99(7):L47 -- L49, 2010. 140. K. He, S. J. Ludtke, H. W. Huang, and D. L. Worcester. Antimicrobial peptide pores in membranes detected by neutron in-plane scattering. Biochem., 34:15614 -- 15618, 1995. 141. K. He, S. J. Ludtke, D. L. Worcester, and H. W. Huang. Neutron scattering in the plane of the membranes: Structure of alamethicin pores. Biophys. J., 70:2659 -- 2666, 1996. 142. L. Yang, T.M. Weiss, T.A. Harroun, W.T. Heller, and H.W. Huang. Supramolecular structures of peptide assemblies in membranes by neutron off-plane scattering: Method of analysis. Biophys. J., 77:2648 -- 2656, 1999. 143. D. Constantin, G. Brotons, A. Jarre, C. Li, and T. Salditt. Interaction of alamethicin pores in DMPC bilayers. Biophys. J., 92:3978 -- 3987, 2007. 144. Doru Constantin, Brigitte Pansu, Marianne Imp´eror, Patrick Davidson, and Franc¸ois Ribot. Repulsion between inorganic particles inserted within surfactant bilayers. Physical Review Letters, 101:098101, 2008. 145. Doru Constantin. The interaction of hybrid nanoparticles inserted within surfactant bilayers. The Journal of Chemical Physics, 133:144901, 2010. 146. D. Constantin. Membrane-mediated repulsion between gramicidin pores. Biochim. Biophys. Acta: Biomembr., 1788:1782 -- 1789, 2009.
1512.04217	1	1512	2015-12-14T08:29:55	Physical Limits on Bacterial Navigation in Dynamic Environments	[ "physics.bio-ph", "cond-mat.soft", "q-bio.CB" ]	Many chemotactic bacteria inhabit environments in which chemicals appear as localized pulses and evolve by processes such as diffusion and mixing. We show that, in such environments, physical limits on the accuracy of temporal gradient sensing govern when and where bacteria can accurately measure the cues they use to navigate. Chemical pulses are surrounded by a predictable dynamic region, outside which bacterial cells cannot resolve gradients above noise. The outer boundary of this region initially expands in proportion to $\sqrt{t}$, before rapidly contracting. Our analysis also reveals how chemokinesis - the increase in swimming speed many bacteria exhibit when absolute chemical concentration exceeds a threshold - may serve to enhance chemotactic accuracy and sensitivity when the chemical landscape is dynamic. More generally, our framework provides a rigorous method for partitioning bacteria into populations that are "near" and "far" from chemical hotspots in complex, rapidly evolving environments such as those that dominate aquatic ecosystems.	physics.bio-ph	physics	Physical Limits on Bacterial Navigation in Dynamic Environments Andrew M. Hein1,∗, Douglas R. Brumley2,3, Francesco Carrara2,3, Roman Stocker2,3 & Simon A. Levin1 1Department of Ecology and Evolutionary Biology, Princeton University, Princeton, NJ 08544, USA 2Ralph M. Parsons Laboratory, Department of Civil and Environmental Engineering, Massachusetts Institute of Technology, Cambridge, MA 02139, USA 3Department of Civil, Environmental and Geomatic Engineering, ETH Zurich, 8093 Zurich, Switzerland ∗ Email for correspondence: [email protected] Abstract Many chemotactic bacteria inhabit environments in which chemicals appear as localized pulses and evolve by processes such as diffusion and mixing. We show that, in such environments, physical limits on the accuracy of temporal gradient sensing govern when and where bacteria can accurately measure the cues they use to navigate. Chemical pulses are surrounded by a predictable dynamic region, outside which bacterial cells cannot resolve gradients above noise. The outer boundary of this region initially t, before rapidly contracting. Our analysis also reveals how chemokinesis -- expands in proportion to the increase in swimming speed many bacteria exhibit when absolute chemical concentration exceeds a threshold -- may serve to enhance chemotactic accuracy and sensitivity when the chemical landscape is dynamic. More generally, our framework provides a rigorous method for partitioning bacteria into populations that are "near" and "far" from chemical hotspots in complex, rapidly evolving environments such as those that dominate aquatic ecosystems. √ Keywords: chemotaxis, navigation, heterogeneity, chemokinesis, microbial ecology. 1 Introduction In natural environments such as oceans and lakes, bacteria and other microbes navigate chemical land- scapes that can change dramatically over the timescales relevant to their motility [1]. Such environments differ in fundamental ways from the static chemical gradients typically considered in studies of microbial chemotaxis (e.g., [2, 3]). From the perspective of microbes, chemical cues in nature often appear as lo- calized pulses with short duration [4, 5]. For example, oil droplets from spills and natural seeps, organic matter exuded by lysed phytoplankton or excreted by other organisms, and marine particles are common sources of short-lived, micro-scale (∼10-1000 µm) chemical pulses [4]. Motile bacteria respond to such cues by swimming up the gradients that are generated when pulses diffuse (e.g., [5 -- 8]). When a pulse ap- pears, for example through the lysis of a phytoplankton cell, the distribution of chemoattractants (often, dissolved organic matter) changes rapidly over both space and time [9]. Because background conditions are highly dilute, bacteria experience the early stages of a spreading pulse as a noisy chemical gradient with low absolute concentration. In marine environments, ephemeral, micro-scale pulses of dissolved chemicals provide a substantial and perhaps dominant fraction of the resources used by heterotrophic bacteria [4, 9, 10]. The advantage that chemotaxis confers cells in such dynamic environments [1, 11, 12] may help explain why chemotactic responses to transient nutrient sources are so common among marine bacteria [5, 6, 8, 10]. Although chemotaxis appears to be an important driver of bacterial competition [1], evolution [11, 12], and nutrient cycling [4, 9], the details of bacterial chemotaxis behaviour are poorly characterized for all but a few well-studied species of bacteria. An important shared feature of bacterial chemotaxis systems, however, is that the measurements of chemical concentration that underpin chemotaxis behavior are subject to considerable noise [13, 14]. In particular, stochasticity in the times at which individual molecules of chemoattractant arrive at the bacterium's surface sets an upper bound on the precision with which the cell can measure changes in concentration [15, 16]. Here, we demonstrate how this physical limit on the precision of temporal gradient sensing constrains when and where bacteria can respond to chemical pulses. Using this approach, we develop a general theory to predict the fundamental length and timescales over which chemotactic bacteria can respond to chemical pulses. Because it requires few assumptions about the underlying mechanisms responsible for chemotactic behaviour, the theory can be applied to the diverse assemblages of bacteria that occur in natural marine and freshwater environments. We first discuss gradient estimation by a cell in a dynamic chemoattractant field. We then derive theoretical bounds on the regions of the environment in which bacteria can respond to gradients, and characterize the spatio-temporal evolution of these regions as a function of physical and biological pa- rameters. Finally, we show that changes in swimming speed in response to measurements of absolute concentration -- a bacterial behaviour known as chemokinesis [10, 17] -- can greatly enhance a cell's ability to measure gradients in a dynamic chemoattractant field. 2 Model development Signal and noise in temporal gradient sensing Unlike large eukaryotic cells, which can directly measure spatial gradients in chemical concentration [18], many chemotactic bacteria navigate by measuring temporal changes in concentration as they swim [19, 20]. They use these measurements to detect concentration gradients and to navigate toward more favourable conditions (toward resources, away from noxious substances). Regardless of the biochemical and behavioural mechanisms a cell uses to navigate, gradient-based navigation can only be as precise as a cell's estimate of the gradient itself; downstream transduction will, in general, only add noise [16]. One can, therefore, establish performance bounds within which real bacterial cells must operate by consider- ing physical limits on the accuracy and precision of gradient sensing by an idealized cell. We begin by considering gradient detection by such a cell: the perfectly absorbing sphere originally described by Berg and Purcell [15]. This cell swims through a dynamic chemoattractant landscape, absorbing all molecules that reach its surface (Fig. 1a). In reality, bacteria absorb some ligands they use for chemotaxis, whereas others are bound only temporarily. However, absorbing ligand always leads to more accurate measure- ment of both absolute concentration and changes in concentration over time because molecules cannot be re-bound once they have been absorbed [13, 18]. We therefore assume molecules are absorbed yielding an upper limit on measurement accuracy [18]. Figure 1: Measurement of ramp rate c1 by an idealized cell. (a) During a time interval of length T , a cell travels from a region of low concentration to a region of higher concentration, absorbing chemoattractant molecules at times {ti} (red spikes in time series). (b) In a static concentration field C(x), c1 is equal to concentration slope g (slope of orange line) times swimming speed v. (c) In a dynamic concentration field C(x, t), c1 ≈ vg + ∂C/∂t; g is confounded with temporal changes in concentration (∂C/∂t) and the cell may perceive a decreasing concentration (red dashed line) although the true concentration slope is positive. Figures in colour online. 3 molecule absorption(a)(b)absorption time seriesconcentrationconcentration(c)time ( )position ( )Like the well-studied enteric bacterium, Escherichia coli, marine bacteria perform chemotaxis by altering the length of relatively straight "runs", which are interspersed with random re-orientation events ("tumbles" for E. coli [21], "flicks" for marine bacteria [22, 23]). As a cell swims, receptors on the cell's surface bind chemoattractant molecules and a signal from the receptors is transduced through a biochemical network to one or more flagellar motors, which control the speed and direction of the flagel- lar rotations that drive locomotion. Changes in receptor occupancy alter the probability that the direction of flagellar rotation will reverse, leading to a re-orientation [24], and the outcome of this is that bacte- ria extend runs when they perceive an increasing concentration of chemoattractant. A requirement for chemotaxis, therefore, is that the cell is capable of detecting meaningful changes in mean concentra- tion [14] over some measurement interval of length T . This task is complicated by significant stochastic variation in the times at which molecules arrive at the cell's surface. The length of the measurement interval T is bounded above by the characteristic timescale of stochastic re-orientations (e.g., rotational diffusion, active re-orientation [15]), which for cells in the size range of E. coli and many marine bacteria, ranges from hundreds of milliseconds [5] to several seconds [25]. A cell has little to gain by using the history of molecule encounters that extends beyond this timescale because rotational diffusion and active stochastic reorientation (e.g., tumbles, flicks) cause random changes in the cell's trajectory, decorrelat- ing the cell's orientation, and rendering old information useless to the cell for determining whether it is currently travelling up or down a chemoattractant gradient (this issue is discussed in detail in [15]). We therefore assume the measurement timescale T is shorter than the timescale of stochastic reorientation and neglect processes such as rotational diffusion. For such short T , the chemoattractant concentration along the swimming cell's path, c(t), can be linearized to c(t) ≈ c0 + c1(t − t0) over the time interval (t0 − T /2, t0 + T /2). The cell experiences this concentration as a noisy time series of encounters with chemoattractant molecules (Fig. 1a), from which it must estimate the concentration ramp rate, c1, to determine whether concentration is increasing or decreasing. Using maximum likelihood, one can show that the optimal way for a perfectly absorbing sphere of radius a to estimate c1 (concentration × time−1) using a sequence of molecule absorptions is, to leading i(ti−t0) order [13]: c1 = i(ti−t0)2 , where c1 is the cell's estimate of the ramp rate, n is the number of molecules absorbed over the measurement interval, D is the diffusivity of the chemoattractant, and ti is the absorption time of the ith molecule. Importantly, c1 has typical measurement variance no less than: n(cid:80) 4πDaT(cid:80) V ar( c1) = 3c0 πDaT 3 , (1) where c0 is the true background concentration in the vicinity of the cell at time t0, and the variance of c1 does not depend on the true ramp rate c1 as long as c0 (cid:29) c1T (Supplementary Text, see also Equa- tion (S44) in ref. [13]). This formulation assumes that a cell can "count" many molecules in a typical observation window, which amounts to assuming that the timescale at which receptors bind chemoattrac- tant molecules is fast relative to the length of the observation window, T . Receptor binding kinetics are typically very fast (millisecond timescales, e.g. [24, 26]), so this assumption will generally hold unless T is extremely short. To summarize, measurements of concentration involve three timescales that are 4 relevant to our model formulation, which are naturally separated in chemotactic bacteria [24]: (1) the timescale of absorptions, which is typically short (∼1 ms [24]), (2) the measurement window T , which is of intermediate length, and (3) the timescale of active re-orientations, which must be longer than T if the bacterium is to perform chemotaxis [15]. Variance in the ramp rate estimate (Eq. (1)) is solely due to stochastic arrivals of chemoattractant molecules and does not include additional sources of noise resulting, for example, from noise in the biochemical network responsible for ramp rate estimation [16, 27]. Eq. (1) thus provides a lower bound on uncertainty about the true ramp rate and a constraint within which real cells must operate, regardless of the precise biochemical mechanism though which they implement ramp rate estimation. Below we use Eq. (1) to define the regions of space where it is possible for cells to use measurements of concentration to climb chemoattractant gradients. Outside these regions, cells may attempt to perform chemotaxis; however, we will show that for several ecologically relevant types of pulses, the signal-to-noise ratio of a cell's estimate of the concentration slope decays sharply (like a Gaussian) far from the origin of a chemoattractant pulse. This strong decrease in the signal-to-noise ratio with increasing distance implies that chemotactic cells far from the origin of a pulse will be responding primarily to noise and will not exhibit biased motion. Gradient estimation in a time-varying environment For a cell swimming at speed v, the instantaneous local slope of the concentration profile along the cell's path, which we will refer to as the concentration slope g, is given by g = ∇C(x) · v/v, where v is the cell's velocity. The concentration slope is the quantity that is useful for climbing gradients, for example, by providing a signal for cells to lengthen runs in run-and-tumble chemotaxis [21]; however, a cell the size of a bacterium (∼1 µm) cannot measure g directly [13]. It must instead infer g from its estimate of the ramp rate c1. In a time-invariant concentration field c1 = gv, and the maximum likelihood estimator of g is proportional to the ramp rate estimator: g = c1/v (Fig. 1b, Supplementary Text). In a time-varying environment the concentration that a swimming cell experiences, c(t) ≈ c0 + (vg + ∂C/∂t)(t − t0), is influenced by local temporal changes in concentration, ∂C/∂t (Fig. 1c); the ramp rate is given by c1 = vg + ∂C/∂t. In this case, the time series of molecule absorptions does not contain the information needed to estimate both g and ∂C/∂t and any estimator the cell uses to measure the concentration slope g will be biased (Supplementary Text). For example, estimating g as g = c1/v means that g → g + (∂C/∂t)/v in the limit of many molecule absorptions. Correcting this bias would require that the cell have an independent estimate of ∂C/∂t. In the absence of such an estimate, the cell can reduce bias by travelling faster, but not by increasing the length of its measurement window T (Supplementary Text). This highlights an important connection between swimming speed and measurement accuracy that we explore in more detail below. Bias in the concentration slope estimate becomes important far from the origin of a pulse where cells can perceive an increasing concentration even if they are travelling down the concentration gradient, and near the origin, where cells can perceive a falling concentration even if they are travelling up a gradient (Fig. 1c). 5 Conditions for chemotaxis and responses to chemical pulses If a cell is to use measurements of ramp rate to climb a concentration gradient, two conditions must be met. First, the cell must be in a region of the environment where typical values of the perceived ramp rate exceed noise: i.e., the signal-to-noise ratio (SNR) of the ramp rate estimator, c1V ar( c1)−1/2 ≥ δ0, where δ0 is a constant threshold on the SNR (Supplementary Text). Second, the ramp rate c1 = vg + ∂C/∂t and the concentration slope g must have the same sign. Applying Eq. (1) and rearranging, these conditions are: (cid:114) ∂t vg + ∂C c0 √ ≥ δ := δ0 3 πDaT 3 , and sign(c1) = sign(g). (2) For a chemoattractant field with concentration C(x, t), Conditions (2) define the regions where cells can reliably determine the sign of the concentration slope, a requirement for gradient-based navigation. Using Conditions (2), we explore how bacteria perceive three types of pulses that occur in natural environments: pulses that arise from surfaces, pulses that arise as thin chemical filaments, and pulses created by small point releases. Localized point pulses are created by many natural sources, including the lysis of small cells and excretions by larger organisms [4, 5]. Thin chemical filaments and sheets occur when turbulence stirs dissolved chemicals. The distribution of chemicals is stretched and folded into sheets and filaments at length scales down to the Batchelor scale [4]. Mixing below the Batchelor scale is dominated by diffusion. This length scale is lB = (νD2/)1/4, where ν is kinematic viscocity, D is mass diffusivity, and is the turbulent dissipation rate. As changes, lB changes slowly implying that small point pulses and filaments or sheets spread primarily by diffusion across a broad range of flows. Across a range of realistic levels of turbulence ( ∼ 10−9 to 10−6 W kg−1 [28]) the average shear rate is of order 10−3 to 1 s−1. Except for the highest values in this range, these shear rates are typically too low to cause significant re-orientation of bacteria as they swim [29]. We therefore focus on the regime in which the environment is steady over the length scales considered here. To illustrate the utility of our theory, we consider how bacteria respond to chemical point pulses, filaments, and sheets. These canonical geometries can be viewed as basic components of more complex chemical landscapes at larger scales (e.g., the types of landscapes considered in [1]). Extending our results to alternative geometries follows from straightforward calculations. At time, t = 0, a single pulse appears with planar (N = 1, sheet), cylindrical (N = 2, filament), or spherical (N = 3, point pulse) symmetry. The size of the pulse is M (molecules per unit area of sheet [N = 1], per unit filament length [N = 2], or per individual point pulse [N = 3]). The three-dimensional chemoattractant field C is governed by ∂C/∂t = D∆C and the concentration is: C(r, t, N ) = M (4πDt)N/2 e − r2 4Dt , (3) 6 centration ((cid:112) C(r, t)) at t = t0 (orange) and individual estimates of this concentration ((cid:112) Figure 2: Gradient estimation in a dynamic environment. (a) Solid orange curve shows the true con- centration profile at t = t0. Solid green curve shows the signal-to-noise ratio (SNR) of c1 a cell would experience if this concentration profile were static. Dotted red curve shows SNR for a cell swimming directly toward origin of pulse. Dashed blue curve shows SNR for a cell swimming directly away from origin of pulse. Concentration and SNR normalized to maximum value of one. (b) Square-root of con- c(t), grey) made by a cell swimming toward pulse origin. Each individual estimate is computed by calculating c0 and c1 (see Supplementary Text for equations) from a time series of random Poisson molecule ar- rivals [31] with an arrival rate given by the true instantaneous concentration at the bacterium's position C(x, t). (c) Relative bias of concentration slope estimate (∂C/∂t/[vg + ∂C/∂t]) measured by slow (solid curve; v = 30 µm s−1) and fast swimming cells (dotted curve; v = 96 µm s−1). In all panels, concentration governed by Eq. (3) with N = 3, M = 1011 molecules, v = 30 µm s−1, a = 1 µm, T = 0.1 s, t0 = 45 s, and δ0 = 1. Pulse sizes in all figures correspond roughly to the quantity of free amino acids released from a lysed phytoplankton cell of ∼ 10 µm in diameter [5]. where D (µm2 s−1) is diffusivity, r (µm) is the distance from the surface (N = 1), filament axis (N = 2), or centre of the point source (N = 3). A cell moving in this chemoattractant field with velocity v (µm s−1) will experience a typical rate of change in concentration of c1 ≈ ∇C · v + ∂C/∂t. For chemoattractant pulses with concentration described by Eq. (3) (Fig. 2a, solid orange curve), the signal-to-noise ratio (Fig. 2a, solid green curve) divides the domain surrounding a pulse into three regions. Far from the pulse, the concentration gradient is shallow and the absolute concentration is low: 7 cells cannot accurately measure changes in concentration because they encounter few molecules during a typical observation window (Fig. 2b, bottom panel). At an intermediate distance from the pulse origin, the gradient is largest in magnitude and cells encounter many molecules during a typical observation window: the SNR is greatest in this region (Fig. 2b, middle panel). Near the pulse origin the gradient is again shallow and variance in the concentration slope estimate is substantial (Fig. 2b, top panel). Moreover, in this region, concentration changes rapidly over time and the concentration slope and ramp rate may differ in sign (i.e., bias in the concentration slope estimate is large, Fig. 2b, top panel; Fig. 2c). Results Cells far from a chemoattractant pulse cannot resolve true changes in concentration above noise (Fig. 2a, SNR drops below threshold δ0 for large distance). The distance beyond which c1 becomes dominated by noise is given implicitly by (4) (cid:12)(cid:12)(cid:12)(cid:12)vg(r, t) + δ = (cid:12)(cid:12)(cid:12)(cid:12) C(r, t) −1/2, ∂C(r, t) ∂t where the term in brackets is the magnitude of the true ramp rate c1 that a cell at distance r with local concentration slope g(r, t) experiences. Because the chemoattractant field is changing, the magnitude of the ramp rate a cell measures will depend on its direction of travel. Far from the pulse, a cell travelling directly inward (Fig. 2a, red dotted curve) will experience a greater SNR than a cell travelling outward (Fig. 2a, blue dot-dash curve). Beyond the inflection point in the concentration profile, the r.h.s. of Eq. (4) is maximized for cells travelling directly up the concentration gradient (i.e., toward the pulse center; Fig. 2a, red dotted curve). The outer boundary beyond which cells cannot reliably perceive changes in concentration is given implicitly by Eq. (4) with g = −∂C/∂r. We refer to the largest distance that satisfies this equation as the outer boundary of sensitivity, ro (Fig. 2a, red point). At distances r > ro, perceived changes in concentration are dominated by noise, regardless of a cell's direction of travel. Bacteria use gradients to navigate toward regions of high attractant concentration, but also to maintain position near local maxima [12]. In order to do this, a cell travelling down the concentration gradient must experience a decreasing concentration, which provides the signal the cell uses to modify swimming behavior [23]. Near the origin, the SNR is maximized for a cell that is travelling directly down the concentration gradient (Fig. 2a blue dash-dot curve). For t greater than a critical time, ts, there is an inner boundary at a distance ri from the origin of the pulse (Fig. 2a, blue point), within which the SNR drops below threshold. For t > ts the location of this inner boundary is given implicitly by Eq. (4) with g = ∂C(r, t)/∂r (Supplementary Text). The boundaries ro and ri define a dynamic region (Fig. 3, blue region in inset), outside of which bacteria cannot reliably respond to chemoattractant gradients because either the ramp rate is too noisy to resolve, or the ramp rate and the concentration slope have different signs (i.e., Conditions (2) are violated). Figure 3 shows the dynamics of ro and ri for bacteria swimming at three different speeds. For all swimming speeds, the outer boundary ro initially expands before rapidly contracting (Fig. 3, red curves). The time dependence of this boundary can be obtained by substituting Eq. (3) into Eq. (4), 8 Figure 3: Inner (blue) and outer (red) boundaries of the region in which cells reliably perceive gradients. Dashed line shows v = 30 µm s−1, maximum swimming speeds of E. coli [10]; dash-dot line shows v = 66 µm s−1, typical cruising speed of Vibrio coralliilyticus; dotted line shows v = 96 µm s−1, maximum speed of V. coralliilyticus after initiating chemokinesis [17]. Other parameters as in Fig. 2. Solid grey curve is outer boundary, rc, of region within which cells can resolve absolute concentration. Solid black curve is 4Dt, the radius at which the SNR is maximized for a static profile (green curve in Fig. 2). Inset shows relative sizes of region where cells can detect gradients (ri < r < ro, blue region), and region where cells can resolve absolute concentration (r < rc, grey region inward) at t = 90 s (v = 66 µm s−1). √ solving for ro, and expanding the resulting product-log solution (Supplementary Text): (cid:115) ro ≈ 4Dt log (cid:20)− log(kt1+N/2) (cid:21) kt1+N/2 , (5) where k = (4πD)N/2δ2 0/(2πaM v2T 3). Swimming speeds of motile bacteria typically range from 30 µm s−1 to over 100 µm s−1 [10]. For many relevant chemoattractants, D ∼ 103 µm2 s−1, and the number of molecules released in a pulse, M, is generally large; for example, a point pulse created by the lysis of even a small phytoplankton cell (a common source of nutrients for marine bacteria) contains upwards of 1011 free amino acid molecules [5]. This means that k (cid:28) 1 such that the logarithmic term in √ t. Eq. (5) varies slowly with time for early times, and leading-order behaviour is initially governed by Pulse size, M occurs only inside the logarithmic terms in Eq. (5) indicating that ro scales weakly with pulse size. For example, doubling the size of a small point pulse (N = 3) increases the volume of water 9 0501001502000100020003000time ( )distance from pulse origin ( )in which gradients are perceived by only 50% (assuming M increases from 1011 to 2 × 1011 molecules, δ0 = 1, and v = 66 µm s−1). Figure 4 shows the dynamics of ro for surface, filament, and point pulses. Eq. (5) agrees well with the exact solution for ro obtained by solving Eq. (4) numerically (Fig. 4 compare solid and dashed lines). Eventually the inner and outer boundaries of sensitivity intersect (Fig. 3), and cells can no longer reliably glean navigational information from the chemoattractant field. We refer to the time at which this occurs as t∗. Finding the time when the SNR falls below threshold δ0 everywhere shows that ∗ ≈ α(M v2T 3) t 2 N +2 , (6) where α = (π(1−N/2)ae−1)2/(N +2)[3(4D)N/2δ0]−2/(N +2) and the approximation assumes vg (cid:29) ∂C/∂t at the point in space where the SNR is maximized (Supplementary Text). This relation illustrates the relative contribution of measurement time T and speed v to the time scale of perceptible changes in concentration, t∗. Moreover, Eq. (6) shows that t∗ is proportional to M 2/(N +2); the scaling of t∗ with pulse size is sublinear for all pulse geometries meaning that doubling the size of a pulse always less than doubles the time over which it can be perceived. The locations of inner and outer boundaries (Fig. 3) are governed, in part, by swimming speed. Many bacteria alter swimming speed in response to stimuli, and a natural question, therefore, is whether a cell could adjust its speed adaptively to achieve high sensitivity to chemical gradients. Some species exhibit a behaviour known as chemokinesis: cells swim at a speed that depends on the local concentration of chemoattractant, often swimming at a high speed when absolute concentration is high, and a low speed when concentration is low [10, 17]. In the presence of a resolvable gradient, the interpretation of chemokinesis is straightforward: cells can climb the gradient faster if they swim at a higher speed (at the expense of a higher energetic cost of motility). However, chemokinesis may also have a second role. The SNR of the ramp rate is smaller than the SNR of the absolute concentration, c0, implying that cells may be able to accurately detect whether absolute concentration has crossed a threshold before they can resolve changes in concentration over time. The mean rate of arrival of molecules to the surface of a sphere of radius a is 4πDac(t) [15]. Poisson molecule arrivals imply that the SNR of absolute concentration c0 is c0V ar( c0)−1/2 = c0[4πDaT c0]−1/2. Using this ratio, we define a third boundary, rc, beyond which the SNR of c0 falls below threshold, δ0: (cid:113) rc = 8Dt log(ηt−N/2), (7) −1 0 (M aT )1/2(4πD)1/2−N/4. This boundary has the same leading order behaviour in time where η = δ as ro, but extends well beyond ro (Fig. 3, solid grey curve); for example, assuming ro is at its maximum value (Fig. 3), the volume within which cells can accurately measure absolute concentration in the water surrounding a small point pulse (N = 3) is six times larger than the volume in which cells can resolve changes in concentration (assuming M = 1011 molecules [5], δ0 = 1, v = 66 µm s−1). Note that we use the same threshold (δ0) on the SNR of c0 and c1 for the purpose of comparison but thresholds on these ratios need not be equal. 10 Figure 4: Scaling of the outer boundary of sensitivity ro for pulses emitted from surfaces (light grey), filaments (grey), and point sources (black). Solid curves are numerical solution to Eq. (4). Dashed curves given by Eq. (5). Solid black line is proportional to t. Solid curves truncated when the SNR falls below δ0. Dashed curves truncated at t∗ (Eq. (6)). M scaled so that pulses with different geometries have the same concentration profile at t = 10 s (M = 8.0 × 105 molecules per µm2 surface for surface source; M = 2.8× 108 molecules per µm length for line source; M = 1011 molecules for point source); v = 66 µm s−1; other parameters as in Fig. 2. √ By increasing their swimming speeds when concentration exceeds a threshold, cells can increase their sensitivity to changes in concentration (first Condition (2); Fig. 3) and reduce bias in estimation of the concentration slope (Fig. 2c). The effect of increasing swimming speed is to expand the region of space over which the cell can resolve gradients, ri < r < ro, and to extend the time t∗ beyond which gradients become too noisy for the cell to measure (Fig. 3, compare curves for different swimming speeds; Fig. 5). Effects of changes in speed may be substantial. For example, the coral pathogen Vibrio coralliilyticus increases its speed by as much as 45% when chemoattractant concentration is high [17]. The temporal evolution of a chemoattractant pulse appears very different to a bacterium swimming at 66 µm s−1 (typical cruising speed of V. coralliilyticus and other Vibrio spp.; Fig. 5, blue regions) than it does to a bacterium travelling at speeds closer to 100 µm s−1 (swimming speeds of chemokinetic V. coralliilyticus [10, 17]; Fig. 5, orange regions). 11 tohundredsofmicrons,viscousforcesdominateandnutrientsaredispersedviamolecularditurbulentmixing.Smallpulsesofresources(10-100micronrange)arealsodominatedbythislaminarregime.Wecanthinkaboutthreecannonicalstructuresthatarerelevanttothetimeandlengthscalesofbacterialchemotaxis:nutrientsheets,filaments,andlocalpulses(Fig.1).Wewouldexpectthesestructurestoappearintheneighborhoodofapopulationofbacteria(viaturbulentmixing)andthenbegintodiffuse.Itthereforeseemssensibletomodelthearrivalofsuchstructuresusingastochasticprocessandmodeltheinteractionsbetweenthesestructuresandapopulationofbacteriausingdiffusiontheory.Inthisdocument,wefocusoninteractionsbetweenapopulationofbacteriaandasinglestructure,whichwewillrefertoasapulse.Itturnsoutthatclassicproblemsofdiaplanesheet,cylinder,andsphereareextremelyusefulwhenmodelingbacterialinteractionswithsheets,filaments,andlocalpulses(respectively)ofnutrient.sheetfilamentFigure1:Sheets,filaments,andlocalpulses:threetypesofstructurebacteriacaninteractwith.Blacklinesindicatelevelsurfacesatwhichnutrientconcentration(orconcentrationgradient,orsomefunctionofgradient)isequaltoacriticalvalue(seebelow).2Bacterium-nutrientinteractionsatthelocalscaleWewanttomodelhowchemotacticbacteriainteractwithasingleresourcepatchoffinitemassthatappearsattimet=0andsubsequentlydiffuses.Ourultimategoalistoobtainanexpressionforthemean(andpossiblyothermoments)uptakerateofthebacterialpopulationoverthelifespanofthepulse(i.e.,beforethepulseisdepletedbelowsomecriticalconcentrationeverywhere).Inwhatfollows,wewillfocusonresourcepulseswithcylindricalor1001000100000.11101001000time (s)distance from pulse origin ( )Figure 5: Effect of swimming speed on the time evolution of the region where chemotaxis is possi- ble. Colored regions show a two-dimensional cross-section of the region in which cells can resolve chemoattractant gradients (i.e. Conditions (2) are satisfied). Blue regions are those experienced by a cell travelling at a cruising speed typical of the bacterium V. coralliilyticus (∼ 66 µm s−1). Orange regions are those experienced by a V. coralliilyticus cell travelling at a high speed after initiating chemokinesis (∼ 96 µm s−1) [17]. Other parameters as in Fig. 2. Note the blind spot that forms at the centre of the region as the inner boundary of sensitivity, ri, expands. Discussion Bacteria must cope with considerable noise and estimation bias when navigating dynamic chemical land- scapes. The advantage conferred by an early response to chemical pulses suggests that there may be selection for high accuracy and sensitivity in the chemotaxis response [1, 4]. Our framework provides a means of studying how the basic components of bacterial navigation strategies (swimming speed, mea- surement time) and physical parameters (e.g., chemoattractant diffusivity, pulse size) influence when and where bacteria can perform chemotaxis. Expressions for the outer boundary of sensitivity, ro (Eq. 5), and the time after which gradients created by a pulse are no longer perceptible, t∗ (Eq. 6), may prove particularly useful as they constrain the length and timescales over which bacteria can perceive individ- ual chemical pulses. The relationship between the size of the pulse, pulse geometry, and the length and timescales over which the pulse is perceptible provides a basis for modeling more realistic environments where many pulses appear with characteristic sizes, geometries, and temporal statistics. For example an empirical estimate of typical inter-pulse-interval in, say, a marine environment [4], can be compared to t∗ to determine whether the environment is highly granular or relatively homogeneous from the per- spective of bacteria. For the canonical pulse geometries considered here (Eq. (3)), the signal-to-noise ratio of the concentration ramp rate decays sharply far from the origin of a pulse (Fig. 2a, blue, red, and green curves). In particular, substituting Eq. (3) into the expression for the SNR of c1 (r.h.s. of Eq. (4)) shows that the SNR decays like a Gaussian for large r (SNR ∝ exp[−r2/(8Dt)] for large r). This sharp transition in the SNR means that, near the outer boundary of sensitivity, there is a stark division between cells that have access to useful chemotactic information (r < ro) and cells that do not (r > ro). Using ro 12 region where chemotaxis is possible= 5= 33= 60= 88= 120time (s)900= 140to partition bacterial cells into subpopulations that are near and far from chemical pulses could greatly simplify models of bacterial competition and population dynamics in complex environments [1]. Our theory makes a number of predictions that could be tested with chemotaxis experiments. First, the theory predicts that for times t < t∗, the mean orientation of bacterial swimming trajectories outside the region ri < r < ro should be unbiased. Because the conditions considered in this work correspond to an upper bound on sensory accuracy, the region within which cells exhibit biased motion may be a sub-region of ri < r < ro. A second prediction is that, for times greater than t∗, bacteria should not exhibit biased motion anywhere in the environment because each cell's estimate of the gradient will be dominated by noise, regardless of where it is located relative to the origin of the pulse. Again, because of the assumptions used to derive t∗, the observed time at which the average directional bias of a bacterial population drops to zero may be shorter than t∗. One of the implications of our model for temporal gradient sensing is that sensory acuity is intimately linked to swimming speed (Eq. 4, Fig. 5). Because swimming at high speed is costly [1, 15], bacteria likely benefit by changing speed in an adaptive way, cruising at low speed in the absence of a chemical signal, and speeding up when concentration exceeds a threshold. The connection between speed and measurement accuracy may explain the counterintuitive observation that some species of marine bacteria swim at high speeds even near local maxima in chemoattractant concentration [10]; bias in the concen- tration slope estimate is high near local maxima (Fig. 2b). A cell cannot decrease bias by lengthening measurement time, but it can reduce bias by swimming faster, suggesting that bacteria may use chemoki- nesis to enhance chemotactic accuracy near the blind spot that forms at the centre of spreading chemical pulses (Fig. 5 t = 120 s, t = 140 s; Supplementary Text). More generally, our framework suggests that bacteria can improve chemotactic performance by using chemokinesis and chemotaxis in concert. The hypothesis that bacteria initiate chemokinesis in response to absolute concentration to enhance sensitiv- ity to gradients could be investigated by independently varying the concentration gradient and absolute concentration of a chemoattractant, for example using a microfluidic device [30]. Our framework uses fundamental limits on the accuracy of chemical sensing [13, 16] to determine when and where chemotaxis is feasible, and provides a tool for modeling bacterial behaviour in more realistic dynamic environments. Importantly, it is agnostic to the details of bacterial movement patterns and chemosensory machinery and can therefore provide general principles that apply to the broad range of bacterial species in real ecological communities that navigate using temporal gradient sensing. Acknowledgements This work was supported by Army Research Office Grants W911NG-11-1-0385 and W911NF-14-1-0431 to S.A.L., a James S. McDonnell Foundation Fellowship to A.M.H., a Swiss National Science Foundation postdoctoral fellowship to F.C., a Human Frontier Science Program Cross-Disciplinary fellowship to D.R.B., and a Gordon and Betty Moore Marine Microbial Initiative Investigator Award (GBMF3783) to R.S. 13 References and Notes 1. Taylor JR, Stocker R. Trade-offs of chemotactic foraging in turbulent water. Science. 2012;338:675 -- 679. 2. Kalinin Y, Jiang L, Tu Y, Wu M. Logarithmic sensing in Escherichia coli bacterial chemotaxis. Biophys J. 2009;96:2439 -- 2448. 3. Ahmed T, Shimizu TS, Stocker R. Bacterial chemotaxis in linear and nonlinear steady microfluidic gradients. Nano Letters. 2010;10:3379 -- 3385. 4. Stocker R. Marine microbes see a sea of gradients. Science. 2012;338:628 -- 633. 5. Blackburn N, Fenchel T, Mitchell J. Microscale nutrient patches in planktonic habitats shown by chemotactic bacteria. Science. 1998;282:2254 -- 2256. 6. Seymour JR, Marcos, Stocker R. Resource patch formation and exploitation through the marine microbial food web. Am Nat. 2009;1:E15 -- E29. 7. Stocker R, Seymour JR, Samadani A, Hunt DE, Polz MF. Rapid chemotactic response enables marine bacteria to exploit ephemeral microscale nutrient patches. Proc Natl Acad Sci USA. 2008;105:4209 -- 4214. 8. Seymour JR, Sim´o R, Ahmed T, Stocker R. Chemoattraction to Dimethylsulfoniopropionate throughout the marine microbial food web. Science. 2010;329:342 -- 345. 9. Fenchel T. Microbial behavior in a heterogeneous world. Science. 2002;296:1068 -- 1071. 10. Barbara GM, Mitchell JG. Marine bacterial organisation around point-like sources of amino acids. FEMS Microb Ecol. 2003;43:99 -- 109. 11. Frankel NW, Pontius W, Dufour YS, Long J, Hernandez-Nunez L, Emonet T. Adaptability of non- genetic diversity in bacterial chemotaxis. eLife. 2014;3:e03526. 12. Celani A, Vergassola M. Bacterial strategies for chemotaxis response. Proc Natl Acad Sci USA. 2010;107:1391 -- 1396. 13. Mora T, Wingreen NS. Limits of sensing temporal concentration changes by single cells. Phys Rev Lett. 2010;104:248101. 14. Andrews BW, Yi TM, Iglesias PA. Optimal noise filtering in the chemotactic response of Escherichia coli. PLoS Comp Biol. 2006;2:1407 -- 1418. 15. Berg JC, Purcell EM. Physics of chemoreception. Biophys J. 1977;20:193 -- 219. 16. Bialek W, Setayeshgar S. Physical limits to biochemical signaling. Proc Natl Acad Sci USA. 2005;102(10040-10045). 17. Garren M, Son K, Raina JB, Rusconi R, Menolascina F, Shapiro OH, et al. A bacterial pathogen uses dimethylsulfoniopropionate as a cue to target heat-stressed corals. ISME J. 2013;8:999 -- 1007. 14 18. Endres RG, Wingreen NS. Accuracy of direct gradient sensing by single cells. Proc Natl Acad Sci USA. 2008;105:15749 -- 15754. 19. Macnab RM, Kochland DE. The gradient sensing mechanism in bacterial chemotaxis. Proc Natl Acad Sci USA. 1972;69:2509 -- 2512. 20. Segall JE, Block SM, Berg HC. Temporal comparisons in bacterial chemotaxis. Proc Natl Acad Sci USA. 1986;83:8987 -- 8991. 21. Brown DA, Berg HC. Temporal stimulation of chemotaxis in Escherichia coli. Proc Natl Acad Sci USA. 1974;71:1388 -- 1392. 22. Xie L, Altindal T, Chattopadhyay S, Wu XL. Bacterial flagellum as a propeller and as a rudder for efficient chemotaxis. Proc Natl Acad Sci USA. 2011;108:2246 -- 2251. 23. Xie L, Lu C, Wu XL. Marine bacterial chemoresponse to a stepwise chemoattractant stimulus. Biophys J. 2015;108:766 -- 774. 24. Jiang L, Ouyang Q, Tu Y. Quantitative modeling of Escherichia coli chemotactic motion in environ- ments varying in space and time. PLoS Comp Biol. 2010;6:e1000735. 25. Alon U, Cmaerena L, Surette MG, Aguera y Arcas B, Liu Y, Leibler S, et al. Response regulator output in bacterial chemotaxis. EMBO J. 1998;17:4238 -- 4248. 26. Zhang W, Olson JS, Phillips Jr GN. Biophysical and kinetic characterization of HemAT, an aerotaxis receptor from Bacillus subtilis. Biophys J. 2005;88:2801 -- 2814. 27. Lestas I, Vinnicombe G, Paulsson J. Fundamental limits on the suppression of molecular fluctua- tions. Nature. 2010;467:174 -- 178. 28. Doubell MJ, Prairie JC, Yamazaki H. Millimeter scale profiles of chlorophyll flourescence: deci- phering the microscale spatial structure of phytoplankgon. Deap-Sea Res Pt II. 2014;101:207 -- 215. 29. Rusconi R, Guasto JS, Stocker R. Bacterial transport suppressed by fluid shear. Nature Phys. 2014;10:212 -- 217. 30. Son K, Brumley DR, Stocker R. Live from under the lens: exploring microbial motility with dynamic imaging and microfluidics. Nat Rev Microbiol. 2015;13:761 -- 755. 31. Asmussen S, Glynn PW. Stochastic Simulation: Algorithms and Analysis. New York, NY USA: Springer; 2007. 15 Supplementary Text Ramp rate and concentration slope estimation Estimating the ramp rate from a series of molecule absorptions. Here we discuss constraints on the estimation of the ramp rate c1 and the concentration slope g. Following [S1, S2, S3], we approximate a cell as an idealized measuring device: a sphere of radius a that absorbs all molecules that come in contact with its surface. We begin by recalling relevant results of [S3]. The average flux of molecules (molecules × time−1) arriving at the surface of the sphere at position x, time t is (cid:104)I(x, t)(cid:105) = 4πDaC(x, t) [S1, S3] , which is equivalent in our notation to (cid:104)I(t)(cid:105) = 4πDac(t). Again, the far-field chemoattractant concentration in the medium surrounding the sphere c(t) can be linearized to c(t) ≈ c0 + c1(t − t0). The question discussed in [S3] is: given a series of absorption times {ti}, i = 1, 2, ..., n measured during the interval ti ∈ (t0 − T /2, t0 + T /2), what is the minimum variance in the estimate of c1 that a cell could possibly achieve? A natural tool for answering this question is the statistical framework known as Maximum Likelihood. Given a set of data (the time series {ti}) and a generating model for those data -- in this case, that c(t) = c0 + c1(t − t0), and absorptions are Poisson with rate (cid:104)I(t)(cid:105) = 4πDac(t) -- one seeks values of the parameters of the generating model (c0 and c1) that maximize the probability, or "likelihood", of the data. Maximum likelihood estimates are optimal in the sense that, as the number of observations becomes large, the variance of these estimators approaches a theoretical minimum variance for any unbiased estimator, which is given by the Cram´er-Rao theorem [S4]. This lower bound can be used to establish a bound on the accuracy with which cells can measure changes in concentration. In the context considered in [S3] and in our study, molecule absorptions are assumed to be indepen- dent Poisson events. Let t = 0 be the time at which the pulse appears, and t = t0 be a reference time marking the midpoint of the measurement interval (t0− T /2, t0 + T /2). The lengths of the time intervals between molecule arrivals, σi = ti − ti−1 [S5] obey P(σi) = (cid:104)I(ti)(cid:105) exp − (cid:104)I(s)(cid:105)ds , (S1) where σ1 is defined as t1 − (t0 − T /2) and the integral in Eq. (S1) is taken from t0 − T /2 (the start of the measurement interval) to t1 for σ1. The probability of observing the set {ti} is P({ti}) = P({σi}) = (cid:104)I(ti)(cid:105)e . (S2) By solving ∂ log(P[{ti})]/∂c0 = 0 and ∂ log[P({ti})]/∂c0 = 0, one can show that the values of c0 and c1 that maximize Eq. (S2) are: (S3) (cid:34) (cid:90) ti ti−1 (cid:35) n(cid:89) i=1 −(cid:82) t0+T /2 t0−T /2 (cid:104)I(t)(cid:105)dt c0 = n 4πDaT 16 (cid:80) (cid:80) i(ti − t0) i(ti − t0)2 , and c1 = c0 (S4) where n is the number of molecules absorbed during the observation interval. Note that (cid:104)n(cid:105) ≈ 4πDac0T ti(ti − t0)(cid:105) = ti(ti − t0)2(cid:105) = t0−T /2 [c0+c1(t−t0)](t−t0)2dt = πDac0T 3/3, indicating that maximum likelihood estimators, as the measurement interval becomes long, and as n becomes large,(cid:80) 4πDa(cid:82) t0+T /2 t0−T /2 [c0 + c1(t − t0)](t − t0)dt = πDac1T 3/3 and (cid:80) 4πDa(cid:82) t0+T /2 i(ti − t0) ≈ (cid:104)(cid:80) ti(ti − t0)2 ≈ (cid:104)(cid:80) c0 and c1, are asymptotically unbiased, i.e., c0 → c0 and c1 → c1 for large n. (S5) We derive Eq. estimator, c1. The Cram´er-Rao theorem states that the variance of c1 is bounded by the relation [S4]: (1) in the Main Text by calculating a lower bound on the variance of the ramp rate (cid:21)−1 Employing the assumption that c0 (cid:29) c1T , and using(cid:80) (cid:20) ∂2 log(P({ti}; c1)) var( c1) ≥ −E ∂c1 2 = E (cid:34)(cid:88) ti (ti − t0)2 [c0 + c1(ti − t0)]2 . (S6) ti[ti − t0]2 ≈ πDac0T 3/3 as the number of (cid:35)−1 absorptions becomes large implies var( c1) (cid:38) 0(cid:80) ti[ti − t0]2 ≈ 3c0 πDaT 3 , c2 (S7) which is the relation given in Eq. (1) of the Main Text. Estimating the concentration slope in static and dynamic concentration fields. We are concerned with cells that use their estimate of the ramp rate c1 to estimate the spatial gradient in chemical concentra- tion, which we refer to as the concentration slope. We also assume that the concentration field C changes over time as a pulse spreads. In this setting, the cell must estimate the concentration slope g along its path using some estimator g that can be computed from a series of observed absorption times. The con- centration experienced by the cell can still be written c(t) = c0 + c1(t − t0), but now c1 ≈ gv + ∂C/∂t. If we begin by assuming ∂C/∂t = 0, the maximum likelihood estimator for g follows from the estimator for c1: In the limit of many molecule absorptions,(cid:80) (S8) i(ti − t0)2 ≈ πDac0T 3/3 (using the assumption that c0 (cid:29) c1T ), which imply that g approaches the true concentration slope g as the number of molecule absorptions becomes large (i.e., g is asymptotically unbiased). g = c0 (cid:80) v(cid:80) i(ti − t0) i(ti − t0) ≈ πDavgT 3/3 and(cid:80) i(ti − t0)2 . When ∂C/∂t is not equal to zero and the cell is swimming at speed v > 0, the absorption time series {ti} does not contain the information necessary to estimate both g and ∂C/∂t. This can be shown by combining Eq. (S4) and (S5): 17 n(cid:80) 4πDaT(cid:80) c1 = i(ti − t0) i(ti − t0)2 → c1 = gv + ∂C/∂t. concentration slope in a dynamic environment. For ∂C/∂t (cid:54)= 0, the sum(cid:80) Equation (S9) is clearly underdetermined; an infinite number of g and ∂C/∂t value pairs can satisfy Eq. (S9). Without additional information, any estimator of the concentration slope g will be biased. For instance, a cell could implement the maximum likelihood estimator g defined above to estimate the i(ti − t0) ≈ πDa(gv + (S9) ∂C/∂t)T 3/3, which implies that g → g + (∂C/∂t)/v. (S10) Equation (S10) illustrates two important points: the bias in the concentration slope estimate (second term on the r.h.s. of Eq. (S10)) is reduced as swimming speed increases; and this bias does not depend on measurement time T . One could propose alternative estimators for the concentration slope g that satisfy Eq. (S9), but these basic conclusions are unchanged. (cid:20) r2 4Dt2 − N 2t (cid:21) ∂C ∂t = (cid:112)−4Dt W (−16Dtd2), Dynamics of the outer boundary Derivation of the outer boundary, ro. To derive the time-scaling of the outer boundary ro, we consider a cell that is travelling directly toward the origin of the pulse at speed v. We define ro as the largest radius, r that satisfies Eq. (4) in the Main Text. To approximate this value, note that temporal changes in concentration are described by C, (S11) √ for the concentration profile studied in the Main Text, which implies that near r = 2N Dt, temporal changes in the concentration field are small. We assume ro is in this region and therefore neglect contri- butions of ∂C/∂t to the ramp rate measured by a swimming cell. This implies that the condition for the signal-to-noise ratio to rise above δ0 is −v ∂C ∂r C−1/2 = vrC1/2/(2Dt) ≥ δ. Solving for r gives: r = (S12) √ M v)−1, and W (·) is the product log function. In general, M will be large where d = δ(4πDt)N/4(4 so the argument of the product log function will be negative and close to zero (because d is small). An approximation for the branch of the product log function that corresponds to ro in this regime is W (x) ≈ ln(−x) − ln(− ln(−x)) [S6], which yields the approximation for ro given by Eq. (5) in the Main Text. Derivation of the time when chemotaxis ceases, t∗. The signal-to-noise ratio takes its maximum 4Dt. Near this radius the contribution of temporal changes in the chemical field to the value at r = cell's perceived ramp rate are small and, again, the signal-to-noise ratio is approximately proportional to −v ∂C ∂r C−1/2. Solving for the time at which the maximum signal-to-noise ratio falls below threshold yields the expression for t∗ given by Eq. (6) in the Main Text. √ 18 Dynamics of the inner boundary. The inner boundary ri is given implicitly by (cid:12)(cid:12)(cid:12)(cid:12)vg(r, t) + (cid:12)(cid:12)(cid:12)(cid:12) C(r, t) −1/2 − δ = 0, (S13) ∂C(r, t) ∂t with g = ∂C/∂r. Eq. (S13) has zero, one, or two positive roots. When this expression has no positive roots, cells travelling down the concentration gradient experience a signal-to-noise ratio of the ramp rate estimator that is below threshold δ0 everywhere. When this expression has one positive and one nega- tive root, there is a maximum distance, beyond which cells travelling down the concentration gradient typically fail to detect a signal that is resolvable above noise, but any cell within this outer radius can typically resolve the ramp rate (Fig. 3 of the Main Text, early time). When Eq. (S13) has two positive roots, there exists an inner boundary, ri > 0 within which, cells cannot resolve the ramp rate. This latter case is shown in Fig. 2a (dashed blue curve) and Fig. 3 inset in the Main Text. References and Notes S1. Berg JC, Purcell EM (1977) Physics of chemoreception. Biophysical Journal 20:193 -- 219. S2. Endres RG, Wingreen NS (2008) Accuracy of direct gradient sensing by single cells. Proc. Natl. Acad. Sci. U.S.A. 105:15749 -- 15754. S3. Mora T, Wingreen NS (2010) Limits of sensing temporal concentration changes by single cells. Physical Review Letters 104:248101. S4. Rice J (1995) Mathematical Statistics and Data Analysis (Duxbury, Belmont), 2nd edition. S5. Asmussen S, Glynn PW (2007) Stochastic Simulation: Algorithms and Analysis (Springer, New York, NY USA). S6. Abramowitz M, Stegun I (1964) Handbook of Mathematical Functions (US Government Printing Office, Washington, DC, USA). 19
1706.01541	1	1706	2017-06-05T21:16:43	Calculation of the Sun exposure time for the synthesis of vitamin D in Urcuqu\'i, Ecuador	[ "physics.bio-ph" ]	The synthesis of vitamin D is strongly linked to the availability of solar energy. For a long time, it was clear that using the erythemal irradiances is a good choice to calculate the exposure time. Several authors argue that this method of minimum exposure time overestimates the calculation of the solar irradiance of vitamin D. In this paper, the vitamin D and erythemal irradiances are calculated on the basis of the spectral solar irradiance in Urcuqu\'i (Ecuador). From these results, we obtain the minimal exposure time. It was found that there is a difference between the use of the erythemal irradiance and the vitamin D one; namely, the exposure times were too large (almost a 100% difference) for the erythemal irradiance.	physics.bio-ph	physics	Calculation of the Sun exposure time for the synthesis of vitamin D in Salum GM1, García Molleja J1, Regalado Díaz BA1, Guerrero León LA1 and Berrezueta C1 Urcuquí, Ecuador Yachay Tech University, School of Physical Sciences and Nanotechnology, 100119-Urcuquí, Ecuador [email protected], [email protected], [email protected], [email protected], [email protected] Abstract. The synthesis of vitamin D is strongly linked to the availability of solar energy. For a long time, it was clear that using the erythemal irradiances is a good choice to calculate the exposure time. Several authors argue that this method of minimum exposure time overestimates the calculation of the solar irradiance of vitamin D. In this paper, the vitamin D and erythemal irradiances are calculated on the basis of the spectral solar irradiance in Urcuquí (Ecuador). From these results, we obtain the minimal exposure time. It was found that there is a difference between the use of the erythemal irradiance and the vitamin D one; namely, the exposure times were too large (almost a 100% difference) for the erythemal irradiance. Keywords. Vitamin D synthesis – erythemal effect – exposure time – action spectrum 1 Introduction Human beings do not produce the required daily dose of vitamin D (calciferol) on their own, other mechanisms must be used to obtain this minimum dose, i.e., exposure to sunlight or dietary supplements [1]. Vitamin D is directly involved in the body's Ca-P homeostasis and its deficiency can cause rickets in children and osteomalacia in adults' homeostasis. Similarly, other important physiological processes linked to the immune system have been attributed to vitamin D deficiency. 25(OH) levels of vitamin D in the human body maintain healthy bones and the integrity of the immune and muscular systems [2]. Studies related to the emergence of erythema due to Sun exposure have demonstrated that short exposure times promote vitamin D synthesis but, on the other hand, long exposure times can promote skin cancer, cataracts and the suppression of the immune system [3]. It is worth mentioning that erythema is the reddening of the skin due to Sun exposure and for each skin phototype there is a minimum dose of ultraviolet energy that produces erythema (minimal erythema dose or MED). This MED depends on the irradiated skin zone and the number of exposures [4]. In 1975, Thomas Fitzpatrick changed the classification of the skin based on the color and responses to sun exporsure in terms of degrees of burning and tanning. This classification comprises six different skin types, from very fair (phototype "I") to very dark (phototype "VI") [5]. Some of the physical factors that have significant influence on the solar radiation are: altitude above sea level, albedo (reflectivity) of the ground, amount and types of aerosols (any solid or liquid particles suspended in the atmosphere) and the total amount of atmospheric ozone. The latter has an important influence on the level of UVB radiation [6]. If the altitude above sea level is increased, the total and UV (ultraviolet) solar irradiance increase, too. In particular, there is a mean increase of the global radiation ≈ 9% for every 1000 m of altitude and 23.7% in the erythema UV solar radiation [7]. In particular, it was found that in the Andean mountains the effect of altitude for UVB irradiance was only 7 – 9%, because in this region the amount of aerosols and tropospheric ozone is very small [8]. Then, it is worth noting that the levels of erythemal and vitamin D solar irradiance in Urcuquí are different from the ones in cities on the coast, at sea level. For example, with the satellite database of the OMI/Aura satellite on the Giovanni/NASA webpage we extracted the daily average UV Index (erythemal solar irradiance multiplied by 40 [m2/W]) in Urcuquí and Muisne (Ecuadorian city approximately at the same latitude than Urcuquí but at 100 masl) for the year 2014: IUVUrcuquí = 11.7 and IUVMuisne = 9.7. Thus, if we analyze the erythemal solar irradiance we obtain an erythemal solar irradiance of 0.2925 [W/m2] for Urcuquí and 0.2425 [W/m2] for Muisne which is equivalent to a difference of 20.2% and a positive variation of 9.4% per 1000 m of altitude increase. The Institute of Medicine (US) has recommendations for the adequate daily intake of vitamin D according to age range, but most experts agree that without adequate Sun exposure, children and adults require approximately 800 to 1000 IU per day [9]. Holick's rule says that ¼ MED generates enough energy to produce vitamin D3, i.e., equivalent to 1000 IU (international units) of vitamin D taken orally. This amount of energy is calculated over the erythemal irradiance and not over the spectral irradiance [10]. There is a significant correlation between the minimal exposure time to the Sun for the production of previtamin D3 and the maximum exposure time for the production of erythema. This exposure time range gives us a time interval in which sunbathing is healthy without developing erythema, providing enough energy for the vitamin D synthesis. This time interval was studied by Salum et al. [11] for Concepción del Uruguay (Argentina) in a whole year but only considered the erythemal irradiance. In that study the minimum time interval obtained for the minimum dose of vitamin D was 6.6 – 20.7 min. This paper uses the synthesis of previtamin D3 for the phototype "III" skin, according to the classification by Fitzpatrick [6]. We apply this to residents of Urcuquí (0.51° N, 78.2° W) in Ecuador, which is at an altitude of 2384 meters above sea level therefore the incidence of solar radiation is much higher than elsewhere in the country. 2 Materials and methods For a better understanding, it is necessary to review some methodological concepts used in this paper. First, the way to represent the response of each tissue for the wavelength of radiation is through the action spectrum. The action spectrum describes the relative effectiveness of the biological response for different wavelengths [3]. Among all action spectra of biological effects, the most popular is the CIE erythemal action spectrum (Commission Internationale de l'Éclairage) because erythema is the most common effect on the population. This erythemal irradiance is obtained by integrating the spectral irradiance weighted by CIE reference action spectrum up to 400 nm and normalized at 297 nm. In this paper we need to use the action spectra of vitamin D [12] and erythema [13]. In Figure 1, these two action spectra are presented. It is possible to see that the highest sensibility for vitamin D synthesis is at 298 nm (value equal to 1) and it has a range of response from 250 to 330 nm. In the case of the erythema production, its total response is between 285 and 300 nm. Fig. 1. Vitamin D action spectrum [12] (blue line) and erythema action spectrum [13] (red line). Second, the intensity of solar radiation for each wavelength that reaches the Earth's surface is called solar spectral irradiance, (units of W/(m2 nm)). The solar spectral irradiance on March 15th, 2015 can be seen in Figure 2 (top). The determination of the solar irradiation from models is very useful when there is no equipment for measuring it. In the present article, the SMARTS model [14] allows us to obtain the solar intensity (irradiance) for each wavelength over the terrestrial surface in any instant of time (of day or year). This model requires input data of the weather conditions and the atmospheric constituents. This information was obtained from satellite databases, like: SSE/NASA (Solar meteorology and Solar Energy) which has solar and meteorological data; ESRL/NOAA (Earth System Research Laboratory of National Oceanic and Atmospheric Administration) about the average concentration (per month) of atmospheric CO2; and Giovanni/NASA (Goddard Earth Sciences Data and Information Services Center) which has a website with access to different satellites. The ozone data have been obtained from OMI/NASA and the aerosol data from MODIS-Aqua Deep Blue. When the spectral irradiance of biological effects is needed, the mathematical convolution (the summation of the products for each wavelength) between the solar spectral irradiance I(,t) with the action spectrum ()of the biological effect, must be calculated. We will study the vitamin D (IvitD) and the erythemal (Iery) irradiances. The mathematical equations used to obtain each irradiance's biological effect are shown in the equations (1) and (2). (cid:2872)(cid:2868)(cid:2868) (cid:3041)(cid:3040) (cid:1835)(cid:3032)(cid:3045)(cid:3052)((cid:2019), (cid:1872)) = (cid:3533) (cid:1835)((cid:2019), (cid:1872)). (cid:2013)(cid:3032)(cid:3045)(cid:3052)((cid:2019)) (1) (cid:3090)(cid:2880)(cid:2870)(cid:2876)(cid:2874) (cid:3041)(cid:3040) (cid:2871)(cid:2871)(cid:2868) (cid:3041)(cid:3040) (cid:1835)(cid:3049)(cid:3036)(cid:3047)(cid:3005)((cid:2019), (cid:1872)) = (cid:3533) (cid:1835)((cid:2019), (cid:1872)). (cid:2013)(cid:3049)(cid:3036)(cid:3047)(cid:3005)((cid:2019)) (2) (cid:3090)(cid:2880)(cid:2870)(cid:2873)(cid:2870) (cid:3041)(cid:3040) As can be seen from the equations (1) and (2), the erythemal action spectrum extends from 250 to 400 nm [15] and the spectrum of the production of pre-vitamin D3 from 252 to the 330 nm [12]. Now, with Iery(,t) and IvitD(,t), the irradiance of biological effect can be obtained (in W/m2) by integrating the spectral irradiance with regard to wavelength. The integral is a mathematical operation similar to the calculation of the area under a curve (eqs. 3 y 4). This process can be seen in Figure 2. (cid:1835)(cid:3049)(cid:3036)(cid:3047)(cid:3005)((cid:1872)) = (cid:3505) (cid:1835)(cid:3049)(cid:3036)(cid:3047)(cid:3005)((cid:2019), (cid:1872)) (cid:1856)(cid:2019) (3) (cid:1835)(cid:3032)(cid:3045)(cid:3052)((cid:1872)) = (cid:3505) (cid:1835)(cid:3032)(cid:3045)(cid:3052)((cid:2019), (cid:1872)) (cid:1856)(cid:2019) (4) (cid:3090)(cid:3118) (cid:3090)(cid:3117) (cid:3090)(cid:3118) (cid:3090)(cid:3117) Each calculation of spectral irradiance and the following integration calculates only one point. In the present case, six points were calculated: one for every hour from 5:00 to 12:00 (local time). Since the solar irradiance is like a Gaussian curve, it must be symmetrical with regard to the solar noon. Then, the same six points calculated, until the solar noon, were used for 12:00 to 19:00 (local time). Finally, we did an interpolation of all points, in order to obtain the curve of the irradiance of the biological effect (Figure 2 in the right). In Figure 2, the arrow shows the correspondence between the area (upper panel) and the point on the curve (lower panel). 3 Results Following the above-mentioned considerations, the interpolation was carried out using a Gaussian distribution (cf. eq. (5)) with parameters selected for each case. These ones are shown in Table 1. (cid:1835)(cid:3032)(cid:3045)(cid:3052)((cid:1872)) = (cid:1877)(cid:2868) + (cid:1827). (cid:1857)(cid:2879) (cid:2869) (cid:2870).(cid:4672) (cid:3047)(cid:2879)(cid:3047)(cid:3278) (cid:3050) (cid:4673) (cid:3118) (5) Table 1. Selected parameters for each day and each biological effect for the Gaussian distribution. Paramete r Erythemal irradiance Vitamin D irradiance Erythemal irradiance Vitamin D irradiance December 15th, 2015 March 15th, 2015 y0 A tc w 0.001 0.403 12:20 0.087 0.001 0.830 12:20 0.088 0.001 0.348 12:20 0.095 0.001 0.677 12:20 0.090 Because of the variability of the solar irradiance, due to the atmospheric parameters, each irradiance of biological effect will be different. We hope to later automate this calculation using software with routines calculating areas under curves. The calculated vitamin D and erythemal irradiances from March 15th and December 15th, 2015, are presented in Figure 3 (left). Once the irradiances of biological effect for erythema and vitamin D are calculated, we proceeded to calculate the needed exposure times to obtain the ¼ MED for phototype "III" skin. According to Fitzpatrick [6], the MED for UVB radiation is between 30-50 mJ/cm2 for phototype "III". This is about 400 J (considering a medium value), so the ¼ MED is 100 J. Fig. 2. Top: Solar spectral irradiance (red line), Vitamin D action spectrum (blue points) and Vitamin D spectral irradiance (green line). Vitamin D irradiance is shown as the shaded green area which is equal to the value of the integral. Bottom: Vitamin D irradiance calculated from the integrals (blue points) and a curve of Vitamin D irradiance using these points (blue line). Table 2. Time of Sun exposure (min) that produces 100 J of energy. Local Time Vitamin D 5:00 6:00 7:00 8:00 9:00 10:00 11:00 12:00 125 73 34 15 7 4 3 3 March 15th, 2015 erythema % Dif 28.0 42.5 70.6 86.7 100.0 100.0 66.7 66.7 160 104 58 28 14 8 5 5 December 15th, 2015 Vitamin D erythema 124 73 36 17 8 5 3 3 142 89 49 25 13 8 6 5 % Dif 14.5 21.9 36.1 47.1 62.5 60.0 100.0 66.7 We calculated the minimum exposure time for each hour, starting at 5:00, until 12:00 local time. For this, the integral of the irradiance was calculated to reach the condition of 100J. The minimum exposure time for each analyzed day and irradiance is presented in Table 2 and Figure 3 (right). In Table 2, you can see the difference (in percentage) between the calculations made with the two irradiances (vitamin D and erythema). In some cases, the difference is 100% (cf. at 09:00 and 10:00 hours on March 15th, 2015). If other phototypes are considered for study, the MED for each one can be consulted on the scale proposed by Fitzpatrick. After an inspection of this scale it is easy to see that lower phototype means lower MED, and vice-versa. For example, a skin of phototype "I" can receive a MED of UVB radiation of 15-30 mJ/cm2, on the other hand, a phototype "VI" can receive 90-150 mJ/m2 before erythema emergence. To sum it up, if we want a model for all phototypes, we need to extend the present calculation to other skin types [16]. Figure 3 (right) depicts the minimum exposure time for obtain an energy of 100 J, using the erythemal (red) and the vitamin D (blue) irradiances on two different days (i.e., March 15th, 2015 –solid circles- and December 15th, 2015 –empty circles-). According to Figure 3 (right), there is an important difference between the erythemal or vitamin D irradiances for both days, namely, the former irradiance predicts higher exposure times. Fig. 3. Left: Vitamin D and erythemal irradiances for the two days analyzed. Right: Minimum exposure times using vitamin D and erythemal irradiances. Holick's rule for Vitamin D is based on the MED concept related to the erythemal effect, but the spectral action of this effect is different from the one for the production of vitamin D. When Table 2 is analyzed it is clear that the minimum exposure time for the synthesis of vitamin D has greater values when the erythemal spectrum is used. Therefore, we can conclude that the calculation of the exposure time with the erythemal model produces larger minimum time values. 4 Conclusions In this paper the estimation of the Sun exposure time in order to reach the minimum energy for the synthesis of vitamin D was studied. For comparison purposes, the solar spectral irradiance was calculated for two different days. The effect of the atmospheric constituents was considered consulting several satellite databases. The minimum exposure times using the erythemal irradiance were higher than those using the vitamin D irradiance. We will next ivestigate phototype "IV", which is very common in Ecuador. Furthermore, it is important to compare these results with other regions around the world with different values of latitude and altitude than the ones of Urcuquí. Finally, another interesting research line will be the effect of creams for different solar factors if we modify the minimum exposure time. Acknowledge The authors wish to thank Professor Elisabeth Griewanck for their selfless contributions to our work. References 1. Gilaberte, Y., Aguilera, J., Carrascosa, J., Figueroa, M., Romaní de Gabriel, J., Nagore, E.: La vitamina D: evidencias y controversias. Elsevier Doyma. Actas Dermo-Sifiliográficas. 2011;102(8):572-588 (2011) 2. Valero, M.Á., Hawkins, F.: Metabolismo, fuentes endógenas y exógenas de vitamina D. Hospital Universitario 12 de Octubre. Madrid-España. REEMO, 16(4):63-70 (2007) 3. Lovergreen, C., Álvarez, J.L., Fuenzalida, H., Aritio, M.: Radiación ultravioleta productora de eritema en Valdivia. Comparación entre inferencias satelitales, modelo de transferencia radiativa y mediciones desde Tierra. Revista médica chilena. 130 (1) (2002) 4. Rivas, M., Rojas, E., Méndez, J., Contreras, G.: Dosis eritémicas, sobreexposición a la radiación solar ultravioleta y su relación con el cáncer de piel en Arica, Chile. INTERCIENCIA. Jul 2014, Vol. 39, N° 7. 0378-1844/14/07/506-05 (2014) 5. Sachdeva, S. Fitzpatrick skin typing: Applications in dermatology. Indian J Dermatol Venereol Leprol, 75:93-6 (2009). 6. Sachdeva, S. Fitzpatrick skin typing: Applications in dermatology. Indian J Dermatol Venereol Leprol, 75:93-6 (2009). 7. Blumthaler, M.: Solar Radiation of the High Alps. In: Cornelius Lütz (eds) Plants in Alpine Regions. Springer-Verlag/Wien 2012, pp 11-20 .DOI: 10.1007/978-3-7091-0136-0_2 (2012) 8. Blumthaler, M., Rehwald, W., and Ambach, W. Seasonal Variations of Erythema Dose at Two Alpine Stations in Different Altitudes. Arch. Met. Geoph. Biocl., Ser. B, 35,389-397 (1985) 9. Holick, M.F.: Vitamin D Deficiency, The New England Journal of Medicine, 357;3, 266-281 (2007) 10. Dowdy, J.C., Sayre, R.M., Holick, M.F.: Holick's rule and vitamin D from sunlight. Journal of Steroid Biochemistry & Molecular Biology, 121, 328–330 (2010) 11. Salum, G.M., Ipiña, A., Ernst, M.J.: Estimación del tiempo de exposición al sol necesario para asegurar la producción de vitamina D y prevenir eritema para la ciudad de Concepción del Uruguay. Avances en Energías Renovables y Medio Ambiente. 12, 11.57-11.61 (2008) 12. CIE Technical Report. Action spectrum for the production of previtamin D3 in human skin. CIE 174:2006. ISBN 3901906509 (2006). 13. McKinlay, A.F., Diffey, B.L.: A reference action spectrum for ultraviolet induced erythema in human skin. In: Human Exposure to Ultraviolet Radiation: Risks and Regulations. 83-87, Elsevier, Amsterdam (1987) 14. Gueymard, C.A.: SMARTS2, A Simple Model of the Atmospheric Radiative Transfer of Sunshine: Algorithms and Performance Assessment. Technical Report No. FSEC-PF-270-95. Cocoa, FL: Florida Solar Energy Center (1995) 15. CIE Technical Report. Rationalizing nomenclature for UV doses and effects on humans. CIE 209:2014. ISBN 978-3-902842-35-0 (2014). 16. Astner, S., Anderson, R.: Skin Phototypes 2003. Journal of Investigative Dermatology, 122, 2, xxxxxxi (2004)
1003.3036	1	1003	2010-03-15T21:10:04	Investigating the Morphological Categories in the NeuroMorpho Database by Using Superparamagnetic Clustering	[ "physics.bio-ph", "q-bio.NC" ]	The continuing neuroscience advances, catalysed by multidisciplinary collaborations between the biological, computational, physical and chemical areas, have implied in increasingly more complex approaches to understand and model the mammals nervous systems. One particularly important related issue regards the investigation of the relationship between morphology and function of neuronal cells, which requires the application of effective means for their classification, for instance by using multivariated, pattern recognition and clustering methods. The current work aims at such a study while considering a large number of neuronal cells obtained from the NeuroMorpho database, which is currently the most comprehensive such a repository. Our approach applies an unsupervised clustering technique, known as Superparamagnetic Clustering, over a set of morphological measurements regarding four major neuronal categories. In particular, we target two important problems: (i) we investigate the coherence between the obtained clusters and the original categories; and (ii) we verify for eventual subclusters inside each of these categories. We report a good agreement between the obtained clusters and the original categories, as well as the identification of a relatively complex structure of subclusters in the case of the pyramidal neuronal cells.	physics.bio-ph	physics	Investigating the Morphological Categories in the NeuroMorpho Database by Using Superparamagnetic Clustering Krissia Zawadzki1, Mauro Miazaki1 and Luciano da F. Costa12 1 Institute of Physics at Sao Carlos - University of Sao Paulo, Avenue Trabalhador Sao Carlense 400, Caixa Postal 369, CEP 13560-970, Sao Carlos, Sao Paulo, Brazil 2 National Institute of Science and Technology for Complex Systems, Brazil E-mail: [email protected], [email protected], [email protected] Abstract. The continuing neuroscience advances, catalysed by multidisciplinary collaborations between the biological, computational, physical and chemical areas, have implied in increasingly more complex approaches to understand and model the mammals nervous systems. One particularly important related issue regards the investigation of the relationship between morphology and function of neuronal cells, which requires the application of effective means for their classification, for instance by using multivariated, pattern recognition and clustering methods. The current work aims at such a study while considering a large number of neuronal cells obtained from the NeuroMorpho database, which is currently the most comprehensive such a repository. Our approach applies an unsupervised clustering technique, known as Superparamagnetic Clustering, over a set of morphological measurements regarding four major neuronal categories. In particular, we target two important problems: (i) we investigate the coherence between the obtained clusters and the original categories; and (ii) we verify for eventual subclusters inside each of these categories. We report a good agreement between the obtained clusters and the original categories, as well as the identification of a relatively complex structure of subclusters in the case of the pyramidal neuronal cells. 0 1 0 2 r a M 5 1 ] h p - o i b . s c i s y h p [ 1 v 6 3 0 3 . 3 0 0 1 : v i X r a 2 1. Introduction The progress in neuroscience [1] aimed at understanding the complexity of the nervous system has continuously stimulated interdisciplinary integration between several scientific fields, such as biology, computer science [2], chemistry, and engineering. Despite the advances in these fields, there are still many remaining challenges. In this context, studies focused on morphological characterization and classification of neuronal cells have contributed substantially to enhance the neuroscientific knowledge about the relations between neuronal shape and physiology [3, 4, 5]. These discoveries are invaluable for several related areas, such as compared anatomy, neurobiology and diagnosis. The relationship between neuronal shape and function was first suggested and investigated by Cajal [6]. Subsequently, the attention from the neuroscientists shifted to electrophysiological analysis, which focused on neuronal electrical responses to stimulus. However, evidences about morphophysiological relationships have accumulated, such as the correspondence between morphological and physiological classifications of retinal ganglion cells [7, 8]. At the same time, advances in scientific visualization and shape analysis [9] paved the way to more comprehensive and sophisticated morphological approaches, giving rise to the area of computational neuromorphometry, whose aim is to quantify and study geometrical features of neurons [10]. The onset of the free-content initiatives, such as in software, artistic works, and research papers, have motivated the creation and expansion of public databases. Indeed, a free server of data allows researchers of several fields to have immediate access to the raw materials they need in their studies. The use of these databases also facilitate the replication of experiments and results, as anyone can access the same data [11]. Another important point is the prevention of data loss, as the servers cater for backup and data maintenance. Currently, the largest repository of neuronal cells is NeuroMorpho [12]. It contains not only the complete geometrical representation of the cells, but also several morphological measurements and data information, such as cell type, species, region, staining method, etc. The cell categories appearing in such databases frequently take into account both morpho or eletrophysiological properties. However, there is no consensus in the classification of neurons, to the extent that reclassifications of cells have been reported periodically (e.g. [13]). The difficulties in obtaining a more definitive taxonomy are, ultimately, a consequence of the incipient situation of neuromophological research. To begin with, there is no established set of geometrical measurements which should be used [10]. In addition, the lack of a representative number of cells allied to the choice of specific stochastic and classification methods also tends to yield varying taxonomies. Another important problem regarding the classification of neuronal cells is the identification of meaningful subcategories. In this context, the availability of significant amounts of neurons with their 3 morphological properties in the NeuroMorpho repository motivated an underlying and more systematic study of classification based on morphometric features. Though several categories of neuronal cells have been traditionally adopted in the literature [3, 14, 15, 5, 10], as a consequence of the largely subjective and incomplete methods used for their definition, it is not clear how homogeneous these classes are. Therefore, the investigation of the cell distribution within each of the main classes provides a particularly important issue. Especially for the more heterogeneous cases, it is possible that the existing categories are composed of subcategories which have been overlooked as a consequence of the subjectiveness and coarseness of the previous applied measurements and classification methods. The current work sets out to investigate this important issue, namely by investigating the uniformity and possible presence of subcategories inside well-established morphological categories. In this work, we explore the morphological measurements in NeuroMorpho by using a established classification procedure of statistical physics, known as Superparamagnetic Clustering (SPC) [16]. This method is inspired in a natural magnetic phenomenon presented by materials due to temperature variations. Described by non-homogeneous Potts model, where spins at the same state are grouped together, this physical phenomenon can be applied to data clustering applications. We used a software available in VCCLAB [17]. We chose four large categories of neurons according to the cell type-region classification and clustered them. The results were then compared to the original classification. The main objective in this work is to compare the obtained clusters with the original classification in the repository, checking the agreement between the original categories and our clustering. Since SPC is an unsupervised method, it will explore data and find clusters analysing their features without any subjective judgment. This approach can either confirm the strength of the established classification or reveal unnoticed subcategories and problems in classifications. Our analysis presents interesting results regarding the classification of the selected categories as well as their homogeneity, leading to the necessity to investigate the possible factors which are responsible to the observed subcategories and suggesting a new classification based in the inner relations in classes that exhibits this behaviour. This article starts by presenting the adopted database, followed by the morpho- logical concepts relative to the measurements used in the neuronal characterization. Afterwards, the theoretical method of SPC and the software used to analyse the Neu- roMorpho data are explained. Next, the results are presented and discussed. 2. Materials and Methods 2.1. NeuroMorpho Database NeuroMorpho [18] is a contributory database containing information about digitally reconstructed neuronal cells, collected by laboratories worldwide. Publicly available, 4 this data is intended for researches working on issues such as studies of neuronal system complexity, visualization and neuronal modeling. The maintenance of the database is provided by the Computational Neuroanatomy Group of the Krasnow Institute for Advanced Study, from George Mason University. This initiative is part of the Neuroscience Information Framework project [19], endorsed by the Society for Neuroscience, and includes institutes such as Cornell Univ., Yale Univ., Stanford Univ., and Univ. of California. The database has been periodically updated. Currently, it contains data relative to 5673 neurons (version 4.0, released in 02/16/2010). We can access the original and standardized cell morphological reconstructions, as well as several data and properties such as cell type, species, brain region, animal width and weight, development age, gender, used reconstruction methods, magnification, date of upload, images of spatial structures of neurons, and references to the related literature. In this work, we separated the data according to the cell type and brain region and selected the most numerous classes from the database. This yelded the following categories: • Pyramidal cells from Hippocampus (Pyr-Hip); • Medium Spiny cells from Basal Forebrain (Spi-Bas); • Ganglion cells from Retina (Gan-Ret); • Uniglomerular Projection neurons from Olfactory Bulb (Uni-Olf). Table 1 presents the distribution of species within each type-region category. Since this is a public database, all data used in this article is available and easily accessible through the NeuroMorpho website. The original category names are the same as found in the database. Table 1. Number of species in each type-region category. Species Spi-Bas Pyr-Hip Uni-Olf Gan-Ret Rat Mouse Salamander Drosophilla 232 1 0 0 209 0 0 0 0 0 0 233 0 181 64 0 2.2. Measurements In order to study the morphology of neurons, it is necessary to represent and characterize them in some way suitable for processing and analysis. NeuroMorpho provides L- Measure [20], a tool to extract several measurements from the neurons in the database. The measurements used in this work are illustrated in Figure 1, numbered from 1 to 19 and named as in the software documentation. 5 Figure 1. NeuroMorpho measurements. The concepts of compartment, branch and bifurcation are illustrated in Figure 1. Compartments are segments represented as cylinders with diameter and extremity points coordinates. Branches are formed with one or more compartments between the soma, the bifurcations and the tips. Bifurcations are points where a branch splits into two other branches. Measurements 1, 2 and 3 are the height, width and depth of a neuron, calculated after its alignment along the principal axis using PCA. The number of bifurcations and branches in a neuron correspont to the measurements 4 and 5. The features related to the compartment are from 6 to 9, respectively: diameter, length, surface area and volume. The branches have their associated measurements numbered from 10 to 14. Measure 10 is the Euclidean distance between a compartment and the soma, while the path distance (11) is the sum of the lengths of the compartments between two endpoints. Contraction (12) is the ratio between the Euclidean distance and its path distance. Measure 13 is the order of the branch regarding the soma, which has order 0. The 6 branches attached to the soma have order 1. The branches connected to these branches have order 2, and so on. Fragmentation (14) is the number of compartments in a branch. Only compartments between bifurcations or between a bifurcation and a tip are considered. Measurement 15 is the soma surface area. The soma can be of two types: a sphere or a set of compartments. In the latter case, the area is calculated as the sum of the area surfaces of the soma compartments. 2 1+dr dr br The other measurements are related to bifurcations. Pk classic (16) is the ratio , where r is the Rall's power law value, set in this measure as 1.5, and b, d1 and d2 are the diameters of the bifurcation compartments (the parent and the two daughters, respectively). The partition asymmetry (17) considers the number of tips on the left and on the right daughter subtrees of a bifurcation as n1 and n2 in the expression n1−n2 n1+n2−2. In Figure 1, the analysed bifurcation has vertical stripes, while the left daughter subtree has horizontal stripes and the right one has a pattern of squares. Then, in this example, 3−2 3+2−2 = 0.33. Measure 18 is the calculation of the angle n1 = 3 and n2 = 2 gives between two daughter compartments in a bifurcation, while measure 19 is the angle regarding the endpoints of two daughter branches. Table 2 shows the mean values of the measurements in each category. Table 2. Mean of the measurements in each type-region category. Measurement Spi-Bas Pyr-Hip Uni-Olf Gan-Ret Height Width Depth N. Bifurcations N. Branches Diameter Length Surface Volume Euclid. Dist. Path Dist. Contraction Branch Order Fragmentation Soma Surface Pk Classic Part. Asymmetry Local Bif. Ampl. Remote Bif. Ampl. 167.43 182.23 66.94 8.59 24.79 0.78 1266.38 2958.64 999.01 179.08 224.35 0.86 3.28 1003.34 715.14 1.30 0.45 63.55 56.30 614.88 576.74 353.68 99.29 209.06 0.71 23654.53 36941.94 9826.63 974.86 2727.95 0.85 18.96 3311.95 1005.75 1.60 0.54 69.76 55.24 51.78 117.13 56.54 25.42 52.17 1.07 499.74 1764.18 655.88 112.58 158.07 0.92 12.35 446.99 0.00 1.77 0.60 93.24 91.62 249.11 247.25 21.01 60.34 131.47 0.75 4339.40 9305.34 2455.30 218.86 291.82 0.86 8.97 2629.73 780.84 1.62 0.49 82.75 68.52 7 2.3. Superparamagnetic Clustering Method The superparamagnetic method is a clustering procedure based on a physical model of a real material exhibiting magnetic response to some external parameter. Different from classical approaches, which are restricted to statistical and mathematical analysis of the system, SPC allows to evaluate how efficient is the grouping in terms of its intrinsical properties and has as additional advantages insensitivity to the initial conditions and robustness to noise. Therefore, it is necessary to understand the theoretical foundation of the physical concepts underlying the approach, and how it can be applied to data clustering. The technique by Blatt, Wiseman and Domany [16] suggests an analogy with the generalized Ising model, also known as non-homogeneous Potts model. In this method, the temperature acts as a parameter controlling the spin configurations of a two-dimesional atoms network. It is possible to identify three reactions of the material in response to temperature variations: low, medium and high temperature intensities. They characterize the respective ferromagnetic, paramagnetic and superparamagnetic behaviour. The obtained arrangements of spin orientation is understood as defining the clustering structure of the data. The Potts model gives a reference to the energy E of the configuration, so that stability requires low energy: H(s) = −J N X<i,j> xixj (1) The transition between the magnetic phases due to variation of temperature can be characterized by the susceptibility χ, a physical parameter that reflects the respective overall magnetization of the system relative to the number of spins with a value between 1 and q, calculated as: where the variance of magnetization is defined as: The lowest energy probabilistic distribution of requires predominance of configurations with similar spins for low temperatures and strong interactions. This distribution is expressed as: the system states In order to apply these concepts to data sets and optimize the execution of the algorithm, the Swendsen-Wang method was adopted. The motivation of this N χ = m = T (cid:16)Dm2E − hmi2(cid:17) (Nmax/N)q − 1 q − 1 P (s) = 1 Z T ! exp H(s) where Z is a normalization constant: T ! exp −H(s) Z = XS (2) (3) (4) (5) 8 approach is to analyse different configurations of the system based on spin neighborhood interactions. The method proceeds as described in the following. First, we assume a data set containing N variables xi whose d components are measurements of the system features. In analogy to atoms in a two-dimensional network, we assign a random state si among the q possibilities to each respective point xi. The ranges and steps of temperature variation are predefined in order to choose the number of interactions in which we want to locate the superparamagnetic behaviour of system. Then, we analyse the probability (4) of connection between the sites with the same spin based on the mutual interaction and neighbourhood criterion. The latter suggests a maximum number K such that the interaction Jij between xi and xj is computed only if they are K-nearest neighbors of each other. This interaction is inversely proportional to the average nearest-neighbor distance a and is mathematically defined as: Jij = 1 K exp − dij 2a! (6) By changing the configuration, we the physical parameters of magnetization (3) and the susceptibility (2). This is repeated until the number M of interactions is reached. estimate The threshold temperatures Tf s and Tps, for which the system is in the superparamagnetic phase, can be located by identifying the points of maximal susceptibility and sharp decrease, respectively. In the transition to paramagnetic regime a guess to Tps can be T = exp1/2 /4 ln(1 + √q). Aimed at quantifying the ordering properties of the new system configuration, we , given in Equation 7, which need to estimate the spin-spin correlation Gij = (q−1) Cij +1 is typically performed by a Monte Carlo procedure. q Cij = Iij(ℓ) M (7) M Xℓ=1 where Iij = 1 if the points are at the same group or 0 otherwise. So, it is necessary to identify the Swendsen-Wang groups with connected sites, and then construct the data clusters, compute the magnetization average h ¯mi = Nmax repeat the above procedure until the maximal range of temperature is reached. N We can verify the different regimes of the superparamagnetic phase through the susceptibility measure. After locating Tf s and Tps, we can analyse the superparamagnetic sub-phases assuming their mean temperature Tclus = (Tf s + Tps)/2 as the point where the clusters are formed. The SPC algorithm follows all these steps and concepts, and has many applications. Currently, there are effective and optimized implementations available on the web, such as Tetko's program [21]. We used this software in the current article, described in the next section. 2.4. SPC software 9 is developed in Java, For our data analysis, we used the SPC software implemented by Tetko [21]. It freely available through the VCCLAB (Virtual Computational Chemistry Laboratory) web site at <http://www.vcclab.org/lab/spc>. VCCLAB aims to provide free on-line tools to analyse chemical data [17]. runs on-line as an applet and is In order to satisfy the required input format, we calculated the Euclidean distance between all data points and saved them in a text file to upload into the SPC program. We used all parameters in the default configuration, since these values are recomended by the author. The output is analysed and discussed in the following section. 3. Results and Discussions In order to compare the results of SPC and other approaches of morphological analysis, we applied PCA and LDA to the set of neuronal cells, which was selected based in the cell type and brain region. We also verified the agreement between each of the obtained clusters and the original categories. y t i l i b i t p e c s u S 0.016 0.014 0.012 0.01 0.008 0.006 0.004 0.002 0 0 Pyr-Hip Uni-Olf Gan-Ret Spi-Bas 0.05 0.1 0.15 0.2 0.25 Temperature Figure 2. Evolution of the susceptibility parameter to each type-region. Observe the points of maximal and decreasing with which we can identify the superparamagnetic regime. In both PCA and LDA, the Pyr-Hip cells are scattered, as we can see in Figures 3(a) and 4(a), respectively. This suggests that these neurons exhibit morphological features overlapping the other categories, instead of presenting more homogeneous characteristics which would otherwise imply in their separation from the other groups. The most homogeneous category is the Spi-Bas, which appears as a little, compact region in both methods. The Uni-Olf seems to have a more central position, intermediate to the other clusters. This behaviour is also observed for the Gan-Ret, whose location in LDA shows it to be distinct from the latter category. Pyr-Hip Spi-Bas Gan-Ret Uni-Olf 1 x 8 6 4 2 0 -2 -4 -4 -3 -2 -1 0 x2 (a) 10 Rat Mouse Salamander Drosophilla 1 x 6 4 2 0 -2 -4 -6 1 2 3 4 -4 -3 -2 -1 0 1 2 3 x2 (b) Figure 3. PCA of the four selected cell categories considering the type-region (a) and the species (b) classification. x1 and x2 correspond to the first and second components, respectively. 1 x 0.3 0.2 0.1 0 -0.1 -0.2 -0.3 -0.4 Pyr-Hip Spi-Bas Gan-Ret Uni-Olf Rat Mouse Salamander Drosophilla 1 x 2 1.5 1 0.5 0 -0.5 -1 -1.5 -2 -0.2 -0.15 -0.1 -0.05 0 0.05 0.1 -0.1 -0.05 x2 (a) 0.05 0.1 0 x2 (b) Figure 4. LDA of the four selected cell categories considering the type-region (a) and the species (b) classification. x1 and x2 correspond to the first and second components, respectively. In order to analyse the internal composition of the categories, we applied the SPC method in each one of them and verified the formation of inner clusters by using PCA (Figure 5) and LDA (Figure 6) approaches. The Pyr-Hip category revealed more heterogeneity, splitting in more clusters than the others during the SPC process (five clusters at the mean temperature Tclus = 0.07). This category also presented a susceptibility curve with a longer superparamagnetic phase, characterizing a behaviour different from those observed for the other classes. On the other hand, the Spi-Bas category presented three clusters not too far from each other and a big and sparse set of non-clustered cases. In the LDA, the homogeneity is indicated by the fact that the respective individuals appeared compacted in a specific region. Regarding the susceptibility curve, its main difference from the other classes is that it has only one peak at the superparamagnetic phase. 11 In the Gan-Ret category, we observed a single cluster in the superparamagnetic process, and two peaks of susceptibility with approximate values, relatively lower than the peaks on the other classes. The resulting clusters agree with the multivariated methods, especially in a region which overlaps with other categories. We can also see many non-clustered individuals. The Uni-Olf category presented the most characteristic susceptibility curve, giving rise to two clusters in the obtained distribution. The points are relatively sparse, and a more numerous subcategory can be seen, which is surrounded by many unclassified cases and another small cluster. Another interesting approach regarding the distribution of the neuronal cells and their categories concerns the investigation about the relationship between type-region and species, as well as the obtained clusters. This study was done for both PCA and LDA (Figures 3(b) and 4(b)) and compared with the direct references of the data set. We identified the species of the selected cells, finding them to correspond to rats, mice, salamanders and drosophillas. So, we count the number of these animals in each category (see Table 1) in order to verify the possible agreement with the results yielded by LDA and PCA. In most part of the measurements, it is possible to verify that the Pyr-Hip category has a significative higher mean value (see Table 2). Figure 5. SPC clusters of each category, visualized with PCA: (a) Spi-Bas, (b) Uni- Olf, (c) Gan-Ret, and (d) Pyr-Hip. In each pair of graphics, the left-hand graph shows the components 1 and 2, and the right-hand one exhibits the components 1 and 3. In order to check about possible influences of the original data properties (e.g. researcher, staining method, etc) on the clusters obtained in the case of the Pyr-Hip cells, we visualized the PCA and LDA multivariated projections marked accordingly to these properties. This category presented five well defined and distinguished clusters that could indicate influence of the original data properties or new distinct subcategories. The available information about the data is: researcher who provided the data, animal strain, minimum age, maximum age, age scale, gender, minimum weight, 12 Figure 6. SPC clusters of each category, visualized with LDA: (a) Spi-Bas, (b) Uni- Olf, (c) Gan-Ret, and (d) Pyr-Hip. In each pair of graphics, the left-hand graph shows the components 1 and 2, and the right-hand one exhibits the components 1 and 3. maximum weight, development, secondary brain region, tertiary brain region, original format, experimenting protocol, staining method, slicing direction, slice thickness, objective type, magnification, reconstruction method, date of deposition and date of upload. In Figure 7, we show only the properties for which we found some relationship with the clusters. The unclustered elements were eliminated and the data was reprojected with LDA, in order to obtain a better visualization of the clusters for this analysis (Figure 7(a)). Table 3. Pyr-Hip subclusters size. Cluster Number of elements Blue square Red diamond Green square Magenta x Cyan star 111 29 25 18 6 Table 3 shows the number of elements in the clusters of the Pyr-Hip category. The blue square cluster concentrates more than half of the elements, being underlain by a variety of property classes. On the other hand, the small cyan star cluster always presented its elements stable in the same property class through different information properties, but did not distinguish from the others since its classifications were never exclusive. The green square cluster is formed by the data from researchers (Figure 7(b)): Larkman (black cross), Barrionuevo (green squares) and part from Turner (red triangles). This cluster did not lead to a unique and concise classification. 13 Figure 7. LDA visualization of the clusters found by SPC in the Pyr-Hip category (a), and the original data properties: researcher (b), strain (c), development (d), staining method (e), and reconstruction method (f). In each pair of graphics, the left-hand graph shows the components 1 and 2, and the right-hand one exhibits the components 1 and 3. The neuronal data produced by the researcher Gulyas constitute almost all elements in the red diamond cluster and has unique classification in the original data properties: staining method (classified as Biotinylated dextran amine - Figure 7(e)) and reconstruction method (classified as Arbor - Figure 7(f)). In the strain category (Figure 7(c)), the Gulyas data shares its classification with the Spruston data. Both are classified as using the strain 'Wistar Rat'. All but one of the four Spruston elements are in the red diamond cluster. The magenta x cluster comprises all old rats, as we can see in the development data graphic in Figure 7(d). But, it also includes three exceptions (young rats). Examining the data, it was possible to find out that these old rats were 24 months old, while the young exceptions were 1 day old. Thus, although it gets all old rats, there are exceptions at the other side of the age cluster. Generally, we found no clear correspondence with any of those a priori properties. The fact that the obtained clustering structure could not be clearly explained by the original data properties suggests that the obtained subclusters are a consequence of some intrinsic morphological variation among the considered cells, possibly implying the definition of new categories. 14 4. Conclusion Among several challenges in neuroscience, the morphophysiological relationship of the neuronal cells has figured as an enhanced approach in order to establish the characterization and classification of these structures. Currently, the development of these studies have taken advantage of information avaliable in public databases, allowing the application of many pattern recognition and clustering methods. The problem has been continuosly investigated, leading to new techniques to classify a data set. The theoretical bases of these procedures can be established in physical real models, as Superparamagnetic Clustering. Aimed at investigating the clustering structures of the neuronal cells, we extracted many morphological measurements from NeuroMorpho, which is the largest repository currently, and compared the results of multivariated and clustering methods in the most numerous categories, whose analysis was performed considering all category elements and individual categories. In the first case, our purpose was to verify for agreement between the original classification and the categories obtained by the PCA and LDA. Afterwards, we isolated each selected category in order to locate internal clusters and the respective information that could explain their organization. The Pyr-Hip cells seemed to form the most heterogeneous category in both PCA and LDA results, in which remained sparse, as well as in the SPC results, where it had the higher number of clusters. The Gan-Ret category was located as intermediate among other categories and presented two species, despite the result of the SPC that revelead a single cluster surrounded by many unclassified cases. Given that the subclusters obtained for the Pyr-Hip category could not be explained by specific features of the original cells, we understand that they potentially imply in a revision of the current classification in order to account for possible new types of neuronal cells. New approaches can be used to complete these results, expanding the analysis to more properties, performing correlations and eliminating redundancy between the measurements, as well as the application of other clustering methods, such as the hierarchical Ward method, motivating further studies. 5. Acknowledgments Mauro Miazaki thanks FAPESP (07/50988-1) for financial support. Luciano da F. Costa is grateful to FAPESP (05/00587-5) and CNPq (301303/06-1) for sponsorship. 15 References [1] E. R. Kandel, J. H. Schwartz, and T. M. Jessel. Principles of neural science. McGraw-Hill, New York, 2000. [2] T. Trappenberg. Fundamentals of Computational Neuroscience. Oxford University Press, USA, 2002. [3] R. H. Masland. Neuron cell types. Curr. Biology, 14:497 -- 500, 2004. [4] Q. Wen and D. B. Chklovskii. A cost-benefit analysis of neuronal morphology. J. Neurophysiology, 99:497 -- 500, 2008. [5] M. Botaa and L. W. Swanson. The neuron classification problem. Brain Research Reviews, 56:79 -- 88, 2007. [6] S. R. Cajal. Recollections of my life. MIT Press, Massachussets, 1989. [7] B. B. Waessle, H. Boycott and R. B. Illing. Morphology and mosaic of on- and off-beta cells in the cat retina and some functional considerations. The Royal Society, 212:177 -- 195, 1981. [8] H. Waessle and L. Peichl. The structural correlate of the receptive field centre of alpha ganglion cells in the cat retina. J. Physiology, 341:309 -- 324, 1983. [9] C. R. Hosking and J. L. Schwartz. The future's bright: Imaging cell biology in the 21st century. Trends in Cell Biology, 9(11):553 -- 554, 2009. [10] L. da F. Costa. Computer vision based morphometric characterization of neural cells. Review of Scientific Instruments, 66:3770 -- 3773, 1995. [11] A. E. Dashti, S. Ghandeharizadeh, J. Stone, L. W. Swanson, and R. H. Thompson. Database challenges and solutions in neuroscienfic applications. NeuroImage, 97:97 -- 115, 1997. [12] NeuroMorpho.Org. [13] L. da F. Costa and T. J. Velte. Automatic characterization and classification of ganglion cells from salamander retina. J. Comp. Neurol., 404:33 -- 51, 1999. [14] J. E. Cook. Getting to grips with neuronal diversity. what is a neuronal type? Plenun Press, pages 91 -- 120, 1998. [15] J. Stone. Parallel processing in the visual system: the classification of retinal ganglion cells and its impact on the neurobiology of vision. Plenum Press, New York, 1983. [16] M. Blatt, S. Wiseman, and E. Domany. Superparamagnetic clustering of data. Physical Review Letters, 76(18):3251 -- 3254, 1996. [17] I. V. Tetko, J. Gasteiger, R. Todeschini, A. Mauri, D. Livingstone, P. Ertl, V. A. Palyulin, E. V. Radchenko, N. S. Zefirov, A. S. Makarenko, V. Y. Tanchuk, and V. V. Prokopenko. Virtual computational chemistry laboratory -- design and description. J. Comput. Aid. Mol. Des., 19:453 -- 63, 2005. [18] G. A. Ascoli, D. E. Donohue, and M. Halavi. NeuroMorpho.Org: a central resource for neuronal morphologies. Journal of Neuroscience, 27(35):9247 -- 9251, 2007. [19] M. Halavi, S. Polavaram, D. E. Donohue, G. Hamilton, J. Hoyt, K. P. Smith, and G. A. Ascoli. NeuroMorpho.Org implementation of digital neuroscience: Dense coverage and integration with the NIF. Journal of Neuroinformatics, 6(3):241 -- 252, 2008. [20] R. Scorcioni, S. Polavaram, and G. Ascoli. L-Measure: a web-accessible tool for the analysis, comparison and search of digital reconstructions of neuronal morphologies. Nature Protocols, 3:866 -- 76, 2008. [21] I. V. Tetko, A. Facius, A. Ruepp, and H.-W. Mewes. Super paramagnetic clustering of protein sequences. BMC Bioinformatics, 6:82, 2005.
1912.11777	1	1912	2019-12-26T04:53:18	Photonics study of probiotic treatment on brain cells exposed to chronic alcoholism using molecular specific nuclear light localization properties via confocal imaging	[ "physics.bio-ph", "physics.med-ph" ]	Molecular specific photonics localization technique, the inverse participation ratio (IPR), is a powerful technique to probe the nanoscale structural alterations due to abnormalities or chronic alcoholism in brain cells using the confocal image. Chronic alcoholism is correlated with medical, behavioral, and psychological problems including brain cell damage. However, probiotics such as Lactobacillus Plantarum has shown the promising result in soothing the human brain. This report, using the Confocal-IPR technique, nano to submicron scale structural abnormalities of the glial cells and the nuclei of alcoholic mice brain in the presence of probiotics. The increase in the structural disorder of alcoholic brain cells while the decrease or normalcy in the structural disorder of brain cells of mice fed with probiotics and alcohol simultaneously indicates that alcohol stimulates probiotics and enhances brain function.	physics.bio-ph	physics	Photonics study of probiotic treatment on brain cells exposed to chronic alcoholism using molecular specific nuclear light localization properties via confocal imaging Prakash Adhikari,1 Pradeep K. Shukla,2 Mehedi Hasan,1 Fatemah Alharthi,1 Binod Regmi,1and Radhakrishna Rao,2 Prabhakar Pradhan 1,* 1Department of Physics and Astronomy, Mississippi State University, Mississippi State, MS, USA, 39762 2Department of Physiology, University of Tennessee Health Science Center, Memphis, TN, USA, 38103 [email protected] and [email protected] and Abstract: Molecular specific photonics localization technique, the inverse participation ratio (IPR), is a powerful technique to probe the nanoscale structural alterations due to abnormalities or chronic alcoholism in brain cells using the confocal image. Chronic alcoholism is correlated with medical, behavioral, and psychological problems including brain cell damage. However, probiotics such as Lactobacillus Plantarum has shown the promising result in soothing the human brain. This report, using the Confocal-IPR technique, nano to submicron scale structural abnormalities of the glial cells and the nuclei of alcoholic mice brain in the presence of probiotics. The increase in the structural disorder of alcoholic brain cells while the decrease or normalcy in the structural disorder of brain cells of mice fed with probiotics and alcohol simultaneously indicates that alcohol stimulates probiotics and enhances brain function. 1. Introduction Photonics/light is an important probe for the characterization of the structural properties of cells. The structural properties of cells and tissues change from the normal with the progress of disease or abnormalities, such as cancer, stress, drugs, etc. Cells and tissues are weakly disordered dielectric media, therefore their structure, as well as any change in the structural properties, can be characterized by using light. The light probing is generally done by measuring the scattering signals to quantify the optical parameters of the samples. Different types of structural changes can happen in a cell with the progression of diseases such as carcinogenesis or due to the effects of drugs[1,2]. These changes in a cell may range from bulk structural change to nanoscale molecular specific structures alterations. The nanoscale structural changes in cells/tissues due to diseases or abnormalities are associated with the mass density fluctuation, in turn, the refractive index fluctuations that are explored using recently introduced mesoscopic-physics based sensitive imaging technique called the partial wave spectroscopy (PWS) [2 -- 6]. Although the overall change in a cell/nucleus has been studied for a long time to probe structural changes in a cell, however, these changes appear at the later stages of abnormalities or diseases which may be too late for treatment. The significant changes occur at the early stages of the abnormalities or diseases, where the molecular specific nanoscale structural changes occur due to the rearrangements of the macromolecules present in a cell, which is the major interest of study. Therefore, we hypothesize that probing the molecular specific structural changes can explain the physical state of a system. The concept of characterizing the molecular specific structural changes has been introduced recently to quantify the nanoscale changes in cells in abnormalities [1,3,7]. The initial case studies ranging from cancer to abnormalities in brain cells due to chronic alcoholism show promising results to characterize the physical state of the cell [7 -- 9]. Chronic alcoholism affects the brain, in cells and tissue levels. It has been shown by transmission emission microscopy (TEM) imaging that structural properties of the cells changes in chronic alcoholism at the nanoscales, and changes are prominent at around the length scales of ~100nm, which is related to the building blocks of the cells. Glial cells such as astrocytes, and microglia, and chromatin of the brain cells are specifically found to be highly affected due to chronic alcoholism which totals the vital portion of the central nervous system (CNS). Significance of literature showed that alcohol consumption damages the structure of the brain, from cellular to the molecular level. Chronic alcoholism can produce sustainable damage in the brain. This structural damage can bring neuropathological disorders, such as memory loss, dysfunction of the brain resulting in cognitive and behavioral deficits [10 -- 12]. In particular, astrocytes are star-shaped and most abundant cells i.e. glial cells in the central nervous system (CNS) and largely play as a regulator in the CNS immunity. In the same way, chromatin in all brain cell nuclei is a component that plays important role in genetic inheritance. In humans, a single astrocyte can interact with two million neurons at a time [13]. These astrocytes perform various tasks such as axon guidance, synoptic support, controlling blood flow and blood brain barrier as well. Along with their regulating behavior, they play a vital role in neuroinflammation in both beneficial and detrimental ways depending on the stimuli they receive from their inflamed environment. Alcohol is one of the well- known detrimental stimuli for neuroinflammation. Pathogenesis of many CNS disorders and several neurodegenerative diseases are crucially caused by brain inflammation [14]. This inflammation activates the glial cell, mainly the microglia and astrocytes that liberate the free radicals, cytokines, and inflammatory mediators that can damage the normal brain function [15]. A recent study showed that alcohol treatment in the microglial cells and chromatin altered their morphology [16]. In some cases, these structural changes may be reversible by different types of treatments. It has been reported that probiotic treatment has a good effect on brain health. In particular, it can reduce brain structural damage to a certain extent [17,18]. Initial results also show that probiotic treatment such as Lactobacillus Plantarum can be more effective in the presence of alcohol. Lactobacillus Plantarum is a Gram-positive lactic acid bacterium present in fermented food and in the gastro intestinal tract which has ample application in the medical field. The recently developed mesoscopic physics-based light localization technique, inverse participation ratio (IPR), can quantify the structural change in cells/tissues as the degree of structural disorder due to diseases or any other abnormalities [1,3,8,9]. In the IPR technique, an optical lattice is constructed using the pixel intensities of the confocal image and solved it for eigenvalues and eigenfunction under the closed boundary conditions. Finally, the eigenfunction is used to calculate the light localization properties of samples by the average of the IPR, <IPR> and the standard deviation of the IPR, (<IPR>). The earlier IPR analysis results show that the degree of disorder strength i.e. Ld is proportional to <IPR> or (IPR) [7,8]. This technique has been used to quantify the extent of aggressive cancer in biological cells, especially to detect the progress of carcinogenesis and to study the anti- cancerous drug effect [8,19]. We want to extend this method now to probe the molecular specific spatial structural changes in brain cells and organelles due to chronic alcoholism. In this work, using the IPR technique on a confocal image, we will first study the effect of alcohol in glial cells astrocytes, and microglia, and nuclei of mice brain cells by probing molecular specific structural changes at the nanoscale level. Then the effect of probiotic treatment in astrocytes, microglia, and nuclei in the brain cells of control fed, and lastly, the effect of probiotic in chronic alcoholism is studied. 2. Method 2.1 Molecular specific structural disorder analysis of confocal images by applying the inverse participation ratio (IPR) analysis technique: Confocal imaging is a technique used to capture the images with high optical contrast and resolution, ranging at submicron length scales. The main principle of the confocal microscopy imaging is that a spatial pinhole is used to block out-of-focus light to acquire a controlled depth of field and reduced background lights in images. Here, the samples are first treated with fluorescence that can emit a broadband light intensity at a specific wavelength. The amount of fluorophore dye that binds a molecular mass is proportional to the molecular mass at any point or a small voxel volume. Therefore, the fluorescence light collected at any point is approximately proportional to the fluorophore dye/elements, specific molecular mass present at that point or voxel. As dyes are independent of each other, and treating a cell with different molecular binding dyes at the same time and then probing appropriate wavelengths can provide us the different molecular specific spatial structural mass density fluctuations in a cell. The nanoscale mass density fluctuations can be quantified by calculating the degree of structural disorders in confocal images using the inverse participation ratio (IPR) technique. In the IPR technique, an optical lattice is formed using the pixel intensities of the confocal images. The optical lattice is a representation of the 'mass density fluctuations' that are scanned voxel-wise. Then Anderson tight binding model is used to obtain the Hamiltonian of the closed system and the eigenfunctions of optical lattices of light waves are used to analyze the localization properties of the sample. The light localization strength, therefore, indicates the level of structural disorder in the abnormal cells. An increase in the mass density fluctuations in the cells at the nanoscale level is represented by a higher value of the IPR. The <IPR> or (<IPR>) ultimately quantify the degree of structural disorder in the medium. As mentioned in [7,8], the pixel intensity I(r) of the confocal image at position r and ρ is the density of molecule at voxel amount dV(dV=dxdydz) of the image can be defined as: (1) It has been shown in [8,20] that the local refractive index of the cell slice is directly proportional to the local mass density of the cell at that point (x,y) and can be written as: ()()IrdV (2) where no is the average refractive index of the confocal images, dn(x,y) is the refractive index fluctuation at position (x,y) of the voxel dV. Now, If I0 is the average pixel intensity of the confocal images over the surface and dI(x,y) is the intensity fluctuation of the pixel at position (x, y) of the sample. Then, I(x,y) represents the confocal image intensity at any voxel point (x,y) of the sample cell given as: (3) It is to be noted that the value of intensity fluctuation, dI(x,y) is always less than the average intensity I0 (dI << I0). In the same way, the refractive index fluctuation, dn(x,y) is also less than the average refractive index n0 (dn << n0). The refractive index n(x,y) of thin scattering substances such as biological cells has a linear relation to the mass density [3,21]. The pixel intensity values at position (x, y) can be correlated with the refractive index fluctuation due to the mass density variation of the fluorescence molecules. Therefore, the confocal image's intensity I(x, y) is linearly proportional to the mass density M(x, y), and refractive index n(x, y) of the voxel. Also, a representative refractive index matrix constructed using the pixel intensity value [22] can be correlated with optical potential εi(x,y) as: (4) The pixel intensity values in the confocal image give the onsite optical potential, which is a representation of the spatial refractive index fluctuations of the fluorescent molecules inside the sample [21,23]. Anderson's tight binding model (TBM) is a widely recognized model studied in condensed matter physics and can be used to explain the disorder properties of any optical geometrical systems [8]. The Hamiltonian approach of the Anderson Tight Binding Model (TBM) produces the eigenfunctions to analyze the spatial structural properties of an optical lattice that has been produced from the confocal image. The optical potential at every point can be obtained from equation (4). If we consider one optical state per lattice site and the inner lattice site hopping restricted to the nearest neighbors then the Hamiltonian of Anderson tight-binding model [22,24,25] generated as: (5) Here, εi(x,y) is the optical potential energy of the ith lattice site, i> and j> are the eigenvectors of the ith and the jth lattice sites, and t is the inter-lattice site hopping strength. By the diagonalization method, we can generate the eigenfunctions (Ei's) from the Hamiltonian, H of the system. Using these eigenfunctions (Ei's) we calculate the average IPR value, <IPR> of the entire sample images as defined in [1,3,26,27]: (6) Where Ei is the ith eigenfunction of the Hamiltonian H of the optical lattice size L×L, and N is the total number of potential points on the refractive index matrix. For the heterogeneous light transparent medium such as biological cells, two parameters namely the refractive index fluctuations dn and its spatial fluctuation correlation length lc are used to specify the disorder of the system. This collective measurement provides the refractive index fluctuations inside the system termed as the disorder strength, Ld [7,8,23]. It is shown in [7,8] that the average IPR value <IPR> is directly proportional to the structural disorder strength i.e. , and can be expressed as: (7) (8) Lastly, the average (<IPR>) and standard deviation ((<IPR>)) of the IPR values or Ld indicated as potential biomarkers of the confocal images of biological cells to understand the morphological condition under any abnormalities. 2.2 Brain Cell Samples Preparation using a mouse model: 0(,)(,)nxyndnxy0(,)(,)IxyIdIxy00(,)(,)(,)idnxydIxyxynI(,)ixyiiijHijtijji41001(,)LLNiiIPRExydxdyNdcLdnldcIPRLdnldcIPRLdnlHere, we study the effect of probiotics on chronic alcoholic's brain cells and nuclei using a mouse model. In particular, we have studied the glial cells: astrocytes, and microglia which include the major portion of CNS, molecular structural properties of astrocytes cells using glial fibrillary acidic protein (GFAP) antibody and microglia cells using Anti- TMEM119 transmembrane protein antibody; DNA molecular structural properties of chromatin structure in the nuclei of brain cells using DAPI. To study the effect of alcohol and probiotics in the mice brain cells, 8-10 weeks mice bought from Jackson Laboratory were divided into 3-5 mice into 4 different groups. They are divided into four groups, equal males and females: (1) Control (PF), (2) Ethanol Fed (EF), (3) Probiotic Fed (PF+LP), (4) Probiotic and Alcohol Fed (EF+LP). To be more specific, we observed two glial cells and nuclei of the brain cells, namely: (I) Astrocytes, (II) Microglia, (III) Chromatin. The rationale for choosing the mentioned glial cells and chromatin. (I) Astrocytes: These cells represent the major portion of the CNS and can interact millions of neurons at a time. They get exposed to the external simulator such as alcohol and probiotic. This can be easily tagged by the glial fibrillary acidic protein (GFAP) antibody marker. (II) Microglia: These cells, similar to the astrocytes are the major part of CNS system and considered as the macrophages of the CNS which clean up the cellular debris and participate in neuroinflammation. They show their over-expression due to the presence of stimulators like alcohol and probiotic, and the cell can be tagged by the dye called Anti-TMEM119 transmembrane protein antibody. (III) Chromatin: In all brain cell nuclei, chromatin is a major part. Chromatin is the main DNA molecular component of the cell. It has been shown that chromatin spatial structures changes with the progress of disease and any abnormalities. A previous study using TEM probing has shown that nanoscale structural disorder increases with the progress of alcoholism. The chromatin can be easily stained by the standard DAPI dye. The nanoscale changes in these types of cells are shown well for the detection of cancer stages. We hypothesize that this will also work well for the brain disorder case. Bio-Marker: The main marker is to probe the confocal images using the IPR technique to quantify the molecular specific nanoscale molecular spatial mass density fluctuations, as mentioned in section 2.1. As described above mice were randomly divided into 4 categories and fed ethanol in a standard protocol in Leiber DiCarli liquid diet (0-6% v/v stepwise increase) with or without Lactobacillus Plantarum (106 cfu/ml) for 4 weeks. (I) The first group was saline fed or vehicle (II) the second group was fed with ethanol (III) the third group was fed with the probiotic only (IV) fourth group was fed with probiotics in the presence of ethanol. Brain tissues: When the mice were treated for 4 weeks, they were sacrificed and brains were removed. The brains were then cryo freeze and sectioned to 10µm slices by a microtome. antibody and DAPI. Confocal imaging: Brain cryo sections (10μm thick) were fixed in acetone: methanol (1:1) at −20°C for 2 min and rehydrated in PBS (137 mM sodium chloride, 2.7 mM potassium chloride, 10 mM disodium hydrogen phosphate, and 1.8 mM potassium dihydrogen phosphate). Sections were permeabilized with 0.2% Triton X-100 in PBS for 10 min and blocked in 4% nonfat milk in Triton-Tris buffer (150 mM sodium chloride containing 10% Tween 20mM and 20mM Tris, pH 7.4). It was then incubated for 1 hour with the primary antibodies (mouse monoclonal anti- Alexa Fluor 488 -- conjugated GFAP and rabbit polyclonal anti -- TMEM119), followed by incubation for 1 hour with secondary antibodies Cy3- conjugated anti-rabbit IgG antibodies) and 10 min incubation with Hoechst 33342. The fluorescence was examined using a Zeiss 710 confocal microscope (Carl Zeiss GmbH, Jena, Germany), and images from x -- y sections (1μm) were collected by LSM 5 Pascal software (Carl Zeiss Microscopy). Images were stacked by ImageJ software (Image Processing and Analysis in Java; National Institutes of Health, Bethesda, MD, USA) and processed by Adobe Photoshop (Adobe Systems, San Jose, CA, USA). All images for tissue samples from different group were collected and processed under identical conditions. discussed above. Staining: The different sections of the brain were treated with three types of dyes: GFAP antibody, Anti-TMEM119 IPR analyses: IPR analyses were performed to quantify the nanoscale structural change in the cells and nuclei, as 3. Results Confocal images of the mice brain cells stained with different protein/dye were obtained to study the effect of alcohol and probiotics in astrocytes, microglia and nuclei of the mice brain cells. IPR analyses were performed on the individual confocal images at (l  l)nm2 as described. The degree of <IPR(l)> values at different length scales for all molecular specific components of the mice brain cells were calculated. At each length scale l, at least 5 confocal micrographs of each category: astrocytes, microglia, and nuclei, of mice brain cells were analyzed separately and the ensemble averaging IPR calculated for their STD values. Finally, the calculated std(<IPR>) of astrocytes, microglia, and nuclei of the mice's brain cells of control, ethanol, and probiotics and ethanol fed were compared. PF EF PF+LP EF+LP (a) (b) (c) (d) l a c o f n o C s e g a m I > R P < I (a') (b') (c') (d') s e g a m I Fig. 1: Confocal and IPR images of astrocytes (PF, EF, PF+LP, EF+LP): (a)- (d) are the representative confocal images of control (PF), ethanol fed(EF), probiotics fed(PF+LP), and ethanol and probiotics fed(EF+LP) astrocytes of mice brain cells while (a')-(d') are their corresponding IPR images respectively. The IPR images are distinct from the confocal images and shows that structural disorder in astrocytes increases in the presence of ethanol as well as probiotics, but when treated together they decreases the alcohol damage. Fig. 2: Bar graph represents the relative study of molecular specific light localization property or (<IPR>) of astrocytes, brain cells of control (PF) mice with respect to ethanol fed (EF), probiotic fed (PF+LP), and ethanol and probiotics fed (EF+LP) mice. The IPR analysis of astrocytes show that the std of disorder strength or (<IPR>) of EF mice increase by 280% relative to PF mice. The (<IPR>) of astrocytes cells of mice fed with only probiotics is relatively higher compared to the control mice. This increase in aggressiveness indicates that there is some effect of probiotics in the astrocytes. However, the (<IPR>) of astrocytes cells of mice fed with EF+LP at the same time decreases by 100% relative to ethanol fed mice. This implies probiotics in the presence of ethanol get more stimulated and are beneficial for brain cells, reducing the (<IPR>) of astrocytes of the brain cells. PF EF PF+LP EF+LP (a) (b) (c) (d) l a c o f n o C s e g a m I > R P < I (a') (b') (c') (d') s e g a m I Fig: 3. Confocal and IPR images of microglia (PF, EF, PF+LP, EF+LP): (a)- (d) are the representative confocal images of control (PF), ethanol fed (EF), probiotics fed(PF+LP), and ethanol and probiotics fed(EF+LP) microglia of mice brain cells while (a')-(d') are their corresponding IPR images respectively. The IPR images are distinct than the confocal images and shows that structural disorder increases in the microglia of mice brain cells in the presence of ethanol. Fig. 4: Bar graph represents the relative study of molecular specific light localization property or (<IPR>) of microglia, brain cells of control (PF) mice with respect to ethanol fed (EF), probiotic fed (PF+LP), and ethanol and probiotics fed (EF+LP) mice. The IPR analysis of microglia shows that the std of disorder strength, (<IPR>) of EF mice increase by 120% in reference to PF mice. The (<IPR>) of microglia of mice fed with only probiotics is relatively higher than the control mice. This increase in aggressiveness indicates that there is some effect of probiotics in microglia of the brain cells. However, the (<IPR>) of microglia cells of mice fed with ethanol and probiotics at the same time decreases by 100% relative to ethanol fed mice. This implies probiotics in the presence of ethanol lead to more stimulated and soothed brain cells, reducing the (<IPR>) of microglia of the brain cells. PF EF (a) (b) PF+L P (c) EF+L P (d) l a c o f n o C s e g a m I > R P < I (a') (b') (c') (d') s e g a m I Fig: 5. Confocal and IPR images of nuclei chromatin (PF, EF, PF+LP, EF+LP): (a)-(d) are the representative confocal images of control (PF), ethanol fed(EF), probiotics fed(PF+LP), and ethanol and probiotics fed(EF+LP) of chromatin nuclei in the mice brain cells while (a')-(d') are their corresponding IPR images respectively. The IPR images are distinct than the confocal images and show that the structural disorder increases in the nuclei of mice brain cells in the presence of ethanol. Fig. 6. Bar graphs represents the relative values of DNA molecular specific (or chromatin) light localization property or (<IPR>) of nuclei in the brain cells of control (PF) mice with respect to ethanol fed (EF), probiotic fed (PF+LP), and ethanol and probiotic fed (EF+LP). The IPR analysis of nuclei in the brain cells of mice shows that the std of disorder strength, (<IPR>) of EF mice increase by 25% relative to PF mice. The (<IPR>) of nuclei in the brain cells of mice fed with probiotics only is relatively lower compare to the control mice. This decrease in aggressiveness indicates that there is some effect of probiotics in nuclei of the brain cells structure. However, the (<IPR>) of nuclei in mice brain cells fed with ethanol and probiotics at the same time decreases by 40% relative to EF mice. This implies probiotics in the presence of ethanol are more active and help in soothing brain and reducing the (<IPR>) of nuclei in the brain cells. Fig.3. (a)-(d), similar to Fig.1, shows the confocal image of microglia treated with red dye while Fig. 3(a')-(d') are their Fig.5. (a)-(d), similar to Figs. 1 and 3, shows the confocal image of nuclei treated with DAPI while Fig. 3(a')-(d') are Fig.1. (a)-(d) show the confocal image of astrocytes treated GFP while Fig. 1(a')-(d') are their corresponding IPR The representative confocal images of astrocytes from the thin section of mice brain cells fed with: (i) Control Fed (PF), (ii) Ethanol Fed (EF), (iii) Probiotics Fed (PF+LP), and (iv) Probiotics Fed simultaneous with Ethanol Fed (EF+LP) respectively represented in the above figures. images, respectively. In Fig. 2, the bar graphs of relative change in the std of disorder strength of astrocytes in the brain cells of control (PF) mice, ethanol fed (EF), probiotic fed (LP), and ethanol and probiotic fed (EF+LP) are shown. corresponding IPR images, respectively. Fig. 4, similar to Fig. 2, the bar graphs of relative change in the std of disorder strength of microglia cells of control (PF) mice, ethanol fed (EF), probiotic fed (LP), and ethanol and probiotic fed (EF+LP) are shown. their corresponding IPR images, respectively. cells of control (PF) mice, ethanol fed (EF), probiotic fed (LP), and ethanol and probiotic fed (EF+LP) are shown. The intensity variation in the IPR images represents structural disorder patterns in the different components of mice brain cells. Here, the higher intensity fluctuations are represented by the red color and lower intensity with blue color in all the IPR images. As can be visualized from the IPR images, the std of disorder strength or (<IPR>) increases in astrocytes, microglia, and chromatin in brain cells fed with ethanol relative to control indicates that alcohol has an adverse effect in the brain cells. However, when fed with only probiotics, the probiotics have some sort of effect on the brain so the disorder strength is slightly higher than the normal. This increase in the disorder strength of glial cells astrocytes, and microglia, and chromatin of the brain cells in the presence of probiotics only and go reaction to the brain cells to some extent. When the mice were fed with both probiotics and ethanol simultaneously, the (<IPR>) decrease back to normal or less than the normal, suggesting probiotics drug functions well in the presence of alcohol and increases the efficiency of the brain cells. Probiotics are considered good for brain cells and help to soothe and increase the cognitive function of the brain. Fig. 6, similar to Figs. 2 and 4, the bar graphs of relative change in the std of disorder strength of nuclei in the brain As can be seen from Figs. 1 and 2, the relative study of (<IPR>) shows that the structural disorder of astrocytes brain cells increases when mice were fed with ethanol, whereas it decreases significantly when mice were subsequently fed with probiotics and alcohol at the same time. Also, IPR analysis shows that the probiotics themselves have some interaction with the astrocytes and increases the structural disorder of astrocytes in the brain cells. The (<IPR>) of astrocytes brain cells of mice fed with only probiotics is relatively higher compared to the PF. Although probiotics are believed to boost mood and increase cognitive function of the brain, the result shows the brain can be reactive to some specific components of brain cells like astrocytes. In particular, the bar graphs show that the (<IPR>) of astrocytes in the brain cells of EF mice increase by 280% relative to PF mice. This implies that alcohol has an adverse effect in the brain cells, especially in astrocytes which perform a variety of tasks such as axon guidance, maintenance of redox potential, regulation in neurotransmitter and ion concentrations, synaptic support, blood flow control, removal of toxins and debris from the cerebrospinal fluid, etc. [28]. Astrocytes, being the most numerous cells within the central nervous system (CNS) of the brain, chronic alcoholism exacerbate neuronal dysfunction and advances mechanisms in potentiating or nullifying the pathway of neuropathologic injury [29]. On the other hand, the (IPR) of astrocytes cells decrease by 100% relative to EF mice when the mice were fed with probiotic and ethanol at the same time. This decrease in the (<IPR>) of astrocytes cells of EF mice supports that probiotics in the presence of ethanol are good for brain cells and help in maintaining proper brain functions. In Figs. 3 and 4, molecular specific light localization, the IPR analysis was performed in microglia, neuronal support cells present in the CNS which primarily function as the immune system of the brain. The relative study of the (<IPR>) of microglial cells of PF mice with EF, PF+LP, and EF+LP are presented. The IPR analysis performed at the same sample length of all cases shows that the (<IPR>) of EF mice increases by 120% relative to PF mice. This increase in the std of disorder strength of microglia brain cells of EF mice suggests that alcohol has a negative effect on the immune system of brain cells. Microglia are considered as the macrophages of the CNS which clean up the cellular debris and participate in neuroinflammation to various intrinsic and extrinsic stimuli. In addition to well established phagocytic function, and innate immune function microglia involve in the development of CNS immunopathology [30]. It is known that alcohol enhances immunomodulatory molecules such as corticosterone and endotoxin which degrades the neuroimmune cells of the brain and selectively modulates the intracellular signal transductions of microglia [31]. The IPR analysis shows that (<IPR>) of microglial cells of mice fed with only probiotics is relatively higher than in PF mice. This might be due to the interaction of probiotics with microglia brain cells in an adverse way resulting in increasing the degree of disorder strength. Probiotics used to soothe and increase the cognitive function of the brain may be sometimes reactive and harmful to the neuroimmune system of the brain cells; however, the (<IPR>) of microglia brain cells of mice fed with alcohol and probiotics simultaneously decreases by 100% relative to EF mice. That means probiotics have increased efficacy in the presence of alcohol which helps in reducing the (<IPR>) or disorder strength of microglia brain cells. Therefore, probiotics with alcohol are good for brain cells and enhance the development of CNS immunopathology. In Figs. 5 and 6, the color maps and bar graphs of relative change in the disorder strength of chromatin in brain cells of PF mice with respect to EF, PF+LP, and EF+LP are shown. The IPR analysis of DAPI stained confocal images of mice brain cells at sample length l nm shows that the (<IPR>) of EF mice increase by 25% in reference to PF mice. This increase in the (<IPR>) is due to the adverse effect of alcohol in the chromatin of brain cells. The increase in mass density fluctuations of the chromatin due to alcohol is responsible for an increase in the std of disorder strength. Persistent alterations to the chromatin structure are factors for epigenetics inheritance and can have a long-lasting influence on the activity and connectivity functions of the brain [32]. The (<IPR>) of nuclei in the brain cells of mice fed with only probiotics is relatively lower compared to the PF mice indicate that probiotics are good and soothe the chromatin of brain cells to some extent with the increase in chromatin's efficiency. Further, the (<IPR>) of nuclei in brain cells of mice fed with alcohol and probiotics at the same time decreases by 40% relative to ethanol fed mice. This decrease in the disorder strength of the chromatin of mice fed with alcohol and probiotics at the same time is due to the fact that probiotics in the presence of alcohol get more stimulated allowing them to function better which are good for the nucleus of the brain cells. The nanoscale quantification of alcohol effect on brain cells using the IPR analyses shows that alcohol has an adverse effect on astrocytes, microglia, and chromatin. Probiotics alone are also responsible for an increase in the std of disorder strength in astrocytes and microglia cells while a decrease in the std of disorder strength in the nuclei of the brain cells. Probiotics alone affect the astrocytes, microglia, and nucleus of the brain cells differently. However, the results show that probiotics in the presence of ethanol get stimulated and function better in the brain cells, resulting in soothing the cell structure and increasing to the cognitive function of the brain. 4. Conclusion The molecular specific light localization technique, Confocal-IPR is used to study the effect of probiotics in chronic alcoholism in brain cells and chromatin using a confocal image and a mouse model. The nanoscale mass density fluctuations quantified by std of the average IPR show an increase in disorder strength fluctuations or (<IPR>) from control fed to ethanol fed mice in astrocytes, microglia, and chromatin of the brain cells. This increase in structural disorder confirms the adverse effect of alcohol in astrocytes, microglia, and chromatin of the brain cells. On the other hand, an increase in the disorder strength of brain cells when control mice were fed with probiotics to improve the efficacy of brain might be due to brain cells reactions which are excited in the presence of probiotic drugs in some specific cells such as astrocytes and microglia. In detail, the study might be required to elucidate plausible reasons for the increase in (<IPR>) and have adverse effects in astrocytes and microglia although probiotic drugs are used to improve cognitive function of the brain. Interestingly, decrease in the (<IPR>) in astrocytes, microglia, and chromatin of the brain cells fed with ethanol and probiotics simultaneously relative to normal and ethanol fed indicate that probiotics in the presence of alcohol get highly stimulated and helps in soothing the brain cells physical structure and increasing the multifunctionality of brain. That means the astrocytes, microglia, and chromatin which includes a major portion of the CNS of the brain and perform vital brain functions such as blood flow control, removal of toxins and debris from the cerebrospinal fluid, immune system, genetic inheritance, neuroinflammation to various intrinsic and extrinsic stimuli, etc. get improved. The almost reversible effect of probiotics given with the alcohol in the brain cells could be a useful way to mitigate or improve abnormalities in the brain cells due to alcohol, stress or any other harmful sedative drugs, a major concern of modern life. As an illustration of the potential application of the mesoscopic physics-based IPR technique, we have successfully measured the structural alterations at the nanoscale level in the mice brain cells and chromatin due to probiotics in chronic alcoholism. Acknowledgments PPradhan acknowledges NIH and MSU for financial supports. References P. Pradhan, D. Damania, H. M. Joshi, V. Turzhitsky, H. Subramanian, H. K. Roy, A. Taflove, V. P. Dravid, and V. Backman, "Quantification of nanoscale density fluctuations by electron microscopy: probing cellular alterations in early carcinogenesis," Phys. Biol. 8(2), 026012 (2011). P. Adhikari, P. K. B. Nagesh, F. Alharthi, S. C. Chauhan, M. Jaggi, M. M. Yallapu, M. M. Yallapu, P. Pradhan, and P. Pradhan, "Optical detection of the structural properties of tumor tissue generated by xenografting of drug-sensitive and drug-resistant cancer cells using partial wave spectroscopy (PWS)," Biomed. Opt. Express, BOE 10(12), 6422 -- 6431 (2019). 1. 2. 3. 4. 5. 6. 7. 8. 9. P. Pradhan, D. Damania, H. M. Joshi, V. Turzhitsky, H. Subramanian, H. K. Roy, A. Taflove, V. P. Dravid, and V. Backman, "Quantification of nanoscale density fluctuations using electron microscopy: Light-localization properties of biological cells," Appl. Phys. Lett. 97(24), 243704 (2010). H. Subramanian, P. Pradhan, Y. Liu, I. R. Capoglu, J. D. Rogers, H. K. Roy, R. E. Brand, and V. Backman, "Partial-wave microscopic spectroscopy detects subwavelength refractive index fluctuations: an application to cancer diagnosis," Opt Lett 34(4), 518 -- 520 (2009). P. Adhikari, F. Alharthi, and P. Pradhan, "Partial Wave Spectroscopy Detection of Cancer Stages using Tissue Microarrays (TMA) Samples," in Frontiers in Optics + Laser Science APS/DLS (2019), Paper JW4A.89 (Optical Society of America, 2019), p. JW4A.89. S. Bhandari, P. Shukla, H. Almabadi, P. Sahay, R. Rao, and P. Pradhan, "Optical study of stress hormone-induced nanoscale structural alteration in brain using partial wave spectroscopic (PWS) microscopy," Journal of Biophotonics 0(ja), e201800002 (n.d.). P. Sahay, A. Ganju, H. M. Almabadi, H. M. Ghimire, M. M. Yallapu, O. Skalli, M. Jaggi, S. C. Chauhan, and P. Pradhan, "Quantification of photonic localization properties of targeted nuclear mass density variations: Application in cancer-stage detection," Journal of Biophotonics 11(5), e201700257 (2018). P. Sahay, H. M. Almabadi, H. M. Ghimire, O. Skalli, and P. Pradhan, "Light localization properties of weakly disordered optical media using confocal microscopy: application to cancer detection," Opt. Express, OE 25(13), 15428 -- 15440 (2017). S. Bhandari, P. K. Shukla, P. Sahay, A. S. Meena, P. Adhikari, R. Rao, and P. Pradhan, "Photonic localization probe of molecular specific intranuclear structural alterations in brain cells due to fetal alcoholism via confocal microscopy," arXiv:1812.10652 [physics] (2018). 10. M. A. Ron, "Brain damage in chronic alcoholism: a neuropathological, neuroradiological and psychological review," Psychological Medicine 7(1), 103 -- 112 (1977). C. Harper and I. Matsumoto, "Ethanol and brain damage," Curr Opin Pharmacol 5(1), 73 -- 78 (2005). 11. 12. A. Topiwala and K. P. Ebmeier, "Effects of drinking on late-life brain and cognition," Evidence-Based Mental Health 21(1), 12 -- 15 (2018). 13. R. D. Fields, A. Araque, H. Johansen-Berg, S.-S. Lim, G. Lynch, K.-A. Nave, M. Nedergaard, R. Perez, T. Sejnowski, and H. Wake, "Glial Biology in Learning and Cognition," Neuroscientist 20(5), 426 -- 431 (2014). S. Hunot and E. C. Hirsch, "Neuroinflammatory processes in Parkinson's disease," Annals of Neurology 53(S3), S49 -- S60 (2003). S.-M. Lucas, N. J. Rothwell, and R. M. Gibson, "The role of inflammation in CNS injury and disease," British Journal of Pharmacology 147(S1), S232 -- S240 (2006). B. J. Crenshaw, S. Kumar, C. R. Bell, L. B. Jones, S. D. Williams, S. N. Saldanha, S. Joshi, R. Sahu, B. Sims, and Q. L. Matthews, "Alcohol Modulates the Biogenesis and Composition of Microglia-Derived Exosomes," Biology 8(2), 25 (2019). C. Park, E. Brietzke, J. D. Rosenblat, N. Musial, H. Zuckerman, R.-M. Ragguett, Z. Pan, C. Rong, D. Fus, and R. S. McIntyre, "Probiotics for the treatment of depressive symptoms: An anti-inflammatory mechanism," Brain, Behavior, and Immunity 73, 115 -- 124 (2018). L. Desbonnet, L. Garrett, G. Clarke, J. Bienenstock, and T. G. Dinan, "The probiotic Bifidobacteria infantis: An assessment of potential antidepressant properties in the rat," Journal of Psychiatric Research 43(2), 164 -- 174 (2008). P. Adhikari, M. Hasan, V. Sridhar, D. Roy, and P. Pradhan, "Nanoscale structural alterations in cancer cells to assess anti-cancerous drug effectiveness in cancer treatment using TEM imaging," arXiv:1909.02665 [physics] (2019). P. Pradhan and S. Sridhar, "Correlations due to localization in quantum eigenfunctions of disordered microwave cavities," Phys. Rev. Lett. 85(11), 2360 -- 2363 (2000). P. Pradhan and S. Sridhar, "From chaos to disorder: Statistics of the eigenfunctions of microwave cavities," Pramana - J Phys 58(2), 333 -- 341 (2002). B. Kramer and A. MacKinnon, "Localization: theory and experiment," Rep. Prog. Phys. 56(12), 1469 -- 1564 (1993). J. Beuthan, O. Minet, J. Helfmann, M. Herrig, and G. Müller, "The spatial variation of the refractive index in biological cells," Phys Med Biol 41(3), 369 -- 382 (1996). P. A. Lee and T. V. Ramakrishnan, "Disordered electronic systems," Rev. Mod. Phys. 57(2), 287 -- 337 (1985). E. Abrahams, P. W. Anderson, D. C. Licciardello, and T. V. Ramakrishnan, "Scaling Theory of Localization: Absence of Quantum Diffusion in Two Dimensions," Phys. Rev. Lett. 42(10), 673 -- 676 (1979). 14. 15. 16. 17. 18. 19. 20. 21. 22. 23. 24. 25. 26. V. N. Prigodin and B. L. Altshuler, "Long-Range Spatial Correlations of Eigenfunctions in Quantum Disordered Systems," Phys. Rev. Lett. 80(9), 1944 -- 1947 (1998). 27. N. C. Murphy, R. Wortis, and W. A. Atkinson, "Generalized inverse participation ratio as a possible measure of localization for interacting systems," Phys. Rev. B 83(18), 184206 (2011). 28. M. Sidoryk-Wegrzynowicz, M. Wegrzynowicz, E. Lee, A. B. Bowman, and M. Aschner, "Role of Astrocytes in Brain Function and Disease," Toxicol Pathol 39(1), 115 -- 123 (2011). F. Aloisi, "Immune function of microglia," Glia 36(2), 165 -- 179 (2001). 29. M. Aschner, U. Sonnewald, and K. H. Tan, "Astrocyte Modulation of Neurotoxic Injury," Brain Pathology 12(4), 475 -- 481 (2002). 30. 31. K. Suk, "Microglial signal transduction as a target of alcohol action in the brain," Curr Neurovasc Res 4(2), 131 -- 142 (2007). 32. C. Dulac, "Brain function and chromatin plasticity," Nature 465(7299), 728 -- 735 (2010).
1008.0737	1	1008	2010-08-04T10:18:44	Protein-DNA computation by stochastic assembly cascade	[ "physics.bio-ph", "q-bio.BM" ]	The assembly of RecA on single-stranded DNA is measured and interpreted as a stochastic finite-state machine that is able to discriminate fine differences between sequences, a basic computational operation. RecA filaments efficiently scan DNA sequence through a cascade of random nucleation and disassembly events that is mechanistically similar to the dynamic instability of microtubules. This iterative cascade is a multistage kinetic proofreading process that amplifies minute differences, even a single base change. Our measurements suggest that this stochastic Turing-like machine can compute certain integral transforms.	physics.bio-ph	physics	Protein–DNA computation by stochastic assembly cascade Roy Bar-Ziv, Tsvi Tlusty, and Albert Libchaber* Center for Physics and Biology, The Rockefeller University, 1230 York Avenue, New York, NY 10021 The assembly of RecA on single-stranded DNA is measured and interpreted as a stochastic finite-state machine that is able to discriminate fine differences between sequences, a basic compu- tational operation. RecA filaments efficiently scan DNA sequence through a cascade of random nucleation and disassembly events that is mechanistically similar to the dynamic instability of micro- tubules. This iterative cascade is a multistage kinetic proofreading process that amplifies minute differences, even a single base change. Our measurements suggest that this stochastic Turing-like machine can compute certain integral transforms. E ver y computation requires a reliable recogn ition of its input data. Any scheme for computation based on protein–DNA binding must attain this recogn ition w ithin the physical proper- ties of this interaction, its specificity, af fin ity, and cooperativ ity. These properties define biochemical networks such as those used by the cell to process information received from stimuli and to compute its response. The resulting computations are inherently stochastic due to the ‘‘noisy’’ nature of biochemical pathways that resemble more a probabilistic pinball machine than a determin- istic desktop PC. So far, artificial, in vitro biomolecular computing strategies relied mainly on Watson–Crick complementarity of DNA (or RNA). These schemes, which were used to solve a certain class of hard-to-compute problems, are almost determin istic due to the relatively high hybridization energ y. The typical algorithm, combinatorial search, encodes potential solutions as DNA se- quence librar y and then selects correct solution(s) v ia parallel filtration, eliminating the wrong solutions by man ipulations based on complementarity (1, 2). Another approach constructs fin ite-state computing machines, the internal states of which are encoded in DNA sequence (3). Here we report an in vitro stochastic biomolecular computa- tion based on low-specificity protein–DNA binding: An assembly cascade of RecA proteins on single-stranded DNA can discrim- inate between similar sequences, thus fulfilling a basic compu- tational task that may be one stage in a more complex compu- tation. The assembly process overcomes the error-prone nature of the single protein binding by constructing a multistage cascade, similar to k inetic proofreading (4), in which many proteins bind and unbind collectively. We find that the dynamics of the cascade is mechan istically similar to the dynamic insta- bility of microtubules, which is used as an ef ficient space search algorithm w ithin the liv ing cell (5). It also resembles a stochastic counter (6), an imper fect digital apparatus that registers the number of certain events (think of a voting machine). The collective, nonlinear mode of operation of the cascade enables sensitive discrimination of minute length and sequence dif fer- ences including a single base change. The hardware of our molecular machine comprises a test-tube filled w ith a solution of single-stranded DNA molecules, RecA proteins, and AT P molecules that fuel the assembly cascade. When the concentration of RecA monomers exceeds some onset value, they start to form helical filaments, one RecA monomer per each base triplet, that envelope and stretch the DNA (7). A filament first forms when a nucleus, a RecA monomer, binds to a random site along the DNA and then extends rapidly by polymerization to the 3⬘ end of the empty strand. When bound to DNA, RecAs hydrolyze AT P and change their conformation into a less stable state. The RecA that is closest to the 5⬘ end, w ith only one neighboring monomer, tends to disassemble back into the solution when hydrolyzing AT P (8). The resulting assembly– disassembly cascade is asymmetric; while nucleation events extend the filament by long chunks, disassembly removes mono- mers one by one. A graphical man ifestation of this stochastic asymmetr y is the irregular saw-tooth form of the filament length (or machine state) dependence on time (Fig. 1A). Rather than further describing the extensively studied bio- chemistr y of RecA assembly (7) we focus on the computational features of this protein–DNA molecular machine, its ‘‘sof tware.’’ We use here the notion of ‘‘machine’’ in the sense of certain physical realization of an abstract computation, sequence dis- crimination in our case. Nucleation and disassembly are the two basic operations of this machine. They change the machine’s internal state, which is determined by the current length of RecA filament. To describe the machine dynamics, we use the tradi- tional state-transition diagram, where circles represent states, and arrows represent transitions between states (Fig. 1B). In state Qn, n binding sites out of total N sites along the DNA are vacant, and the RecA filament length is therefore N ⫺ n (Q0 is a fully covered DNA, and QN is an empty strand). Clearly, this is a f inite-state machine w ith the number of states equal to the number of binding sites, N. The symbols on each arrow represent the probability per un it of time that such transition occurs given that the machine is in the state at the tail of the arrow. Disassembly can take the machine from state Qn to the next state of the cascade Qn⫹1 at rate ␬⫺ whereas at nucleation events the machine jumps from Qn to any of the lower states Qm, m ⬍ n, at rate ␬⫹. We also need an output dev ice that w ill report the machine’s current state. In the experiment, the molecular ma- chine ‘‘reports’’ its state through a change in the rotational motion of the DNA molecule, which is directly related to the number of bound RecA monomers and measured by f luores- cence an isotropy (9). The stochastic state-transition diagram can be expressed as a set of N dif ferential equations for the probabilities pn that the machine is at state Qn. Summing the incoming (first two terms) and outgoing (last two terms) transitions at each state of the diagram we obtain dp n ⫽ ␬⫺pn⫺1 ⫹ ␬⫹ 冘N d t m⫽n⫹1 pm ⫺ ␬ p ⫺ n ⫺ ␬⫹npn. [1] The st ate-transition diag ram couples each polymerization st ate Qn w ith all the lower st ates Qm, m ⬍ n (Fig. 1), and the equivalent master equation (Eq. 1) is therefore integ ro- dif ferential, w ith boundar y and normalization conditions dpN兾dt ⫽ ␬⫺p N N⫺1 ⫺ NpN, 兺 n⫽0 pn ⫽ 1. To reduce the con- nectiv it y of the st ate-transition diag ram, we express it in terms N n ⫽ 兺 m⫽n pm, the probabilit y to find the of the cumulative P machine at a st age higher or equal to Qn. It is also the probabilit y that the filament is shorter than N ⫺ n and that site Master equations such as 1 and 2 are generic in stochastic transition processes, especially in chemical k inetics (10). What makes the computing-machines terminolog y natural in our case is the understanding that the RecA-binding cascade processes information encoded in the DNA sequence. This may be clarified if one considers a concrete machine model of the cascade. This time we think of a Turing-like dev ice, a determin istic machine that is coupled to an infin ite tape through a reading head (Fig. 1C). The internal states of the machine are the same N binding states Qn. The noisy Brown ian dynamics of the cascade is embedded in the tape, which is produced by the follow ing procedure: Time is div ided into an infin ite series of short equal segments that correspond to the squares on the tape. To each square we randomly assign a symbol w ith a probability that matches the transition rates. We denote disassembly from state Qn by dn, nucleation to state Qn by en, and in the rest of the squares we write x to denote that nothing happens during the corresponding time duration. The machine reads the squares sequentially from, say, lef t to right and responds according to the symbol written in the current square. Suppose that the machine is at state Qn, then it responds according to a simple set of rules. (i) If it reads dn, it moves to state Qn⫹1. (ii) If it reads em and m ⬍ n the machine moves to state Qm. (iii) In all other cases, if it reads x or dm w ith m ⫽ n, ore m w ith m ⫽ n, then it stays at state Qn. Af ter its state is determined, the reading head moves one square to the right. Stochastic automata are natural to information processes ever since they emerged in Shannon’s classical study of commun ica- tion channels (14, 15). The notion of stochastic computers was introduced to the molecular realm in Bennett’s discussion of DNA translation and replication, where the computational task is sequence copying (16). We show below that rather than Xerox ing the sequence like RNA- and DNA-polymerase, the RecA cascade carries out another type of computation, the discrimination of close-by sequences. Sequence information is encoded in the random tape through the dependence of the probabilities for disassembly (dn) and nucleation (en) events at a certain site n on the specific base triplet. This information can be equivalently encoded as sequence-specific transition rates, ␬⫺(n) and ␬⫹(n), in the state-transition diagram and the corre- sponding master equation. A lthough RecA is a nonspecific binding protein w ith similar af fin ities for many possible triplets, our measurements show (9) that the collective assembly cascade constructed from these low-specificity components is a highly specific detector that can amplify and discriminate even minute sequence dif ferences (17). The ‘‘sequence-detector’’ machines we construct are assem- bly cascades on single-stranded DNAs, 39 or 78 bases long (13- and 26-st age machines). Any measurement that tries to ‘‘look inside’’ such a stochastic machine, that is to infer its internal dynamics f rom obser vable output, has to rely on st atistical analysis (18). One must collect a suf ficient set of obser vations to overcome the noisiness of the output. We resolve this dif ficult y by simult aneously measuring many identical ma- chines, ⬇105–106 f luctuating DNA–RecA complexes that pro- duce a ver y smooth ensemble-average signal. An alternative approach could be time averaging over a single-molecule signal (19). Our DNAs carr y a f luorescent dye att ached to their 3⬘ end. The f luorescence an isotropy of the dye reports RecA binding as it slows down the rot ational motion of the DNA (9). The response of the cascade is examined as we tune the interplay between nucleation and disassembly by changing the available amount of RecA in the solution. The nucleation rate diffusion remains minor for short enough inhomogeneous sequences such as the sequence with the point mutation measured in the experiment. In contrast, assembly on longer sequences, (␬⫹兾␬⫺)N ⬇ 1, is predicted to exhibit a striking randomness effect of both sequence and diffusion, which may lead to anomalous motion and phase transitions (13). Fig. 1. (A) Simulation of a 26-state stochastic assembly cascade. State Qn is a RecA filament of length 26n. The assembly machine advances to higher states through protein disassembly at the filament end (open circles) and to lower states by nucleation (solid circles). The machine’s state fluctuates strongly around the ensemble average (dashed). (B) State-transition repre- sentation of the machine dynamics. The interplay between N disassembly steps at rate ␬⫺ (Left) and N(N ⫹ 1)兾2 nucleation paths at rate ␬⫹ (Right). (C) A deterministic reading head changes its internal state according to the tape it reads square by square from left to right. The tape is produced in a random process that maps the stochasticity of the assembly (see text). n is empt y. Two processes can alter Pn: (i) disassembly f ronts vacate site n at a rate proportional to the front position distri- bution, ⫺pn ⫽ Pn ⫺ Pn⫹1, and (ii) nucleation at any of the n vacant sites fills site n. The dynamics is simpler, since any possible nucleation from a higher state, Qm, m ⬎ n leaves Pn unchanged. The resulting master equation is local, dPn兾dt ⫽ ⫺␬⫺(Pn⫹1 ⫺ Pn) ⫺ ␬⫹nPn, with the normalization P0 ⫽ 1. Technically, one can obtain this result directly by summing Eq. 1 from n to N w ith the boundar y conditions. Think ing of n as a spatial coordinate, we approx imate the discreet master equation for Pn(t) by a‘‘drif t’’ equation for the continuous cumulative probabilit y P(n, t). ⭸P 共n 兲 ⭸ t ⫽ ⫺␬⫺ ⭸P共n兲 ⭸n ⫺ ␬⫹nP共n兲 [2] The disassembly term is approx imated by a gradient, neglecting higher order derivatives in the Kramers–Moyal expansion (10). In particular, we omit the familiar second-order dif fusive term that plays a minor role as long as disassembly is much faster than nucleation rate, in the regime, (␬⫹兾␬⫺)N ⬍⬍ 1, which includes the large f luctuation regime, (␬⫹兾␬⫺)N2 ⬇ 1, where the RecA- assembly cascade is the most sensitive.† †We note that the assembly dynamics differs essentially from the Langevin dynamics of a particle diffusing in a one-dimensional random force field (11) or the related asymmetric exclusion process (12): While a diffusing particle travels continuously, the filament end can abruptly jump to a new site by nucleation. The master equation, therefore, does not lead to the familiar Fokker–Planck equation, and an equivalent Langevin formulation would require infinite stochastic forces to enable the nucleation jumps (10). The effect of Fig. 2. Increase of fluorescence anisotropy, A, upon RecA binding on (TAC)13 measured in steady state as a function of RecA concentration R (Inset). Thermal rotation of naked DNAs decreases the intrinsic anisotropy of the dye, Am ⬃ 0.25 ⫺ 0.31, to A0 ⬃ 0.05, as reported previously (9). By varying sequence length, we found that A increases linearly with RecA binding near the 3⬘ end before saturating at Am with l0 ⫽ 7 RecA monomers (unpublished data). The anisotropy therefore measures an average over the cascade occupancy, A ⫽ Am ⫺ ( Am ⫺ A0)S; S ⫽ 1兾l0 兰N ⫺ l0 N P(n)dn. Independent kinetic measurement indicated ␬T ⫽ ␬⫹兾R ⫽ 6.4 ⫻ 10⫺3 sec⫺1䡠␮M⫺1 and ␬⫺ ⫽ 4.4 ⫻ 10⫺2 sec⫺1. The normalized anisotropy, S, is plotted as a function of the rate ratio ␬⫹兾␬⫺. Both curves, (TAC)13 (open circles) and (TAC)26 (triangles), decay exponentially close to saturation with a higher slope for the longer sequence that has more cascade stages (solid line). DNA sequences were synthesized, labeled with 3⬘-fluorescein (Midland Certified Reagent, Midland, TX), and HPLC-purified. The DNA concentration, 39 nM in nucleotide (13 nM RecA-binding sites), was much lower than the RecA binding onset (70 nM) and ensured that the free RecA was in practice the total RecA. RecA protein (New England Biolabs) binding assays were done in 25 mM Tris䡠HCl, pH 7.5兾150 mM NaCl兾1 mM MgCl2兾1 mM DTT兾1 mM ATP. Fluorescence was excited in a quartz cuvette (3 ml) by a vertically polarized 488-nm argon laser line. Emission intensity polarized parallel, Iv, and perpendicular, Ip, to the excitation was measured (9), from which the anisotropy was determined: A ⫽ (Iv ⫺ Ip)兾(Iv ⫹ 2Ip). at any vacant site increases w ith RecA concentration, R, like ␬⫹(n) ⫽ ␬T(n)䡠R (where ␬T(n) is the triplet-specific rate const ant), while the disassembly rate remains const ant as the amount of AT P it consumes is kept at saturation level. When we add RecA to the test tube more monomers bind DNA through nucleation-polymerization process, and the chance to find oc cupied sites increases. The binding cur ve is sigmoidal, t ypical of collective chemical k inetics (Fig. 2 Inset). Our measurement suggests an exponential sensitiv ity of the assembly cascade to sequence length and RecA-binding rate constants, ␬⫺ and ␬⫹ (Fig. 2). To understand how such amplified sensitiv ity is accomplished, we reexamine the master equation (Eq. 2). Motivated by our measurements that indicate fast relaxation of the assembly cascade, we study its steady state. For a un iform sequence, ␬⫺(n) ⫽ ␬⫺, ␬⫹(n) ⫽ ␬⫹, the steady-state probability distribution is Gaussian, n2冊. P 共n 兲 ⫽ exp冉⫺ It follows that even a slight dif ference in transition rates of two un iform sequences is exponentially amplified as the cascade ␬⫹ 2␬⫺ [3] Fig. 3. (A) The difference between the fluorescence anisotropy signal of (TAC)N and (TCA)N for n ⫽ 13,26. Sequence separation peaks at a lower rate ratio, ␬⫹兾␬⫺, for longer sequences in agreement with cascade model. (B) A 3 single base change was introduced at the seventh triplet of (TAC)N: (TAC)13 3 (TAC)26 M ⫽ (TAC)6(TAG)(TAC)6, (TAC)26 M ⫽ (TAC)19(TAG)(TAC)6. This (TAC)13 change induces a weak hairpin secondary structure in the DNA, which forms M toward a barrier for RecA binding (9) and shifts the binding curves of (TAC)N higher RecA concentration with respect to (TAC)N. The difference in signal M is peaked at a lower rate ratio for longer sequences. between (TAC)N and (TAC)N advances to its higher states (large n). The max imal enhance- ment increases exponentially like the square of the states num- ber, actually the number of DNA base triplets. Similarly, the cascade can discriminate between lengths, N1 and N2, of two sequences made of the same triplets, since the probability ratio at the highest states is P(N1)兾P(N2) ⫽ exp[⫺(␬⫹兾2␬⫺)(N1 2 ⫺ N2 2)]. The exponential amplification is the result of the iterative, multistage structure of the cascade. It is the same design principle that underlies industrial distillation (20) and the k inetic proof- reading pathway of protein synthesis (4). The exponential am- plification of the cascade is ev ident from the behav ior of the normalized f luorescence an isotropy of un iform triplet repeats at the saturated regime (Fig. 2). In this regime of high nucleation rate, it becomes harder for the machine to climb up to higher states through successive disassembly steps (Fig. 1B). However, this helps the cascade to discriminate lengths, because shorter sequences need less disassembly steps to reach higher states, and indeed the cur ve for the longer (TAC)26 triplet-repeat sequence is steeper than that of the half-size (TAC)13. We test the sequence-discrimination capability of the cascade by comparing the binding cur ves of two un iform single-stranded DNA molecules made of ver y similar triplet repeats, TAC and TCA (Fig. 3A). The dif ference between the two binding signals behaves similarly to the relative entropy of the two probability distributions (sometimes called ‘‘information for discrimina- tion’’; ref. 21) and therefore gives a good idea about their distinguishability. For both lengths n ⫽ 13,26 we find that the dif ference peaks at a certain rate ratio ␬⫹兾␬⫺ that corresponds to the max imal slope of the binding cur ves (Fig. 2 Inset), where cooperativ ity is highest (Eq. 2). The peak indicates optimal tun ing of the back and forth scann ing motion that is used by the stochastic machine to ‘‘read’’ the sequence (a process that was mapped to sequential reading of a random tape). Think ing of the serriform time series (Fig. 1 A) as a‘‘sentence’’ composed of an N-state alphabet printed by a stochastic typewriter (something like . . . Q10Q11Q8Q9Q3Q3Q3Q4. . . ), then the appearance of the ‘‘letter’’ QN corresponds to a completed scan. Interpreting QN as a ‘‘space bar,’’ the max imal rate of completed scans corresponds to the most informative reading w ith the max imal rate of ‘‘words.’’ This occurs at the ‘‘work ing point’’ of the cascade, t⫹兾t⫺ ⬃ 1, when the time inter val between nucleation events, t⫹ ⬃ 1兾(N␬⫹), is matched w ith the time required to climb back to state QN, t⫺ ⬃ N兾2␬⫺. With the rates measured independently by k inetic assays we find that optimal separation occurs indeed in the optimal regime, t⫹兾t⫺ ⬃ 1–3. The protein assembly cascade dynamics can detect also local- ized dif ferences in nonun iform sequences. A stringent test for our machine is the discrimination of a single base change. We therefore introduced a change C 3 G at the seventh triplet of the two un iform (TAC)N sequences and measured the discrim- ination (Fig. 3B). Similar to the un iform sequences, the dif fer- ence in binding between a sequence and its variant peaks at an optimal ␬⫹兾␬⫺ that is lower for the longer sequence, consistent w ith the work ing point. The machine’s ability to discriminate localized changes sug- gests a basis for certain mathematical computations, integral transforms. Consider an ensemble of un iform sequences made of N RecA-favored triplets (relatively high ␬⫹; ref. 9). Within each sequence, we encode a ‘‘defect’’ in the form of a single unfa- vorable triplet placed at one of the N possible sites. Let w(n) designate the fraction of sequences w ith defect at site n. A test tube w ith a mixture of all these sequences encodes the vector [w(1), w(2), . . . w(N)]. A monodisperse solution w ith defect only at site n is one of the N-un it base vectors that span our sequence space. As shown below, the signal from such a base vector is exponential in the position of the defect, S(n, k) ⬃ exp(⫺k n), where k ⫽ ⌬(␬⫹兾␬⫺) is the dif ference in the ratio of reaction rates at the defect. Since the f luorescence an isotropy is an ensemble average, the signal of mixture is a Laplace-like transform, w共n兲S共n, k兲 ⬃ 冘 S 共k 兲 ⫽ 冘 N N n⫽0 n⫽0 w共n兲exp共⫺kn兲. [4] n m ⫽ ⫺ P 共n 兲 ⫽ ⭸P 共n , t 兲 ⭸ t To account for the nonun iformity of a DNA sequence w ith site-dependent nucleation and disassembly rates, ␬⫹(n) and ␬⫹共m兲dm冊P共n, t兲, 关␬⫺共n兲P共n, t兲兴 ⫺ 冉冕 ␬⫺(n), we modify the continuous master equation to ⭸ ⭸n 0 ␬⫹共q兲dq冊. ␬⫺共n兲 exp冉⫺冕 w ith the inhomogeneous steady-state solution ␬⫺共m兲冕 ␬⫺共0兲 n dm 0 0 A mutation at site n0 implies a localized change of reaction rates by ⌬␬⫹ and ⌬␬⫺. When the mutation is in a formerly un iform sequence, variation of the steady-state profile exhibits a change that depends on the position of the site as ⌬P(n)兾P(n) ⯝ ⫺ ␬n0, where the ‘‘wave number,’’ k, is the dif ference in the reaction rates ratio, k ⫽ ⌬共␬⫹兾␬⫺兲 ⫽ 共␬⫹兾␬⫺兲共⌬␬⫹兾␬⫹ ⫺ ⌬␬⫺兾␬⫺兲. The resulting relative change in P(n) depends exponentially on the position of the mutation n0. Integrating over P(n) we find that the an isotropy signal scales like S(n0, k) ⬃ exp(⫺kn0). By choice of other types of sequence base vectors, the sto- chastic cascade machiner y, through the ensemble measurement, can encode and decode mixtures in terms of other transforms. It is tempting to speculate that w ith additional operations to man ipulate sequences at hand, such as recombination, one could construct a molecular architecture for more complex computa- tions. The question of whether RecA assembly is used for natural computation requires in vivo testing (22). We thank K. Adzuma and B. Shraiman for discussions and suggestions and D. Thaler for a fruitful and inspiring collaboration. 1. Adleman, L. M. (1994) Science 266, 1021–1024. 2. Landweber, L. F. & Kari, L. (1999) BioSystems 52, 3–13. 3. Winfree, E., Liu, F. Wenzler, L. A & Seeman, N. C. (1998) Nature (London) 394, 539 –544. 4. Hopfield, J. J. (1974) P roc. Natl . Acad . Sci . USA. 71, 4135– 4139. 5. Mitchison, T. & K irschner, M. (1984) Nature (London) 312, 232–242. 6. K illeen, P. R. & Taylor, T. J. (2000) Psychol . Rev. 107, 430 – 459. 7. Kowalczykowsk i, S. C., Dixon, D. A., Eggelston, A. K., Lauder, S. D. & Rehrauer, W. M. (1994) Microbiol . Rev. 58, 401– 465. 8. Shan, Q., Bork, J. M., Webb, B. L., Inman, R. B. & Cox, M. M. (1997) J. Mol . Biol . 265, 519 –540. 9. Bar-Ziv, R. & Libchaber, A. (2001) P roc. Natl . Acad . Sci . USA 98, 9068 –9073. 10. van Kampen, N. G. (2002) Stochastic P rocesses in Physics and Chemistr y (Elsev ier, Amsterdam). 11. Bouchaud, J. P, Comtet, A., Georges, A. & Le Doussal, P. (1990) Ann. Phys. 201, 285–341. 12. Schutz, G. M. (1997) J. Stat . Phys. 88, 427– 452. 13. L u b e n s k y , D . K . & N e l s o n , D . R . ( 2 0 0 0 ) P h y s . R e v . L e t t . 8 5 , 1572–1575. 14. Shannon, C. E. (1948) Bell Syst . Tech. J. 27, 379 – 423. 15. Shannon, C. E. (1948) Bell Syst . Tech. J. 27, 623– 656. 16. Bennett, C. H. (1982) Int . J. Theor. Phys. 21, 905–940. 17. von Neumann, J. (1956) in Automata Studies, eds. Shannon, C. E. & McCarthy, J. (Princeton Un iv. Press, Princeton), pp. 43–98. 18. Grenander, U. (1966) Res. Pap. Statist . Testchr ift J. Neyman 107–123. 19. Hegner, M., Smith, S. B. & Bustamante, C. (1999) P roc. Natl . Acad . Sci . USA 96, 10109 –10114. 20. Lord Rayleigh (1896) Philos. Mag. 42, 493– 498. 21. Cover, T. M. & Thomas, J. A. (1991) Elements of Information Theor y (Wiley, New York). 22. Matic, I., Rayssiguier, C. & Radman, M. (1995) Cell 80, 507–515.
1607.06683	1	1607	2016-07-22T14:00:18	Identification of a membrane-bound prepore species clarifies the lytic mechanism of actinoporins	[ "physics.bio-ph" ]	Pore-forming toxins (PFT) are cytolytic proteins belonging to the molecular warfare apparatus of living organisms. The assembly of the functional transmembrane pore requires several intermediate steps ranging from a water-soluble monomeric species to the multimeric ensemble inserted in the cell membrane. The non-lytic oligomeric intermediate known as prepore plays an essential role in the mechanism of insertion of the class of $\beta$-PFT. However, in the class of $\alpha$-PFT like the actinoporins produced by sea anemones, evidence of membrane-bound prepores is still lacking. We have employed single-particle cryo-electron microscopy (cryo-EM) and atomic force microscopy (AFM) to identify, for the first time, a prepore species of the actinoporin fragaceatoxin C (FraC) bound to lipid vesicles. The size of the prepore coincides that of the functional pore, except for the transmembrane region, which is absent in the prepore. Biochemical assays indicated that, in the prepore species, the N-terminus is not inserted in the bilayer but exposed to the aqueous solution. Our study reveals the structure of the prepore complex in actinoporins, and highlights the role of structural intermediates for the formation of cytolytic pores by an $\alpha$-PFT.	physics.bio-ph	physics	Identification of a prepore species in actinoporins IDENTIFICATION OF A MEMBRANE-BOUND PREPORE SPECIES CLARIFIES THE LYTIC MECHANISM OF ACTINOPORINS Koldo Morante1,2,3, Augusto Bellomio2,3, David Gil-Cartón4, Lorena Redondo-Morata5, Jesús Sot2,3, Simon Scheuring5, Mikel Valle4, Juan Manuel González-Mañas2, Kouhei Tsumoto1,6,, and Jose M.M. Caaveiro1, 1Department of Bioengineering, Graduate School of Engineering, The University of Tokyo, Bunkyo-ku, Tokyo 113-8656; 2Department of Biochemistry and Molecular Biology, University of the Basque Country, P. O. Box 644, 48080 Bilbao, Spain; 3Biofisika Institute (UPV/EHU, CSIC), University of the Basque Country, P. O. Box 644, 48080 Bilbao, Spain; 4Structural Biology Unit, Center for Cooperative Research in Biosciences, CICbiogune, 48160 Derio, Spain; 5U1006 INSERM, Aix-Marseille Université, Parc Scientifique et Technologique de Luminy, 163 avenue de Luminy, 13009 Marseille, France; 6Institute of Medical Science, The University of Tokyo, Minato-ku, 108-8639 Tokyo, Japan. Running title: Identification of a prepore species in actinoporins *Corresponding author: Kouhei Tsumoto ([email protected]), and Jose M.M. Caaveiro ([email protected]). Keywords: pore forming protein, cytolysin, lipid‐ protein interaction, protein structure, oligomerization, atomic force microscopy, lipid vesicle. The size of the prepore coincides that of the ABSTRACT functional pore, except for the transmembrane region, which is absent in the prepore. Pore-forming toxins (PFT) are cytolytic Biochemical assays indicated that, in the proteins belonging to the molecular warfare prepore species, the N-terminus is not inserted apparatus of living organisms. The assembly of in the bilayer but exposed to the aqueous the functional transmembrane pore requires solution. Our study reveals the structure of the several intermediate steps ranging from a water- prepore complex in actinoporins, and highlights soluble monomeric species to the multimeric the role of structural intermediates for the ensemble inserted in the cell membrane. The formation of cytolytic pores by an α-PFT. non-lytic oligomeric intermediate known as prepore plays an essential role in the mechanism Pore-forming toxins (PFT) are proteins designed of insertion of the class of β-PFT. However, in for defense and attack purposes found throughout the class of α-PFT like the actinoporins the eukaryote and prokaryote kingdoms (1,2). produced by sea anemones, evidence of These proteins function by opening pores across the membrane-bound prepores is still lacking. We cell membranes, triggering processes conducive to have employed single-particle cryo-electron cell death (3). PFT are commonly classified into α- microscopy (cryo-EM) and atomic force and β-types according to the secondary structure of microscopy (AFM) to identify, for the first time, the transmembrane portion of the pore (4-6). The a prepore species of the actinoporin classical route of pore-formation begins with the fragaceatoxin C (FraC) bound to lipid vesicles. interaction of the water-soluble monomer with the 1 Identification of a prepore species in actinoporins outer leaflet of the cell membrane, followed by the to the oligomerization step and, therefore, this in-plane oligomerization of toxin subunits and model does not contemplate the appearance of conformational changes to assemble the lytic stable prepores (Figure 1) (21). transmembrane pore (7). Among the intermediate Here, we have visualized a non-lytic oligomer of species that populate this pathway, prepore particles FraC bound to large unilamellar vesicles (LUVs) by bound to lipid bilayers have been described in the using cryo-EM, and to supported planar bilayers by class of β-PFT (8-12). In the class of α-PFT, prepore AFM. The overall dimensions of the cryo-EM structures of Cytolysin A from Escherichia coli model and the high-resolution AFM images have been proposed (6,13,14), although direct indicates that the prepore is made of eight protein visualization still remains elusive. For subunits, a result consistent with the actinoporins, a class of α-PFTs secreted by sea oligomerization number of the active pore. anemones, it is unclear whether a stable prepore is Biochemical assays indicate that, in the prepore assembled before formation of the lytic species, the first few residues of the N-terminal transmembrane pore. Structurally, actinoporins are region are not embedded in the lipid phase, but composed of a rigid β-sandwich core (mediating exposed to the aqueous environment. Our results binding to the membrane) and two flanking α- reinforce the idea that protein oligomerization helices, as shown for the actinoporin FraC (20 kDa, occurs prior to the complete insertion of the N- 179 residues). The N-terminal region of FraC is terminal region into the membrane, thus clarifying made of an amphipathic α-helix and neighboring a critical aspect of the lytic mechanism of residues that insert collectively in the bilayer and, actinoporins. together with structural lipids from the membrane, line the lytic pore (15). This remarkable MATERIALS AND METHODS metamorphosis is fully reversible under certain Materials. Sphingomyelin (SM) from porcine brain environmental conditions (16). and chicken egg, 1,2-dilauroyl-sn- glycero-3- A variety of membrane-bound species have been phosphocholine (DLPC), 1,2-dipalmi- toyl-sn- proposed in the mechanism of actinoporins (17-19). glycero-3-phosphocholine (DPPC), and 1,2- However, the nature of some of the intermediate dioleoyl-sn-glycero-3-phosphocholine species and the order at which they appear during (DOPC) were from Avanti Polar Lipids (AL, USA). pore-formation remains unclear. Some authors have 8-Aminonaphthalene-1,3,6-trisulfonic acid proposed that the protein subunits should first (ANTS), p-xylene-bis-pyridinium bromide, assemble into an oligomeric prepore, followed by 1,1'-dioctadecyl-3,3,3',3'-tetramethylindocarbocya- the concerted insertion of the N-terminal region in nine perchlorate (DiIC18), and Alexa Fluor 633 the lipid bilayer that gives rise to the functional pore succinimidyl ester were from Thermo Fisher (20). Other authors, on the contrary, have suggested Scientific (MA, USA). Proteinase K (PK) was that the α-helix inserts deeply in the membrane prior 2 Identification of a prepore species in actinoporins purchased from Sigma Aldrich (St. Louis, MO, Liposome preparation and leakage assays. LUVs USA). of 100 nm were formed by extrusion as described Protein expression and purification. Expression previously (24). The lipid concentration was and purification of FraC was carried out as determined according to Bartlett (25). For the previously described (22). Briefly, FraC expression leakage assays, four different populations of LUVs was induced in E. coli BL21 (DE3) cells and then made of DLPC, DPPC, DOPC, and SM/DOPC purified to homogeneity by ion-exchange and size- (1:1) were prepared as described above in a buffer exclusion chromatography. Oxidation of the containing 10 mM HEPES (pH 7.5), 50 mM NaCl, double-cysteine mutein was carried out as described 25 mM ANTS (the fluorescent probe), and 90 mM by Hong et al. (23). p-xylene-bis-pyridinium (the quencher), followed Protein labeling. To visualize FraC by by washing the liposomes with isosmotic buffer 10 fluorescence microscopy, the protein was labeled mM HEPES, 200 mM NaCl, pH 7.5 in a PD-10 with the amine-reactive fluorescent dye Alexa Fluor column (GE Healthcare). Leakage of encapsulated 633 succinimidyl ester. The succinimidyl ester solutes was assayed as described by Ellens et al. moiety of the reagent reacts with non-protonated (26). Briefly, LUVs were incubated with FraC at aliphatic primary amine groups of the protein. The room temperature for 30 minutes to ensure, as much protein-dye conjugate was prepared following the as possible, completion of vesicle lysis at each instructions supplied by the manufacturer. Briefly, protein concentration employed. Upon solute 740 µl of FraC at 180 µM in 90 mM bicarbonate release to the external medium, the dilution of buffer (pH 8.3) were mixed with 45 µl of the quencher and fluorophore results in an increase of fluorescent dye previously dissolved in DMSO. The the emission of fluorescence of ANTS. The mixture was incubated for 1 hour at room fluorescence was measured in a PHERAstar Plus temperature with constant stirring. To stop the microplate reader (BMG LABTECH, Ortenberg, reaction and remove weakly bound probes from the Germany) with excitation/emission wavelengths of unstable conjugates 80 µl of freshly prepared 1.5 M 350/520 nm. Complete release of the ANTS was hydroxylamine was added and incubated for 1 hour achieved by solubilization of the liposomes with at room temperature. Unreacted labeling reagent Triton X-100 (0.1% w/v). The percentage of was separated from the conjugate by elution of the leakage was calculated as: reaction mix through a Sephadex G-15 (GE % leakage = (Ff - F0 / F100 - F0) × 100 where Healthcare) packed column. The protein-conjugate Ff is the fluorescence measured after addition of the was tested for activity using surface pressure toxin, F0 the initial fluorescence of the liposome measurements and hemolysis assays, showing a suspension and F100 the fluorescence after addition similar behavior to that of the unlabeled protein of detergent. (data not shown), ruling out detrimental effects by Surface pressure measurements. Surface pressure the dye. measurements on lipid monolayers made of pure 3 Identification of a prepore species in actinoporins DLPC, DPPC, or DOPC were carried out using a maximum-likelihood and hierarchical clustering Micro-Trough-S instrument (Kibron, Finland) at procedures within the XMIPP software package room temperature with constant stirring. In these (29). The starting 3D model was generated using experiments, the lipid was spread over the air-water reference-free alignment, classification, and interface to the desired initial surface pressure. The common-lines procedures implemented in EMAN. protein (1 µM) was injected into the aqueous This was followed by iterative refinement using a subphase and the increase of surface pressure projection matching scheme in SPIDER package recorded. The maximum surface pressure (πmax) was (30). determined with the following equation: πmax = π0 + The rigid-body fitting was performed by (Δπ·"x / b + x) maximization of the sum of map values at atom where π0 is the initial surface pressure, Δπ is the positions and by improvement in the coefficient of change in surface pressure, x is time, and b is the correlation between simulated maps from the time necessary to reach Δπ/2 (27). The critical atomic structures and the cryo-EM density map in pressure (πc) corresponded to the initial surface Chimera (31). The correlation between the atomic pressure of the lipid monolayer at which the protein structure of FraC and cryo-EM map suggested a no longer penetrates the surface, calculated by least resolution of ~30 Å. squares fitting as the intercept when Δπ= 0. Preparation of giant unilamellar vesicles (GUVs) Cryo-EM of FraC Inserted in Model Membranes. for confocal fluorescence microscopy. GUVs For cryo-EM imaging, LUVs composed of DOPC made of DOPC/DPPC (20:80) were prepared by were incubated with FraC (5 µM) at a protein/lipid electroswelling on a pair of platinum wires by a ratio of 1:160 for 30 minutes. Holey-carbon grids method first developed by Angelova and Dimitrov were prepared following standard procedures and (32), modified as described previously (33). A observed in a JEM-2200FS/CR transmission temperature-controlled chamber was used electron microscope (JEOL Europe, following previous methodology (33). Briefly, a Croissy-sur-Seine, France) operated at 200 kV at mixture of 0.2 µg/µl lipid and 0.2 % DiIC18 was liquid nitrogen temperature. A set of 1,562 spread on the chamber and dried under vacuum. The individual pore particles were manually selected sample was then covered with 10 mM HEPES, 200 and recorded on CCD camera under low-dose mM NaCl, pH 7.5 at 61 ºC. This temperature was conditions at 60,000 × magnification resulting in a selected to prevent lipid demixing. To form the final pixel size of 1.72 Å. An in-column omega vesicles, current was applied in three steps under energy filter was used to improve the signal to noise AC field conditions and a sinusoidal wave function: ratio of the images. (i) 500 Hz, 0.22 V (35 V/m) for 6 minutes, (ii) 500 The images were CTF-corrected by flipping Hz, 1.9 V (313 V/m) for 20 minutes, (iii) 500 Hz, phases after estimation of CTF parameters in 5.3 V (870 V/m) for 90 minutes. After vesicle EMAN (28). The 2D images were classified by 4 Identification of a prepore species in actinoporins formation the chamber was left to settle at room and incubated for 15 minutes at room temperature. temperature. The resulting supported lipid bilayers were Inverted confocal fluorescence microscopy. For carefully rinsed with imaging buffer before image the visualization of GUVs and labeled protein the collection and always kept under aqueous chamber for GUV formation was placed on a D- environment. During imaging, FraC toxin was Ellipse C1 inverted confocal fluorescence injected into the fluid cell to give a final microscope (Nikon, Melville, NY, USA) and the concentration of ~10 µM. samples visualized at room temperature. The excitation wavelengths used were 561 nm (for DiIC18) and 633 nm (for Alexa Fluor 633). The fluorescence signal was collected into two different channels with band pass filters of 593/40 nm and 650 nm long pass. The objective used was a 60 × oil immersion with a NA of 1.45. Image treatment was performed with the EZ-C1 3.20 FreeViewer software. Preparation of GUVs for AFM. GUVs made of SM/DOPC (1:1) were prepared by the electroswelling technique (34). A volume of 30 µl of 1 mg/mL lipids dissolved in chloroform: methanol (3:1) were deposited in two glass plates coated with indium tin oxide (70-100 Ω resistivity, Sigma-Aldrich) and placed in the desiccator at least 120 minutes for complete solvent evaporation. A U- shape rubber piece of ~1 mm thickness was sandwiched between the two indium tin oxide side slides. Then the formed chamber was filled with ca. 400 µl of 200 mM sucrose and exposed to 1.2 V AC current (12 Hz sinusoidal for 2h, 5 Hz squared for 10 minutes). The resulting suspension was collected in a vial and used within several days. Supported lipid bilayer preparation for AFM. A total of 1 µL of a suspension of GUVs was deposited onto freshly cleaved 1 mm2 mica pretreated with 1 µL of 10 mM Tris-HCl, 150 mM KCl, pH 7.4 (imaging buffer) AFM imaging. AFM was performed at room temperature on a high-speed AFM 1.0 instrument (RIBM, Japan) equipped with short high-speed AFM cantilevers (~8 µm, NanoWorld, Switzerland) with nominal resonance frequency of ~1.2 MHz and ~0.7 MHz in air and liquid, respectively, and a nominal spring constant of ~0.15 Nm-1. Image acquisition was operated using optimized feedback by a dynamic PID controller. Small oscillation free (Afree) and set point (Aset) amplitudes of about 1 nm and 0.9 nm, respectively, were employed to achieve minimum tip-sample interaction. Typically, pixel sampling ranges from 100 × 100 pixels and 200 × 200 pixels and frame rate between 500 and 800 ms per frame. AFM data analysis. AFM data was analyzed in ImageJ and with self-written image analysis scripts (movie acquisition piezo drift correction) in ImageJ (35). To obtain the high resolution images shown in Figure 5 and Figure 6, five consecutive frames were time-averaged. All further analysis, i.e. histogram distributions were analyzed in Igor and Origin. Protease susceptibility assay. Proteinase K (PK) (50 µM) was incubated with FraC (50 µM) for 24 hours at room temperature in 50 mM Tris, 200 mM NaCl, 5 mM CaCl2 at pH 7.4. In the assays with lipids, FraC was incubated with the appropriate LUVs (7.5 mM) made of either DOPC or 5 Identification of a prepore species in actinoporins SM/DOPC (1:1) for 30 minutes prior to the addition interface (Figure 2A). We determined the surface of PK. The reaction was stopped by adding pressure at which the protein will no longer phenylmethylsulfonyl fluoride at a final penetrate, known as critical pressure (πc) (27). The concentration of 5 mM for 10 minutes and analyzed lipid composition at which πc was highest by SDS-PAGE. corresponded to that of monolayers composed of N-terminal sequencing. SDS-PAGE protein bands DOPC (πc = 36.1 ± 1.6 mN/m) followed by that of of the PK reaction products were transferred to a monolayers made of DLPC (πc = 31.1 ± 1.3 mN/m). polyvinylidene fluoride membrane (Bio-Rad The insertion of FraC in DPPC monolayers was Laboratories, CA, USA), stained with Ponceau 3R meager (23.6 ± 1.0 mN/m). These data suggest that solution for 1 hour, and washed with water until the toxin associates more readily with lipids in the complete removal of the excess stain. The red liquid-expanded phase such as DOPC and DLPC colored protein bands were excised from the than those in the liquid-condensed phase (DPPC) membrane and their N-terminal sequenced (37-39). However, the protein does not generate employing standard techniques (36). pores in LUVs when SM was absent regardless of RESULTS the lipid phase (Figure 2B). A close association between actinoporins and lipid monolayers thus Interaction of FraC with PC membranes − does not guarantee effective formation of pores in a Because actinoporins are specifically activated by lipid bilayer system as there are other membranes containing the lipid SM, the use of lipid physicochemical properties involved in pore compositions in which SM is absent but where the formation, such as lipid phase coexistence and the toxin maintain a strong interaction with the vesicles presence of SM (40). may reveal structural intermediates not detectable Additional evidence describing the lipid preference by other means. We first evaluated the interaction of FraC was gathered by visualizing the binding of of FraC with vesicles made of various types of the the fluorescently-labeled toxin to GUVs composed lipid phosphatidylcholine (PC, a lipid displaying the of DOPC/DPPC (20:80). In these experiments, the same phophocholine headgroup moiety as that of fluorescent dye conjugated to the toxin co-localized SM) to determine the optimum PC species yielding with the DOPC domains (dark areas in Figure 2C) the highest possible association between protein indicating that the toxin preferentially binds to the and liposomes. The three PC lipid species fluid phase domains over the gel domains, a result examined were DLPC, DPPC, and DOPC each consistent with the observations made with differing in the length and degree of saturation of monolayers for FraC and leakage assays for their acyl chains. To evaluate the degree of sticholysin II (41). Based on these results, interaction of FraC with these lipids we measured membranes composed of DOPC were selected for the magnitude of the insertion of the protein in a structural studies analyzing the conformation of monolayer of lipid molecules at the water-lipid 6 Identification of a prepore species in actinoporins FraC bound to membranes in a non-lytic the electron density map suggesting that it is either environment. resting on the surface of the membrane or inserted Cryo-EM − To visualize the structure of membrane- in the hydrophobic core of the membrane bound FraC, vitrified samples of toxin-treated (21,23,44). A nonamer of FraC mimicking the DOPC liposomes were imaged by cryo-EM. The structure of a crystallized oligomer of FraC in the observed ring-shaped particles covering the lipid presence of detergents (20,45) was also fitted in the vesicles were attributed to protein oligomers cryo-EM maps, yielding a cross-correlation (Figure 3A). A total of 1,562 top- and side-view coefficient only slightly worse (cc = 0.81) than that images were selected to build a three-dimensional of the octamer. The fitted nonamer displayed a few model of the protein oligomer. The model was built clashes between protomers, in contrast to an by common-lines procedures, followed by iterative oligomer made of ten units in which the numerous refinement using projection matching of the class- collisions between protein chains made the decamer averaged images and the density map projections prepore unfit for this electron density. From these (Figure 3B). A second classification method based data we cannot rule out the existence of a minor on maximum-likelihood (42) and hierarchical population of nonameric prepore species preceding clustering approaches (43) (Figure 3C) rendered a hypothetical nonameric pore, as discussed images similar to those used to generate the final previously (15). density map. A comparative analysis of the electron density The reconstructed image consisted of a doughnut- distribution along the central section of the shaped ring with an external and internal diameter oligomer of FraC in DOPC membranes and in of ~11 and ~ 5 nm, respectively (Figure 3D and 3E). SM/DOPC (1:1) clearly shows that the differences These dimensions are very similar to those of the between pores and prepores occur in the critical crystallized pore in the active state (15). However, transmembrane region. To perform this comparison unlike the cryo-EM model of the pore bound to we employed the previously reported cryo-EM map SM/DOPC (1:1) liposomes (20), the oligomer of the active pore (20). For the analysis, 3D volumes bound to DOPC vesicles does not span the lipid focused on the pore regions (shown within yellow membrane (see below). This architecture is rectangles in Figures 4A and 4B) were projected consistent with a non-lytic oligomeric species into 2D images, and the gray values of the images resembling a prepore. A rigid-body fitting of an (resulting from the accumulation of 3D density octameric model of FraC based on the atomic values) were plotted in 1D profiles (Figures 4C and structure of the transmembrane pore of FraC 4D). The structure of FraC in DOPC membranes achieved a high cross-correlation coefficient (cc = (Figure 4A) reveals a peak of higher density values 0.82, Figure 3D, E) (15). The oligomeric model fits in a central lobe at the membrane level below the well within the perimeter of the cryo-EM map, vestibule of the oligomer (Figure 4C), whereas in except for the N-terminal region, which lies outside SM/DOPC (1:1) membranes (Figure 4B), the same 7 Identification of a prepore species in actinoporins region is characterized by lower density values Under oxidizing conditions, the N-terminal (Figure 4D). These features are consistent with the segment of this mutein is covalently attached to the absence (in DOPC membranes) or the presence (in protein core by means of a disulfide bond, SM/DOPC (1:1) membranes) of a transmembrane preventing the protein from generating a pore. In the former, the high-density central region transmembrane pore, and thus inactivating the likely represents the accumulation of membrane toxin (15,23). As with WT FraC, the construct 8- lipids and N-terminal α-helices detached from the 69OX also gave rise to a dense array of pore-like β-core of the toxin. In contrast, in membranes particles (Figure 6), indicating that the protein containing SM the central region of the oligomer readily oligomerizes in the presence of membranes displays lower electron density because the α- even if the N-terminal region remains attached to helices span the membrane from top to bottom, and the protein. The average diameter of these particles consequently the lipids are cleared off producing an (6.2 ± 0.7 nm) is somehow smaller than that of WT aqueous pore. protein, reflecting the influence of the N-terminal AFM − In the presence of supported lipid bilayers region attached to the β-core region. Because of the composed of the equimolar mixture SM/DOPC constrains imposed by the disulfide bond, the (1:1), WT FraC assembles in a dense array of conformation of the N-terminal region in 8-69OX is closely packed oligomers as determined by AFM likely to differ from that of WT FraC bound to (Figure 5). These oligomers, presumably liposomes made of DOPC (Figure 3D, E). To corresponding to pore particles, cover the SM-rich further investigate this question we employed domains in an arrangement previously observed in biochemical assays (see below). FraC and other actinoporins (45,46) or the SM- Protease susceptibility of the membrane-bound specific PFT lysenin (47). The cross-section toxin − It was shown that FraC bound to LUVs profile of the oligomeric complexes reveals an exhibits different susceptibility to proteinase K average diameter of 7.5 ± 0.6 nm, a value in good (PK) depending on the lipid composition of the agreement with the mean diameter (average of membrane (15). The incubation of FraC with PK outer and inner diameter) of the pore determined generated a product of smaller size when the toxin by X-ray crystallography (~ 8 nm). Eight protein was bound to DOPC vesicles as compared to those chains are observed in three well-resolved pore generated in the presence of SM/DOPC (1:1) particles encountered (see for example Figure 5C). vesicles (15), although the basis of this difference Because prepores of FraC were not resolved in was not explained. In view of the new prepore DOPC, probably caused by high-diffusivity oligomeric species described herein, we preventing AFM contouring, a construct of FraC hypothesized that the N-terminus of this prepore is bearing a double cysteine mutation (V8C/K69C, located in a solvent-exposed environment termed 8-69OX) was instead examined on accessible to PK, whereas in the pore the N- supported membranes made of SM/DOPC (1:1). terminus is deeply inserted in the membrane and 8 Identification of a prepore species in actinoporins thus inaccessible to the protease. To verify this transmembrane pore in PFT are key species that can hypothesis and determine the extent of the help to elucidate the details of pore formation. digestion, we incubated samples of FraC with PK Membrane-bound oligomeric structures poised for followed by their separation by SDS-PAGE and N- membrane disruption are commonly referred to as terminal sequencing. prepores and have been visualized in lipid bilayers The incubation of PK with FraC in the presence of only for β-PFT. In contrast, the existence of DOPC vesicles yields a fragment of smaller prepores in α-PFT is controversial. An example is molecular weight than that of the untreated protein the family of actinoporins, where a strong debate is (shown in the 20 kDa region). In contrast, in the held about the existence or not of these non-lytic presence of SM/DOPC (1:1) vesicles, the bands of oligomers (20,21,48). Until now, the evidence treated and untreated toxin display the same supporting a prepore in actinoporins was based on molecular mass (Figure 7A). The mutein 8-69OX the crystal structure of a non-lytic nonameric bound to membranes was also employed, since its ensemble solved for FraC (20). N-terminus remains exposed to the solvent Herein, we have described a low-resolution constrained by the disulfide bond. As expected, 8- membrane-bound oligomer consistent with the 69OX was also susceptible the proteolytic activity of ability of FraC to assemble as a prepore on PK in DOPC and SM/DOPC (1:1) vesicles. To biological membranes. We employed a protein determine the cleavage point the proteins were concentration above physiological levels to ensure sequenced from their N-terminus. The sequencing a large and homogeneous population of pre-pore data revealed that, in the presence of vesicles of species bound to the liposomes, thus facilitating DOPC, FraC WT and 8-69OX were cleaved at the N- their visualization by cryo-EM and AFM. The pre- terminus by PK, rendering products in which the pore structure could explain the readiness of first four and first eleven residues, respectively, actinoporins to induce lysis in liposomes made of were missing (Figure 7B, C). FraC bound to PC upon generation of lipid domains in situ (40). SM/DOPC (1:1) was not digested by PK as The size and stoichiometry of the prepore in the expected from the position of the band in the SDS- cryo-EM (Figures 3, 4) and AFM (Figure 6) images PAGE gel, whereas 8-69OX was cleaved at the same were in the range of those of the crystallized pore position seen in vesicles of DOPC. These results species (15). The cryo-EM reconstruction map is demonstrate that the N-terminus of FraC in DOPC not consistent with a transmembrane pore, an vesicles (prepore configuration) is accessible to PK, argument strengthened by the comparison side-by- i.e. this region is not embedded in the lipid bilayer. side with cryo-EM data of pores of FraC embedded DISCUSSION in vesicles of SM/DOPC (1:1) (20). Electron density gradient analysis and protease Structural intermediates that populate the pathway digestion assays suggest a close association of the leading to the formation of a functional N-terminal region with the membranes in a position 9 Identification of a prepore species in actinoporins approximately parallel to the plane of the membrane thermodynamic considerations suggest that would as was described before for other actinoporins not be the case: The penetration of individual α- (44,49). Evidence that this oligomer precedes pore helices containing a large number of charged formation is inferred from a previous study carried residues (FraC displays three Asp and one Glu in out with the actinoporin equinatoxin II. It was shown that the addition of phospholipase C to ACKNOWLEDGEMENTS vesicles of PC decorated with toxin promoted this region) in the hydrophobic core of biological vesicle lysis by the in situ generation of lipid membranes would be strongly disfavored (50-52). domains (40). Our results suggest a model where In conclusion, our study clarifies the structure of a the N-terminal α-helices penetrate the bilayer in a key intermediate, known as prepore, in the route of concerted manner (Figure 8), an alternative pore formation by actinoporins belonging to the mechanism to that in which helix penetration occurs group of α-PFTs. The characterization of the before protein oligomerization (21). Although pore- prepore in actinoporins highlights similarities with formation by the successive insertion of single α- the mechanism for pore formation of the group of helices cannot be completely ruled out in β-PFT, despite these two groups having quite membranes made of SM/DOPC (1:1), simple distinct architectures at the transmembrane region. AB is a staff scientist from the CONICET (Argentina) and received a visiting scientist fellowship from the Basque Government while conducting this work. We thank Dr. S. Kudo for expert advice. This work was supported by a Grant-in-Aid for Scientific Research A (25249115 to KT) and a Grant-in-Aid for Scientific Research C (15K06962 to JMMC). KM was a recipient of a fellowship from the Spanish Ministerio de Ciencia e Innovación during the beginning of this study. Work in the Scheuring-Lab was supported by a European Research Council Grant (#310080, MEM-STRUCT-AFM). This work was also supported by grant BFU2015-66326-P from the Spanish Ministry of Economy and Competitiveness to MV. FOOTNOTE The present address of AB is Instituto Superior de Investigaciones Biológicas (INSIBIO, CONICET-UNT) e Instituto de Química Biológica "Dr. Bernabé Bloj", Facultad de Bioquímica, Química y Farmacia, Universidad Nacional de Tucumán, Chacabuco 461, San Miguel de Tucumán. Argentina. CONFLICT OF INTEREST The authors declare that they have no conflicts of interest with the contents of this article. AUTHOR CONTRIBUTIONS KM, JMGM, and JMMC conceived the study. KM, JMGM, KT and JMMC coordinated the study. MV and DGC analyzed and performed cryo-EM studies. SS and LRM analyzed and performed AFM studies. KM, 10 AB, DGC, LRM, and JS performed experiments. All authors designed experiments and analyzed the data. KM and JMMC wrote the paper with input from all other authors. All authors reviewed the results and Identification of a prepore species in actinoporins approved the manuscript. REFERENCES 1. 2. 3. 4. 5. 6. 7. 8. 9. Anderluh, G., and Lakey, J. H. (2008) Disparate proteins use similar architectures to damage membranes. Trends Biochem. Sci. 33, 482-490 Dal Peraro, M., and van der Goot, F. G. (2016) Pore-forming toxins: ancient, but never really out of fashion. Nat. Rev. Microbiol. 14, 77-92 Tilley, S. J., and Saibil, H. R. (2006) The mechanism of pore formation by bacterial toxins. Curr. Opin. Struct. Biol. 16, 230-236 De, S., and Olson, R. (2011) Crystal structure of the Vibrio cholerae cytolysin heptamer reveals common features among disparate pore-forming toxins. Proc. Natl. Acad. Sci. USA 108, 7385-7390 Gilbert, R. J. C., Serra, M. D., Froelich, C. J., Wallace, M. I., and Anderluh, G. (2014) Membrane pore formation at protein-lipid interfaces. Trends Biochem. Sci. 39, 510-516 Mueller, M., Grauschopf, U., Maier, T., Glockshuber, R., and Ban, N. (2009) The structure of a cytolytic alpha-helical toxin pore reveals its assembly mechanism. Nature 459, 726-730 Rojko, N., and Anderuh, G. (2015) How lipid membranes affect pore forming toxin activity. Acc. Chem. Res. 48, 3073-3079 Fang, Y., Cheley, S., Bayley, H., and Yang, J. (1997) The heptameric prepore of a Staphylococcal alpha-hemolysin mutant in lipid bilayers imaged by atomic force microscopy. Biochemistry 36, 9518-9522 Hotze, E. M., Heuck, A. P., Czajkowsky, D. M., Shao, Z. F., Johnson, A. E., and Tweten, R. K. (2002) Monomer-monomer interactions drive the prepore to pore conversion of a beta-barrel-forming cholesterol- dependent cytolysin. J. Biol. Chem. 277, 11597-11605 10. Leung, C., Dudkina, N. V., Lukoyanova, N., Hodel, A. W., Farabella, I., Pandurangan, A. P., Jahan, N., Damaso, M. P., Osmanovic, D., Reboul, C. F., Dunstone, M. A., Andrew, P. W., Lonnen, R., Topf, M., Saibil, H. R., and Hoogenboom, B. W. (2014) Stepwise visualization of membrane pore formation by suilysin, a bacterial cholesterol-dependent cytolysin. eLife 3, e04247 11. Sonnen, A. F. P., Plitzko, J. M., and Gilbert, R. J. C. (2014) Incomplete pneumolysin oligomers form membrane pores. Open Biol. 4, 140044 12. Gilbert, R. J. C. (2016) Protein-lipid interactions and non-lamellar lipidic structures in membrane pore formation and membrane fusion. Biochim. Biophys. Acta 1858, 487-499 13. Eifler, N., Vetsch, M., Gregorini, M., Ringler, P., Chami, M., Philippsen, A., Fritz, A., Muller, S. A., Glockshuber, R., Engel, A., and Grauschopf, U. (2006) Cytotoxin ClyA from Escherichia coli assembles to a 13-meric pore independent of its redox-state. EMBO J. 25, 2652-2661 11 Identification of a prepore species in actinoporins 14. Fahie, M., Romano, F. B., Chisholm, C., Heuck, A. P., Zbinden, M., and Chen, M. (2013) A Non-classical assembly pathway of Escherichia coli pore-forming toxin cytolysin A. J. Biol. Chem. 288, 31042-31051 15. Tanaka, K., Caaveiro, J. M. M., Morante, K., Gonzalez-Manas, J. M., and Tsumoto, K. (2015) Structural basis for self-assembly of a cytolytic pore lined by protein and lipid. Nat. Commun. 6, 6337 16. 17. Tanaka, K., Caaveiro, J. M. M., and Tsumoto, K. (2015) Bidirectional Transformation of a Metamorphic Protein between the Water-Soluble and Transmembrane Native States. Biochemistry 54, 6863-6866 Subburaj, Y., Ros, U., Hermann, E., Tong, R., and Garcia-Saez, A. J. (2015) Toxicity of an alpha-pore- forming toxin depends on the assembly mechanism on the target membrane as revealed by single molecule imaging. J. Biol. Chem. 290, 4856-4865 18. Wacklin, H. P., Bremec, B. B., Moulin, M., Rojko, N., Haertlein, M., Forsyth, T., Anderluh, G., and Norton, R. S. (2016) Neutron reflection study of the interaction of the eukaryotic pore-forming actinoporin equinatoxin II with lipid membranes reveals intermediate states in pore formation. Bba-Biomembranes 1858, 640-652 19. Antonini, V., Perez-Barzaga, V., Bampi, S., Penton, D., Martinez, D., Serra, M. D., and Tejuca, M. (2014) Functional characterization of sticholysin I and W111C mutant reveals the sequence of the actinoporin's pore assembly. Plos One 9 20. Mechaly, A. E., Bellomio, A., Gil-Cartón, D., Morante, K., Valle, M., González-Mañas, J. M., and Guerin, D. M. A. (2011) Structural insights into the oligomerization and architecture of eukaryotic membrane pore- forming toxins. Structure 19, 181-191 21. Rojko, N., Kristan, K. C., Viero, G., Zerovnik, E., Macek, P., Dalla Serra, M., and Anderluh, G. (2013) Membrane damage by an alpha-helical pore-forming protein, equinatoxin II, proceeds through a succession of ordered steps. J. Biol. Chem. 288, 23704-23715 22. Morante, K., Caaveiro, J. M., Viguera, A. R., Tsumoto, K., and González-Mañas, J. M. (2015) Functional characterization of Val60, a key residue involved in the membrane-oligomerization of fragaceatoxin C, an actinoporin from Actinia fragacea. FEBS Lett 589, 1840-1846 23. Hong, Q., Gutiérrez-Aguirre, I., Barlic, A., Malovrh, P., Kristan, K., Podlesek, Z., Macek, P., Turk, D., González-Mañas, J. M., Lakey, J. H., and Anderluh, G. (2002) Two-step membrane binding by Equinatoxin II, a pore-forming toxin from the sea anemone, involves an exposed aromatic cluster and a flexible helix. J. Biol. Chem. 277, 41916-41924 24. Morante, K., Caaveiro, J. M. M., Tanaka, K., Gonzalez-Manas, J. M., and Tsumoto, K. (2015) A pore-forming toxin requires a specific residue for its activity in membranes with particular physicochemical properties. J. Biol. Chem. 290, 10850-10861 Bartlett, G. R. (1959) Phosphorus assay in column chromatography. J. Biol. Chem. 234, 466-468 Ellens, H., Bentz, J., and Szoka, F. C. (1985) H+- and Ca2+-induced fusion and destabilization of liposomes. Biochemistry 24, 3099-3106 12 25. 26. Identification of a prepore species in actinoporins 27. Caaveiro, J. M., Echabe, I., Gutiérrez-Aguirre, I., Nieva, J. L., Arrondo, J. L., and González-Mañas, J. M. (2001) Differential interaction of equinatoxin II with model membranes in response to lipid composition. Biophys. J. 80, 1343-1353 28. Ludtke, S. J., Baldwin, P. R., and Chiu, W. (1999) EMAN: Semiautomated software for high-resolution single-particle reconstructions. J. Struct. Biol. 128, 82-97 29. 30. Scheres, S. H. W., Nunez-Ramirez, R., Sorzano, C. O. S., Carazo, J. M., and Marabini, R. (2008) Image processing for electron microscopy single-particle analysis using XMIPP. Nature Protoc. 3, 977-990 Baxter, W. T., Leith, A., and Frank, J. (2007) SPIRE: The SPIDER Reconstruction Engine. J. Struct. Biol. 157, 56-63 31. Pettersen, E. F., Goddard, T. D., Huang, C. C., Couch, G. S., Greenblatt, D. M., Meng, E. C., and Ferrin, T. E. (2004) UCSF chimera - A visualization system for exploratory research and analysis. J. Comput. Chem. 25, 1605-1612 32. 33. Angelova, M. I., and Dimitrov, D. S. (1986) Liposome electroformation. Farad. Discuss. 81, 303-311 Montes, L. R., Ahyayauch, H., Ibarguren, M., Sot, J., Alonso, A., Bagatolli, L. A., and Goñi, F. M. (2010) Electroformation of giant unilamellar vesicles from native membranes and organic lipid mixtures for the study of lipid domains under physiological ionic-strength conditions. Methods Mol. Biol. 606, 105-114 34. Chiaruttini, N., Redondo-Morata, L., Colom, A., Humbert, F., Lenz, M., Scheuring, S., and Roux, A. (2015) Relaxation of loaded ESCRT-III spiral springs drives membrane deformation. Cell 163, 866-879 35. Husain, M., Boudier, T., Paul-Gilloteaux, P., Casuso, I., and Scheuring, S. (2012) Software for drift compensation, particle tracking and particle analysis of high-speed atomic force microscopy image series. J. Mol. Recogn. 25, 292-298 36. 37. Edman, P., and Begg, G. (1967) A protein sequenator. Eur. J. Biochem. 1, 80-91 Calvez, P., Bussieres, S., Demers, E., and Salesse, C. (2009) Parameters modulating the maximum insertion pressure of proteins and peptides in lipid monolayers. Biochimie 91, 718-733 38. Kakiuchi, T., Kotani, M., Noguchi, J., Nakanishi, M., and Senda, M. (1992) Phase-transition and ion permeability of phosphatidylcholine monolayers at the polarized oil-water interface. J. Colloid. Interface Sci. 149, 279-289 39. Pedrera, L., Fanani, M. L., Ros, U., Lanio, M. E., Maggio, B., and Alvarez, C. (2014) Sticholysin I-membrane interaction: An interplay between the presence of sphingomyelin and membrane fluidity. Biochim. Biophys. Acta 1838, 1752-1759 40. Barlic, A., Gutiérrez-Aguirre, I., Caaveiro, J. M. M., Cruz, A., Ruíz-Arguello, M. B., Pérez-Gil, J., and González-Mañas, J. M. (2004) Lipid phase coexistence favors membrane insertion of equinatoxin-II, a pore- forming toxin from Actinia equina. J. Biol. Chem. 279, 34209-34216 41. Palacios-Ortega, J., Garcia-Linares, S., Astrand, M., Al Sazzad, M. A., Gavilanes, J. G., Martinez-del-Pozo, A., and Slotte, J. P. (2016) Regulation of Sticholysin II-Induced Pore Formation by Lipid Bilayer Composition, Phase State, and Interfacial Properties. Langmuir 32, 3476-3484 13 Identification of a prepore species in actinoporins 42. Scheres, S. H. W., Valle, M., Nunez, R., Sorzano, C. O. S., Marabini, R., Herman, G. T., and Carazo, J. M. (2005) Maximum-likelihood multi-reference refinement for electron microscopy images. J. Mol. Biol. 348, 139-149 43. Sorzano, C. O. S., Bilbao-Castro, J. R., Shkolnisky, Y., Alcorlo, M., Melero, R., Caffarena-Fernandez, G., Li, M., Xu, G., Marabini, R., and Carazo, J. M. (2010) A clustering approach to multireference alignment of single-particle projections in electron microscopy. J. Struct. Biol. 171, 197-206 44. Gutiérrez-Aguirre, I., Barlic, A., Podlesek, Z., Macek, P., Anderluh, G., and González-Mañas, J. M. (2004) Membrane insertion of the N-terminal alpha-helix of equinatoxin II, a sea anemone cytolytic toxin. Biochem. J. 384, 421-428 45. Mechaly, A. E., Bellomio, A., Morante, K., Agirre, J., Gil-Cartón, D., Valle, M., González-Mañas, J. M., and Guerin, D. M. A. (2012) Pores of the toxin FraC assemble into 2D hexagonal clusters in both crystal structures and model membranes. J. Struct. Biol. 180, 312-317 46. Mancheño, J. A., Martin-Benito, J., Gavilanes, J. G., and Vázquez, L. (2006) A complementary microscopy analysis of Sticholysin II crystals on lipid films: Atomic force and transmission electron characterizations. Biophys. Chem. 119, 219-223 47. Yilmaz, N., Yamada, T., Greimel, P., Uchihashi, T., Ando, T., and Kobayashi, T. (2013) Real-time visualization of assembling of a sphingomyelin-specific toxin on planar lipid membranes. Biophys. J. 105, 1397-1405 48. Rojko, N., Dalla Serra, M., Macek, P., and Anderluh, G. (2016) Pore formation by actinoporins, cytolysins from sea anemones. Biochim. Biophys. Acta 1858, 446-456 49. Alegre-Cebollada, J., Onaderra, M., Gavilanes, J. G., and del Pozo, A. M. (2007) Sea anemone actinoporins: The transition from a folded soluble state to a functionally active membrane-bound oligomeric 50. 51. 52. pore. Curr. Protein Pept. Sci. 8, 558-572 Bayley, H. (2009) Membrane-protein structure: Piercing insights. Nature 459, 651-652 von Heijne, G. (2006) Membrane-protein topology. Nat. Rev. Mol. Cell Biol. 7, 909-918 Wimley, W. C., and White, S. H. (1996) Experimentally determined hydrophobicity scale for proteins at membrane interfaces. Nat. Struct. Biol. 3, 842-848 53. Metkar, S. S., Wang, B. K., Catalan, E., Anderluh, G., Gilbert, R. J. C., Pardo, J., and Froelich, C. J. (2011) Perforin rapidly induces plasma membrane phospholipid flip-flop. Plos One 6 54. Scherfeld, D., Kahya, N., and Schwille, P. (2003) Lipid dynamics and domain formation in model membranes composed of ternary mixtures of unsaturated and saturated phosphatidylcholines and cholesterol. Biophys. J. 85, 3758-3768 55. Bellomio, A., Morante, K., Barlic, A., Gutiérrez-Aguirre, I., Viguera, A. R., and González-Mañas, J. M. (2009) Purification, cloning and characterization of fragaceatoxin C, a novel actinoporin from the sea anemone Actinia fragacea. Toxicon 54, 869-880 14 Identification of a prepore species in actinoporins ABBREVIATIONS PFT, pore-forming toxins; FraC, fragaceatoxin C; SM, sphingomyelin; DLPC, 1,2-dilauroyl-sn-glycero-3-phosphocholine; DPPC, 1,2-dipalmitoyl-sn-glycero-3-phosphocholine; DOPC, 1,2-dioleoyl-sn-glycero-3-phosphocholine; DiIC18, 1,1'-dioctadecyl-3,3,3',3'-tetramethylindocarbocyanine perchlorate; LUVs, large unilamellar vesicles; GUVs, giant unilamellar vesicles; AFM, atomic force microscopy; cryo-EM, single-particle cryo-electron microscopy; PK, Proteinase K; CTF, contrast transfer function. FIGURES Figure 1. Two alternative routes for pore formation in actinoporins. The binding of the water-soluble monomer to the cell or model membranes leads to a lytic (active) pore by at least two alternative routes, as shown 15 in the figure. Top, formation of a non-lytic oligomer (prepore) precedes insertion into the membrane (20). Bottom, insertion of the N-terminal region into the membrane occurs prior oligomerization of the functional pore (21). Identification of a prepore species in actinoporins Figure 2. Interaction of FraC with model membranes. (a) Change in surface pressure of lipid monolayers composed of DOPC (red), DLPC (green), or DPPC (blue) after treatment with FraC (1 µM). The parameter πc corresponds to the value of π0 where the regression line intersects the abscissa. The inset shows a representative example of the kinetic profile of insertion of FraC in DOPC monolayers (π0 = 20 mN/m, gray trace). The experimental data was fitted to a hyperbola (red line) from which the value of Δπ was determined. (b) Lytic activity of FraC in LUVs made of DOPC (red) or DPPC (blue). The data obtained with SM/DOPC (1:1) represents a positive control (black). The LUVs made of DLPC are permeable to encapsulated dyes in the absence of protein (spontaneous leakage) and thus the data obtained with them was not considered. For the experiments in panels (a) and (b) the mean and standard deviation of three independent measurements was plotted. (c) Binding of FraC to GUVs made of DOPC/DPPC (20:80) supplemented with 0.2% DiIC18. This probe partitions in the ordered phase regions (yellow domains) (53,54). Protein (red) was added to a final concentration of 1.3 µM. Lipid and protein were visualized with a 593/40 nm band pass filter (yellow, left panel), or with a 650 nm long pass filter (center panel), respectively. Merged images are shown on the right panel. The white arrows point at liquid disordered regions (dark domains) where FraC is preferentially located. The scale bar represents 5 µm. 16 Identification of a prepore species in actinoporins Figure 3. Structure of the oligomeric prepore of FraC bound to DOPC vesicles. (a) Representative image of FraC in DOPC vesicles obtained by cryo-EM. Top- and side-views of the protein oligomers were selected (red squares) for subsequent classification analysis. The scale bar corresponds to 100 nm. (b) Density map projections (top row) and 2D class-averaged particles (bottom row) employed to build a 3-dimensional model of the protein oligomer (see below). (c) Set of particles obtained by maximum-likelihood (ML2D, top row) and hierachical clustering (CL2D, bottom row) procedures. (d) Top- and (e) side-views of the 3-dimensional model of the prepore of FraC bound to vesicles of DOPC. The atomic model of FraC was built as an octamer using the coordinates of the protomer of FraC prior to pore formation (entry code 4TSL). 17 Identification of a prepore species in actinoporins Figure 4. Electron density of FraC bound to vesicles. Side view (Z-projection) of oligomers of FraC bound to vesicles of (a) DOPC, or (b) SM/DOPC (1:1). The yellow square indicates the region where the 1-dimensional profile of the Z-projection (shown in c, d) was calculated. The intensity of the electron density is expressed in gray values. Panels (b) and (d) correspond to the analysis carried out with published data (20), although we note that the analysis presented here has not been shown elsewhere. 18 Identification of a prepore species in actinoporins Figure 5. Visualization of pores of WT FraC with AFM. (a) Two-dimensional packing of ring-shaped oligomers of WT FraC on supported lipid bilayers composed of the lipid mixture SM/DOPC (1:1). (b) Diameter distribution analysis (peak-to-peak distances of the protein protrusion in the height profile). The average diameter of the particles was 75 ± 6 Å (mean ± SD from the Gaussian distributions). Inset: detail of the particles inside the white dashed rectangle in panel (a). (c) Magnification (13 nm frame size) of a single FraC oligomer in panel (a) (white dashed square). (d) Cross-section profile (left to right) of FraC oligomers shown in panel (a) (white dashed line). The molecules are packed with a center-to-center distance of ~112 Å. Figure 6. Visualization of prepores of 8-69OX FraC with AFM. (a) Two-dimensional packing of ring-shaped oligomers of 8-69OX FraC on supported lipid bilayers composed of the lipid mixture SM/DOPC (1:1). (b) Diameter distribution analysis (peak-to-peak distances of the protein protrusion in the height profile). The average diameter was 62 7 Å (mean SD from the Gaussian distributions). Inset: detail of the particles inside the white dashed rectangle in panel (a). The slightly smaller diameter compared to the WT suggests a tighter association of the subunits in the 8-69OX FraC mutant. (c) Magnification (12 nm size frame) of a single prepore of FraC 8-69OX (white square in panel a). (d) Cross-section profile (left to right) of prepore particles of FraC 8-69OX (white line in panel a). The molecules packed with a center-to-center distance of ~108 Å. 19 Identification of a prepore species in actinoporins Figure 7. Protection of FraC from PK in the presence of liposomes. (a) SDS-PAGE of the products obtained after the incubation of WT and 8-69OX FraC with DOPC vesicles in the absence and in the presence of PK (lanes 1-4) or with SM/DOPC (1:1) (lanes 5-8). (b) N-terminal sequence of FraC after digestion with PK. The circled number before the sequence corresponds to the lane of the same number in the SDS-PAGE. Residues highlighted in red were digested by PK. The first 16 residues of the recombinant WT protein expressed in E. coli are ADVAGAVIDGAGLGFD (55). (c) The location of the residues digested by PK are depicted in the three dimensional structure of the monomer of FraC (PDB code 3VWI). 20 Identification of a prepore species in actinoporins Figure 8. Model for pore formation by FraC. A toxin monomer binds the membrane. The membrane promotes protein-protein interactions between monomers to produce a dimer (15) leading to prepore upon successive addition of monomer and/or dimers to the growing oligomer. The N-terminal α-helices in the prepore embedded on the surface on the membrane with the N-terminus exposed to the aqueous solution. The conversion to the transmembrane pore would be achieved by the concerted penetration and elongation of the helices across the lipid bilayer. The structures of the monomer, dimer, and pore were retrieved from the PDB with entry codes of 3VWI, 4TSL, and 4TSY, respectively. The structure of the prepore at high resolutions has not been determined experimentally. In this figure the structure of the prepore is drawn to illustrate the model consistent with the experimental data reported in our study. 21
1410.3992	2	1410	2015-01-29T09:55:05	Efficient transmission of subthreshold signals in complex networks of spiking neurons	[ "physics.bio-ph", "cond-mat.dis-nn", "q-bio.NC" ]	We investigate the efficient transmission and processing of weak, subthreshold signals in a realistic neural medium in the presence of different levels of the underlying noise. Assuming Hebbian weights for maximal synaptic conductances -- that naturally balances the network with excitatory and inhibitory synapses -- and considering short-term synaptic plasticity affecting such conductances, we found different dynamic phases in the system. This includes a memory phase where population of neurons remain synchronized, an oscillatory phase where transitions between different synchronized populations of neurons appears and an asynchronous or noisy phase. When a weak stimulus input is applied to each neuron, increasing the level of noise in the medium we found an efficient transmission of such stimuli around the transition and critical points separating different phases for well-defined different levels of stochasticity in the system. We proved that this intriguing phenomenon is quite robust, as it occurs in different situations including several types of synaptic plasticity, different type and number of stored patterns and diverse network topologies, namely, diluted networks and complex topologies such as scale-free and small-world networks. We conclude that the robustness of the phenomenon in different realistic scenarios, including spiking neurons, short-term synaptic plasticity and complex networks topologies, make very likely that it could also occur in actual neural systems as recent psycho-physical experiments suggest.	physics.bio-ph	physics	Efficient transmission of subthreshold signals in complex networks of spiking neurons Joaquin J. Torres1,∗, Irene Elices1,∗∗, J. Marro1 1 Department of Electromagnetism and Physics of the Matter, University of Granada, Granada, Spain ∗ E-mail: Corresponding [email protected] ∗∗ Presently at Grupo de Neurocomputaci´on Biol´ogica, Dpto. de Ingenier´ıa Inform´atica, Escuela Polit´ecnica Superior, Universidad Aut´onoma de Madrid, Madrid, Spain Abstract We investigate the efficient transmission and processing of weak, subthreshold signals in a realistic neural medium in the presence of different levels of the underlying noise. Assuming Hebbian weights for maximal synaptic conductances -- that naturally balances the network with excitatory and inhibitory synapses -- and considering short-term synaptic plasticity affecting such conductances, we found different dynamic phases in the system. This includes a memory phase where population of neurons remain synchronized, an oscillatory phase where transitions between different synchronized populations of neurons appears and an asynchronous or noisy phase. When a weak stimulus input is applied to each neuron, increasing the level of noise in the medium we found an efficient transmission of such stimuli around the transition and critical points separating different phases for well-defined different levels of stochasticity in the system. We proved that this intriguing phenomenon is quite robust, as it occurs in different situations including several types of synaptic plasticity, different type and number of stored patterns and diverse network topologies, namely, diluted networks and complex topologies such as scale-free and small-world networks. We conclude that the robustness of the phenomenon in different realistic scenarios, including spiking neurons, short-term synaptic plasticity and complex networks topologies, make very likely that it could also occur in actual neural systems as recent psycho-physical experiments suggest. Author Summary In this work we have investigated how weak, subthreshold stimuli can be efficiently processed in an auto-associative spiking neural network with dynamic synapses, in the presence of noise. We have described different non-equilibrium phases including a memory phase, an oscillatory phase, with random transitions among memory attractors, and a non-memory phase. Our main finding is that system is able to efficiently process the relevant signals at several well defined levels of the underlying stochasticity in the system, which are related with phase transitions points. In addition, we have proved the robustness of this intriguing phenomenology in different situations which include several types of long term and short-term synaptic plasticity and different network structural properties. This makes very likely that this phenomenology can appear also in actual neural systems. Introduction Many physical systems present ambient and intrinsic fluctuations that often are ignored in theoretical studies to obtain simple mean-field analytical approaches. Nevertheless, these fluctuations may play a fundamental role in natural systems. For instance, they may optimize signals propagation by turning the medium into an excitable one -- e.g., the case of ionic channel stochasticity in neurons that can affect the first spike latency [1, 2] or enhance signal propagation through different neuronal layers [3] 1 -- , originate order at macroscopic and mesoscopic levels [4, 5] or induce coherence between the intrinsic dynamics of a system and some weak stimuli it receives, a phenomenon known as stochastic resonance (SR) (see for instance [6] for a review). Precisely, this intriguing phenomenon has attracted the interest of the computational neuroscience community for its possible implications in the complex processing of information in the brain [7 -- 11], or as a way to control specific brain states [12]. In fact, most neural systems naturally include the main factors involved in stochastic resonance, namely, different sources of intrinsic and external noise and complex nonlinear dynamic processes associated, for instance, to neuron excitability and information transmission through the synapses. Thus, several experimental, theoretical and numerical studies concerning the efficient transmission of weak signals in noisy neural systems have been reported recently [7 -- 11, 13 -- 17]. Particularly interesting are recent works that show that, differently to what happen in traditional SR phenomena, noisy neural systems can optimally process relevant information at more than one level of the ambient noise [15, 17]. However, it is not clear yet what are the main factors responsible for this new phenomenology. It has been reported, for instance, that this new phenomenon can appear in single perceptrons with dynamic synapses [15] as a consequence of the complex interplay between dynamic synapses and adaptive threshold mechanisms affecting neuron excitability. Also, similar phenomena has been reported in binary networks with scale-free topologies [13], so that these resonances may emerge as a consequence of topological disorder. Finally, very recently, it has been reported that optimal processing of weak relevant signals at different levels of the underlying noise also occurs in auto-associative networks of binary neurons with dynamic synapses [17]. Most of these studies, however, consider very simple models at the neuron, network and synapse levels. This makes difficult to extrapolate their results and conclusions to actual neural systems. Here, we present a full computational study of how weak relevant subthreshold signals can be processed by neural systems in a more realistic scenario, that is, a complex auto-associative network of spiking neurons with dynamic synapses. We consider a network of N integrate and fire neurons, and assume a long-term synaptic plasticity mechanism, due to Hebbian learning, affecting the synapses connecting the neurons. In order to increase the biological relevance of our study we also consider different types of short-term synaptic plasticity, such as the so called short-term depression and short-term facilitation. These mechanisms introduce synaptic changes at short time scales, as it is expected to occur due to the existence of a limited amount of neurotransmitters at each synapse that can be released after some presynaptic stimulus, which needs some time to recover after this stimulus. In addition, to mimic any source of intrinsic or external ambient noise that can affect neuron dynamics and its excitability, we add to each neuron dynamics a source of uncorrelated Gaussian noise together with a weak stimulus. In this way, by controlling the intensity of noise we can monitor the levels of the noise at which the processing of the weak stimulus by the system is more efficient. For simplicity, we assume in most of the cases reported here a form of sinusoidal signal for the stimulus, which simulates some relevant information to be processed by neurons, but other type of more realistic signals can also be considered with the same results (see Results section). To see how efficient this weak stimulus in each neuron is processed by the system, one can compare, for instance, the coherence in time between the mean network activity and the stimulus by means some information transfer measurement as a function of the noise intensity. In general for low noise, the activity of the system cannot correlate with the stimulus due to its small amplitude. If the intensity of noise is increased sufficiently, then the noise is able to enhance the stimulus temporal features in such a way that the network activity can start to correlated with it and a peak of information transfer will appear. Nevertheless, if the noise intensity is too high, the network activity will be dominated by the noise preventing the input stimulus from being detected by the system. On the other hand, activity dependent synaptic mechanisms, such as short-term depression and short- term facilitation, may be highly relevant in signal detection in noisy environments and can play a main role, for instance, in SR [15, 17]. These synaptic mechanisms may modify the postsynaptic neural response 2 in a nontrivial way. Synapses can present short-term depression when the amount of neurotransmitters that are available for release whenever an action potential arrives is limited, and consequently, the synapse may not have time to recover them if the frequency of arriving spikes is too high. On the contrary, short-term facilitation is determined by the excess of calcium ions in the presynaptic terminal which can increase the postsynaptic response under repetitive stimulation. Both synaptic processes could interact with noise and some neuron excitability and adaptive mechanisms to induce a strong influence during the processing of relevant signals or stimulus in the brain. In particular, it has been recently reported in single perceptrons and in network of binary neurons that the complex interplay among these synaptic mechanisms allow for efficient detection of weak signals at different levels of the underlying noise [15, 17] and maintain coherence for a wide range of the intensity of such noise [8]. In this work, we demonstrate that these intriguing emergent phenomena appear also during the processing of weak subthreshold signals in more realistic neural media and in many different conditions. Therefore, it is highly likely that also they may appear in the actual neural systems, where different types of signals and stimulus are continuously processing in the presence of different sources of intrinsic and external noise. Moreover, the fact that the processing of weak subthreshold signals occurs at well defined different levels of noise -- normally one relatively low and the other relatively high -- can have strong implications concerning how the signal features are being processed. This can be clearly depicted in the case of more realistic Poissonian signals (see Results section) To demonstrate the robustness of our findings, we performed a complete analysis of the emergent phenomena changing many variables in our system. This confirms that the same interesting phenomena emerges in all these situations, including, for instance the case in which the number of neurons in the network and the number of stored patterns is increased. In addition, the phenomenon of interest also remains for non-symmetric stored patterns provided that there is a phase of transitions between a high activity state (up state) and a low activity state (down state) in the network activity. Including short-time synaptic facilitation at the synapses, competing with synaptic depression, causes also intriguing features. This includes a dependency with the level of facilitation, of the level of noise at which the subthreshold signals are processed and detected and an enhancing of the detection quality for large facilitation. Finally, we checked the robustness of our finding for more realistic network topologies, such as diluted networks, and complex scale-free and small-world topologies confirming that phenomenon is robust also in these cases. Materials and Methods The system under study consists of a spiking network of N integrate and fire neurons interconnected each other. The membrane potential of the i − th neuron then follows the dynamics τm dVi(t) dt = −Vi(t) + RmIi(t) 0 < Vi(t) < Vth, (1) where τm is the cell membrane time constant, Rm is the membrane resistance and Vth is a voltage threshold for neuron firing. Thus, when the input current Ii(t) is such that depolarizes the membrane potential until it reaches Vth = 10mV an action potential is generated. Then, the membrane potential is reset to its resting value -- that for simplicity we assume here to be zero -- during a refractory period of τref = 5 ms. We can assign binary values si = 1, 0 to the state of the neurons, depending if they have their membrane voltage above or below the voltage firing threshold Vth. Furthermore, we assume that synapses between neurons are dynamic and described by the Tsodyks-Markram model introduced in [18]. Within this framework, we consider that the total input current Ii(t) in the equation (1) has four i + Dζ(t) where I0 is a constant input current. The second represents an external input weak signal which encodes relevant information and for simplicity components, that is, Ii(t) = I0 + I ext term I ext i i + I syn 3 we assume to be sinusoidal, that is I ext i = ds sin(2πfst), (2) with frequency fs and a small amplitude ds. The fourth component of Ii(t) is a noisy term that tries to mimic different sources of intrinsic or external current fluctuations, where ζ(t) is a Gaussian white noise of zero mean and variance σ = 1, and D is the noise intensity. Finally, the third component I syn is the sum of all synaptic currents generated at neuron i from the arrival of presynaptic spikes on its neighbors. Following the model of dynamic synapses in [18], we describe the state of a given synapse j by variables yj(t), zj(t) and xj(t) representing, respectively, the fraction of neurotransmitters in active, inactive and recovering states. Within this framework, active neurotransmitters yj(t) are the responsible for the generation of the postsynaptic response after the incoming presynaptic spikes, and become inactive after a typical time τin ∼ 2 − 3ms. On the other hand, inactive neurotransmitters can recover during a typical time τrec which is order of a half second for typical pyramidal neurons [18], a fact that induces short-term synaptic depression. Recovered neurotransmitters become immediately active with some probability U -- the so called release probability -- every time a presynaptic spike arrives to the synapses. In actual synapses, U can increases in time with a typical time constant τf ac -- due to some cellular biophysical processes associated to the influx of calcium ions after the arrival of presynaptic spikes -- which induces the so called short-term synaptic facilitation. i The synaptic current generated at each synapse then is normally assumed to be proportional to the fraction of active neurotransmitters, namely yj(t), so the total synaptic current generated in a postsynaptic neuron i is: I syn i = A yj(t) Jijij. (3) N(cid:88) j N(cid:88) i=1 mµ(t) = 1 aN (1 − a) 4 i − a)si(t), (ξµ (5) P(cid:88) µ Here A is the maximum synaptic current that can be generated at each synapses, ij is the adjacency matrix that accounts for the connectivity matrix in the neural medium, and Jij are fixed parameters modulating the synaptic current which can be related, for instance, with maximal synaptic conductance modifications due to a slow learning process. In this way, one can choose these synaptic weights Jij following, for instance, a Hebbian learning prescription, namely: Jij = κ (cid:104)k(cid:105)a(1 − a) i − a)(ξµ (ξµ j − a), Jij = Jji, Jii = 0. (4) i = 0, 1}, with Here, Jij contains information from a set of P patterns of neural activity, namely {ξµ µ = 1, ..., P and i = 1, ..., N that are assumed to have been previously stored or memorized by the system during the learning process. Here ξµ i denotes the firing (with membrane voltage above Vth) or silent (with Vm below Vth) state of a given neuron in the pattern µ. The parameter a measures the excess of firing over silent neurons in these learned patterns, or more precisely a = (cid:104)ξµ i (cid:105)i,µ. Since Jij can be in general very small and it is multiplying the single synapse currents, we have considered in (4) an amplification factor κ = 2000 to ensure a minimum significant effect of the resulting synaptic current (3) in the excitability of the postsynaptic neuron. Moreover, we also choose a mean node degree factor (cid:104)k(cid:105), instead of N in the denominator of Jij which is more appropriate since it gives a similar mean synaptic current per neuron for all type of network topologies considered in this study, including fully connected networks, diluted networks and complex networks such as scale-free and the classical Watts-Strogatz small-world networks [19]. Following standard techniques from binary attractor neural networks, we can measure the degree of similarity between a state of the network and a certain stored activity pattern by means of an overlap function mµ(t) defined as: (cid:88) i 0 (7) (8) as well as describe the activity of the system through the mean firing rate: ν(t) = 1 N si(t). (6) In order to visualize if our system is able to respond efficiently to some input weak stimulus, it is useful to quantify the intensity of the correlation, during a time window T, between the weak input signal and the network activity by computing, for instance, the Fourier coefficient at a given frequency f , of the network mean firing rate, that is, Cf = lim T→∞ 1 T T ν(t)eif tdt. The relevant correlation, denoted C(D) in the following, it then is defined as the value of C(D) ≡ Cfs2 d2 s , that is, the ration between the power spectrum computed at the frequency of the input signal fs and the amplitude of this weak signal. Results The effect of short-term synaptic depression As it was stated above, it is important to investigate the mechanisms involved in the processing of different stimulus by a neural system, in the presence of noise. This would determine the conditions of ambient or intrinsic noise at which the transmission of information can be more efficient mainly, when the relevant information of the stimulus is encoded in weak signals. This is particularly important in a complex neural system as it is the brain, where certain brain areas have to respond adequately to some signals, for instance, arriving from other specific brain areas or the senses, within a background of noisy activity. Following this aim, we have first studied how efficient is the processing of noisy weak signals in a network of N spiking neurons, when it stores a single pattern of neural activity, and where the synapses among neurons present short-term synaptic depression. Our study reveals the relevant signals can be processed by the systems at more than one level of the underlying noise, as it is depicted in figure 1. More precisely, the correlation measure C(D) presents, two well defined maxima, one at relatively low noise D1 = 97.5 pA and the second at relatively large noise intensity D2 = 265 pA. Model parameter values are indicated in the caption of the figure 1. A full description of collective behavior of the network, by means of the temporal evolution of the mean firing rate and the overlap function compared with the weak sinusoidal input, and for increasing values of the noise parameter D along the curve C(D), is depicted in figure 2. Moreover, raster plots of the network activity, for the same cases shown in figure 2, are presented in figure 3. Both figures show (more clearly illustrated in figure 3), that for relatively low noise the system is able to recall the stored pattern, which becomes an attractor of the system dynamics. The system, therefore, shows the associative memory property. When noise intensity is increased to some given value D1 (around 97.5 pA in this figure), the dynamic regime of the system changes sharply to an oscillatory phase where the network activity periodically switches between a pattern and anti-pattern configurations. Around this phase-transition point D1, these oscillations start to be driven by the weak signal which causes the first appearing maxima in C(D). This periodically switching behavior correlated with the weak signal is clearly reflected in the overlap function and the mean firing rate (see figure 2), and relatively large-amplitude oscillations in these order parameters with the same frequency as the sinusoidal weak input signal appear. However, by increasing further the level of noise D, we observe that the correlation with the input signal 5 is lost. For a further increase of noise around a given value D2 (which in the simulations performed in the figure is about 265 pA), a second peak in C(D) appears, where a strong correlation of the neural activity with the input weak signal is recovered. This noise level corresponds to the critical value of noise at which a second order phase transition between the oscillatory phase and a disordered phase emerges. To check the influence of short term depression in the appearance of these maxima of the correlation function C(D), we have varied the neurotransmitter recovery time constant τrec, which is a well know parameter that permits the tuning of the level of depression at the synapses. In fact, large recovering time constants are associated to stronger synaptic depression because the synapses need more time to have available neurotransmitter vesicles in the ready releasable pool. Therefore, we repeat the numerical study for several values of the time recovery constant τrec = 250, 300, 350 ms, considering a network of N = 2000 neurons. The results are depicted in figure 4, where C(D) is shown for different values of τrec. Two main intriguing effects are observed. First, the maxima of the C(D), at which there is a high correlation with the weak signal, appear at lower noise intensities when τrec is increased. This is due to the fact that, when the level of synaptic depression is increased, the transitions between ordered and oscillatory phases and between oscillatory and disordered phases appear at lower values of noise intensity. This is due to the extra destabilizing effect over the memory attractors consequence of synaptic depression [20] and to the fact the maxima of C(D) occur precisely at these transitions points [17]. The second effect is that the correlation with the weak signal (the height of the maxima) increases with the level of depression, that is, the weak signal is processed with less noise, which is consequence of the phase transitions points -- and therefore the maxima of C(D) -- appear at lower noise values when synaptic depression is increased. The effect of short-term facilitation In general, synapses in the brain and, in particular, in the cortex can present -- in addition to synaptic depression -- the so called synaptic facilitation mechanism, that is an enhancement of the postsynaptic response at short time scales [18, 21]. Both opposite mechanisms can interact at the same time scale during the synaptic transmission in a complex way whose computational implications still are far of being well understood. The study of the influence of both mechanisms during the processing of weak signals in a neural medium, constitutes a very suitable framework to investigate this interplay. With this motivation, we present in this section a computational study of how synaptic facilitation competing with synaptic depression influences the detection of weak stimuli in a network of spiking neurons. In the following computational study, we consider a fixed time recovery constant τrec = 300 ms, which is within the physiological range of the actual value measured in cortical neurons with depressing synapses [21]. Also, we take several values for the characteristic facilitation time constant, namely, τf ac = 100, 200, 500 ms, and U = 0.02. The results obtained for the correlation function C(D) for a network of N = 800 neurons are depicted in figure 5. In this figure, we can observe a clear dependence between the level of noise at which maxima in the correlation function appear and the characteristic facilitation time constant τf ac. In particular, the figure shows that, as τf ac increases, the maxima of C(D) emerge at lower noise intensities. Moreover, one observes that the intensity of the correlation at its low noise maximum grows whenever τf ac is increased. A possible explanation for this phenomenology is the following: it is well known, that facilitation favors to reach the stored attractors and their subsequent destabilization, in auto-associative neural networks [22]. In other words, synaptic facilitation favors the appearance of the oscillatory phase. So then, for the same level of noise D, more facilitated synapses induce an easy recovery and posterior destabilization of the attractors, and therefore, an easy transition to the oscillatory phase from the memory phase. This, in practice, means that the transition point between the two phases, appears at lower values of the noise for more facilitated synapses, and it is precisely at this transition point, where the low noise maximum of C(D) appears. On the other hand, synaptic facilitation favors the recovery of the memory attractors with less error [22]. In this way, when the transition to the oscillatory phase occurs, attractors are periodically 6 and transiently recovered with less error during some time so that, the coherence of the activity of the system with the weak signals is larger since it is not affected by this extra source of noise. These findings provide a simple mechanism to control the processing of relevant information by changing the level of facilitation in the system which can be done, for instance, controlling the level of calcium influx into the neuron or by the use of calcium buffers inside the cells. The effect of network size In order to verify the robustness of the results reported above and to observe the possible effects that may arise due to the finite size of the system used in our simulations, we have carried out a study of the system increasing the number of neurons in the network as N = 400, 800, 1600, 2000, but maintaining the rest of the parameters and considering spiking neurons with pure depressing synapses. The computed correlation C(D) for all these cases is depicted in figure 6, which reveals that the main findings of our previous study remain and are independent of the number of neurons in the system. In fact, different C(D) curves for different values of N do not present significantly changes neither in their shape and intensity, nor in the level of noise at which the different maxima of C(D) appear. These results permit us to hypothesize that our main findings here are enough general and could also appear in large populations of neurons as in cortical slices or even in some brain areas. The effect of storing many patterns in the network In the studies reported in the above sections, we have considered just one activity pattern of information stored in the maximal synaptic conductances. We study now, how robust are these findings when the number P of activity patterns stored in the system increases with all other parameters of the model unchanged. In our study, we have varied P from 1 to 10 in a network of N = 2000 neurons with pure depressing synapses (τrec = 300 ms) and the corresponding correlation functions C(D) for all these cases are depicted in figure 7. One can appreciate that the phenomenon remains when P is increased, and while the maximum of C(D) that appears at high noise -- around D2 = 265 pA -- does not dramatically change with P , the number of stored patterns has a strong effect on the maximum of C(D) appearing at low noise. In fact, this maximum appears at lower level of noise and with more intensity as P is increased. A possible explanation for this intriguing behavior can be understood as follows. It is well known that in Hopfield binary neural networks the increase of the number of stored patterns induces interference among the memory attractors and therefore constitutes an additional source of noise that tries to destabilize them [23]. This also occurs in our spiking network and, in the presence of dynamic synapses, this destabilizing effect results in an early appearance of the transition between the memory phase and the oscillatory phase and therefore, in the appearance of the first low-noise maximum of C(D). Then, at relatively low levels of ambient noise D the transition from below to the oscillatory phase will occur at lower values of D in a network that stores a larger number of patterns. At relatively large values of the ambient noise, however, the main destabilizing effect is due to the ambient underlying noise, and therefore, the effect of increasing P , although is also present, it is less determinant. On the other hand, the amplitude of the maximum of C(D) increases with P because, as explained above, the transition among memory and oscillatory phases occurs at low value of the ambient noise for P larger. Then, the thermal fluctuations in the memory phase and during the phase transition to the oscillatory phase are lower. In this way, during the oscillations starting at the transition point, the attractors are recovered transiently with less error which induces the coherence with the weak signal to be larger. The effect of the asymmetry of the stored pattern The features of the pattern-antipattern oscillations which characterize the oscillatory phase in neural networks with dynamic synapses (including short-term facilitation and depression), are highly dependent 7 on the particular symmetry in the number of active and silent neurons in the stored pattern. This is controlled by the parameter a (introduced in the definition of the synaptic weights (4)). In all the results reported in previous sections, we have considered a = 0.5, which causes an oscillatory phase characterized by a regime of symmetric oscillations between an activity state correlated with the stored pattern and If we consider a (cid:54)= 0.5, an another with the same level of activity correlated with the antipattern. asymmetry will be induced in the mean network activity, that is, there will be an excess of 1's over 0's or vice versa during pattern-antipattern oscillations. In fact, oscillations occurs between an high activity (Up) state and a low activity (Down) state. Moreover, this asymmetry in the activity of the stored pattern can have a strong influence in the phase diagram of the system and can cause even that the oscillatory phase does not emerge. Since the phenomenology of interest here -- namely the emergence of an network activity correlated with a weak subthreshold stimulus -- is highly dependent on the transition points at which the system moves over different phases, it is reasonable to think that the parameter a will have a strong influence on it. We have performed a computational study in a network of N = 800 neurons with pure depressing synapses (τrec = 300 ms) to investigate this particularly interesting issue, for which we have considered a single stored pattern P = 1 with a = 0.40, 0.42, 0.43, 0.45, 0.47, 0.48 and whose results are depicted in figure 8. We observe here a very interesting and intriguing effect in the shape of the correlation C(D) when a is varied. The maximum in the correlation between the network activity and the weak signal at low noise tends to disappear as the value of a decreases from the symmetric value a = 0.5. In fact, when a < 0.45 the correlation C(D) drops abruptly at that point, around D1 = 100 pA. As we have explained above, a large level of asymmetry in the stored pattern could impede the appearance of the Up/Down transitions characteristic of the oscillatory phase which, therefore is absent. The consequence is that, there is not a transition point between a memory phase and an oscillatory phase which impedes the emergence of the low noise maximum of C(D). On the contrary, the second maximum of C(D), which appears at high noise around D2 = 265 pA, remains invariant for all values of a studied here. The explanation to this second situation is also simple because, although asymmetry in the stored pattern is present, the phase diagram of the system still presents a phase of memory retrieval at low noise and a non-memory phase at large noise, separated by a second order phase transition point around which the second maximum of C(D) is originated. The effects of the underlying network topology In previous sections, we have considered for simplicity -- as our system under study -- a fully connected network of spiking neurons. This is far to be the situation in actual neural systems, where neurons are not all connected to each others. In fact, biological neural systems are characterized by a underlying complex network topology which is consequence of different biophysical processes during their developing, including among others, exponential growth at early stages of developing and posterior synaptic pruning processes [24]. All these processes are also influenced by limitations in energy consumption in the system. In this section, we explore if the emergence of several maxima, as a function of the underlying noise, in the correlation between the network activity and some weak subthreshold stimulus, is altered when more realistic network topologies are considered. We have considered first the case of a random diluted network. We can configure this network topology starting, for instance, starting with a fully connected network and then removing randomly a certain fraction δ of the synaptic connections. In figure 9A, it is depicted the resulting correlation function C(D) for single realizations of diluted networks generated in this way with N = 800 and δ =10%, 20%, 30% and 40%. The figure illustrates two main findings. First, the robustness of the main emergent phenomena described in the above sections also in this type of diluted networks, and second that, as the dilution grows and a higher fraction of connections is removed, both maxima of C(D) appear respectively at lower levels of noise. Moreover, if dilution is too high, it seems that the maximum appearing at low noise tends to disappear. This only can be consequence that the stable memory attractors lose stability -- due to 8 strong dilution -- and disappear in the presence of ambient noise, in such a way that only an oscillatory phase and non-memory phases are present. In our analysis with a diluted network, we performed dilution starting with a fully connected network where (cid:104)k(cid:105) = N. To avoid the possible effect of a factor 1/N normalizing the synaptic weights (4) during dilution, we have done an additional analysis considering a diluted network with a normalizing factor in the weights (cid:104)k(cid:105) = (1 − δ)2N, which is the mean connectivity degree in the resulting diluted network with δ being the probability of a link to be removed during the dilution process. The corresponding results are summarized in the figure 9B. One can observe also that results are similar for this second type of dilution, that is, the low noise maximum of C(D) moves toward lower values of the ambient noise and even can disappear as dilution is increased. The main difference with the first type of dilutions is, however, that the level of noise at which the high noise maximum of C(D) is not dramatically affected by dilution. In figure 9B, the correlation curves C(D) have been obtained after averaging over 10 realizations of a network of N = 200 neurons with pure depressing synapses (with τrec = 200 ms). Diluted networks, however, are homogeneous and do not introduce complex features which could induce more intriguing behavior in the system. Even more interesting and realistic concerns the case of networks with complex topology such as the so called scale-free networks, where the node degree probability distribution is p(k) ∼ k−γ, with k being the node degree. In fact, it has been recently reported that these complex topologies can induce additional correlation with the network activity and the processed weak stimulus due to the network structural heterogeneity [13]. Figure 10 summarizes our main results concerning the case of complex networks with scale-free topology. Correlation curves C(D) have been obtained after averaging over 10 realizations of a scale-free network of N = 200 neurons with pure depressing synapses (with τrec = 200 ms) and all other model parameters as in figure 1. We see that similarly to the cases studied above, two maxima in the correlation function C(D) emerge - for two well defined values of the underlying noise -- also in scale-free networks. However, we do not observe the emergence of an additional maximum of C(D) which could be induced only by the topology. As it has been reported in [13], this maximum should appear at low values of the ambient noise, for γ ∼ 3. The existence of a robust oscillatory phase when dynamic synapses are considered and a phase transition between memory and this oscillatory phase at relatively low values of the ambient noise could hidden the appearance of the this maximum due to the existence of a low noise maximum of C(D) around this phase transition. In any case, as it is depicted in figure 10, the emergence of two maxima in C(D) is a robust phenomenon also in these complex scale-free network topologies for a wide range of the relevant network parameters such as the exponent of the network degree distribution (figure 10A) and the mean connectivity in the network (figure 10B). Interestingly is that the low noise maximum seem to start to emerge for values of γ (cid:38) 3 and values of the mean connectivity (cid:104)k(cid:105) (cid:38) 15 − 20. These values corresponds to realistic ones, since, for instance, most of the actual complex networks in nature have degree distributions with γ between 2 and 3 [25]. Moreover neurons in the brain of mammals have large connectivity degrees and realistic values of the mean structural connectivity in cortical areas of mammals has been reported to be around (cid:104)k(cid:105) ≈ 20 [26]. Finally, we have consider in our study the case of a complex network with the small-world property. A prominent example of such type of networks is the so called Watts-Strogatz (WS) network [19]. These networks are generated starting with a regular network where each node, normally placed in a circle, has k0 neighbors. Then, with some probability pr, know as probability of rewiring, each link among nodes in this regular configuration is rewired to a randomly chosen node in the network avoiding self-connections and multiple links among two given nodes. In this way for pr = 0, one has a regular network with p(k) = δ(k − k0) and for pr = 1 one has a totally random network with p(k) being a Gaussian distribution centered around k0. Note that for varying pr one always has (cid:104)k(cid:105) = k0. We have placed neurons defined by the dynamics (1) in such WS networks and studied as a function of the underlying noise the emergence of correlations between the network activity and some subthreshold weak signals by means C(D). The results are summarized in figure 11 where each C(D) curve has been obtained after averaging over 10 9 realizations of a network with N = 200 neurons and pure depressing synapses with τrec = 100 ms. One can see that the appearance of several maxima for C(D) occurs for values of pr (cid:38) 0.5. More precisely, the low noise maximum does not emerge for low rewiring probabilities, which clearly indicates that the memory phase does not appear for such small values of pr for the whole range of noise D considered here. Also this finding suggests the positive role of long range connections - that only can emerge with high probability when pr is high -- for the existence of such low noise maximum in C(D). In fact, the emergence of a memory phase can be only understood when in the network appear such long range connections since the stored memory patterns involve these type of spatial correlations among active and inactive neurons. Use of more realistic weak signals In actual neural systems is expected that relevant signals arrive to a particular neuron in the form of a spike train, with relevant information probably encoded in the timing among the spikes. In this sense, the use of a sinusoidal weak current to explore how the system detect it in all cases considered above could not be the most realistic assumption (it could be enough realistic if relevant information will be encoded in subthreshold oscillations instead that in the precise timing of the spikes). To investigate the ability of the system to detect and process the with more realistic weak signals in the presence of noise we have considered the input weak signal as an inhomogeneous Poisson spike train with mean firing rate λ(t) = λ0[1 + asin(2πfst)], being λ0, a positive constants. In this way, relevant information is encoded as a sinusoidal modulation of the arrival times of the spikes in the train. Figure 12A depicts the coherence among mean firing rate in the network and this weak signal (which is shown in the top graph of figure 12B). The correlation curve C(D) has been obtained after averaging over 20 realizations of a fully connected network with N = 400 neurons and pure depressing synapses with τrec = 300 ms. The figure clearly illustrates that also in this more realistic case the system present a strong correlation with the weak signal at different levels of noise at which phase transitions among different non-equilibrium phases appear (see time series for increasing level of noise from top to bottom in figure 12B). These are a memory phase where active neurons in the stored memory pattern are strongly synchronized (D = 60 pA) (population burst regime), a transition point characterized by signal driven high-activity (Up state)/low-activity (Down state) oscillations (D1 = 85.4 pA), a phase of intrinsic Up/down oscillations (D = 160 pA), a critical point toward a non-memory phase characterized by signal driven fluctuations plus thermal fluctuations (at Dc = D2 = 261 pA), and a non-memory phase, or asynchronous phase, characterized by constant firing rate with Gaussian thermal fluctuations (for instance at D = 500 pA). In fact, in figure 12C it is depicted the difference in steady state features of the behavior for the two last cases. That is, at the critical point D2 the stationary distribution of the resulting mean firing rate has a bias toward positive fluctuations induced at the exact arrival time of the weak signal spikes. This as evidenced by the disagreement between this distribution (red curve in figure 12C) and the shaded red area, which represents the best fit to a Gaussian distribution. On the other hand, during the asynchronous state for D = 500 pA the same steady state distribution (green line in figure 12C) is clearly Gaussian (shaded blue area) without presenting a bias with the timing of signal spikes. Discussion We investigated in great detail by computer simulations the processing of weak subthreshold stimuli in a auto-associative network of spiking neurons - N integrate and fire neurons connected to each others by dynamic synapses using different network configurations -- competing with a background of ambient noise. In particular, we studied the role of short-term synaptic depression in the efficient detection of weak periodic signals by the system as a function of the noise. Our results show the appearance of several well defined levels of noise at which there is a strong correlation between the mean activity in the network and the weak signal. More precisely, in the range of noise intensities considered in this study, the transmission 10 of the information encoded in the weak input signal to the network activity is maximum when noise intensity reaches two certain values. The maximum or peak appearing at relatively low levels of ambient noise D1, corresponds to a transition point where the activity of the network switches from a memory phase -- in which a stored memory pattern is retrieved - to an oscillatory phase where the system alternatively is recalling the stored pattern and its anti-pattern in a given aperiodic sequence. Thus, at this level of noise and in the presence of the weak signal this oscillatory behavior of the network activity becomes correlated with the signal oscillating at its characteristic frequency. This fact provides an efficient mechanism for processing of relevant information encoded in weak stimuli in the system since at this transition point, for instance, the system could efficiently recall different sequences of patterns of information according to predefined input signals. On the other hand, the second maximum which appears at relatively high level of ambient noise, namely D2, emerges around a second order phase transition between the oscillatory phase explained above and a disordered or non-memory phase, where the system is not able to recall any information stored in the patterns. Although the resulting network activity around this maximum is highly noisy, it is strongly correlated with the weak signal, appearing a modulation of the noisy activity that follows the signal features (see the case D2 = 265 pA in figure 2). We also studied in detail, the influence that the particular level of synaptic depression could have in the appearance of the two maxima in the correlation between the network activity and the weak stimulus. By changing the recovery time constant of the active neurotransmitters, for instance, we observe that the longer the synapses take to recover the neurotransmitters (larger τrec), the lower level of ambient noise is needed to induce to reach the different maxima of the corresponding correlation function (see the figure 4). Furthermore, we have observed that short-term facilitation competing with short-term depression at the synapses induces also additional intriguing effects in the way system responds to the weak stimulus in the presence of noise. Our results reveal, that for larger values of the characteristic time constant characterizing the facilitation mechanism, namely τf ac, the maxima in the correlation C(D) between the network activity and the weak signal appear at lower levels of ambient noise than in the case that only synaptic depression is considered. In addition, we can observe that the correlation at the low noise maximum amplifies when the level of synaptic facilitation increases (see figure 5). Both phenomena can be understood taking into account that facilitation favors the retrieval of information in the attractors and their posterior destabilization, which is the origin of the oscillatory phase. Then, an increase of facilitation moves the transition point between the memory phase and the oscillatory phase towards lower values of the ambient noise D. Also, facilitation favors the recovery of the memory attractors with less error, which implies that, when the coherence with the weak signals emerges in the network activity, it is affected by less sources of noise, and therefore, it increases for large facilitation. We proof as well the robustness of the results reported here, by checking that the appearance of several maxima in the correlation C(D) between the network activity and weak subthreshold stimuli, around some phase transitions points remains for larger number neurons N or number of stored patterns P . Our study reveals that our main findings are independent of the network size (see figure 6) and therefore could probably also obtained in actual neural media. Nevertheless, when we study the correlation C(D) for different number of stored patterns, some new effects appear. In fact, the low noise maximum of C(D) tends to disappear when the number of stored patterns is increased. The reason is, that an increase in P induces the appearance of the oscillatory phase at lower values of the ambient noise D and, consequently, the development of this maximum of C(D) occurs at lower noise (see figure 7). On the other hand, the appearance of the second high noise maximum of C(D) is not significantly affected by an increase of P (see also figure 7). Another interesting result is obtained when we consider an asymmetric stored memory pattern, that is, a (cid:54)= 0.5, so that there is an excess of 1's over 0's in the stored pattern or vice versa. In figure 8, we can see 11 that by increasing the pattern asymmetry (reducing a), the low noise maximum of C(D) decreases. This is mainly due to the fact that oscillations between the high-activity and low activity states, characterizing the oscillatory phase, tend to be less visible or even disappear -- the network activity is quasi clamped in the memory attractor -- for more asymmetric stored patterns. In fact, if we consider a < 0.45, the low noise maximum of C(D) drops abruptly and the system is not able anymore to process the information encoded in the weak signal at this level of ambient noise (see figure 8). We have performed also a complete study of how our main findings are influenced by the given network topology. As a first step in this research line, we have considered the case of diluted networks. We built different types of diluted networks by erasing a fraction of links at random in a fully-connected network. In all cases that we have considered, the main results still emerge, that is the existence of several maxima in the correlation C(D) between the network activity and the weak stimulus at some precise levels of noise around non-equilibrium phase transitions. If the fraction of erased synaptic connections is larger, these maxima of C(C) appear at lower level of ambient noise. Moreover, there is a tendency that leads to the complete disappearance of the low noise maximum for strong dilution (see figure 9). These findings are due to the fact that, dilution of synaptic links diminishes the memorization and recall abilities of the network (since memory pattern features are stored in these links). The consequence is that, in more diluted networks, the memory phase appears at lower values of ambient noise, and can even disappear in absence of noise for very diluted networks. Secondly, we have consider also the case of complex networks with scale-free properties in the degree distribution and with the small-world property. In all cases there is a wide range of the relevant networks parameters, such as the exponent of the scale-free degree distribution, the mean connectivity in the network and the rewiring probability in the the case of the WS small world network, for which C(D) shows also similar maxima at given noise levels where the system efficiently process the weak stimulus. Moreover this range of parameters is consistent with those measured in actual neural systems. Note that the consideration here of some complex networks with scale-free topology, introduces node-degree heterogeneity in the system. This fact induces also neuron heterogeneity in the sense that more connected neurons can be more excitable than less connected neurons, so one can have different levels of neuron excitability in the system. However, we have seen that even in this case there is a wide range of model parameters where the relevant phenomenology here still emerges. This tell us that in a more general scenario where different types of neuron heterogeneity are considered, the phenomena reported here also will emerge. After our analysis in the present work, we conclude that the efficient detection or processing of weak subthreshold stimuli by a neural system in the presence of noise can occur at different levels of this noise intensity. This fact seems to be related to the existence of phase transitions in the system precisely at this levels of noise, a suspicion which is presently been analyzed in greater detail. In the cases studied here, within the range of noise considered, there is a maximum in the correlation C(D) of the network activity with the stimulus which seems to correspond to a discontinuous phase transition (the maximum at relatively low ambient noise) as well as another maximum appearing around a continuous phase transition (the maximum at relatively high noise). The difference in the type of the emerging phase transition determines the way the weak subthreshold stimulus is processed by the neural medium. We hope to study next the computational implications that each one of theses maxima induces and their possible relation with high level brain functions. In particular, in some preliminary simulations reported in the present work with more realistic Poissonian signals, a detailed inspection of the network activity temporal behavior, compared with the weak signal time series around the low noise maximum, show a strong resemblance with working memory tasks, where relevant information encoded in the input signal is maintained in the network activity during some time, even when the input signal has disappeared. Also in the case of several memory pattern stored in the system and around this low noise maximum, the system could process a given sequence of patterns encoded in the stimulus. Precisely at this level of noise, a non-equilibrium phase emerges characterized by continuous sequence of jumps of the network 12 activity between different memories, which could be correlated with a particular sequence of memories in the presence of an appropriate stimulus. On the other hand, at the second resonance peak is the precise timing between input spikes which are detected and processed by the network activity. Finally, we mention that it would be interesting to investigate if the relevant phenomenology reported in this work could emerge naturally in actual systems. In fact, recent data from a psycho-technical experiment in the human brain [8] can be better interpreted, using different theoretical approaches and dynamic synapses, considering the existence of several levels of noise at which relevant information can be processed [17, 27]. In figure 13 it is shown how these experimental data can be also interpreted in terms of the correlation function C(D) obtained within the more realistic model approach reported in this paper, that is, a complex network of spiking neurons. This should serve as motivation to study in depth how neural systems process weak subthreshold stimuli in a more biological and realistic scenario. For instance, one could consider conductance based neuron models instead of the simplified integrate and fire model used here or conceive more realistic stimuli, and other complex network topologies. The last could include, for instance, different type of node degree-degree correlations [28 -- 30] or network-network correlations constituting a multiplex structure as a recent work suggests to occur in the brain [31]. All these additional considerations could provide some new insights in order to design a possible experiment easily reproducible by biologists to investigate the emergence the phenomena reported here in actual neural systems. On the other hand, the relation of the relevant phenomenology reported in this study with the existence of different phase transitions in our system could be of interest for neuroscientists to investigate the existence of phase transitions in the brain. Acknowledgments We acknowledge useful comments by Matjaz Perc and Claudio Mirasso, and financial support from the Spanish MINECO project FIS2013-43201-P. References 1. Uzun R, Ozer M, Perc M (2014) Can scale-freeness offset delayed signal detection in neuronal networks? EPL 105: 60002. (document) 2. Ozer M, Uzuntarla M, Perc M, Graham LJ (2009) Spike latency and jitter of neuronal membrane patches with stochastic Hodgkin-Huxley channels. J Theor Biol 261: 83 -- 92. (document) 3. Ozer M, Perc M, Uzuntarla M, Koklukaya E (2010) Weak signal propagation through noisy feedforward neuronal networks. Neuroreport 21: 338 -- 343. (document) 4. Sancho JM, Garcia-Ojalvo J (2000) Noise-induced order in extended systems: A tutorial. In: Freund JA, Poschel T, editors, Stochastic Processes in Physics, Chemistry, and Biology, Springer Berlin Heidelberg, volume 557 of Lecture Notes in Physics. pp. 235-246. (document) 5. Yoshimoto M, Shirahama H, Kurosawa S (2008) Noise-induced order in the chaos of the Belousov- Zhabotinsky reaction. J Chem Phys 129: 014508. (document) 6. Gammaitoni L, Hanggi P, Jung P, Marchesoni F (1998) Stochastic resonance. Rev Mod Phys 70: 223 -- 287. (document) 7. Perc M (2007) Stochastic resonance on excitable small-world networks via a pacemaker. Phys Rev E 76: 066203. (document) 13 8. Yasuda H, Miyaoka T, Horiguchi J, Yasuda A, Hanggi P, et al. (2008) Novel class of neural stochastic resonance and error-free information transfer. Phys Rev Lett 100: 118103. (document), 13 9. Ozer M, Perc M, Uzuntarla M (2009) Stochastic resonance on Newman-Watts networks of Hodgkin- Huxley neurons with local periodic driving. Phys Lett A 373: 964 -- 968. (document) 10. Chakravarthy VS (2013) Do basal ganglia amplify willed action by stochastic resonance? a model. PLoS ONE 8: e75657. (document) 11. Droste F, Schwalger T, Lindner B (2013) Interplay of two signals in a neuron with heterogeneous synaptic short-term plasticity. Front Comput Neurosci 7. (document) 12. Schmidt S, Scholz M, Obermayer K, Brandt S, Stephan A (2013) Patterned brain stimulation, what a framework with rhythmic and noisy components might tell us about recovery maximization. Front Human Neurosci 7. (document) 13. Krawiecki A (2009) Structural stochastic multiresonance in the Ising model on scale-free networks. The European Physical Journal B 69: 81 -- 86. (document) 14. Palomino JM (2009) Short-term synaptic plasticity: computational implications in the emergent behavior of neural systems. Ph.D. thesis, Universidad de Granada. (document) 15. Mej´ıas J, Torres J (2011) Emergence of resonances in neural systems: the interplay between adaptive threshold and short-term synaptic plasticity. Plos One 6: e17255. (document) 16. Torres J, Marro J, Mej´ıas J (2011) Can intrinsic noise induce various resonant peaks? New J of Physics 13: 053014. (document) 17. Pinamonti G, Marro J, Torres J (2012) Stochastic resonance crossovers in complex networks. Plos One 7: e51170. (document) 18. Tsodyks M, Markram H (1997) The neural code between neocortical pyramidal neurons depends on neurotransmitter release probability. PNAS 94: 719 -- 723. (document) 19. Watts DJ, Strogatz SH (1998) Collective dynamics of 'small-world' networks. Nature 393: 440 -- 442. (document) 20. Pantic L, Torres J, Kappen H, Gielen SC (2002) Associative memory with dynamic synapses. Neural Computation 14: 2903 -- 2923. (document) 21. Tsodyks M, Pawelzik K, Markram H (1998) Neural networks with dynamic synapses. Neural Computation 10: 821 -- 835. (document) 22. Torres J, Cortes J, Marro J, Kappen H (2008) Competition between synaptic depression and facilitation in attractor neural networks. Neural Computation 19: 2739 -- 2755. (document) 23. Amit D, Gutfreund H, Sompolinsky H (1987) Information storage in neural networks with low levels of activity. Phys Rev A 35: 2293 -- 2303. (document) 24. Johnson S, Marro J, Torres J (2010) Evolving networks and the development of neural systems. J Stat Mech P03003. (document) 25. Newman MJ (2003) The structure and function of complex networks. SIAM Review 45: 167 -- 256. (document) 14 26. Sporns O, Honey CJ, Kotter R (2007) Identification and classification of hubs in brain networks. PLoS ONE 2: e1049. (document) 27. Torres JJ, Marro J, Mejias JF (2011) Can intrinsic noise induce various resonant peaks? New J Phys 13: 053014. (document) 28. Johnson S, Torres J, Marro J, Munoz M (2010) Entropic origin of disassortativity in complex networks. Phys Rev Lett 104: 108702. (document) 29. de Franciscis S, Johnson S, Torres J (2011) Enhancing neural-network performance via assortativity. Phys Rev E 83: 036114. (document) 30. Luccioli S, Ben-Jacob E, Barzilai A, Bonifazi P, Torcini A (2014) Clique of functional hubs orchestrates population bursts in developmentally regulated neural networks. PLoS Comput Biol 10: e1003823. (document) 31. Reis SDS, Hu Y, Babino A, Andrade-Jr JS, Canals S, et al. (2014) Avoiding catastrophic failure in correlated networks of networks. Nature Phys 10: 762 -- 767. (document) 15 Figure Legends Figure 1. Correlation function C(D) for a single realization of a fully connected network of N = 1600 spiking neurons with pure depressing synapses with τrec = 300 ms and one stored pattern P = 1. Other parameters were a = 0.5, κ = 2000, I0 = 100 pA, ds = 20 pA, fs = 1.5 Hz, U = 0.5, A = 45 pA, τin = 3 ms. Figure 2. Temporal behavior of the mean firing rate and the overlap function for five different values of the noise intensity D along the curve C(D) depicted in figure 1. This shows a tendency to the coherence between network activity and the weak stimulus around two values of the noisy intensity, namely D1 = 97.5 pA and D2 = 265 pA. The weak sinusoidal stimulus is also shown on the top panels. Figure 3. Raster plots of the network activity for different values of the noise intensity D for the network realization with the correlation C(D) depicted in figure 1. The stored pattern in such that for i=0,...,799 si = 1, that is neurons are active, and for i=800,...,1600 si = 0 that is, these neurons are silent in the pattern. At the first maximum of C(D) that occurs around D1 = 97.5 pA there are oscillations of the network activity around the stored pattern and its anti-pattern which are correlated with the weak stimulus. At a second critical level noise D2 = 265 pA a transition between the above oscillatory phase and a disordered phase appears, and a second maximum of C(D) emerges where a very noisy mean activity in the network also is correlated with the weak stimulus. All depicted panels corresponds to the same cases shown in figure 2. Figure 4. The behavior of the correlation C(D) for a single realization of a network of N = 2000 neurons with different levels of synaptic depression at the synapses monitored by varying τrec. The figure shows that an increase of τrec causes the maxima of C(D) to appear at lower noise intensities. Other parameters were as in figure 1. Figure 5. The correlation C(D) obtained for a single realization of the system with N = 800 neurons considering both short-term depression and short-term facilitation processes at the synapses with U = 0.02. The figure illustrates that both maxima of C(D) appear at lower noise intensities as well as that the amplitude of the first maximum grows when τf ac is increased. Other parameters were as in figure 1. Figure 6. The correlation C(D) for different number of neurons in the network. The figure shows the same maxima and shape for C(D) independently of N which confirms that the results obtained are independent from the size of the network considered. Each curve has been obtained with a single realization of the corresponding network. Other parameters were as in figure 1. 16 Figure 7. Correlation function C(D) for different number of stored patterns P in a network with N = 2000 neurons. This shows that as P increases the low-noise maximum of the correlation C(D) increases in intensity and appears at lower level of noise while the second correlation maximum remains unchanged. Each curve has been obtained with a single realization of the corresponding network. Other parameters values were as in figure 1. Figure 8. Changes in the shape of the correlation C(D) for a network with N = 800 neurons considering different levels of asymmetry in the network activity encoded in the stored pattern as measured with the parameter a (see main text for the explanation). Panel A depicts correlation curves for a ≤ 0.45, showing that the correlation C(D) drops around D = 100 pA and another maximum starts to emerge at D = 65 pA. Panel B illustrates C(D) curves for a ≥ 0.45, showing that the low noise maximum at D = 100 pA looses intensity as a decreases. Each curve has been obtained with a single realization of the corresponding network. Other parameters were as in figure 1. Figure 9. Behavior of the correlation C(D) for a diluted network. (A) As the fraction δ of removed connections increases both maxima of C(D) appear at lower noise intensity. The network size was set to N = 800 and a factor N instead of (cid:104)k(cid:105) is considered in the definition (4). Each curve has been obtained with a single realization of the corresponding network. Other parameters were as in figure 1. (B) In this case simulations has been performed considering diluted networks using the same procedure than in panel A, but with less level of synaptic depression, τrec = 200 ms, and synaptic weights normalized with a factor (cid:104)k(cid:105) = (1 − δ)2N. Also in this case, the corresponding C(D) curves has been obtained for N = 200 and averaging over 10 different networks. Figure 10. Detection of weak subthreshold stimulus as a function of noise D in spiking networks with complex scale-free topologies. (A) Correlation curves C(D) obtained for different values of the exponent γ of the degree connectivity distribution p(k) ∼ k−γ in a network with (cid:104)k(cid:105) = 20. The figure shows that around γ = 3 multiple maxima in C(D) start to emerge. (B) The same correlation curves C(D) for scale free networks with p(k) ∼ k−3 and different values of the mean connectivity degree (cid:104)k(cid:105). Again for more realistic large values of (cid:104)k(cid:105) around 20 multiple maxima in C(D) start to appear. All curves have been obtained for τrec = 200 ms in a network with N = 200 neurons and after averaging over 10 different network realizations. Other parameters as in figure 1. Figure 11. Detection of weak subthreshold stimulus as a function of noise D in a WS small-world network a function of the rewiring probability pr. The network has N = 200 neurons each one with k0 = (cid:104)k(cid:105) = 20 neighbors and depressing synapses with τrec = 100 ms. Different correlation functions C(D) have been obtained after averaging over 10 different network realizations. Other parameters as in figure 1. 17 i Figure 12. Detection of more realistic Poissonian weak stimuli as a function of the noise noise D. (A) Corresponding correlation C(D) for a network of N = 400 neurons depicting several maxima similarly to the case of sinusoidal stimuli. The curve has been obtained after averaging over 20 network realizations. The weak signal current I ext is the resulting of an inhomogeneous Poissonian train of spikes of amplitude ds = 10 pA injected in each neuron (see panel B top graph) at frequency λ(t) = λ0[1 + asin(2πfst)] with λ0 = 70 Hz and a = 0.75. (B) Time-series of the network mean firing rate for increasing values of the noise intensity D along the correlation curve depicted in panel A. (C) Normalized histograms of the network mean-firing rate at the high noise maximum of C(D) (red curve) corresponding to the critical point Dc = 261 pA and for a higher level of noise, namely D = 500 pA, above this critical point (green curve). The colored areas corresponds to the best Gaussian fitting of both histograms. Other parameters were as in figure 1. Figure 13. Fitting experimental data in the auditory cortex to our model. The experimental data (symbols with the corresponding error bars) reported in [8] are compared as a function of noise with the correlation function C(D) described in figure 1 corresponding to a single realization of a spiking network of N = 2000 integrate and fire neurons (red solid line) and dynamic synapses. The experimental data C (in arbitrary units) has been multiplied by a factor 104, and the noise amplitude M (in dB) has been transformed in our noise parameter D using the nonlinear relationship D = D0 + M η 2 σM = 140 dB. (cid:3)(cid:9) with D0 = 0.1 pA, η = 0.71 (pA/dB), M0 = 50dB and 2σM (cid:8)1 + erf(cid:2)(M − M0)/ √ 18 Figure 1 19 Figure 2 20 Figure 3 21 Figure 4 22 Figure 5 23 Figure 6 24 Figure 7 25 Figure 8 26 Figure 9 27 Figure 10 28 Figure 11 29 Figure 12 30 Figure 13 31
1210.4017	4	1210	2015-01-16T01:57:51	Single molecule thermodynamics of ATP synthesis by F$_1$-ATPase	[ "physics.bio-ph", "cond-mat.stat-mech", "q-bio.SC" ]	F$_\mathrm{o}$F$_1$-ATP synthase is a factory for synthesizing ATP in virtually all cells. Its core machinery is the subcomplex F$_1$-motor (F$_1$-ATPase) and performs the reversible mechanochemical coupling. Isolated F$_1$-motor hydrolyzes ATP, which is accompanied by unidirectional rotation of its central $\gamma$-shaft. When a strong opposing torque is imposed, the $\gamma$-shaft rotates in the opposite direction and drives the F$_1$-motor to synthesize ATP. This mechanical-to-chemical free-energy transduction is the final and central step of the multistep cellular ATP-synthetic pathway. Here, we determined the amount of mechanical work exploited by the F$_1$-motor to synthesize an ATP molecule during forced rotations using methodology combining a nonequilibrium theory and single molecule measurements of responses to external torque. We found that the internal dissipation of the motor is negligible even during rotations far from a quasistatic process.	physics.bio-ph	physics	Single molecule thermodynamics of ATP synthesis by F1-ATPase Faculty of Science and Engineering, Chuo University, Tokyo 112-8551, Japan and Graduate School of Engineering, Tohoku University, Sendai 980-8579, Japan Shoichi Toyabe Faculty of Science and Engineering, Chuo University, Tokyo 112-8551, Japan Eiro Muneyuki∗ FoF1-ATP synthase is a factory for synthesizing ATP in virtually all cells. Its core machinery is the subcomplex F1-motor (F1-ATPase) and performs the reversible mechanochemical coupling. Isolated F1-motor hydrolyzes ATP, which is accompanied by unidirectional rotation of its central γ-shaft. When a strong opposing torque is imposed, the γ-shaft rotates in the opposite direction and drives the F1-motor to synthesize ATP. This mechanical-to-chemical free-energy transduction is the final and central step of the multistep cellular ATP-synthetic pathway. Here, we determined the amount of mechanical work exploited by the F1-motor to synthesize an ATP molecule during forced rotations using methodology combining a nonequilibrium theory and single molecule measurements of responses to external torque. We found that the internal dissipation of the motor is negligible even during rotations far from a quasistatic process. FoF1-ATP synthase comprises an Fo-motor embedded in a membrane (inner membrane of the mitochondria in the eukaryotic cells) and an F1-motor protruding from the membrane (Fig. 1a). Proton translocation through the Fo-motor driven by the transmembrane electrochemical potential unidirectionally rotates the c-ring of the Fo-motor. Because the c-ring is connected to the γ-shaft of F1-motor, the c-ring imposes torque on the γ-shaft causing it to rotate. The forced rotation of the γ-shaft induces the F1-motor's stator α3β3 ring to synthesize ATP from ADP and Pi (inorganic phosphate). The F1-motor converts the mechanical work transferred from the c-ring to the chemical free energy of ATP synthesis from ADP and Pi, ∆µ [1, 2]. Approximately 95% of cellular ATP is synthesized by this rotary mechanochemical transduction. In contrast, the isolated F1-motor hydrolyzes ATP to ADP and phosphate and rotates the γ-shaft by converting ∆µ to mechanical motion [3, 4]. However, the rotational direction is opposite to that of the ATP-synthetic rotation. The γ-shaft rotates 120◦ per ATP hydrolysis [2, 4]. Thus, the F1-motor reversibly transduces free-energy between mechanical work and ∆µ [3 -- 6]. The detailed kinetics and reaction scheme of the F1-motor's rotations have been studied intensively. In contrast, knowledge regarding the energetics is limited [4, 7 -- 11] due to the difficulty in studying the energetics of such nanosized engines working at an energy scale comparable to thermal energy. An experimental methodology was recently estab- lished by combining measurements of single molecule responses with nonequilibrium theory [8 -- 11]. Previous research revealed the high efficiency of the ATP-hydrolytic rotations by the isolated F1-motor as follows: (i) The maximum work performed by the F1-motor during a 120◦ rotation is equal to ∆µ within experimental error [9]. Specifically, the F1-motor achieves a unity thermodynamic efficiency at the stalled state where the mean rotational rate vanishes because of a hindering external torque. (ii) The F1-motor converts ∆µ to mechanical work with negligible internal ∗Electronic address: [email protected] 2 heat dissipation even during rotations far from quasistatic process [8]. (iii) The motor minimizes the internal heat dissipation by switching its chemical state depending on the angular position of the γ-shaft [10]. However, the primary physiological role played by the F1-motor is not hydrolysis but mechanochemical ATP synthesis driven by the forced-rotations of the γ-shaft. This raises the fundamental unanswered question regarding the free-energy transduction as follows: How much work is required to drive the F1-motor to synthesize an ATP molecule? To address this question, we improved available methodology [8] and investigated the energetics of the forced ATP-synthetic rotations by the isolated F1-motor. Experiment An isolated F1-motor molecule was adhered to an upper glass surface (Fig. 1b) [3, 4]. A submicron particle was attached to the γ-shaft with an elastic protein linker to probe the γ-shaft rotations by an optical microscope. We imposed a torque on the probe using an electrorotation method [8, 9, 12 -- 14]. The electrorotation method uses a high-frequency alternating-current electric field generated by quadrupolar electrodes and imposes a torque with a controlled magnitude (see Methods). Thus, we can measure the rotational response of a single motor molecule as a function of external torque. This system mimics the mechanochemical transduction inside the FoF1-ATP synthase; the upper glass surface, the probe, and the torque imposed correspond to the stator, the c-ring, and the proton-motive force that drives the c-ring, respectively. Consider the energetics of this system as follows (Fig. 1c): Under sufficiently strong torque in the ATP-synthetic direction, the probe-shaft complex rotates in that direction [9]. The amount of the external torque multiplied by 120◦ with a sign depending on the rotational direction is the work, Wext, performed on the probe per ATP cycle. A portion of Wext, W , is transferred to the motor by rotating the γ-shaft with the elastic linker and the remainder, Qprobe, dissipates through the viscous friction of the probe. W = Wext − Qprobe. (1) The motor receives W , converts it to the chemical free energy change ∆µ by synthesizing an ATP molecule at its α3β3 ring. The remainder, Qmotor ≡ W − ∆µ, dissipates irreversibly from the motor. The motor molecule itself is not accessible by experiments. Nevertheless, the heat dissipation from the nanosized motor Qmotor can be estimated from the thermodynamic quantities regarding the probe Wext and Qprobe using the energy balance (1). Wext is easily calculated as noted, whereas it is usually difficult to measure heat Qprobe in a microscopic system subjected to thermal fluctuations. This is because, although heat can be defined as the microscopic energy exchange with the solvent [15], the thermal fluctuating force is inaccessible using experimental manipulations. However, a nonequilibrium equality derived by Harada and Sasa [16 -- 18] enables us to calculate Qprobe from quantities that can be derived experimentally: the fluctuation and the response to a small external torque perturbation of the rotational rate (see Materials and Methods for details). Fluctuation and response are related by the fluctuation 3 response relation (FRR) around the equilibrium state [19]. The FRR is generally violated far from equilibrium. Therefore, the extent of the FRR violation indicates how far the system is driven to nonequilibrium. The Harada and Sasa's relation relates the extent of the FRR violation to the heat dissipation at a nonequilibrium steady state. This relation is valid in systems described by Langevin equations such as a colloidal particle in an aqueous solution. In the following experiment, we evaluated Qprobe by Harada-Sasa relation, obtained W from [1], and compared it with ∆µ to evaluate the energetics of mechanical-to-chemical free energy transduction. We also discuss the energetics of the chemical-to-mechanical transduction in the ATP-hydrolytic rotations. Results Single-molecule response measurement and violation of the fluctuation-response relation. The γ-shaft's rotations were assessed using a large dimeric probe (diameter = 300 nm) at a relatively high ATP concentration (10 µM ATP, 10 µM ADP, 1 mM Pi). Under this condition, the probe rotated smoothly without clear steps in the absence of external torque (Fig. 2a). When we applied external torque to the probe using the electrorotation method in the direction opposite to the rotations, the rotational rate decreased (Fig. 2a and b). At torques greater than ∆µ/120◦, the rotation was reversed, and the γ-shaft rotated in the ATP-synthetic direction. This is consistent with a previous result showing that the maximum work that the F1-motor can exert per ATP cycle is similar to ∆µ [9]. Specifically, the thermodynamic efficiency of the F1-motor is nearly 1 at the stalled state where the mean rotational rate vanishes. The FRR is violated at low frequencies (Fig. 2c) where the motor drives the probe to a nonequilibrium state [8], whereas the FRR is held at high frequencies where the probe's motion is expected to behave as a freely rotating Brownian particle at equilibrium. The extent of the FRR violation, or twice that of the shaded area in Fig. 2c, corresponds to the integral in Harada-Sasa relation (see Materials and Methods). When we increased the magnitude of the load from zero, the extent of the FRR violation decreased in the ATP-hydrolytic rotations, nearly vanished at the stalled state, and increased again in the ATP-synthetic rotations. Energetics of mechanical-to-chemical free energy transduction. In Fig. 3a, Qprobe evaluated by Harada-Sasa equality and Wext (= ±(torque) × 120◦) are shown. Qprobe linearly decreased as we increased the load and vanished around the stalled state. This indicates that, at the stalled state, the probe behaved like a rotational Brownian motion at equilibrium in a 120◦-spacing periodic potential. In the ATP-synthetic rotations, the value of Qprobe increased as the torque increased further, mainly because the rotational rate increased with torque. In Fig. 3c, Qprobe is shown separately; the contribution from the steady motion Qs and that from nonequilibrium fluctuations around the steady motion Qv, which correspond to the first and second terms of the rhs of the Harada and Sasa's relation (2), respectively. Qs dominated Qv in the present experimental condition. An interesting observation is that Qv as well as Qs increases linearly as the torque increases. The fraction of Qs and Qv are similar independent of the torque magnitude except around the stalling state. The fraction of Qs seems higher around the stalling state. 4 However, since Qs and Qv are small at this region, more precise measurement is necessary to test this dependency. The difference W = Wext − Qprobe is the work received by the motor (Fig. 4a). In the ATP-synthetic rotations, we found that W = ∆µ over a broad range of torque, where ∆µ is the thermodynamic limit for the work necessary to synthesize an ATP molecule. This implies that the motor's internal dissipation Qmotor = W − ∆µ is negligible and the F1-motor can convert most of W to ∆µ even during rotations at a high rotational rate such as 50Hz. Note that the rate of the proton-driven rotations of the thermophilic FoF1-ATP synthase is 3-4 Hz at saturated ADP and Pi concentrations at a room temperature [20]. When the motor rotates in the ATP-hydrolytic direction under weak external torque, −W corresponds to the work performed by the motor transferred to the probe. A part of −W is used to increase the potential of the probe against load −Wext, and the remainder, Qprobe, dissipates. In Fig. 4a, we show that W is similar to −∆µ in good agreement with a previous result [8] that the motor can consume ∆µ in the rotational degree of freedom with negligible internal irreversible heat production even during rotations. However, we observed a small deviation of W from −∆µ at a small torque magnitude (Fig. 4a inset), which was not observed in the previous experiment. W increased with torque and reached ∆µ around the stalled state. We also measured W of a mutant F1-motor molecule with mutations around the nucleotide binding site (βT165S and βY341W) and the β-subunit's hinge region (βG181A) [7] (Fig. 3b and 4b). This mutant produced a smaller maximum work than the wild-type, supposedly caused by weak binding of ATP [9]. Figure 4b shows that W is significantly less than ∆µ; the mutant produces a finite amount of internal dissipation. Discussion Here we evaluated the energetics of the free-energy conversion by the F1-motor during rotations far from a qua- sistatic process in mechanical-to-chemical and chemical-to-mechanical energy transduction. This was achieved by combining a single molecule response measurement mimicking the mechanical coupling of FoF1-ATP synthase with a nonequilibrium theory. Our main finding is that, in the ATP-synthetic rotations, the internal dissipation from the motor is negligible even during rotations far from a quasistatic process. In the ATP-hydrolytic rotations, we observed a finite amount of the internal dissipation from the motor, which was not observed in the previous experiment. In case the probe and the γ-shaft are thermally insulated in that the heat transfer between them is negligible, ∆µ/W ≤ 1 in the ATP synthesis and W/∆µ ≤ 1 in the ATP hydrolysis because ∆µ is the thermodynamic limit for the work necessary to synthesize an ATP molecule and the thermodynamic limit for the work extractable from an ATP hydrolysis. If this is the case, the present results, ∆µ ≃ W in both the ATP synthesis and ATP hydrolysis, imply that the F1-motor is a highly efficient motor working at almost 100% efficiency even within finite-time. The thermal insulation between the probe and the shaft is supposedly valid when the linker connecting the probe and the shaft is sufficiently flexible and ∆µ and ATP concentration are not extremely small[21, 22] although detailed simulations with 5 explicit separation of the degrees of freedom of the motor and probe are necessary to deduce the exact conditions. Otherwise, the probe and the γ-shaft move in tandem to some extent and the heat dissipated from the system cannot be separated well into Qmotor and Qprobe; for example, a part of the energy absorbed by the motor to overcome an energy barrier for a chemical reaction may dissipate through the probe's motion as a part of Qprobe. W evaluated by (1) is no longer work that is transferred to the motor, and ∆µ/W can be greater than 1 [21 -- 23]. The spring constant of the linker is measured in ref[24] although their probe is different from this work. We expect that future detailed simulation will reveal the implications of the intriguing balance ∆µ ≃ W obtained in this work. In the present study, we assumed tight coupling between ATP synthesis/hydrolysis and the 120◦ rotation and compared W with ∆µ to assess the efficiency. Although, the 120◦ rotational step and the ATP hydrolysis are tightly coupled [2, 4], this has yet to be firmly established for ATP synthesis. The results of a pioneering experiment on the forced rotations of the γ shaft that were induced using magnetic tweezers imply that less than one ATP molecule is synthesized in each 120◦ synthetic-direction rotation [2]. In that study, the tweezers were rotated at a high rate of 10 Hz. Magnetic tweezers impose a trapping force, in contrast to the constant torque imposed by the electrorotation method. At a high manipulation rate the angular positions of the probe and tweezers can be apart. Then, magnetic tweezers can impose a very strong torque, which may induce slippage of the mechanochemical coupling. The results of a more recent experiment using magnetic tweezers manipulated at a low rate of 0.05-0.8 Hz suggests that attachment and detachment of nucleotides and the angular position of the γ shaft are tightly coupled [25]. Further, the previous result showing that the maximum work of the F1-motor is similar to ∆µ over a broad range of ∆µ, suggests tight coupling [9]. The mutant's finite internal dissipation implies that it fails to completely couple mechanical rotation and ATP hydrolysis and synthesis. Some externally-forced mechanical steps may not accompany ATP-synthetic reactions, and some ATP hydrolytic reactions may not accompany mechanical steps. Macroscopic engines operated at a finite rate inevitably generates turbulence, and additional energy dissipates through microscopic degrees of freedom as irreversible heat. In contrast, the F1-motor is itself microscopic and possibly utilizes thermal fluctuations. Therefore, some of the microscopic degrees of freedom might not be hidden and are accessible to the F1-motor. In the ATP-hydrolytic rotations, previous studies suggest that the F1-motor shifts the mechanical potentials discontinuously depending on the γ-shaft's angular position [10, 26]. Only by nanosized machines can perform such an operation to minimize irreversible heat within a finite-time. This highlights the remarkable property of nanosized engines. Further studies will be required to elucidate the molecular mechanism of the highly efficient mechanochemical couplings of ATP-synthetic rotations. Materials and methods Single molecule response measurement. The experimental setup is essentially the same as that in the previous studies [8 -- 10, 14]. F1 molecules derived from a thermophilic Bacillus PS3 with mutations for the rotation assay 6 (His6-αC193S/W463F, His10-β, γS107C/I210C, denoted as wild type) [2] or a mutant with βT165S, βY341W, and βG181A[7] were adhered on a cover slip functionalized by Ni2+-NTA. Rotations of the γ shaft were probed by streptavidin-coated dimeric polystyrene particles (diameter = 300 nm, Seradyn) attached to the biotinylated γ shaft in a buffer containing 5 mM MOPS/KOH, 10 µM MgATP, 10 µM MgADP, 1 mM Pi, and 1 mM MgCl2 (pH 6.9). Observation was performed on a phase-contrast upright microscope (Olympus) with a 100× objective and a high-speed camera (Basler) at 2,000 Hz. The data including a long pause presumably due to the MgADP-inhibited state are excluded from the analysis. We applied torque on the probe by using rotating electric field at 15 MHz generated with the quadrupolar electrodes patterned on the glass surface of the chamber [8, 9, 12, 13, 27]. The torque magnitude was controlled by controlling the electrodes voltage. In Fig. 3 and 4, 69 points of 19 wild-type molecules and 20 points of 9 mutant molecules are shown. Fluctuation, response, and Harada-Sasa equality. Let C(t) = h[v(t) − vs] [v(0) − vs]i and R(t) be the fluctuation and the response function against small external torque, respectively, of the probe's rotational rate v(t) around the mean rotational rate vs. R(t) is defined as hv(t) − vsiN = R ∞ −∞ dsR(t − s)N (s), where h·iN is the ensemble average under a sufficiently small probe torque N (t). Because of the causality, R(t) = 0 if t < 0. Around an equilibrium state, C(t) and R(t) are related by the fluctuation response relation (FRR): C(t) = kBT R(t) [19]. The FRR is generally violated far from equilibrium. The equality by Harada and Sasa relates the extent of the FRR violation to heat dissipation at a nonequilibrium steady state [16 -- 18]. In the frequency space, the equality is expressed as Qprobe = Γ 3vs (cid:20)v2 s +Z ∞ df [ C(f ) − 2kBT R′(f )](cid:21) ≡ Qs + Qv, (2) −∞ where C(f ) and R(f ) are the Fourier transforms of C(t) and R(t), respectively, at frequency f , and R′(f ) is the real part of R(f ). Qs ≡ Γvs/3 and Qv ≡ Γ/(3vs)R ∞ −∞ df [ C(f ) − 2kBT R′(f )] correspond to the dissipation through steady rotations and that through nonequilibrium fluctuations in a 120◦ rotation. This relation is valid in the systems described by Markovian Langevin equations. The FRR becomes C(f ) = 2kBT R′(f ) in the frequency space, Γ is the rotational frictional coefficient, and 3vs is the mean stepping rate that corresponds to the rate of ATP hydrolysis or synthesis. C(f ) was calculated from the rotational trajectories by a fast Fourier transform method and Wiener-Khintchine theorem. Γ was obtained by taking the average of C(f ) around 300 Hz since the FRR, limf →∞ C(f ) = 2kBT /Γ, is supposed to hold in such a high frequency region [8, 9]. Γ was 0.075±0.012 kBT s/rad2 (mean±SD, N=70). For evaluating R(f ), we added a small torque N (t) = N0Pi sin(2πfit) , where fi = 1, 4, 10, 20, 40, 80, 160, 250, and 400 Hz, in addition to the constant load. N0 is unknown a priori. We measured hv(t)iN, performed a Fourier transform, and obtained R(fi) N0 at multiple frequencies fi. Then, we obtained N0 by comparing C(f ) and R(f )N0 around at high frequency regions because of the FRR [8, 9]. The frequency regions for calibration were determined by eye (around 300Hz typically). Finally, we obtained R(fi). N0 was 0.94±0.19 kBT/rad (mean±SD, N=70). We calculated the integration in (2) between -400 Hz and 400 Hz. The contribution from the outside of this region is supposed to be negligible because of the FRR at high frequencies. We omitted data without apparent convergence of C(f ) and R′(f ) at the high frequency region (8 out of 78). When torque multiplied by 120◦ was greater than 80 kBT , the FRR was violated at high frequencies even around 300 Hz, where the torque magnitude was calibrated according to the FRR. 7 Acknowledgments We thank valuable discussions with Kyogo Kawaguchi, Takahiro Sagawa, Masaki Sano, Shin-ichi Sasa, and Hiroshi Ueno. This work is supported by a Grant-in-Aid for Scientific Research on Innovative Areas "Fluctuation & Structure" (No. 26103528) from the Ministry of Education, Culture, Sports, Science, and Technology of Japan. [1] Itoh H, Takahashi A, Adachi K, Noji H, Yasuda R, Yoshida M and Kinosita Jr K 2004 Nature 427 465 -- 468 [2] Rondelez Y, Tresset G, Nakashima T, Kato-Yamada Y, Fujita H, Takeuchi S and Noji H 2005 Nature 433 773 -- 777 [3] Noji H, Yasuda R, Yoshida M and Kinosita Jr K 1997 Nature 386 299 -- 302 [4] Yasuda R, Noji H, Kinosita Jr K and Yoshida M 1998 Cell 93 1117 -- 1124 [5] Abrahams J P, Leslie A G W, Lutter R and Walker J E 1994 Nature 370 621 -- 628 [6] Boyer P D 1993 Biochim. Biophys. Acta 1140 215 -- 250 [7] Muneyuki E, Watanabe-Nakayama T, Suzuki T, Yoshida M, Nishizaka T and Noji H 2007 Biophys. J. 92 1806 [8] Toyabe S, Okamoto T, Watanabe-Nakayama T, Taketani H, Kudo S and Muneyuki E 2010 Phys. Rev. Lett. 104 198103 [9] Toyabe S, Watanabe-Nakayama T, Okamoto T, Kudo S and Muneyuki E 2011 Proc. Nat. Acad. Sci. USA 108 17951 -- 17956 [10] Toyabe S, Ueno H and Muneyuki E 2012 EPL 97 40004 [11] Toyabe S and Muneyuki E 2013 Biophys. 9 91 -- 98 [12] Berry R M, Turner L and Berg H C 1995 Biophys. J. 69 280 -- 286 [13] Washizu M, Kurahashi Y, Iochi H, Kurosawa O, Aizawa S, Kudo S, Magariyama Y and Hotani H 1991 IEEE Trans. Ind. Appl. 29 286 [14] Watanabe-Nakayama T, Toyabe S, Kudo S, Sugiyama S, Yoshida M and Muneyuki E 2008 Biochem. Biophys. Res. Comm. 366 951 [15] Sekimoto K 2010 Stochastic Energetics (Lecture Notes in Physics) (Berlin: Springer) [16] Harada T and Sasa S I 2005 Phys. Rev. Lett. 95 130602 [17] Harada T and Sasa S I 2006 Phys. Rev. E. 73 026131 [18] Toyabe S, Jiang H R, Nakamura T, Murayama Y and Sano M 2007 Phys. Rev. E 75 011122 [19] Kubo R, Toda M and Hashitsume N 1991 Statsitical Physics II 2nd ed (Berlin: Springer) [20] Soga N, Kinosita Jr K, Yoshida M, and Suzuki T 2011 FEBS J. 278 2647 -- 2654 [21] Zimmermann E and Seifert U 2012 New J. Phys. 14 103023 [22] Kawaguchi K, Sasa S I and Sagawa T 2014 Biophys J. 106 2450 -- 2457 [23] Wang H and Oster G 2002 Europhys. Lett. 57 134 -- 140 [24] Okuno D, Iino R and Noji H 2010 Eur. Biophys. J. 39 1589 -- 1596 [25] Adachi K, Oiwa K, Yoshida M, Nishizaka T and Kinosita Jr K 2012 Nat. Comm. 3 1022 [26] Watanabe R, Okuno D, Sakakihara S, Shimabukuro K, Iino R, Yoshida M and Noji H 2011 Nat. Chem. Biol. 8 86 -- 92 [27] Toyabe S, Sagawa T, Ueda M, Muneyuki E and Sano M 2010 Nature Phys. 6 988 [28] Guynn R W and Veech R L 1973 J. Biol. Chem. 248 6966 [29] Krab K and van Wezel J 1992 Biochim. Biophys. Acta 1098 172 [30] Panke O and Rumberg B 1997 Biochim. Biophys. Acta 1322 183 [31] Rosing J and Slater E C 1972 Biochim. Biophys. Acta 267 275 8 a ATP ADP+Pi (cid:1038) (cid:1039) b2 (cid:77)(cid:74)(cid:81)(cid:74)(cid:69)(cid:1)(cid:67)(cid:74)(cid:77)(cid:66)(cid:90)(cid:70)(cid:83) a c-ring F1 Fo + H c ATP ∆µ ADP+Pi b (cid:74)(cid:84)(cid:80)(cid:77)(cid:66)(cid:85)(cid:70)(cid:69)(cid:1)(cid:39)(cid:18)(cid:14)(cid:78)(cid:80)(cid:85)(cid:80)(cid:83) (cid:34)(cid:53)(cid:49)(cid:14)(cid:84)(cid:90)(cid:79)(cid:85)(cid:73)(cid:70)(cid:85)(cid:74)(cid:68)(cid:1) (cid:83)(cid:80)(cid:85)(cid:66)(cid:85)(cid:74)(cid:80)(cid:79) (cid:81)(cid:83)(cid:80)(cid:67)(cid:70) external torque (cid:82)(cid:86)(cid:66)(cid:69)(cid:83)(cid:86)(cid:81)(cid:80)(cid:77)(cid:66)(cid:83)(cid:1)(cid:70)(cid:77)(cid:70)(cid:68)(cid:85)(cid:83)(cid:80)(cid:69)(cid:70)(cid:84) (cid:9493)(cid:78)(cid:80)(cid:85)(cid:80)(cid:83)(cid:1)(cid:9473)(cid:9444)(cid:9499)(cid:9444)(cid:9457)(cid:9444)∆µ (cid:9499)(cid:9444)(cid:9473)(cid:9444)(cid:9499)(cid:70)(cid:89)(cid:85)(cid:9444)(cid:9457)(cid:9444)(cid:9493)(cid:81)(cid:83)(cid:80)(cid:67)(cid:70) (cid:9493)(cid:81)(cid:83)(cid:80)(cid:67)(cid:70) (cid:9499)(cid:70)(cid:89)(cid:85)(cid:9444)(cid:9473)(cid:9444)(cid:694)(cid:9)(cid:85)(cid:80)(cid:83)(cid:82)(cid:86)(cid:70)(cid:10)(cid:695)(cid:18)(cid:19)(cid:17)(cid:707) FIG. 1: The structure of FoF1-ATP synthase and a model for catalysis. a, FoF1-ATP synthase. Proton flux through the Fo- motor rotates the c-ring and the γ-shaft of the F1-motor and drives ATP synthesis. b, Single molecule response measurement of the isolated α3β3γ subcomplex of the F1-motor as the model experiment system of the FoF1-ATP synthase. The F1-motor was fixed on the upper cover slip. A dimeric polystyrene particle (diameter = 300 nm) was attached to the γ-shaft with an elastic protein linker (streptavidin). Torque was imposed on the particle by an electrorotation method using quadrupolar electrodes patterned on the lower glass slide. A strong hindering torque rotates the γ-shaft in the opposite direction and drives the F1-motor to synthesize ATP. c, Energy flow through the F1-motor. Torque (cid:695) 120(cid:707) 0 kBT 7.36 c (cid:73)(cid:90)(cid:69)(cid:83)(cid:80)(cid:77)(cid:90)(cid:85)(cid:74)(cid:68)(cid:1)(cid:83)(cid:80)(cid:85)(cid:66)(cid:85)(cid:74)(cid:80)(cid:79)(cid:84)(cid:1)(cid:9)(cid:9530)(cid:84)(cid:1)(cid:30)(cid:1)(cid:24)(cid:15)(cid:19)(cid:1)(cid:41)(cid:91)(cid:10) 9 ) s / 2 d a r ( e s n o p s e R , n o i t a u t c u F l 40 20 0 40 20 0 40 20 0 (cid:113)(cid:86)(cid:68)(cid:85)(cid:86)(cid:66)(cid:85)(cid:74)(cid:80)(cid:79) (cid:83)(cid:70)(cid:84)(cid:81)(cid:80)(cid:79)(cid:84)(cid:70) 10 1 100 1000 (cid:79)(cid:70)(cid:66)(cid:83)(cid:1)(cid:84)(cid:85)(cid:66)(cid:77)(cid:77)(cid:1)(cid:9)(cid:17)(cid:15)(cid:25)(cid:19)(cid:1)(cid:41)(cid:91)(cid:10) 1 10 100 1000 (cid:84)(cid:90)(cid:79)(cid:85)(cid:73)(cid:70)(cid:85)(cid:74)(cid:68)(cid:1)(cid:83)(cid:80)(cid:85)(cid:66)(cid:85)(cid:74)(cid:80)(cid:79)(cid:84)(cid:1)(cid:9)(cid:14)(cid:21)(cid:15)(cid:21)(cid:1)(cid:41)(cid:91)(cid:10) 1 10 Frequency (Hz) 100 1000 14.3 24.0 64.2 1 Time (s) 36.7 2 ∆µ (cid:88)(cid:74)(cid:77)(cid:69) 20 0 (cid:78)(cid:86)(cid:85)(cid:66)(cid:79)(cid:85) ∆µ 12 8 4 0 0 10 20 a s n o i t t a o R 20 10 0 -10 -20 -30 0 b ) z H t ( e a r l a n o i t t a o R -20 -40 -60 0 20 40 60 80 (cid:53)(cid:80)(cid:83)(cid:82)(cid:86)(cid:70)(cid:1)(cid:695)(cid:1)(cid:18)(cid:19)(cid:17)(cid:707)(cid:9)(cid:76)(cid:35)(cid:53)(cid:10) FIG. 2: a, Rotational trajectory of the wild-type F1-motor under an external load. b, Rotational rate of the wild-type (circle) and mutant containing βG158A, T165S and Y341W[7] (square) in the presence of torque. The thick vertical line indicates ∆µ = ∆µ◦ + kBT ln [ATP]/[ADP][Pi] (17.5-18.9 kBT ). The value of ∆µ◦ was calculated according to values adopted from the literature [8, 28 -- 31]. The width of the line indicates the variation of ∆µ. The inset is the magnified view around the stalled state. The dashed lines were fitted by eye as guides. c, Examples of fluctuations, C(f ), and responses, 2kBT R′(f ). The torques multiplied by 120◦ are 4.15, 14.0, and 26.3 kBT , respectively, from top to bottom. The value of twice the shaded area corresponds to the integral in (2). a (cid:88)(cid:74)(cid:77)(cid:69) (cid:73)(cid:90)(cid:69)(cid:83)(cid:80)(cid:77)(cid:90)(cid:84)(cid:74)(cid:84) (cid:84)(cid:90)(cid:79)(cid:85)(cid:73)(cid:70)(cid:84)(cid:74)(cid:84) b (cid:78)(cid:86)(cid:85)(cid:66)(cid:79)(cid:85) 80 60 40 20 0 (cid:10) (cid:53) (cid:35) (cid:76) (cid:9) (cid:1) (cid:1) (cid:1) (cid:70) (cid:67) (cid:80) (cid:83) (cid:81) (cid:9493) (cid:1) (cid:1) (cid:1) (cid:13) (cid:1) (cid:85) (cid:89) (cid:70) (cid:9499) (cid:1) ∆µ (cid:9499)(cid:70)(cid:89)(cid:85) (cid:9493)(cid:81)(cid:83)(cid:80)(cid:67)(cid:70) 80 60 40 20 0 -20 0 20 40 ∆µ c 80 60 40 20 0 -20 80 60 0 (cid:53)(cid:80)(cid:83)(cid:82)(cid:86)(cid:70)(cid:1)(cid:695)(cid:1)(cid:18)(cid:19)(cid:17)(cid:707)(cid:9)(cid:76)(cid:35)(cid:53)(cid:10) d n o i t c a r f 1 0.8 0.6 0.4 0.2 0 (cid:9493)(cid:84) (cid:9493)(cid:87) (cid:10) (cid:53) (cid:35) (cid:76) (cid:9) (cid:1) (cid:1) (cid:1) (cid:87) (cid:9493) (cid:1) (cid:1) (cid:1) (cid:13) (cid:1) (cid:84) (cid:9493) (cid:1) -20 0 20 40 60 0 20 80 (cid:53)(cid:80)(cid:83)(cid:82)(cid:86)(cid:70)(cid:1)(cid:695)(cid:1)(cid:18)(cid:19)(cid:17)(cid:707)(cid:9)(cid:76)(cid:35)(cid:53)(cid:10) 10 ∆µ (cid:9499)(cid:70)(cid:89)(cid:85) (cid:9493)(cid:81)(cid:83)(cid:80)(cid:67)(cid:70) 20 40 (cid:9493)(cid:84) (cid:9493)(cid:87) 40 60 80 FIG. 3: Heat dissipation through the probe. a, b, Amount of heat dissipation, Qprobe (circle and square) and the potential increase against an external load, Wext (solid line), of wild-type (a) and mutant (b) F1-motors. Wext is calculated as the torque multiplied by 120◦ with the sign depending on the rotational direction, that is, positive in the ATP synthetic rotations and negative in the ATP hydrolytic rotations. 69 points of 19 wild-type molecules and 20 points of 9 mutant molecules are shown. We excluded one aberrant point due to the vanishingly small rotational rate from the graphs displayed in a (torque × 120◦ and Qprobe are 20.0 kBT and -40.5 kBT , respectively). c, Amount of heat dissipation of wild-type molecules was separately plotted as the heat contributed by the steady rotations Qs and by the nonequlibrium fluctuations Qv, which correspond to the first and second terms of the rhs of the Harada-sasa equation (2), respectively. d, Fraction of Qs and Qv of wild-type molecules. a (cid:88)(cid:74)(cid:77)(cid:69) (cid:73)(cid:90)(cid:69)(cid:83)(cid:80)(cid:77)(cid:90)(cid:84)(cid:74)(cid:84) (cid:84)(cid:90)(cid:79)(cid:85)(cid:73)(cid:70)(cid:84)(cid:74)(cid:84) ∆µ b (cid:78)(cid:86)(cid:85)(cid:66)(cid:79)(cid:85) ∆µ (cid:10) (cid:53) (cid:35) (cid:76) (cid:9) (cid:1) (cid:1) (cid:1) (cid:9499) (cid:1) 20 0 -20 20 0 -20 -15 -20 0 10 20 0 20 40 80 60 (cid:53)(cid:80)(cid:83)(cid:82)(cid:86)(cid:70)(cid:1)(cid:695)(cid:1)(cid:18)(cid:19)(cid:17)(cid:707)(cid:9)(cid:76)(cid:35)(cid:53)(cid:10) 0 20 40 FIG. 4: Work performed on the motor for the wild type (a) and mutant (b). W = Wext − Qprobe. The linear fits of the curves are −16.7 − 0.082x in the hydrolytic rotations (dashed line in the inset of a) and 18.2 + 0.00286x in the synthetic rotations (not shown), where x is the torque multiplied by 120◦. The dashed line in b was fitted by eye. We excluded one aberrant point as noted in the caption of fig. 3.
1508.03249	1	1508	2015-08-13T15:27:37	Using Robotic Fish to Explore the Hydrodynamic Mechanism of Energy Saving in a Fish School	[ "physics.bio-ph" ]	Fish often travel in highly organized schools. One of the most quoted functions of these configurations is energy savings. Here, we verified the hypothesis and explored the mechanism through series of experiments on "schooling" robotic fish, which can undulate actively with flexible body, resembling real fish. We find that, when the school swims in the same spatial arrays as the real one, the energy consumption of the follower mainly depends on the phase difference, a phase angle by which the body wave of the follower leads or lags that of the leader, instead of spatial arrays. Further analysis through flow visualization indicates that the follower saves energy when the phase difference corresponds to the situation that the follower flaps in the same direction of the flow field induced by the vortex dipole shedding by the leader. Using biomimetic robots to verify the biological hypothesis in this paper also sheds new light on the connections among the fields of engineering, physics and biology.	physics.bio-ph	physics	USING ROBOTIC FISH TO EXPLORE THE HYDRODYNAMIC MECHANISM OF ENERGY SAVING IN A FISH SCHOOL LIANG LI, LICHAO JIA, AND GUANGMING XIE ABSTRACT. Fish often travel in highly organized schools. One of the most quoted func- tions of these configurations is energy savings. Here, we verified the hypothesis and ex- plored the mechanism through series of experiments on "schooling" robotic fish, which can undulate actively with flexible body, resembling real fish. We find that, when the school swims in the same spatial arrays as the real one, the energy consumption of the follower mainly depends on the phase difference, a phase angle by which the body wave of the follower leads or lags that of the leader, instead of spatial arrays. Further analysis through flow visualization indicates that the follower saves energy when the phase difference corre- sponds to the situation that the follower flaps in the same direction of the flow field induced by the vortex dipole shedding by the leader. Using biomimetic robots to verify the biolog- ical hypothesis in this paper also sheds new light on the connections among the fields of engineering, physics and biology. 1. INTRODUCTION In the animal kingdom, many species move in highly organized formations [1, 2, 3]. Some well-known systems include the queuing of migrating lobsters [4], organized swarm- ing ducklings [5], flocking birds [6], and schooling fishes [7, 8]. Apart from the social benefits [1, 9, 10] and physiological factors [11], energy saving for individuals is another prevalent hypothesis of why animals prefer to move in groups. In the case of fish schools, since Weihs first predicted that fish swimming in school will save energy [12], this hypoth- esis has been proved through series of methods, such as theoretical analysis [13], compu- tational methods [14, 15, 16, 17], biological experiments experiments [18, 19, 20, 8] and physical models [21, 22, 23]. For theoretical analysis and computational methods, turbu- lent flow is difficult to estimate. These methods can only draw an approximate conclusion. For experiments on real fish, it is difficult to control the fish school and decouple the social factors from physical factors. Recently, with the development of mechanics, electronics, materials and controls, it is possible to build a man-made "fish school" to unravel the hy- drodynamic mechanics of fish school. Previous physical models were passive flapping filaments that shed different vortices from real fish [21, 24], or, rigid foils that have no body shape or body wave [22, 23]. Using a high-fidelity fish-like robot to uncover the hydrodynamic mechanism of the fish school has never been explored. In previous studies, on the energy harvest of fish school and energy expenditure was mainly concerned in relation to lattices [25]. Among the formation hypotheses, a prevalent example is the diamond lattice first proposed by Weihs [12]. Recent studies, based on computational model, predicted that group configurations including diamond, rectangular, phalanx and line are all hydrodynamically advantageous [16, 17]. Further studies on the real fish school demonstrated energy savings regardless of individuals' spatial positions [8]. Key words and phrases. hydrodynamic mechanism; fish school; robotic fish; energy saving; flow visualization. 1 2 L. LI, L. JIA, AND G. XIE Whether spatial location is a crucial factor for energy expenditure when fish are swimming in groups is still controversial. Other factors related to energy consumption remain largely unexplored. Here, we consider a school of two fish-like robots swimming in a laminar flow. The high-fidelity of our robotic fish resides in its body shape, bio-inspired locomotion and sim- ilar Reverse Karman Vortexes (RKV) shedding by the caudal fin. The flow first encounters the rigid head and then the flexible body. Thus, a realistic body shape is critical. The loco- motion should be similar to the real fish to generate a similar flow field. Moreover, since the energy transfer between fish in school is mostly based on the vortices near the caudal fin, thus a similar wake is also significant. In this study, we test and compare the power consumed by two robot flapping fish fixed in a flow tunnel at different relative positions and flapping phase differences. To make fish body undulate in the flow field, the robot should consume power to overcome the resistive force of water. When robots condense into a group state, water-mediated interactions may lead to modifications of energy cost to keep robots undulating in the same body wave. Through series of experiments, we find that (a) the trailing fish will benefit energy saving when swimming in some conditions, and (b) when the schooling robotic fish swim in the similar spatial arrays of the real fish, the energy cost of the trailing fish mostly depends on the phase difference. 2. MATERIAL AND METHODS 2.1. The high-fidelity robotic fish. The robotic fish used in our experiments is designed according to a kind of carangiform fish, which propel themselves by the undulation of the rear body and caudal fin. In order to improve the fidelity of the flow-mediated interactions between schooling fish, we enable our robotic fish to (a) have a controllable flexible body, (b) swim with biomimetic locomotion and (c) shed a similar flow field after the tail. The first step is to endow the robotic fish with a fish-like body. Based on the analyses of biologists, the flexible body has two almost mutually exclusive mechanical properties for swimming: flexible in lateral directions and incompressible in the longitudinal direc- tion [26]. In order to meet these requirements, we construct our robotic fish with three servomotors to generate the lateral rhythmic locomotion. The connections between them are rigid aluminium frames, in longitudinal director. The caudal fin is made with stiffness decreasing from the peduncle to the tail end, resembling to the real one. Most carangiform fish swim forward by passing a body wave down their bodies, a mode known as undulatory propulsion. One of the widely used body wave function is proposed by Lighthill [27]. Here we modified the function as [28]. (1) y = (c1(x− x0) + c2(x− x0)2)sin(kx + 2π f t) where x is the displacement along the boy axis and y is the lateral body displacement (see Figure. 1 (b)), c1 and c2 are linear and quadratic wave amplitude envelope, k denotes the number of whole body wave length, f is the frequency of body wave, x0 is defined as fixed point, which has no oscillation in y axis. The pattern of body undulation is shown in Figure. 1(c). From the body wave function, we find that each segment of the fish body has its own oscillating amplitude as well as coupled phase differences with its adjacent segments. Guided by the body wave undulation, we implemented a similar undulation mode in our robotic fish by a bio-inspired controller, Central Pattern Generator (CPG) controller [29, 30, 31]. The advantages of using CPG controller are that both the phase difference among the joints and between the fish can be well controlled on line. ENERGY SAVING IN FISH SCHOOL 3 FIGURE 1. An introduction of the robotic fish, including (a) three- dimensional mechanism consisting of rigid head, flexible body and cau- dal fin, (b) a snapshot of robotic fish swimming in water in the top view, (c) a cluster of body undulation derived from real carangiform fish, where BL represents body length, (d) a snapshot of vortices shedding by the robotic fish while swimming forward, (e) a typical schematic of the vortices. (online version in color.) Apart from a shape-similar body and bimimetic locomotion, the flow field shedding by the robotic fish is also critical for the interactions between "schooling" robotic fish. If we can generate a similar flow field around the robotic fish to the real one, the energy consumption of real fish school can be well estimated by our robotic fish. We compared the vortex shedding by real fish [32] and our robotic fish to show the similarity between them [30]. From the snapshot of the water flow of robotic fish in Figure. 1 (d), we can find vortices shed to the left-hand side of the robot rotate clockwise, and vortices shed to the right-hand side rotate counterclockwise (see Figure. 1 (e)). This is quite similar to the Reverse Karman Vortex Street (RKVS), by which fish generates thrusts. 2.2. Experimental Apparatus. To experimentally model the hydrodynamic interactions among fish school, we insert a "schooling" robotic fish into a laminar water tunnel, as shown in Figure. 2 (a). The flow velocity of the water tunnel is set as U = 20 cm/s, which corresponds to the free-swimming speed of the robotic fish. The Reynolds number is 9 × 104, which is similar to the natural situation. In this article, we study how the Lateral Spaces (LS), Streamwise Spaces (SS) and Phase Differences (PD) affect the power consumption for the tailing fish (see Figure. 2 (b)). Among the three factors, Lateral 4 L. LI, L. JIA, AND G. XIE FIGURE 2. An introduction of the test apparatus and the factors. (a) Lat- eral view of the experimental set-up. Two robotic fish are fixed in a lam- inar flow with a free surface. A camera captures the interactions among the two robotic fish and the vortices. (b) Top view of fish swimming in group with Lateral Space (LS), Streamwise Space (SS) and anti-phase (Phase Difference (PD) equals π). (online version in color.) Spaces (LS) and Streamwise Spaces (SS) are controlled by group of step motors; and phase difference (PD) is controlled by the CPG controller [30]. Referring to the spatial lattices of real fish school [7] and considering the limitation of the width of the flow tunnel, we arrange the lateral space (LS) ranging from 0.27 to 0.33 BL, the streamwise space (SS) from 0.35 to 0.45 BL and the phase difference from 0 to 2π. As robotic fish is powered by electricity, a measurement and comparison of power con- sumptions of fish swimming alone and in a formation directly reflect the effects of swim- ming in groups. Robotic fish is powered by a stabilized volt supply (Matrix MPS-3005L- 3). The electrical currents are measured by National Instrument Current Acquisition (NI 9227) with a sample rate of 5000 Hz. The electrical current data acquisition lasts 10 s once after robotic fish swimming stably and repeats 5 times for each spatial lattice and phase difference. 2.3. Evaluation of power consumption. Generally, the electrical energy input to the robotic fish will change to mechanical energy and other energy forms such as heat. For the swimming robot, the useful energy is the mechanical energy which is used to overcomes the reactive force of the fluid. We evaluate this part of energy consumption by subtracting the energy consumption when the fish is swimming in air from the energy consumption when fish swim in water [33, 34]. Thus, the power consumption of fish swimming alone in water is evaluated as Pwa − Pa, where Pwa indicates the total power input when the fish ENERGY SAVING IN FISH SCHOOL 5 swims in water alone, and Pa represents the total power consumption while the fish swims in air. Similarly, we calculate the power consumption for a robotic fish swimming in school by Pws − Pa, where Pws is the total power consumption when the fish swims in water in school. To compare the power consumption between fish swimming alone and in groups, we define a power coefficient number η, (2) η = Pwa − Pws Pwa − Pa = (Pwa − Pa)− (Pws − pa) Pwa − Pa A positive value of η indicates a power reduction when fish swimming in school. On the contrary, a negative value of η means fish in school will use more energy. Notably, all the values are time-averaged as the power consumption varies in one oscillation period. For each relative position and phase difference between robotic fish, the tests include √ power acquisitions of robotic fish swimming in air, in water alone and in a school. Uncer- tainties during the test quantities are defined as the standard error of the mean σ = σ / N, for each situation consisting of N =5 measurements. And σ is the sample standard devia- tion. 3. RESULTS 3.1. Energy cost depends more on phase difference (PD). We first test the performance characteristics of a lone robotic fish both in air and water. The power consumptions in four different conditions are shown in Figure. 3 (a) (test in air), Figure. 3 (b) (test in water), Figure. 3 (c) (test in school with 0.4π PD), Figure. 3 (d) (test in school with 1.2π PD). Since the power consumption of robot is periodic at twice the frequency of body wave, the FIGURE 3. Data analysis of power consumptions of the robotic fish. The left column shows the time-domain analysis. And the right column shows the frequency-domain analysis. From the top to the bottom, data in each row are detected in different swimming conditions. (a) and (b), fish swims alone in the air. (c) and (d) fish swims alone in the water. (e) and (f) fish swims in school with 0.4π PD. (g) and (h) fish swims in school with 1.2π PD. power cost is evaluated by the average value. The average power cost of fish swimming 6 L. LI, L. JIA, AND G. XIE in water Pwa = 2.71 W is higher than that in air Pa = 1.44 W . Thus when fish swim alone in the water, the power cost to overcome the resistive force of water is Pwa − Pa = 1.27 W . While fish swim in school with 0.4π and 1.2π, the power costs to overcome the resistive force of water Pws − Pa are respectively 1.32 and 1.21. Further, the efficiency coefficient for the 0.4π PD is −3.94%, thus power increase. The efficiency coefficient for the 1.2π PD is 4.72%, thus power reduction. Note that the noise signal is large in the power values acquired from the experiments. To verify the effectiveness of the data, we apply the Fourier transform to the data. As fish undulates the body periodically and the power consumption mainly depends on the reactive force of the fluid, the power cost value must also be periodic (as the filtered data shown in Figure. 3 (a), (c), (e), (g)). Since the undulation of fish body is symmetrical, the frequency of the power 1.7 Hz is twice that of the body wave 0.85 Hz. Therefore, we prove the effectiveness of the data by this property, as shown in Figure. 3 (b), (d), (f), (h). FIGURE 4. Phase difference is critical for fish swimming in groups to extract energy from the aquatic environment. (a) At the fixed Stream- wise Space (SS) 0.4 BL and each Lateral Space (LS), Power coefficient periodically varies across the whole phase. The shadow area shows the stand deviation for 5 trials at each point. (b) The power coefficient versus the spatial spaces at Phase Difference (PD) of 0.4 π. Most of the coef- ficients are negative denoting power increase. (c) The power coefficient versus the spatial spaces at Phase Difference (PD) of 1.2 π. Most of the coefficients are positive denoting power reduction. (online version in color.) ENERGY SAVING IN FISH SCHOOL 7 We then insert a second robotic fish into the flow with a fixed SS of 0.4 BL and series of LS. The power coefficient is calculated at each spatial lattice as well as phase difference for 5 trials, as shown in Figure. 4 (a). The estimated error are illustrated in shadow. The variation of power coefficient as a function of PD is quite similar to a sine curve and are little related to spatial factors. Fish school swimming with PD around 0.4 π will consume more energy, in schools while around 1.2 π schooling will lead energy saving. Our data show that school fish energy saving is more related to phase differences than spatial lattice. To prove the hypothesis, we compare the power variation by fixing the phase difference 0.4π Figure. 4 (b), and 1.2π Figure. 4 (c). In all the areas, fish swimming with phase difference of 0.4π costs more energy while phase difference of 1.2π consumes less energy, indicating that the power consumption is more related to the phase difference than the spatial lattices. FIGURE 5. Two dimensional map of power coefficient for the follower swimmer in the (Phase Difference (PD), Streamwise Space (SS)) phase space. Lateral Space (LS) is discretized into (a) 0.27, (b) 0.29, (c) 0.31, (d) 0.33 (BL). The power reduction and increase are definitely divided by Phase Difference around 0.8 π. (online version in color.) To further study how the spatial and temporal factors affect the power consumption of group robotic fish. We carry out all the situations within the ranges, totaling 2600 trials. Figure. 5 shows the power coefficient as a function of SS and PD with four different LSs. In contrast to swimming alone, the trailing fish may enjoy energy saving or suffer extra energy cost when swimming in different arrays and Phase Differences (PDs). For almost all the locations around the leader, the follower will benefit of saving energy when swimming with a phase difference ranging of 0.8π ∼1.4π. The larger the SS, the less energy the follower saves. However, we find that, at larger Streamwise Space (SS), the trailing fish need larger Phase Difference (PD) to get a maximum power reduction. This indicates that the spatial array also play a role in the power cost of the trailing fish but with 8 L. LI, L. JIA, AND G. XIE less impact than the Phase Difference (PD). Among all the trails we did here, the maximum power reduction for the follower is 6.9%, and the power increase is 5.6%. 3.2. Hydrodynamic mechanism of energy saving in fish school. To understand why the follower in school saves energy and why the power consumption is strongly related to Phase Difference (PD), we carry out a techchnique of flow visualization to explore the hydrodynamics of fish swimming in school [35]. We carry out the experiments by using laser-induced fluorescence (LIF) technique [36]. Since we would like to only show the interactions between the body wave and the shedding vortices, we put some fluorescent species on the fish body and inject the fluorescence dye from the tail. The fluorescent emission will be captured by the high-speed camera from the bottom of the tunnel at a fre- quency of 125 Hz (see Figure 2). As a result, only the parts with fluorescent species will be high lighted; other parts are black (Movies S4, S5). In order to illustrate the hydrodynamic mechanism of fish school clearly, we show the relationship between the vortices and the body wave in a schematic form in Figure. 6. The left column (Figure. 6(a), (c)) shows why fish in school saves energy in one period. And the right column (Figure. 6(b), (d)) denotes the mechanism of why fish in school costs more energy in one period. The vortex dipole is shed by the leader and have an effect on the follower. The position of the robotic fish and vortices are all based on the Flow Visualization. FIGURE 6. Bottom view of the interactions between the vortices and undulation body of robots. (a), (c) The follower saves energy by flapping in the same direction of the vortex diploe induced by the leader. (b), (d) The follower costs more energy when flapping in the opposite direction of the vortex diploe induced by the leader (See Ref. [30] for further explanation). (online version in color.) When the direction of the induced vortex dipole is the same as that of the flapping tail, the follower will benefits energy saving (see Figure. 6(a), (c)). On the contrary, when the flapping direction is opposite to the direction of the induced vortex dipole, the follower will consume more energy (see Figure. 6(b), (d)). Since PD describes the body wave of the follower leads or lags that of the leader, the relationship between the flapping direction and vortex-dipole direction is also determined by PD. When the schooling fish swim near (around 0.5 1BL), the energy consumption of the follower is greatly controlled by PD ENERGY SAVING IN FISH SCHOOL 9 instead of space array. For all the PDs, PD around 1.2π leads maximum energy saving for the follower swimming in school, and PD around 0.4π make the follower consume more energy. Further, one can definitely find that, with a phase difference of 1.2π, the follower swims in the same direction of the flow field induced by the vortex dipole, thereby saving energy. While the follower with a Phase Difference of 0.4π swims in the opposite direction, and thus it costs more energy. As the flow field induced by the vortex dipole is determined by the leader's flapping tail. Therefore, the energy cost of the trailing fish essentially depends on the Phase Difference (PD), which proves our hypothesis. 4. DISCUSSION In this article, we applied a high-fidelity robotic fish endowed with a controllable, flexi- ble and active body to investigate the hydrodynamic mechanism of power consumption in a fish school. Spatial and temporal factors were both considered in the robotic fish school. The power consumption of the trailing fish is tested and compared when fish are swim- ming alone and in schools. Moreover, we consider the spatial arrays that appeared in the real fish school [7]. And under this spatial condition, we vary the Phase Difference (PD) between the two robotic fish swimming together. After series of experiments, we find that, under the same spatial condition as in real fish school, the trailing robotic fish saves energy depending more on the PD rather than the array spacing. Compared with the previous apparatuses, the robotic fish consists of fish-like body, bio-inspired locomotion and similar Reverse Karman Vortex Street (RKVS). A popular apparatus that was used to uncover the mechanism of energy saving in fish school is the flexible filament in a flowing soap film [21, 24]. As the filament is flexible, the interactions between the body and fluid vortices can be well illustrated [37]. However, because of the passive body, the vortices shedding by the filaments are Karman Vortices instead of Reverse Karman Vortices. Using active flapping foils, B. M. Boschitsch and P. A. Dewey respectively studied the hydrofoils in a in-line configuration [23] and side-by-side [22] configuration. Although the rigid foils are active and controllable, They are quite different from the flexible tails, as real fish tail has a increasing elasticity towards the end. Moreover, a flapping foil can not represent the hydrodynamics of fish swimming in water, as the flow first encounters the rigid head and then interacts with the flexible body. The robotic fish we applied here combines both the advantages. And thus, using robotic fish to mimic the fish school to explore the mechanism of energy saving is feasible and more realistic. Note that the maximum energy saving for the follower is only 6.9%, smaller than pre- vious studies [24, 21, 17, 23]. One explanation of this is that the real school is more than two fish. In this article, we only considered the basic schooling of two fish. As one leader will make the follower save 6.9%, two leaders will make the follower save about 13.8%. Moreover, the frequency of our robotic fish is around 3 times lower than the real fish, there- fore the energy of the vortex field shed by our robotic fish is smaller than that shed by the real fish. This may also lead our data show that the robotic fish only gets a small energy saving. But in the filed of engineering, an energy saving of 6.9% is great for the manmade machines. Our finding in the study may guide the formation and locomotion control for group of bioinspired underwater automatic vehicles. The work presented here is basis and needs more work. The power costs of the leader will be further explored. Nevertheless, this work first implements artificial fish school, and evaluated the mechanism of fish swimming in groups in the aspect of hydrodynamics. 10 L. LI, L. JIA, AND G. XIE Further the connections of biology and physics also lie in other aspects such as group behaviour verifications by using bio-inspired high-fidelity robots. ACKNOWLEDGMENT We thank J. F. Steffensen for supporting the video of real fish. We also thank W. Wang and C. Wang for helpful discussions. REFERENCES [1] J. Krause and G. D. Ruxton, Living in Groups. Oxford University Press, 2002. [2] I. D. Couzin and J. Krause, "Self-organization and collective behavior in vertebrates," in Advances in the Study of Behavior, vol. Volume 32, pp. 1 -- 75, Academic Press, 2003. [3] D. J. Sumpter, Collective Animal Behavior. Princeton University Press, 2010. [4] R. G. Bill and W. F. Herrnkind, "Drag reduction by formation movement in spiny lobsters," Science, vol. 193, no. 4258, pp. 1146 -- 1148, 1976. [5] F. E. Fish, "Kinematics of ducklings swimming in formation: Consequences of position," Journal of Exper- imental Zoology, vol. 273, no. 1, pp. 1 -- 11, 1995. [6] S. J. Portugal, T. Y. Hubel, J. Fritz, S. Heese, D. Trobe, B. Voelkl, S. Hailes, A. M. Wilson, and J. R. Ush- erwood, "Upwash exploitation and downwash avoidance by flap phasing in ibis formation flight," Nature, vol. 505, no. 7483, pp. 399 -- 402, 2014. [7] B. L. Partridge, T. Pitcher, J. M. Cullen, and J. Wilson, "The three-dimensional structure of fish schools," Behavioral Ecology and Sociobiology, vol. 6, no. 4, pp. 277 -- 288, 1980. [8] S. Marras, S. Killen, J. Lindstrom, D. McKenzie, J. Steffensen, and P. Domenici, "Fish swimming in schools save energy regardless of their spatial position," Behavioral Ecology and Sociobiology, pp. 1 -- 8, 2014. [9] N. Miller, S. Garnier, A. T. Hartnett, and I. D. Couzin, "Both information and social cohesion determine collective decisions in animal groups," Proceedings of the National Academy of Sciences, 2013. [10] A. Berdahl, C. J. Torney, C. C. Ioannou, J. J. Faria, and I. D. Couzin, "Emergent sensing of complex environments by mobile animal groups," Science, vol. 339, no. 6119, pp. 574 -- 576, 2013. [11] C. M. Breder, "Studies on the structure of the fish school," Bulletin of the American Museum of Natural History, vol. 98, no. 1, pp. 1 -- 27, 1951. [12] D. Weihs, "Hydromechanics of fish schooling," Nature, vol. 241, no. 5387, pp. 290 -- 291, 1973. [13] D. Weihs, "Some hydrodynamics aspects of fish schooling," in Swimming and flying in nature (T. Y.-T. Wu, C. J. Brokaw, and C. Brennen, eds.), pp. 703 -- 718, Plenum Press, 1975. [14] G.-J. Dong and X.-Y. Lu, "Characteristics of flow over traveling wavy foils in a side-by-side arrangement," Physics of Fluids, vol. 19, no. 5, p. 11, 2007. [15] E. Kanso and P. K. Newton, "Locomotory advantages to flapping out of phase," Experimental Mechanics, vol. 50, no. 9, pp. 1367 -- 1372, 2010. [16] C. K. Hemelrijk, D. A. P. Reid, H. Hildenbrandt, and J. T. Padding, "The increased efficiency of fish swim- ming in a school," Fish and Fisheries, pp. 1467 -- 2979, 2014. [17] X. Zhu, G. He, and X. Zhang, "Flow-mediated interactions between two self-propelled flapping filaments in tandem configuration," Physical Review Letters, vol. 113, no. 23, p. 238105, 2014. [18] J. C. Svendsen, J. Skov, M. Bildsoe, and J. F. Steffensen, "Intra-school positional preference and reduced tail beat frequency in trailing positions in schooling roach under experimental conditions," Journal of Fish Biology, vol. 62, no. 4, pp. 834 -- 846, 2003. [19] S. S. Killen, S. Marras, J. F. Steffensen, and D. J. McKenzie, "Aerobic capacity influences the spatial posi- tion of individuals within fish schools," Proceedings of the Royal Society B-Biological Sciences, vol. 279, no. 1727, pp. 357 -- 364, 2012. [20] J. C. Liao, D. N. Beal, G. V. Lauder, and M. S. Triantafyllou, "Fish exploiting vortices decrease muscle activity," Science, vol. 302, no. 5650, pp. 1566 -- 1569, 2003. [21] L. Ristroph and J. Zhang, "Anomalous hydrodynamic drafting of interacting flapping flags," Physical Review Letters, vol. 101, no. 19, p. 194502, 2008. [22] P. A. Dewey, D. B. Quinn, B. M. Boschitsch, and A. J. Smits, "Propulsive performance of unsteady tandem hydrofoils in a side-by-side configuration," Physics of Fluids (1994-present), vol. 26, no. 4, pp. 041903 -- 041918, 2014. [23] B. M. Boschitsch, P. A. Dewey, and A. J. Smits, "Propulsive performance of unsteady tandem hydrofoils in an in-line configuration," Physics of Fluids (1994-present), vol. 26, no. 5, pp. 051901 -- 051917, 2014. ENERGY SAVING IN FISH SCHOOL 11 [24] L. Jia and X. Yin, "Passive oscillations of two tandem flexible filaments in a flowing soap film," Physical Review Letters, vol. 100, no. 22, p. 4, 2008. [25] C. M. Breder, "Fish schools as operational structures," Fishery Bulletin, vol. 74, no. 3, pp. 471 -- 502, 1976. [26] J. J. Videler, Fish Swimming. Netherlands: Springer Netherlands, 1993. [27] M. J. Lighthill, "Note on the swimming of slender fish," Journal of Fluid Mechanics, vol. 9, no. 02, pp. 305 -- 317, 1960. [28] W. C. Li, Liang and G. Xie, "Modeling of a carangiform-like robotic fish for both forward and backward swimming: Based on the fixed point," in 2014 IEEE International Conference on Robotics and Automation (ICRA), pp. 800 -- 805, 2014. [29] A. J. Ijspeert, "Central pattern generators for locomotion control in animals and robots: A review," Neural Networks, vol. 21, no. 4, pp. 642 -- 653, 2008. [30] See Supplemental Material at [URL will be inserted by publisher] for (a) Body wave comparison between the real fish and our robotic fish; (b) Vortexes shedding by our robotic fish and a comparison to the real one; (c) Introduction of the automatic experiment device; (d) The mechanism of power reduction for robotic fish school shown by flow visualization. (e) The mechanism of power increase for robotic fish school illustrated through flow visualization. [31] L. Li, C. Wang, and G. Xie, "A general cpg network and its implementation on the microcontroller," Neuro- computing, 2015 inpress. [32] Y. Aleyev, Nekton. The Hague, 1977. [33] L. Wen, T. Wang, G. Wu, and J. Liang, "Hydrodynamic investigation of a self-propelled robotic fish based on a force-feedback control method," Bioinspiration & Biomimetics, vol. 7, no. 3, 2012. [34] L. Wen, T. Wang, G. Wu, and J. Liang, "Quantitative thrust efficiency of a self-propulsive robotic fish: Ex- perimental method and hydrodynamic investigation," IEEE/ASME Transactions on Mechatronics, vol. 18, no. 3, pp. 1027 -- 1038, 2013. [35] A. J. Smits and T. T. Lim, Flow visualization: techniques and examples. World Scientific Publishing Com- pany, 2000. [36] D. R. Unger and F. J. Muzzio, "Laser-induced fluorescence technique for the quantification of mixing in impinging jets," Aiche Journal, vol. 45, no. 12, pp. 2477 -- 2486, 1999. [37] J. Zhang, S. Childress, A. Libchaber, and M. Shelley, "Flexible filaments in a flowing soap film as a model for one-dimensional flags in a two-dimensional wind," Nature, vol. 408, no. 6814, pp. 835 -- 839, 2000. INTELLIGENT CONTROL LABORATORY, COLLEGE OF ENGINEERING, PEKING UNIVERSITY, BEIJING 100871, P. R. CHINA E-mail address: [email protected] STATE KEY LABORATORY FOR TURBULENCE AND COMPLEX SYSTEMS, COLLEGE OF ENGINEERING , PEKING UNIVERSITY, BEIJING 100871, P. R. CHINA E-mail address: [email protected] INTELLIGENT CONTROL LABORATORY, COLLEGE OF ENGINEERING, PEKING UNIVERSITY, BEIJING 100871, P. R. CHINA E-mail address: [email protected]
1508.00528	1	1508	2015-08-03T18:55:00	Theory of morphological transformation of viral capsid shells during maturation process	[ "physics.bio-ph", "cond-mat.soft" ]	In the frame of the Landau-Ginzburg formalism we propose a minimal phenomenological model for a morphological transformation in viral capsid shells. The transformation takes place during virus maturation process which renders virus infectious. The theory is illustrated on the example of the HK97 bacteriophage and viruses with similar morphological changes in the protective protein shell. The transformation is shown to be a structural phase transition driven by two order parameters. The first order parameter describes the isotropic expansion of the protein shell while the second one is responsible for the shape symmetry breaking and the resulting shell faceting. The group theory analysis and the resulting thermodynamic model make it possible to choose the parameter which discriminates between the icosahedral shell faceting often observed in viral capsids and the dodecahedral one observed in viruses of the Parvovirus family. Calculated phase diagram illustrates the discontinuous character of the virus morphological transformation and shows two qualitatively different paths of the transformation in a function of two main external thermodynamic parameters of the in vitro and in vivo experiments.	physics.bio-ph	physics	Theory of morphological transformation of viral capsid shells during maturation process O.V. Konevtsova,* V.L. Lorman,† and S.B. Rochal* *Faculty of Physics, Southern Federal University, 5 Zorge str., 344090 Rostov-on-Don, Russia †Laboratoire Charles Coulomb, UMR 5221 CNRS and Université de Montpellier, pl. E. Bataillon, 34095 Montpellier, France In the frame of the Landau-Ginzburg formalism we propose a minimal phenomenological model for a morphological transformation in viral capsid shells. The transformation takes place during virus maturation process which renders virus infectious. The theory is illustrated on the example of the HK97 bacteriophage and viruses with similar morphological changes in the protective protein shell. The transformation is shown to be a structural phase transition driven by two order parameters. The first order parameter describes the isotropic expansion of the protein shell while the second one is responsible for the shape symmetry breaking and the resulting shell faceting. The group theory analysis and the resulting thermodynamic model make it possible to choose the parameter which discriminates between the icosahedral shell faceting often observed in viral capsids and the dodecahedral one observed in viruses of the Parvovirus family. Calculated phase diagram illustrates the discontinuous character of the virus morphological transformation and shows two qualitatively different paths of the transformation in a function of two main external thermodynamic parameters of the in vitro and in vivo experiments. PACS numbers: 62.23.St, 64.70.Nd, 87.15.Zg INTRODUCTION I. Self-assembly and shape transitions in biological nanostructures with non-trivial topology are characterized by unconventional properties and unusual pathways. In contrast to classical condensed media, bionanoassemblies undergo a multistep process which involves several physical transformations. Virus self-assembly is a typical example of the multistep process [1]. Viral capsid, the protein shell which protects viral genome, also passes through several successive steps of the assembly process. At the first step the so-called procapsid shell self- assembles from the aqueous solution of individual viral proteins. Capsid maturation is the final step of the virus self-assembly process. During maturation virus acquires the ability to infect its host cell. The process is characterized by a number of considerable correlated structure changes in the procapsid shell resulting in an infectious virion. One of the most interesting features of maturation process in a whole series of viruses is a morphological variation of the capsid shell resulting in a shape transition from the spherical to the faceted polyhedral geometry. It is worth noting that maturation in the host cell secretory pathway is often accompanied by irreversible biochemical events in capsid proteins including protein domain cleavage or neighboring protein crosslinking [2,3]. However, irreversible events usually take place when the structural modification leading to final protein positions and orientations in the shell is already achieved (see e.g. [4]). Thus, the reversible structural transition during maturation process can be considered as an independent physical phenomenon and treated in the frame of condensed matter physics. To illustrate the notions of the theory developed in our work we choose the example of bacteriophage HK97 as well as other viruses with similar structural characteristics. HK97 is one of the viruses studied in detail by high-resolution structural methods which show eventful 1 maturation scenario. HK97 is a dsDNA virus with the capsid protein shell which has rotational icosahedral symmetry group I and consists of 420 proteins distributed in seven 60-fold general positions of this group. Capsid structure satisfies to the well-known Caspar and Klug geometrical model [5] which limits the number of proteins constituting the shell to N=60T where the number of different positions T should take the value , with h and k being non-negative integers. hT hk 2 k + = 2 + One of the main physical phenomena taking place during the HK97 bacteriophage maturation is the morphological transformation of the spherical procapsid into the faceted capsid accompanied by the increase in the shell volume [6,7,8]. Mean capsid diameter varies during the transformation to more than 20%. The protein shell becomes also much thinner, with more homogeneous shell thickness, and acquires a pronounced icosahedral shape. In the host cell pathway these morphological changes are induced by the ATP-dependent genome packaging into the pre-assembled procapsid protein shell, but in in vitro experiments similar structure and shape variation is realized in the absence of viral genomic DNA by using controlled pH decrease of the buffer [9]. Multistep maturation process with flattening transition at intermediate stages is typical for a whole series of viruses (e.g. P22 phage, Herpes Simplex Virus, etc.) which have structural organization similar to that of bacteriophage HK97 [9]. Universal features of this process represent the fundamental interest for both physics of nanoassemblies and physics of biological systems, and shed new light on principles of phase transitions in nanostructures with non-trivial topology. Detailed study of different maturation steps for HK97 both in in vivo and in vitro experiments is still not complete. As it is often the case, the study of living systems by physical methods encounters certain experimental difficulties. The structures of the procapsid and matured capsid shells are usually determined by means of X-ray crystallography and high- resolution cryoelectron microscopy (cryoEM) [8,10]. However, the intermediate states of the in vivo maturation process are inaccessible by these techniques. Biochemical in vitro experiments study maturation dynamics in buffers with controlled pH level [11,12] and manage to distinguish several intermediate states with faceted shapes and one intermediate state with a spherical one. They are stabilized at different pH levels and characterized by successive increase in the shell volume. In spite of this progress the relation between pH variation in in vitro experiments and the in vivo capsid transformations induced by the genome packaging with the help of motor proteins [3] remains unclear. Experimental problems in virus maturation dynamics studies make of theoretical and numerical modeling an important tool for understanding physical phenomena at the origin of the procapsid shell transformation into the mature infectious virion. Several theoretical approaches have been tested for capsid morphological transformation modeling. Structural changes during maturation have been described in [13] as a condensation of several low-frequency modes in a model system with the icosahedral symmetry. Two types of modes were considered to be responsible for the capsid faceting, the modes of the capsid isotropic expansion (or compression) and the modes of the pentamer displacements. HK97 virus maturation was also modelled [14] in the frame of the simplified Landau-Ginzburg theory of phase transitions. In contrast with the classical Landau-Ginzburg approach, the model proposed in [14] did not take into account the normal mode symmetry, thus reducing the order parameter to a simplified scalar physical quantity and disconnecting the structure from the free energy form. At the same time additional terms dependent on continuous derivatives of the order parameter with respect to variables on the shell surface were introduced in the model free energy, though the shell consists of only 420 particles. The intermediate structures of the maturation obtained in the frame of this approach are spatially inhomogeneous and their interpretation on the basis of available experimental data is not straightforward. The model [14] 2 was then modified [15] by introducing methods of continuous elasticity theory but still remained disconnected from the mode symmetry. The aim of the present work is to perform detailed group theory analysis of low- frequency modes responsible for the morphological changes, and to propose in the frame of the Landau-Ginzburg approach a minimal model of the procapsid-to-capsid transformation for the HK97 bacteriophage with the well-defined physical meaning of the free energy terms. The model developed in the paper is suitable not only for the considered HK97 virus but for a whole series of viruses demonstrating maturation process accompanied by capsid faceting and discontinuous volume jump. II. CRITICAL ORDER PARAMETERS RESPONSIBLE FOR THE MORPHOLOGICAL TRANSFORMATION As it was shown previously in [13], structural changes during maturation in viral capsid shell of the HK97 bacteriophage and in a series of similar viruses involve at least two low- frequency modes. First of them induces isotropic capsid expansion, the second one is responsible for the shell faceting. Let us now discuss symmetry characteristics of the shell and their incidence on the modes responsible for the morphological changes during maturation process. For that aim we will first distinguish the symmetry of the protein density distribution from the symmetry of the shell shape, and, second, compare the symmetry of corresponding distributions in the procapsid and capsid states. Due to the asymmetry (and namely to the chirality) of coat proteins the protein density distribution of both procapsid and capsid shells have the same chiral symmetry group I of all icosahedral rotations [5,16]. In contrast, the shape of the procapsid shell is spherical with a good accuracy while it becomes icosahedral in the faceted capsid state. The morphological transformation we are interested in is related mainly to this shape symmetry breaking. Thus, both modes responsible for the transformation are classified in the model as spanning irreducible representations of the SO(3) symmetry group of a chiral protein shell with the initial spherical shape. All possible modes of the spherical shell transformation are collective displacement fields of material points of the spherical surface. In the most general form, corresponding displacement fields contain both radial and tangent components, tangent components being in turn separated into stretching and shear fields spanning different representations of the SO(3) group (for full mode classification and detailed study of the corresponding mechanical problem see [17]). Radial displacement modes bringing the main contribution to the morphological changes, in the minimal model we will take into account explicitly only the radial part of the displacement fields responsible for the procapsid-to-capsid shape variation. Note also, that the stretching component of the tangent displacement field spans the same irreducible representation of the SO(3) group as the radial component does, and thus, its amplitude is directly determined by the coupling with the radial component [17]. Any field of radial displacement of a spherical shell is usually expanded in scalar spherical harmonics φθ ) ,( = u r = 0 l −= ml ∞ = ml ∑ ∑ YA mlml . . φθ ) ,( , where θ and φ are the angles of a standard spherical coordinate system. Corresponding Cartesian vector R’ of a point on a deformed sphere surface has the form: =< R ' u ( r + R sin) φθ u (, cos + R sin) φθ u (, sin r r + R ) cos θ > , (2) 3 (1) where R is the radius of the initial non-deformed sphere. A normal mode responsible for the shape symmetry breaking spans an irreducible representation of the SO(3) group labeled by a fixed value of the index (i.e. wave number) l. Spherical harmonic Y00 describes isotropic expansion (or compression) of the procapsid shall and plays the role of the first one-dimensional fully symmetrical order parameter of the minimal model developed in our approach. The second order parameter responsible for the shape variation from the spherical to the icosahedral one is not so simple, and needs much more detailed discussion. The experimentally observed capsid shape is characterized by the average shape which has full symmetry of an icosahedron Ih in contrast with the microscopic symmetry of its chiral protein distribution which is invariant with respect to the rotational symmetry of an icosahedron I. Because of the shape invariance with respect to spatial inversion, the expansion (1) is limited to spherical harmonics with even wave numbers l only. Furthermore, there are additional, much stronger rules which select possible index l values for an irreducible mode, which can drive a transition from the spherical shape to the shape with full icosahedral symmetry Ih. The analysis based on the invariant theory (see Appendix) shows that these modes correspond to the functions with indices l satisfying following selection rules: ,(. φθmlY ) = + (3) l ∈ , where i and j are positive integers or zero. The sequence L of the index l values, where L allowed by (3) has the form : L = (6, 10, 12, 16, 18, 20, 22, 24 ...). It selects spherical harmonics, and, consequently, symmetry breaking modes which can give a contribution to the icosahedral faceting of a viral capsid. 10j 6i , l ),( φθru Explicit form of the displacement field ( )ϕθ, obtained in terms of orthogonal functions i lf in the state with the icosahedral shape is with full icosahedral symmetry φθ ),( = u r A 0.0 + = ni t ∑∑ ∈ Ll = 1 i i fD il . l φθ ),( , (4) ilD , are the amplitudes of the orthogonal icosahedral functions in the displacement field. where The index i (i=1,…, tn ) in the sum in (4) runs over all functions with the same fixed wave number l value. The number of tn values is equal to the number of non-negative integer solutions (i, j) of equation (3) for a given allowed l value, and is quite limited. Consequently, the number is also limited. According to Eq. (3) this of linearly independent icosahedral functions with a number is equal to given fixed l value is easily obtained by averaging of spherical harmonics lmY over the full icosahedral symmetry group: for all l<30. Explicit form of the icosahedral functions ( )ϕθ, ( )ϕθ, 1=tn lf lf i i ϕθ ,( = /1) 120 i f l ∧ gY ,(( lm ϕθ )) . ∑ G (5) Number tn can be obtained by a cumbersome but more accessible classical method widely used in condensed matter physics, especially in Raman and IR spectroscopies for active modes determination [18]. It involves the well-known character relation of the representations for the symmetry groups of the symmetric state and the state with the broken symmetry (with 4 i ( )ϕθ, non-zero mode amplitude). For the procapsid spherical shell 2l +1 harmonics with m=-l, -l+1, ... l span one (2l +1)–dimensional irreducible representation of the SO(3) group. The introduced above are mutually orthogonal basic functions of icosahedral functions identity (fully symmetrical) representations of the icosahedral capsid shell symmetry group. with the given l which are formed by the restriction of They are linear combinations of the SO(3) group to the icosahedral Ih group. The number of identity representations tn is given by the character relation in the form [19]: ,(. φθmlY ) lf ,(. φθmlY ) ln t /1)( = ∧ ξ g )( l G ∑ G (6) ∧ g of the icosahedral symmetry group, number of Where the sum runs over the elements glξ are the characters of the SO(3) representation with the fixed l elements is G = 120, and value. Taking into account the fact that all functions with even wave number l are invariant with respect to the spatial inversion we limit sum (6) to 60 rotational elements of the icosahedral symmetry group. After additional partition of the representation characters into conjugacy classes Eq. (6) takes the following form: ∧ )( 1 60 + l = where angle α matrix. αξ ) l ( ln )( t = l 2( ++ 1 15 πξπξ l )3/2( )( 20 + l + 12 πξ l )5/2( + 12 πξ l ))5/4( , (7) sin(( )2/1 αα sin( ) )2/ is the explicit form of the character for a rotation to the The full set of harmonics in the displacement field (4) defines any surface with the icosahedral symmetry Ih. However, to describe the shape symmetry breaking it is sufficient to take into account in the displacement field expansion only minimal set of critical harmonics. According to the Landau theory principles, corresponding modes give a critical contribution to the capsid free energy variation in the vicinity of the morphologic transformation point where the amplitudes of other non-critical harmonics in (4) are negligibly small. Displacement field (4) limited to critical harmonics gives the main contribution to the spherical shape variation to the icosahedral one. The main result of the group theory analysis presented above is the conclusion that the simplest mode responsible for the capsid shape symmetry breaking corresponds to the minimal wave number value l=6 allowed by selection rules (3). Corresponding icosahedral function is the basis function of the (unique in this case) fully symmetric representation of the Ih group resulting from the restriction of the irreducible representation of the SO(3) group with l=6. ( )ϕθ, 6f In Cartesian coordinates the ( f 6 )zyx , , ( f 6 ) = zyx , , 2 ( y − τ 22 z 2 )( z function has the following form: − , x )2 − 5(21/1) τ 22 τ 22 )( − − y x 2 (8) =τ − is the golden mean, and coordinates <x, y, z> of a point on a unit sphere where are related in a standard way to θ and φ variables of the spherical coordinate system. Explicit 2/)15( 5 , 6 f ( )zyx , function amplitude substitution of this function in the displacement field expression leads to the minimal displacement field which breaks the shape symmetry from the spherical to the icosahedral one. iD ,6 lead to two different Note that two opposite signs of the surfaces with the icosahedral symmetry. For one sign the displacement field results in a surface with the icosahedral shape, while for the opposite sign the surface acquires the dodecahedral shape (see Fig. 1). Experimentally, in the case of bacteriophage HK97 considered in our work the capsid shape induced by the displacement field is icosahedral (see Fig. 1,c) [20]. However, there exist a whole series of viruses which display dodecahedral capsid shape. An example of viruses of this type is given by the pathogenic human parvovirus B19 (Fig. 1,d) [21]. It is remarkable that the minimal model developed in our work catches this structural difference and describes it in a very simple way. 6f ( )ϕθ, FIG. 1. (a-b): Spherical surface deformation by radial displacement fields proportional to the iD ,6 of the displacement fields in panels (a) and (b) icosahedral function have opposite signs. (c-d): Experimental realization of the capsid shapes induced by displacement fields (a) and (b). Viral capsid of the bacteriophage HK97 [20] with the icosahedral faceted shape (c), and viral capsid of the pathogenic human parvovirus B19 [21] (d) with the dodecahedral faceted shape. . Amplitudes III. FREE ENERGY OF THE PROCAPSID-TO-CAPSID MORPHOLOGICAL TRANSFORMATION Symmetry and structure analysis performed in the previous Section shows that the main features of the shape variation during the maturation process in the HK97 bacteriophage and similar viruses are described by displacement fields with the minimal set of spherical harmonics with l=0 and l=6. These critical order parameters span two irreducible representations of the SO(3) symmetry group with the corresponding wave number values. The former field is responsible for the isotropic volume changes and the latter one breaks the shape symmetry from the spherical to the icosahedral one. Free energy of the transformation is dependent on the amplitude of the fully symmetric function with l=0, and 13 amplitudes A6,m of the harmonics with l=6. However, group symmetry analysis shows that these 13 amplitudes are linearly 6 6f )ϕθ, ( dependent in the state with the icosahedral symmetry. Consequently, all amplitudes A6,m in this state are proportional to the same value noted below as η. This “effective” order parameter value η defines the D6,1 amplitude of the icosahedral function. The “effective” free energy depends only on η and on the variable describing isotropic volume variation. Value of the order parameter η obtained by the free energy minimization defines the degree of the capsid faceting, and the sign of η discriminate between icosahedral and dodecahedral shape of the resulting capsid. In addition, in what follows, instead of the A00 amplitude of the fully symmetric harmonic Y00 responsible for the isotropic expansion (or compression) we use explicitly capsid’s volume variation ∆V as the corresponding variable in the free energy expansion. This choice is more suitable for physical interpretation of the results obtained in the frame of the developed approach. It is evident that the amplitude A00 and the volume variation ∆V are related linearly. This justifies the change of variable proposed. Then, the minimal expansion of the free energy density describing the morphological transformation during maturation process takes the form: F = A(d)η2 + a2η3 + a3η4 - P∆V + b∆V2 - g∆Vη2, (9) Phenomenological coefficient A(d) in free energy density (9) is dependent on the capsid shell thickness d, which depends in turn on a whole series of biochemical parameters. Detailed discussion of the biochemical processes involved in the maturation and corresponding microscopic parameters are out of the scope of the present phenomenological work focused on the minimal model construction, and based on the full symmetry analysis of the morphological transformation problem. Here we limit our discussion to the experimental fact that decrease in the capsid shell thickness leads to the faceting transition. In the frame of the Landau approach developed in our work the stability limit for the spherical capsid shape corresponds to the coefficient A(d) vanishing. But the faceting instability transition takes place even earlier, when A(d) is still positive. Analysis of the morphological transformation thermodynamics described by free energy (9) shows that the considered shape transition is discontinuous. The procapsid undergoes first order faceting transition before full softening of the normal mode responsible for this shape transition takes place. Discontinuous character of the faceting transition is directly related to the presence of the cubic term a2η3 in free energy density (9). It is worth noting that an invariant term which is cubic in a function of amplitudes of spherical harmonics, exists for any irreducible representation of the SO(3) group with even value of the wave number l. This fact was widely used previously in many different fields of physics. It is the case, for example, of nematic liquid crystal physics. The orientational ordering of rod-like molecules is described by the second-rank symmetric traceless tensor Qij , and the nematic order parameter spans the irreducible representation of the SO(3) group with l=2. It is evident that the determinant of the corresponding tensor is the cubic invariant of the irreducible representation. Due to even value of l the same term is also invariant with respect to the full spherical symmetry group O(3). The same principles applied to other values of l show that cubic invariant exists for all even wave numbers. Direct calculation of the cubic invariant for the irreducible representation with l=6 can be performed explicitly using properties of Clebsch-Gordan coefficients. Because of its cumbersome form we omit it in the present work. Note, that the cubic term in the free energy not only makes the transition discontinuous but also plays the crucial role in the choice between icosahedral and dodecahedral shapes of the resulting capsid shell. Other terms in free energy density (9) have rather straightforward form and physical meaning. Because of the identical symmetry of the mode responsible for the isotropic volume change the free energy contains linear in ∆V term and the coupling term with the symmetry- 7 breaking order parameter η which is also linear in ∆V. The term quadratic in ∆V with the positive coefficient b>0 ensures global stability of the isotropic expansion (or compression) mode. Fourth-degree term in order parameter η multiplied by the positive coefficient a3>0 ensures the existence of a global minimum in the considered system. Coefficient P in free energy density (9) expresses the pressure difference between the inner and the outer regions of the capsid shell. Osmotic pressure difference created during the genome packaging into the capsid shell gives the main contribution to the corresponding term for in vivo maturation process. Negative sign of the coefficient g corresponds to the fact that both packaging-induced osmotic pressure and increase in capsid volume make the spherical shape less stable and favor morphological transformation to the faceted state. In a more complex model it is possible to take into account additional non- linearity of the free energy in a function of the isotropic expansion (or compression) mode represented here by the volume change variable ∆V. Nonlinear in ∆V terms will favor intermediate states between spherical procapsid and icosahedral capsid states. But in the minimal model of the morphological transformation these terms lead to unjustified mathematical complications. Minimization of the free energy functional with the density given by (9) leads to three possible solutions with different symmetries: i) η=0; ii) η<0; and iii) η>0, volume change ∆V being nonzero for all three states. Solutions ii) and iii) with opposite signs of the order parameter correspond to the shell with the icosahedral and the dodecahedral shape, respectively. They are usually called anti-isostructural states in the theory of phase transitions in condensed media [22]. Free energy density being non-linear in η and quadratic in ∆V, it is more convenient to minimize it first with respect to ∆V, and substitute the solution in (9). The resulting “effective” free energy depends only on the symmetry breaking order parameter η and its minimization leads to the same values of η in the ordered states with the icosahedral symmetry of the shell F(η) = (A(d) - gP/2b)η2 + a2η3 + (a3 - g2/4b)η4 - P2/4b, (10) Free energy density (10) has a simpler form and its behavior can be easily illustrated graphically. Its minimization is straightforward. The phase diagram of the model is presented in Fig. 2. Free energy (10) plots for several points typical for corresponding regions of the phase diagram are given in inserts a), b) and c) in Fig. 2. Phase diagram shows that the in-out pressure difference variation and the A(d) coefficient variation, induced by the shell thinning and the pH level decrease, contribute to the same part of the free energy. 8 FIG. 2. Phase diagram of the free energy functional with density (9). Full line 1 and dash-dotted line 2 divide the phase diagram in qualitatively different regions: region of stability of the minimum with the spherical symmetry, region of stability of the minimum with the icosahedral symmetry, and the region where two minima with different symmetries coexist. Minimum with η=0 (spherical shape state without faceting) exists only in the region above line 2. The state with the icosahedral shape (η<0) appears below line 1. At dashed line 3 free energies of spherical and icosahedral faceted states are equalized. Below line 2 the minimum which corresponds to the spherical state disappears leaving the place for a weak metastable minimum with η<0. Inserts a), b) and c) show free energy density (10) as a function of the order parameter η in the typical points given by the same letters in the phase diagram. Two qualitatively different thermodynamic paths of the morphological transformation during maturation process are given by arrowed lines. Line 4 (downward) corresponds to the shell thinning induced by the pH level decrease. Line 5 (left to right) illustrates the result of the pressure difference increase during the in vivo genome packaging into the capsid. The phase diagram in Fig. 2 was calculated for the case a2>0. For negative values of this coefficient, free energy density (10) in inserts a), b) and c) is reflected with respect to the F axis and the ordered state with the icosahedral symmetry changes its shape from the icosahedral to the dodecahedral one. Corresponding lines in the phase diagram separate in this case the regions of stability of the spherical and the dodecahedral shell shapes. Dashed line 3 indicates in this case the equalization of the free energies of the spherical and the dodecahedral states. Minimal model developed here allows us to calculate analytically the equation of line 3 where the free energies of the spherical and the icosahedral (or dodecahedral) states become equal: dA )( = 4 1 2 bg Pa 3 4( 3 + − 2 2 bgba ) 3 Pg − 22 ba 2 . Another important quantitative characteristic which is also obtained analytically in the frame of the minimal model is the value of the volume jump at the transition from the spherical to the faceted state: 1 8 bgP a 16 3 22 b ) + dAba )( 3 − gba 3 2 dbAg 4)( =Δ V 22 ba 2 Pg g 3( ba 32 3 ) + 8 + 2 2 4( − − 9 9 IV. DISCUSSION AND CONCLUSION The model developed in our work relates the faceting transition with two different external thermodynamic parameters of the system. First of them is the capsid shell thinning induced by the pH level decrease observed in in vitro experiments. The second one is connected to the in vivo genome packaging into the capsid shell with the help of motor proteins. Progressive capsid filling with the viral DNA leads to the pressure difference increase between the inner and outer regions of the shell. In the phase diagram of the model independent variation of these two external parameters corresponds to two different thermodynamic paths (shown by arrowed lines 4 and 5 in Fig. 2). The capsid morphological transformation mechanism proposed here is consistent with classical works [23,24] on the capsid shell faceting based mainly on the continuous elasticity theory. These works have used the analogy between the faceting phenomenon and the longitudinal instability of disclinations in two-dimensional crystals. They have shown that in the locally hexagonal protein packing proposed as a model for viral capsid organization by Caspar and Klug [5] disclination instabilities in the vicinity of five-fold axes lead to the capsid faceting for viruses of sufficiently big size. The model proposed in [23] was based on the non-linear physics of thin elastic shells. Using continuous elasticity methods it elucidated the dependence of the viral shell faceting on the value of dimensionless Föppl–von Kármán (FvK) number γ, which characterizes the shell buckling instability. The FvK number is the combination γ=YR2/κ, where Y is the two-dimensional Young modulus of the shell, κ is its bending rigidity, and R is the mean radius of the capsid shell [23]. High bending rigidity favors smooth, practically spherical shell shape while low bending rigidity leads to the faceted shape. In the model developed in our work capsid shell thinning is shown to be one of the two important external parameters of the morphological transformation. It is evident that the shell thinning results in the decrease of its bending rigidity κ. The decrease in κ leads to the shell faceting in a good accord with the predictions of [23]. In the recent work [15] the classical approach developed in [23,24] was submitted to a certain criticism. To propose an alternative theory the authors of [15] used the fact observed in in vitro experiments [3,8,25] that during the transition between two intermediate states EI and EII of the HK97 bacteriophage the protein hexamers constituting (together with pentamers) the viral capsid, change their shape from skewed to more regular one. Continuous elastic model developed in [15] proposed to relate the faceting transition not to the FvK number but to the hexamer shape change. However, X-ray crystallography and high-resolution cryoEM data reveal one more intermediate state of the HK97 maturation process which is characterized by the spherical capsid shape and the regular hexamer shape simultaneously [12]. This make experimental relevance of the model [15] not convincing enough. We expect that further development of cryoEM technique will result in the near future in new high-resolution data on intermediate states of the HK97 morphological transformation. These additional data would constitute the basis for the further development of the minimal model proposed in the present work. The simplest extension would be the model which takes into account in (9) the terms of higher order in ∆V. It is easy to see that by adding fourth–order terms in ∆V in (9) one obtains additional states which differ by their volume values. They might correspond to several spherical shells with different volumes observed in in vitro experiments [3,8,25]. In conclusion, let us note that the minimal model with a clear physical meaning of the free energy parameters proposed in the present work describes the viral capsid morphological transformation during maturation process for the HK97 bacteriophage as well as for a series of 10 similar viruses. Underlying physical processes are driven by the order parameters spanning irreducible representation of the SO(3) symmetry group of the spherical shell constituted by asymmetric viral coat proteins. The morphological transformation during maturation process is understood as a sequence of phase transitions leading to the isotropic shell expansion and the symmetry-breaking faceting. The first fully symmetric order parameter is characterized by the shell volume change ∆V. The second 13-dimensional order parameter responsible for the procapsid shape symmetry breaking describes explicitly icosahedral faceting of the viral shell. It spans the irreducible representation of the SO(3) group with l=6 and represents the linear combination of the spherical harmonics with l=6 invariant with respect to the icosahedral symmetry group Ih. In the state with the icosahedral symmetry the components of this order parameter depend on only one amplitude. In the minimal model of the transformation this fact is described by the effective one-dimensional order parameter η. The model is then reduced to the coupling between the fully-symmetric order parameter responsible for the capsid volume change and one-dimensional order parameter responsible for the shell faceting. It admits third-order term in η in the free energy thus making the morphological transformation discontinuous. The third- order term sign discriminates between icosahedral and dodecahedral final shapes of the faceted capsid. The calculated phase diagram shows two qualitatively different paths of the transformation in a function of two main external thermodynamic parameters of the in vitro and in vivo experiments. The ensemble of the results obtained describes the experimentally observed physical phenomena which accompany maturation process in the HK97 bacteriophage and similar viruses. APPENDIX: DISPLACEMENT FIELDS WITH THE SYMMETRY ON A SPHERICAL SURFACE Following analysis performed in terms of the invariant theory makes it possible to construct an arbitrary function with the icosahedral symmetry. It provides a justification for selection rules (3) which determine the possible wave number l values associated with the order parameter responsible for the procapsid shape symmetry breaking from SO(3) to Ih. ICOSAHEDRAL We start with general properties of an arbitrary scalar function defined on a spherical surface (e.g. function of radial displacements of the capsid shell material points) and invariant with respect to the symmetry group Ih, which contains all symmetry operations of an icosahedron. Let us fix the point group orientation with respect to the coordinate frame and choose Cartesian axes x, y and z along the two-fold symmetry axes of an icosahedron. This choice allows us to express the invariants of the full icosahedral group in a simple way. Another property of the group helps us to construct so-called integrity basis constituted by the generators of the ring of invariant polynomials. Any function with the icosahedral symmetry can then be expanded in series of the finite number of invariant polynomials constituting the integrity basis. Point group Ih belongs to the class of simple mathematical objects called groups generated by reflections. For the groups of this type the number of invariants in the integrity basis is equal to the dimension of its vector representation (i.e. to the dimension of the space in the considered case) and the product of degrees of basis invariants is equal to the number of the group elements G. Consequently, the integrity basis of the Ih group contains only three following invariant polynomials with the degrees 2, 6 and 10, respectively: J 0 = 2 x + 2 y + 2 z , ∏ = 6 J rq , ∏ = 10 J 1 i 2 = 1 i = 1 i rp , i (11) 11 =< > r zyx , , iq and ; vectors herе ip are parallel to the 5-fold and 3-fold icosahedral axes, respectively. From the geometric point of view each term in the products in the expressions for the invariant polynomials 2J is equivalent to the equation of a plane perpendicular to the 5-fold and 3-fold axes. 1J and r fl + + + + 1 > + 2 + 2 1 1 2 =< sin=x sin=y ϕθsin , ϕθcos , On the unit sphere the invariant According to the well-known presentation of spherical harmonics as homogeneous (given by Eq. (5)) with the fixed wave number of the degree l. Taking polynomial functions any icosahedral function lf zyx , , value l can be expressed as a homogeneous polynomial in unit radius-vector length we express r in spherical coordinates as: θcos=z . 0J acquires constant value 0J =1. Then, any function on a spherical surface invariant with respect to the icosahedral symmetry group Ih can be presented in the form: =φθ (12) ) ,( The degree l of the lf function is equal the degree (in a function of the radius-vector components) of the last terms in expansion (12). The number of terms of the degree l in (12) is the number of possible integer linear combinations of numbers 6 and 10 equal to l. Consequently, for any homogeneous function of the degree l invariant with respect to the Ih group, the number l satisfies to the condition l=6i+10j, where i and j are non-negative integers. Due to the function irreducibility, coefficients Ai multiplying the terms with the degrees smaller than l in (12) are univocally defined by the orthogonality of function (12) to basic functions of other irreducible representations with the wave numbers l’<l. Following orthogonality relations hold for a given irreducible function F: JJA 4 JA 2 JA 1 JA 6 JA 3 JA 5 3 1 ... 2 2 A 0 θφθ = dd φθ Y ),( ml ' ∫∫ F ,0 sin for all l’ < l (13) In practice the number of equations in system (13) is much higher than the number of unknown coefficients in function (12). However, the additional equations are either linearly equivalent or become identity. Acknowledgments S.R. and O.K. acknowledge financial support by the RFBR grant 13-02-12085 ofi_m. (Russia). V.L. acknowledges financial support by the Laboratory of Excellence NUMEV (France), partial support by National Science Foundation Grant No. PHYS-1066293 and the hospitality of the Aspen Center for Physics. References [1] S. J. Flint, L. W. Enquist, V. R. Racaniello, and A. M. Skalka, Principles of Virology: Molecular Biology, Pathogenesis, and Control (ASM, Washington, 2000); J.B. Bancroft, Advances of Virus Research (Academic, New York, 1970), Vol. 16, p. 99. [2] J. King and W. Chiu, in Structural Biology of Acknowledgments Viruses, edited by W. Chiu, R.M. Burnett, and R. Garcea (Oxford University Press, New York, 1997), p. 288. [3] J.F. Conway, W.R. Wikoff, N. Cheng, R.L. Duda, R.W. Hendrix, J.E. Johnson, and A.C. Steven, Science 292, 744 (2001). 12 [4] L. Li, S. Lok, I. Yu, Y. Zhang, R.J. Kuhn, J. Chen, and M.G. Rossmann, Science 319, 1830 (2008); I. Yu, W. Zhang, H.A. Holdaway, L. Li, V.A. Kostyuchenko, P.R. Chipman, R.J. Kuhn, M.G. Rossmann, and J. Chen, Science 319, 1834 (2008). [5] D.L.D. Caspar and A. Klug, Cold Spring Harbor Symp. Quant. Biol. 27, 1 (1962). [6] R.L. Duda, J. Hempel, H. Michel, J. Shabanowitz, D. Hunt, and R.W. Hendrix, J. Mol. Biol. 247, 618 (1995). [7] W.R. Wikoff, L. Liljas, R.L. Duda, H. Tsuruta, R.W. Hendrix, and J.E. Johnson, Science 289, 2129 (2000). [8] 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, 646 (2009). [9] J.E. Johnson, Curr. Opin. Struct. Biol. 20, 210 (2010); C.M. Teschke, and K.N. Parent, Virology 401, 119 (2010). [10] C. Helgstrand, W.R. Wikoff, R.L. Duda, R.W. Hendrix, J.E. Johnson, and L. Liljas, J. Mol. Biol. 334, 885 (2003). [11] R. Lata, J.F. Conway, N. Cheng, R.L. Duda, R.W. Hendrix, W.R. Wikoff, J.E. Johnson, H. Tsuruta, and A.C. Steven, Cell 100, 253 (2000). [12] W.R. Wikoff, J.F. Conway, J. Tang, K.K. Lee, L. Gan, N. Cheng, R.L. Duda, R.W. Hendrix, A.C. Steven, and J.E. Johnson, J. Struct. Biol. 153, 300 (2006). [13] F. Tama and C. L. Brooks III, J. Mol. Biol. 345, 299 (2005). [14] T. Guérin and R. Bruinsma, Phys. Rev. E 76, 061911 (2007). [15] A. Aggarwal, J. Rudnick, R.F. Bruinsma, and W.S. Klug, Phys. Rev. Lett. 109, 148102 (2012). [16] V. L. Lorman and S. B. Rochal, Phys. Rev. Lett. 98, 185502 (2007); V. L. Lorman and S. B. Rochal, Phys. Rev. B 77, 224109 (2008). [17] S.B. Rochal, V.L. Lorman, and G. Mennessier, Phys. Rev. E 71, 021905 (2005). [18] J.L. Birman, Theory of Crystal Space Groups and Lattice Dynamics: Infra-Red and Raman Optical Processes of Insulating Crystals (Springer-Verlag, Heidelberg, 1974) [19] J. P. Elliot and P. G. Dawber, Symmetry in Physics (Macmillan, London, 1979). [20] G. Cardone, R.L. Duda, N. Cheng, L. You, J.F. Conway, R.W. Hendrix, and A.C. Steven, MBio. 5, e02067 (2014). [21] B.L. Gurda et al., J. Virol. 84, 5880 (2010). 13 [22] Yu. M. Gufan, Structural Phase Transitions (Nauka, Moscow, 1982); J. C. Tolédano and P. Tolédano, The Landau Theory of Phase Transitions (World Scientific, Singapore, 1987); P.G. de Gennes and J. Prost, The physics of Liquid Crystals, (Clarendon, Oxford, 1993), 2nd ed. [23] J. Lidmar, L. Mirny, and D. R. Nelson, Phys. Rev. E 68, 051910 (2003). [24] M. Widom, J. Lidmar, and D. R. Nelson, Phys. Rev. E 76, 031911 (2007). [25] K. K. Lee, L. Gan, H. Tsuruta, C. Moyer, J. F. Conway, R. L. Duda, R. W. Hendrix, A. C. Steven, and J. E. Johnson, Structure 16, 1491 (2008). 14
physics/0504117	3	0504	2010-03-19T04:21:25	Transport Reversal in a Thermal Ratchet	[ "physics.bio-ph", "cond-mat.stat-mech", "physics.data-an" ]	Transport of a Brownian particle moving in a periodic potential is investigated in the presence of symmetric unbiased external force. The viscous medium is alternately in contact with the two heat reservoirs. We present the analytical expression of the net current at quasi-steady state limit. It is found that the competition of the asymmetric parameter of potential with the temperature difference leads to the phenomena like current reversal. The competition between the two driving factors is a necessary but not a sufficient condition for current reversals.	physics.bio-ph	physics	Transport Reversal in a Thermal Ratchet Baoquan Ai1,2, Liqiu Wang1, and Lianggang Liu3 1 Department of Mechanical Engineering, The University of Hong Kong, Pokfulam Road, Hong Kong 2School of Physics and Telecommunication Engineering, South China Normal University, GuangZhou, China 3 Department of Physics, ZhongShan University, GuangZhou, China Abstract Transport of a Brownian particle moving in a periodic potential is investigated in the presence of symmetric unbiased external force. The viscous medium is alternately in contact with the two heat reservoirs. We present the analytical expression of the net current at quasi-steady state limit. It is found that the competition of the asymmetric parameter of potential with the temperature difference leads to the phenomena like current reversal. The competition between the two driving factors is a necessary but not a sufficient condition for current reversals. PACS numbers: 05. 40. -a, 05. 70. -a, 87. 10. +e Keywords: Current reversal, thermal ratchet, transport processes 0 1 0 2 r a M 9 1 ] h p - o i b . s c i s y h p [ 3 v 7 1 1 4 0 5 0 / s c i s y h p : v i X r a 1 1. INTRODUCTION Transport phenomena play a crucial role in many processes from physical, biological to social systems. There has been an increasing interest in transport properties of nonlinear systems which can extract usable work from unbiased nonequilibrium fluctuations [1 -- 4]. This comes from the desire of understanding molecular motors [5], nanoscale friction [6], surface smoothening [7], coupled Josephson junctions [8], optical ratchets and directed motion of laser cooled atoms [9], and mass separation and trapping schemes at microscale [10]. The focus of research has been on the noise-induced unidirectional motion over the last decade. A ratchet system is generally defined as a system that is able to transport particles in a periodic structure with nonzero macroscopic velocity in the absence of macroscopic force on average. In these systems, directed Brownian motion of particles is generated by nonequilibrium noise in the absence of any net macroscopic forces and potential gradients. Typical examples are rocking ratchets [4, 11], flashing ratchets [12], diffusion ratchets [13], correlation ratchets [4, 14] and white-shot-noise ratchets [2]. In all these studies, the potential is taken to be asymmetric in space. It has also been shown that a unidirectional current can also appear for spatially symmetric potentials if there exits an external random force either asymmetric or spatially-dependent. If spatially periodic structures are exposed to additive Poissonian white shot noise of zero average, a macroscopic current occurs even in the absence of spatial asymmetry [2]. The current reversal is very important in new particle separation devices such as elec- trophoretic separation of micro-particles [15]. It is also of interest in biology [32]. Motions of macromolecules are probably responsible for the vesicle transport inside eukaryotic cells. A typical example is the motion of proteins along a microtubule, modelled usually by a ratchet 2 [29]. It is well known that the two typical proteins, kinesins and dyneins, move along tubulin filaments in opposite directions. This can be explained by the current reversal. The current reversal in ratchet systems can be engendered by varying system parameters [16 -- 28]. The current can be reversed, for example, by a noise of Gaussian force with non- white power spectrum in present of stationary periodic potential [19]. The current reversal can also be obtained in two-state ratchets if the long arm is kinked [20]. Bier and Astumian [21] have also found the current reversal in a fluctuating three-state ratchet. In the presence of a kangaroo process as the driving force, the current reversal can be triggered by varying the noise flatness, the ratio of the fourth moment to the square of the second moment[22]. The current reversal can be induced by both an additive Gaussian white and an additive Ornstein-Uhlenbeck noise in a correlation ratchet [23]. The current reversal also appear in forced inhomogeneous ratchets [17, 18]. The previous works on the current reversal are limited to case of one heat reservoir. The present study extends the study of current reversal to the case of two heat reservoirs. When a positive driving factor competes with a negative one, the current may reverse its direction. The competition between the competitive driving factors is necessary but not sufficient for the current reversal. Our emphasis is on finding conditions of generating current reversal. This is achieved by using a quasi-steady state limit to solve the Fokker-Planck equation. 2. NET CURRENT OF THE THERMAL RATCHET Consider a Brownian particle moving in a sawtooth potential with an unbiased external force where the medium is alternately in contact with the two heat reservoirs. This model was first proposed to describe molecular motor in biological system [29]. The particle motion 3 satisfies the dimensionless Langevin equation of motion [30, 31] m d2x dt2 = −β dx dt − dU(x) dx + F (t) + q2kBT (x)βξ(t), (1) where x stands for the position of Brownian particle, m the mass of the particle, β the viscous friction drag coefficient, kB Boltzmann constant, T (x) absolute temperature. ξ(t) is a randomly-fluctuating Gaussian white noise of zero mean and the autocorrelation function < ξ(t)ξ(s) >= δ(t − s). Here < ... > denotes an ensemble average over the distribution of the fluctuating forces ξ(t). F (t) is an external periodic force (Fig. 1b), satisfy F (t + τ ) = F (t),Z τ 0 F (t)dt = 0. (2) The geometry of symmetric potential U(x) = U(x + L) is displayed in Fig. 1a and U(x) within the interval 0 ≤ x ≤ L is described by U(x) =   U L1 U L2 x, 0 ≤ x < L1; (L − x), L1 ≤ x ≤ L, (3) where L = L1 + L2 is the period of the potential. The temperature T (x) has the same period as the potential U(x). Therefore, T (x) = T (x + L), T (x) =   T + δ, 0 ≤ x < L1; T, L1 ≤ x ≤ L. (4) Because the motion of the ratchet is highly overdamped in general [30], the inertia term can be neglected. Hence, Eq. (1) reduces to, when β = 1 and kB = 1, dx dt = − dU(x) dx + F (t) + q2T (x)ξ(t). The probability density satisfies the associated Fokker-Planck equation [30, 31], ∂P (x, t) ∂t = ∂ ∂x ′ [(U (x) − F (t))P (x, t) + ∂ ∂x (T (x)P (x, t))] = − ∂j(x, t) ∂x , 4 (5) (6) ) x ( U U0 T+ T L1 L2 x (a) F0 ) t ( F -F0 /2 3 /2 2 t (b) FIG. 1: Potential and Driving force: (a)potential U (x) = U (x + L); U (x) is a piecewise linear and periodic potential; the period of the potential is L = L1 + L2; ∆ = L1 − L2; the temperature profiles is also shown. (b) Driving force F(t) which preserved the zero mean < F (t) >= 0; F (t + τ ) = F (t); F0 is amplitude of F (t). 5 j(x, t) = −[U ′ (x) − F (t)]P (x, t) − d dx [T (x)P (x, t)], (7) here j is the probability current density. The prime stands for the derivative with respect to the space variable x. P (x, t) is the probability density for the particle at position x and at time t. It satisfies the normalization condition and the periodicity condition, P (x, t) = P (x + L, t), Z L 0 P (x, t)dx = 1. (8) (9) If F (t) changes very slowly with respect to t, namely, its period is longer than any other time scale of the system, there exists a quasi-steady state. In this case, by following the method in [30, 31], we can obtain the current j(F (t)) from Eq. (7)-Eq. (9), (10) (11) j(F (t)) = −Q G1G2 + HQ , where Q, G1, G2 and H are Q = ea−b − 1, L + ∆ 2a(T + δ) L + ∆ G1 = G2 = (1 − e−a) + L − ∆ 2bT e−a(eb − 1), (ea − 1) + 2a L − ∆ 2b ea(1 − e−b), H = A + B + C, A = 1 T + δ ( L + ∆ 2a 6 )2(a + e−a − 1), B = C = L2 − ∆2 4abT (1 − e−a)(eb − 1), 1 T ( L − ∆ 2b )2(eb − 1 − b), a = b = 2U0 − F (t)(L + ∆) 2(T + δ) 2U0 + F (t)(L − ∆) 2T , . The average current is J = 1 τ Z τ 0 j(F (t))dt, (12) (13) (14) where τ is the period of the driving force F (t). τ is assumed to be longer than any other time scale of the system at quasi-steady state. For the external force F (t) shown in Fig. 1b, J = 1 2 [j(F0) + j(−F0)]. (15) When both the potential and the temperature are symmetric (δ = 0, F0 = 0), the current J reduces to: J = 1 2(2T + δ) ( U0 2L )2( 1 U0 T +δ − 1 e − 1 U0 T − 1 e ). (16) Therefore, the net current is not zero even when both the potential and the temperature are spatially symmetric. The direction of the current is determined by the sign of δ. The particle tends to move from high temperature region to low temperature region. In fact, this agrees with the diffuse law. 3. RESULTS AND DISCUSSION Figure 2 shows the current J as a function of the asymmetric parameter ∆ of the potential at δ = 0. The current is negative for ∆ < 0, zero at ∆ = 0 and positive for ∆ > 0. Therefore, we can have the current reversal by changing the sign of ∆, the asymmetry of the potential. 7 J 0.3 0.2 0.1 0 -0.1 -0.2 -1.0 -0.5 0.5 1.0 FIG. 2: Current J versus asymmetric parameter ∆ of the potential at U0 = 5, F0 = 3.0, L = 1.0, T = 1.0 and δ = 0. J 8 6 4 2 0 -2 -4 -6 -8 -10 10 20 30 FIG. 3: Current J versus temperature difference δ at U0 = 5, F0 = 3.0, L = 1.0, T = 10 and ∆ = 0. 8 Figure 3 shows the current J versus temperature difference δ in a symmetric potential ∆ = 0. The temperature difference δ controls not only the magnitude but also the direction of the current. When δ = 0 and ∆ = 0, there is no current. For the asymmetric potentials, varying temperature difference is another way of inducing a net current. 0.4 0.3 0.2 0.1 0.0 -0.1 -0.2 -0.3 -0.4 J =0.1, =0.9 =0.1, =0.1 2 4 6 8 T 10 =-0.1, =-0.1 =-0.1, =-0.9 FIG. 4: Current J versus temperature T for different asymmetric parameters δ and∆ at U0 = 5, F0 = 3.0 and L = 1.0. The current J as the function of T is shown in Fig. 4 for different combinations of ∆ and δ. The curve is observed to be bell-shaped, which shows the feature of resonance. When T → 0, J tends to zero for all values of δ and ∆. Therefore, the particle can not pass the barrier and there is no currents. When T → ∞ so that the thermal noise is very large, the ratchet effect disappear and J → 0, also. There is an optimized value of T at which the current J takes its maximum value. There is no current reversal at δ = 0.1, ∆ = 0.9; δ = 0.1, ∆ = 0.1; δ = −0.1, ∆ = −0.1; and δ = −0.1, ∆ = −0.9. In fact, the temperature cannot lead to the current reversal if ∆δ > 0 (Fig. 2-Fig. 4). 9 0.14 0.12 0.10 0.08 0.06 0.04 0.02 0.00 -0.02 -0.04 -0.06 -0.08 -0.10 -0.12 -0.14 0.12 0.10 0.08 0.06 0.04 0.02 0.00 J J =0.1, =-0.3 =0.1, =-0.8 2 4 6 8 =-0.1, =0.8 =-0.1, =0.2 (a) 0.0025 0.0020 0.0015 0.0010 0.0005 0.0000 -0.0005 -0.0010 -0.0015 2 0.2 0.3 0.4 0.5 0.6 0.7 4 6 8 (b) T 10 T 10 FIG. 5: (a)Current J versus temperature T for different values of the asymmetric parameters ∆ at U0 = 5, F0 = 3.0, L = 1.0. (b)Current J versus temperature T at U0 = 5, F0 = 3.0, L = 1.0, δ = 0.1 and ∆ = −0.4. 10 In Fig. 5a, we plot the current J as a function of temperature T for different combinations of ∆ and δ. When temperature difference (δ = 0.1) is positive, the current may reverse its direction as increasing temperature for negative ∆ (∆ = −0.8). It is observed that the current reversal may occur for negative δ and positive ∆ (δ = −0.1, ∆ = 0.8). We can also have the current reversal twice at δ = 0.1, ∆ = −0.4 (Fig. 5b). Therefore, there may exist current reversal for ∆δ < 0. However, ∆δ < 0 is not sufficient condition for current reversal. For example, the current is always positive for δ = 0.1, ∆ = −0.3 and negative for δ = −0.1, ∆ = 0.2 (Fig. 5a). 0.8 0.6 0.4 0.2 0.0 -0.2 c -0.4 b -0.6 -0.8 0.26 0.21 0.16 0.14 0.11 -0.030 0 0.065 1 Ta 2 3 T 4 5 FIG. 6: Current contours on ∆-T plane at U0 = 5, F0 = 3.0, L = 1.0 and δ = 0.1: ∆c (-0.3987) is maximum ∆ for the curve J = 0; ∆b (-0.4913) is asymmetric parameter of potential at the cross point of J = 0 and T = 0; Ta is temperature at the cross point of J = 0 and ∆ = −1.0. In order to illustrate the current reversal in detail, the current contours are shown in Figs. 6 and 7, respectively. When T > Ta or ∆ > ∆c, the current is always positive there is no current reversal (Fig. 6, also see the case δ = 0.1, ∆ = −0.3 in Fig. 5a). The current 11 reversal may, however, occur by varying T or ∆ when T < Ta or ∆ < ∆c (Fig. 6, also see the case δ = 0.1, ∆ = −0.8 in Fig. 5a). In particular, the current may reverse its direction twice if ∆b < ∆ < ∆c (Fig6, also see the case δ = 0.1, ∆ = −0.4 in Fig .5b). c b 0.20 0.15 0.10 0.05 a 0.00 -0.13 -0.16 0.18 0.070 0 0.13 -0.051 -0.10 -0.05 -0.22 1 2 3 T 4 5 FIG. 7: Current contours on δ-T plane at U0 = 5, F = 3.0, L = 1.0 and ∆ = −0.6: δa = 0; δb (0.138) is temperature difference at the cross point of J = 0 and T = 0; δc (0.151) is maximum temperature difference for J = 0. The current is always negative and there is no current reversal for δ ≤ 0 and ∆ = −0.6 (Fig. 7). When δa < δ < δc, the current may reverse its direction as increasing temperature. The current may changes its direction twice, in particular, when δb < δ < δc. Therefore, we can not have the current reversal when δ∆ ≥ 0. When δ∆ < 0, the current may reverse its direction. However, δ∆ < 0 is not the sufficient condition for current reversal. 12 4. CONCLUDING REMARKS The transport of a Brownian particle moving in a periodic potential is studied in the presence of an unbiased fluctuation and two heat reservoirs. In a quasi-steady state limit, we obtain the current analytically. It is found that the asymmetric parameter ∆ of the potential and the temperature difference δ are the two pivotal factors for obtaining a net current. For the two positive or the two negative driving factors such that ∆δ > 0, current cannot reverse its direction. The current reversal cannot occur either if there is only one driving factor such that ∆δ = 0. For the two opposite driving factors so that ∆δ < 0, the current reversal may occur. The current reversal can also occur twice at certain conditions. The condition ∆δ < 0 is a necessary but not a sufficient condition for the current reversal. [1] J. L. Mateos, Phys. Rev. Lett. 84, 258 (2000). [2] J. Luczka, R. Bartussek and P. Hanggi, Europhysics Letters 31 (8), 431-436 (1995). [3] L. P. Faucheux et al., Phys. Rev. Lett. 74 (1995) 1504. [4] B. Q. Ai, X. J. Wang, G. T. Liu and L. G. Liu, Phys. Rev. E, 68, 061105 (2003); B. Q. Ai, X. J. Wang, G. T. Liu and L. G. Liu, Phys. Rev. E, 67, 022903 (2003); B. Q. Ai, G. T. Liu, H. Z. Xie, L.G. Liu, Chaos 14(4),957 (2004). [5] J. Maddox, Nature (London) 368 287(1994); R. D. Astumian and I. Derenyi, Eur. Biophys. J. 27, 474 (1998); N. Thomas and R. A. Thornhill, J.Phys. D 31, 253 (1998); C.R. Doering, B. Ermentrout, G. Oster, Biophys. J. 69, 2256(1995); [6] J. Krim, D.H. Solina, R. Chiarello, Phys. Rev. Lett. 66 (1991) 181; J.B. Sokolo2, J. Krim, A. Widom, Phys. Rev. B 48 (1993) 9134; L. Daikhin, M. Urbakh, Phys. Rev. E 49 (1994) 1424; 13 C. Daly, J. Krim, Phys. Rev. Lett. 76 (1996) 803; M.R. Sorensen, K.W. Jacobsen, P. Stoltze, Phys. Rev. B 53 (1996) 2101. [7] I. Derenyi, C. Lee, and A.-L. Barabasi, Phys. Rev. Lett. 80 (1998) 1473. [8] I. Zapata, R. Bartussek, E. Sols, P. HFanggi, Phys. Rev. Lett. 77, 2292 (1996). [9] L.P. Faucheux, L.S. Bourdieu, P.D. Kaplan, A.J. Libchaber, Phys. Rev. Lett. 74,1504 (1995); C. Mennerat-Robilliard, D. Lucas, S. Guibal, J. Tabosa, C. Jurczak, J.-Y. Courtois, G. Gryn- berg, Phys. Rev. Lett. 82, 851(1999). [10] L. Gorre-Talini, J.P. Spatz, P. Silberzan, Chaos 8, 650 (1998); I. Derenyi, R.D. Astumian, Phys. Rev. E 58, 7781 (1998); D. Ertas, Phys. Rev. Lett. 80, 1548 (1998); T.A.J. Duke, R.H. Austin, Phys. Rev. Lett. 80, 1552 (1998). [11] M. O. Magnasco, Phys. Rev. Lett. 71, 1477 (1993). [12] P. Hanggi and R. Bartussek, Nonlinear physics of complex system - Current status and Future Trends, 476, Spring, Berlin, (1996), 294. [13] P. Reimann, R. Bartussek, R. Haussler and P. Hanggi, Phys. Lett. A 215, 26 (1994). [14] C. R. Doering, W. Horsthemke and J. Riordan, Phys. Rev. Lett. 72, 2984 (1994). [15] C. kettner, Phys. Rev. E 61, 312 (2000). [16] R. Tammelo, R. Mankin and D. Martila, Phys. Rev. E 66, 051101 (2002). [17] D. Dan, M. C. Mahato and A. M. Jayannavar, Phys. Rev. E 63, 056307 (2001). [18] B. Q. Ai, X. J. Wang, G. T. Liu, H. Z. Xie, D. H. Wen, W. Chen, and L. G. Liu., Eur. Phys. J. B 37, 523-526 (2004). [19] M. M. Millonas and M. I. Dykman, Phys. Lett. A 185, 65 (1994). [20] J.-F. Chauwin, A. Ajdari, and J. Prost, Europhys. Lett. 32, 373 (1995). [21] M. Bier and R. D. Astumian, Phys. Rev. Lett. 76, 4277 (1996). 14 [22] C. R. Doering, W. Horsthemke, and J. Riordan, Phys. Rev. Lett. 72, 2984 (1994). [23] R. Bartussek, P. Reimann, and P. Hanggi, Phys. Rev. Lett. 76, 1166 (1996). [24] R. Mankin, A. Ainsaar, A. Haljas and E. Reiter, Phys. Rev. E 63, 041110 (2001). [25] J. Kula, T. Czernik and J. Luczka, Phys. Rev. Lett. 80, 1377 (1998). [26] M. Kostur and J. Luczka, Phys. Rev. E 63, 021101 (2001). [27] I. Derenyi and A. Ajdari, Phys. Rev. E 54, R5 (1996). [28] F. Marchesoni, Phys. Lett. A 237, 126 (1998); H. A. Larrondoa, Fereydoon Family, C. M. Arizmendia, Physica A 303, 67-68 (2002). [29] Oster, G., H. Wang. How protein motors convert chemical energy into mechanical work. In Molecular Motors, M. Schliwa, ed. pp. 207-228. ISBN 3-527-30594-7. Wiley-VCH (2002). [30] H. Risken, The Fokker-Planck Equation, Springer-Verlag, Berlin, 1984. [31] M. Asfaw and M. Bekele, Eur. Phys. J. B 38, 457 (2004). [32] U. Henningsen, M. Schliwa, Nature 389, 93 (1997) 15
1209.1179	1	1209	2012-09-06T05:03:47	Statistical Physics of Self-Replication	[ "physics.bio-ph", "cond-mat.stat-mech", "q-bio.PE" ]	Self-replication is a capacity common to every species of living thing, and simple physical intuition dictates that such a process must invariably be fueled by the production of entropy. Here, we undertake to make this intuition rigorous and quantitative by deriving a lower bound for the amount of heat that is produced during a process of self-replication in a system coupled to a thermal bath. We find that the minimum value for the physically allowed rate of heat production is determined by the growth rate, internal entropy, and durability of the replicator, and we discuss the implications of this finding for bacterial cell division, as well as for the pre-biotic emergence of self-replicating nucleic acids.	physics.bio-ph	physics	Statistical Physics of Self-Replication Jeremy L. England Department of Physics, Massachusetts Institute of Technology, Building 6C, 77 Massachusetts Avenue, Cambridge, MA 02139 (Dated: May 2, 2014) Self-replication is a capacity common to every species of living thing, and simple physical intuition dictates that such a process must invariably be fueled by the production of entropy. Here, we undertake to make this intuition rigorous and quantitative by deriving a lower bound for the amount of heat that is produced during a process of self-replication in a system coupled to a thermal bath. We find that the minimum value for the physically allowed rate of heat production is determined by the growth rate, internal entropy, and durability of the replicator, and we discuss the implications of this finding for bacterial cell division, as well as for the pre-biotic emergence of self-replicating nucleic acids. Every living thing bears some resemblance to its ances- tors; this is a basic premise of biology. From the stand- point of physics, however, self-replication presents a chal- lenge. Reduced to its microscopic details, an organism cannot be distinguished from its environment: a priori, a fluctuating cluster of atoms does not “know” it is doing anything to affect the assembly of a similar-looking clus- ter. As biologists, we watch a bacterial cell divide and say that it facilitates its own duplication, but as physi- cists, plotting the course from one microstate to the next, we cannot attribute any more agency to the atoms of the bacterium than we can to the atoms in the sugar it eats; all we see is interacting particles, exploring a series of arrangements permitted by conservation of momentum and energy. To resolve this difficulty, we must recognize that the “self” in self-replication is not anywhere implicit in the atomistic physical description of the system. Rather, it arises only once an observer carries out a further classi- fication of microstates by providing a definition for some pattern of interest. Such a coarse-graining of phase space should be familiar to any student of statistical mechan- ics, except that here, there need not be any way of phys- ically summarizing (with, for example, an order parame- ter) the function of microscopic variables used to define the coarse-graining. To make things explicit, we might imagine a thought- experiment in which we showed every possible microstate for some system to a microbiologist who was asked to des- ignate, in each case, how many live, healthy, wild-type E. coli bacteria were present. Though the microbiologist’s assessment would be based partly or wholly on qualita- tive criteria, we would nevertheless come away from the procedure with a well-defined value for our cell count at each point in phase space. In this way, holistic, bio- logical judgments could be rendered into numbers that might then be incorporated into a quantitative model of the system’s dynamics. Here, we will use a coarse-graining like the one sketched above to investigate the statistical physics of self-replication. Based on a general argument from mi- croscopic reversibility, we will derive a lower bound on the heat output of a self-replicator in terms of its size, growth rate, entropy, and durability. We will further- more show, through analysis of empirical data, that this bound operates on a scale relevant to the functioning of real microorganisms and other self-replicators. We begin by considering the preparation of a large, finite system initially containing a single E. coli bac- terium, immersed in a sample of rich nutrient media in contact with a heat bath held at the bacterial cell’s opti- mal growth temperature (1/β ≡ T ∼ 4.3 × 10−21 joules) [1, 2]. We can assume furthermore that the cell is in ex- ponential growth phase at the beginning of its division cycle, and that, while the volume and mass of the entire system are held fixed, the composition and pressure of the nutrient media mimics that of a well-oxygenated sample open to the earth’s atmosphere. If we summarize the ex- perimental conditions described above with the label I, we can immediately say that there is some probability p(iI) that the system is found in some particular mi- crostate i given that it was prepared in the macroscopic condition I by some standard procedure. Although this probability might well be impossible to derive ab initio, in principle it could be measured through repeated con- sultations of a microbiologist as described above. Now suppose we consider what would happen in our system if we started off in some microstate i and ob- served it again after a time interval of τdiv, the typical duration of a single round of growth and cell division. From the biological standpoint, the expected final state for the system is clear: two bacteria floating in the media instead of one, and various surrounding atoms rearranged into new molecular combinations (e.g. some oxygen con- verted into carbon dioxide). While very likely, however, such an outcome is not certain, and in general we have to consider that each microstate j will have some finite likelihood p(jII). Since our system is coupled to a heat bath, it obeys stochastic dynamics described by the tran- sition matrix π(→ ji), which is the conditional probabil- ity of ending up in microstate j (with energy Ej) at time t = τdiv given that one started out in microstate i (with 2 1 0 2 p e S 6 ] h p - o i b . s c i s y h p [ 1 v 9 7 1 1 . 9 0 2 1 : v i X r a (cid:90) energy Ei) at time t = 0. In these terms, our ensemble II of possible arrangements after time τdiv can be defined via p(jII) = di p(iI)π(→ ji), which we take to have the normalization(cid:82) dj p(jII) = 1. (1) I By stipulation, there are no external driving forces acting on the system, and it can furthermore be as- sumed that the changes in our bacterial incubator over the course of time τdiv are dominated by diffusive mo- tions that lack any sense of momentum [3]. Thus, the transition matrix π(→ ji) must obey a detailed balance relation π(→ ij) π(→ ji) (cid:90) (cid:90) I π(→ III) = di dj = exp[−β∆Qi→j] (cid:19) (cid:18) p(jII) p(iI) p(iI)π(→ ij) = (2) (cid:90) di (cid:90) I to write(cid:28) (cid:29) where (cid:104). . .(cid:105)I→II denotes an average over all paths from some i in the initial ensemble I to some j in the final ensemble II, with each path weighted by its likelihood. the usual manner (S ≡ −(cid:80) Defining the Shannon entropy S for each ensemble in i pi ln pi), we can construct ∆Sint ≡ SII − SI, which measures the internal entropy change for the replication reaction. Since it is generally the case that ex ≥ 1 + x for all x, we may rearrange (4) e−β∆Qi→j−ln π(→III)−ln p(jII)+ln p(iI) = 1 (5) I→II and immediately arrive at β(cid:104)∆Q(cid:105) + ln [π(→ III)] + ∆Sint ≥ 0 (6) In one sense, this result simply says what we might have guessed: that the average total entropy production for the forward process (cid:104)∆Stot(cid:105) ≡ ∆Sint + β(cid:104)∆Q(cid:105) sets a bound on how likely things would be to run in reverse: since the probability π(→ III) ≤ 1, it follows that (cid:104)∆Stot(cid:105) ≥ 0. Put another way, what we have here is simply a precise statement of the Second Law of Ther- modynamics. It should therefore perhaps not be surpris- ing to learn that the formula applies under conditions more general than those for which it was derived: in any diffusive, microscopically reversible system driven from equilibrium by a time-symmetric drive, it can be shown that π(→ ij)/π(→ ji) = (cid:104)exp[−β∆Qi→j](cid:105)i→j by using the irreversibility formula derived by Crooks [3]. Thus, although we have prefaced this investigation with a dis- cussion of self-replication, our bound holds for a range of (cid:90) (cid:90) 2 where ∆Qi→j ≡ Ei − Ej is the heat released into the surrounding bath in the course of going from i to j. This follows from the underlying microscopic reversibility of particle dynamics in our system, along with the fact that the thermal equilibrium distribution p(i) ∝ exp[−βEi] must be stationary under π(→ ji) [4]. To appreciate the thermodynamic consequences of the assumptions we have already made, it is necessary to consider the reverse probability π(→ III) = dj p(jII)π(→ ij), (3) di I that is, the minuscule probability that the system returns to one of the microstates satisfying the macroscopic con- dition I in time τdiv given that we start out with an ini- tial distribution over microstates of p(jII). Substituting from eq. (2), we can rearrange this quantity to obtain dj p(iI)π(→ ji) (cid:32) (cid:33) (cid:29) (cid:28) eln[ p(jII) p(iI) ] eβ∆Qi→j = eln[ p(jII) p(iI) ] eβ∆Qi→j (4) I→II driven, nonequilibrium transitions between ensembles; in this light, we can see that the result is closely related to the well-known Landauer bound for the heat generated by the erasure of a bit of information [5]. Having established a general thermodynamic con- straint on how self-replication can proceed, we now must consider whether or our finding is relevant in particular cases of interest. For the process of bacterial cell division introduced above, our ensemble II is a bath of nutrient- rich media containing two bacterial cells in exponential growth phase at the start of their division cycles. In order to make use of the relation in (6), we need to estimate the likelihood that after time τdiv, we will have ended up in an arrangement I where only one, newly formed bacterium is present in the system and another cell has somehow been converted back into the food from which it was built. Our first hint that this likelihood must be quite small comes from our knowledge of the system’s biology: if we start with two cells and wait a full divi- sion time, we will almost certainly end up with four! The challenge before us is therefore to quantify the extreme unlikelihood of the system doing something else. The first piece is relatively easy to imagine: while we may not be able to compute the exact probability of a bacterium fluctuating to peptide-sized pieces and de-respirating a certain amount of carbon dioxide and water, we can be confident it is less likely than all the peptide bonds in the bacterium spontaneously hydrolyz- ing. Happily, this latter probability may be estimated in terms of the number of such bonds npep, the division 3 We can therefore argue that the likelihood of a spon- taneous, sustained pause (of duration τdiv) in the pro- gression of these reactions is very small indeed: if each enzymatic protein component of the cell were to reject each attempt of a substrate to diffuse to its active site (assuming a diffusion time of small molecules between proteins of τdif f ∼ 10−8 sec [7, 8]), we would expect ln ppause ∝ npep(τdiv/τdif f ) to exceed ln phyd by or- ders magnitude. We must, however, consider an alter- native mechanism for the most likely II → I transition (Fig. 1C): it is possible that a cell could grow and divide in an amount of time slightly less than τdiv [2]. If, sub- sequent to such an event, the daughter cell of the recent division were to spontaneously disintegrate back into its constituent nutrients (with log-probability at most on the order of npep ln[τdiv/(npepτhyd)]), we would complete the interval of τdiv with one, recently divided, processively growing bacterium in our system, that is, we would have returned to the I ensemble. Thus, via a back-door into I provided to us by bacterial biology, we can claim that that ln π(→ III) ≤ 2 ln phyd (cid:39) 2npep ln[τdiv/(npepτhyd)] (7) Having obtained the above result, we can now refer back to the bound we set for the heat produced by this self-replication process and write β(cid:104)Q(cid:105) ≥ 2npep ln[(npepτhyd)/τdiv] − ∆Sint (8) This relation demonstrates that the heat evolved in the course of the cell making a copy of itself is set not only by the decrease in entropy required to arrange molecular components of the surrounding medium into a new or- ganism, but also by how rapidly this takes place (through the division time τdiv) and by how long we have to wait for the newly assembled structure to start falling apart (through τhyd). Moreover, we can now quantify the ex- tent of each factor’s contribution to the final outcome, in terms of npep, which we estimate to be 1.6 × 109, as- suming the dry mass of the bacterium is 0.3 picograms [9]. The total amount of heat produced in a single divi- sion cycle for an E. coli bacterium growing at its maxi- mum rate on lysogeny broth (a mixture of peptides and glucose) is β(cid:104)Q(cid:105) = 220npep [1]. We expect the largest contributions to the internal entropy change for cell divi- sion to come from the equimolar conversion of oxygen to carbon dioxide (since carbon dioxide has a significantly lower partial pressure in the atmosphere), and from the confinement of amino acids floating freely in the broth to specific locations inside bacterial proteins. We can esti- mate the contribution of the first factor (which increases entropy) by noting that ln(υCO2/υO2) ∼ 6. The liber- ation of carbon from various metabolites also increases entropy by shuffling around vibrational and rotational degrees of freedom, but we only expect this to make FIG. 1. A single bacterium in ensemble I is most likely to grow and divide (green arrows) as it progresses to ensemble II (panel A). As panel B depicts, it is possible that a transition from II to I could be achieved by the spontaneous disinte- gration of one bacterial cell (orange arrows) accompanied by the spontaneous pausing of the growth of another one (red arrow). However, it is arguable that the scenario displayed in panel C is more likely. Here, one cell disintegrates, while another one divides into two cells, one of which then subse- quently disintegrates. time τdiv, and the peptide bond half-life τhyd. Assuming Poissonian statistics and a large value for npep, we have ln phyd (cid:39) npep ln[τdiv/(npepτhyd)]. Handling the cell that stays alive is more challenging, as we have assumed this cell is growing processively, and we ought not make the mistake of thinking that such a reaction can be halted or paused (Fig. 1B) by a small perturbation. The onset of exponential growth phase is preceded in E. coli by a lag phase that can last several hours [1], during which gene expression is substantially altered so as to retool the cell for rapid division fueled by the available metabolic substrates [6]. It is therefore appropriate to think of the cell in question as an opti- mized mixture of components primed to participate in irreversible, forward reactions like nutrient metabolism and protein synthesis. ABCIIIIIIIIIsome order unity modification to the entropy per car- bon atom metabolized. At the same time, peptide an- abolism reduces entropy: by assuming that in 1% tryp- tone broth, an amino acid starts with a volume to ex- plore of υi = 100 nm3 and ends up tightly folded up in some υf = 0.001 nm3 sub-volume of a protein, we obtain ln(υf /υi) ∼ −12. In light of the fact that the bacterium consumes during division a number of oxygen molecules roughly equal to the number of amino acids in the new cell it creates [1, 10], we can arbitrarily set a generous upper bound of −∆Sint ≤ 10npep. In order to compare this contribution to that of the irreversibility term in (6), we assume a cell division time of 20 minutes [1, 2], and a spontaneous hydrolysis life- time for peptide bonds of 600 years at physiological pH [11], which yields 2npep ln[τdiv/(npepτhyd)] = 1.2×1011 (cid:39) 75npep, a quantity at least several times larger than ∆Sint. Thus, as we may have expected based on previous evidence [12], the entropic cost for aerobic bacterial respi- ration is relatively small, and is substantially outstripped by the sheer irreversibility of the self-replication reaction as it churns out copies that do not easily disintegrate into their constituent parts. More significantly, these calculations also establish that the E. coli bacterium produces an amount of heat less than three times as large as the absolute physical lower bound dictated by its growth rate, internal entropy production, and durability. In light of the fact that the bacterium is a complex sensor of its environment that can very effectively adapt itself to growth in a broad range of different environments, we should not be surprised that it is not perfectly optimized for any given one of them. Rather, it is remarkable that in a single environment, the organism can convert chemical energy into a new copy of itself so efficiently that if it were to produce even half as much heat it would be pushing the limits of what is thermodynamically possible! This is especially the case since we deliberately underestimated the reverse reaction rate with our calculation of phyd, which does not account for the unlikelihood of spontaneously converting carbon dioxide back into oxygen. Thus, a more accurate esti- mate of the lower bound on β(cid:104)Q(cid:105) in future may reveal E. coli to be an even more exceptionally well-adapted self-replicator than it currently seems. We have seen how this argument plays out for a bac- terium, but the approach applies equally in a broad range of cases where there is some reliable stochastic model of a replicator’s population dynamics [13]. For exam- ple, a recent study has used in vitro evolution to opti- mize the growth rate of a self-replicating RNA molecule whose formation is accompanied by a single backbone ligation reaction and the leaving of a single pyrophos- phate group [14]. With a doubling time of 1 hour, a half-life for RNA of 4 years [15], and the reasonable as- sumption (in this case) that the change in translational entropy is negligible, we can estimate the heat bound as 4 (cid:104)Q(cid:105) ≥ RT ln[(4 years)/(1 hour)] = 7 kcal mol −1. Since experimental data indicate an enthalpy for the reaction −1 [16, 17], it would seem in the vicinity of 10 kcal mol this molecule operates quite near the limit of thermody- namic efficiency set by the way it is assembled. To underline this point, we may consider what the bound might be if this same reaction were somehow achieved using DNA, which is much more kinetically stable against hydrolysis than RNA [18]. In this case, we would have (cid:104)Q(cid:105) ≥ RT ln[(3 × 107 years)/(1 hour)] = −1, which exceeds the estimated enthalpy for 16 kcal mol the ligation reaction and is therefore prohibited ther- modynamically. This calculation illustrates a signifi- cant difference between DNA and RNA, regarding each molecule’s ability to participate in self-catalyzed repli- cation reactions fueled by simple triphosphate building blocks: the far greater durability of DNA demands that a much higher per-base thermodynamic cost be paid in entropy production [19] in order for for the growth rate to match that of RNA in an all-things-equal compari- son. Moreover, the heat bound difference between DNA and RNA should increase roughly linearly in (cid:96), the num- ber of bases ligated during the reaction, which forces the maximum possible growth rate for a DNA replicator to shrink exponentially with (cid:96) in comparison to that of its RNA equivalent. This observation is certainly intriguing in light of past arguments made on other grounds that RNA, and not DNA, must have acted as the material for the pre-biotic emergence of self-replicating nucleic acids [17, 20]. The process of cellular division, even in a creature as ancient and streamlined as a bacterium, is so bewilder- ingly complex that it may come as some surprise that physics can make any binding pronouncements about how fast it all can happen. The reason this becomes possible is that time-symmetrically driven, nonequilib- rium processes in constant temperature baths obey gen- eral laws that relate forward and reverse transition prob- abilities to heat production [3]. Previously, such laws had been applied successfully in understanding thermo- dynamics of copying “informational” molecules such as nucleic acids [13]. In those cases, however, the informa- tion content of the system’s molecular structure could more easily be taken for granted, in light of the clear role played by DNA in the production of RNA and protein. What we have glimpsed here is that the underlying con- nection between entropy production and transition prob- ability has a much more general applicability, so long as we recognize that “self-replication” is something that happens relative to an observer: only once a classification scheme determines how many copies of some object are in the system for each microstate can we talk in proba- bilistic terms about the general tendency for that object to affect its own reproduction, and the same system’s microstates can be classified using any number of dif- ferent schemes. We may hope that this insight spurs fu- ture work that will clarify the general physical constraints obeyed by natural selection in nonequilibrium systems. The author thanks C. Cooney, J. Gore, A. Grosberg, D. Sivak, and A. Szabo for helpful comments. [1] H. P. Rothbaum and H. M. Stone, J. Bacteriol. 81, 172 (1961). [2] P. Wang, L. Robert, J. Pelletier, W. L. Dang, F. Taddei, A. Wright, and S. Jun, Curr. Biol. 20, 1099 (2010). [3] G. E. Crooks, Phys Rev E 60, 2721 (1999). [4] C. W. Gardiner, Handbook of Stochastic Methods, 3rd ed. (Springer, 2003). [5] R. Landauer, IBM J. 5, 183 (1961). [6] D. E. Chang, D. J. Smalley, and T. Conway, Mol. Mi- crobiol. 45, 289 (2002). [7] D. Brune and S. Kim, Proc. Natl. Acad. Sci. U.S.A. 90, 3835 (1993). [8] S. C. Blacklow, R. T. Raines, W. A. Lim, P. D. Zamore, and J. R. Knowles, Biochemistry 27, 1158 (1988). [9] F. C. Neidhardt, E. coli and Salmonella: Cellular and 5 Molecular Biology, Vol. 1 (ASM Press, 1990). [10] C. L. Cooney, D. I. Wang, and R. I. Mateles, Biotechnol. Bioeng. 11, 269 (1969). [11] A. Radzicka and R. Wolfenden, Journal of the American Chemical Society 118, 6105 (1996). [12] U. von Stockar, T. Maskow, J. Liu, I. W. Marison, and R. Patino, J. Biotechnol. 121, 517 (2006). [13] D. Andrieux and P. Gaspard, Proc. Natl. Acad. Sci. U.S.A. 105, 9516 (2008). [14] T. A. Lincoln and G. F. Joyce, Science 323, 1229 (2009). [15] J. E. Thompson, T. G. Kutateladze, M. C. Schuster, and R. T. Raines, F. D. Venegas, J. M. Messmore, Bioorg. Chem. 23, 471 (1995). [16] C. A. Minetti, D. P. Remeta, H. Miller, C. A. Gelfand, G. E. Plum, A. P. Grollman, and K. J. Breslauer, Proc. Natl. Acad. Sci. U.S.A. 100, 14719 (2003). [17] H. J. Woo, R. Vijaya Satya, and J. Reifman, PLoS Com- put. Biol. 8, e1002534 (2012). [18] G. K. Schroeder, C. Lad, P. Wyman, N. H. Williams, and R. Wolfenden, Proc. Natl. Acad. Sci. U.S.A. 103, 4052 (2006). [19] D. Y. Zhang, A. J. Turberfield, B. Yurke, and E. Win- free, Science 318, 1121 (2007). [20] W. Gilbert, Nature 319, 618 (1986).
1805.03794	3	1805	2019-01-17T14:53:13	Diffusion-dynamics laws in stochastic reaction networks	[ "physics.bio-ph", "q-bio.MN" ]	Many biological activities are induced by cellular chemical reactions of diffusing reactants. The dynamics of such systems can be captured by stochastic reaction networks. A recent numerical study has shown that diffusion can significantly enhance the fluctuations in gene regulatory networks. However, the universal relation between diffusion and stochastic system dynamics remains veiled. Within the approximation of reaction-diffusion master equation (RDME), we find general relation that the steady-state distribution in complex balanced networks is diffusion-independent. Here, complex balance is the nonequilibrium generalization of detailed balance. We also find that for a diffusion-included network with a Poisson-like steady-state distribution, the diffusion can be ignored at steady state. We then derive a necessary and sufficient condition for networks holding such steady-state distributions. Moreover, we show that for linear reaction networks the RDME reduces to the chemical master equation, which implies that the stochastic dynamics of networks is unaffected by diffusion at any arbitrary time. Our findings shed light on the fundamental question of when diffusion can be neglected, or (if nonnegligible) its effects on the stochastic dynamics of the reaction network.	physics.bio-ph	physics	Diffusion-dynamics laws in stochastic reaction networks Tan Van Vu∗ and Yoshihiko Hasegawa† Department of Information and Communication Engineering, Graduate School of Information Science and Technology, The University of Tokyo, Tokyo 113-8656, Japan (Dated: January 18, 2019) Many biological activities are induced by cellular chemical reactions of diffusing reactants. The dynamics of such systems can be captured by stochastic reaction networks. A recent numerical study has shown that diffusion can significantly enhance the fluctuations in gene regulatory networks. However, the universal relation between diffusion and stochastic system dynamics remains veiled. Within the approximation of reaction-diffusion master equation (RDME), we find general relation that the steady-state distribution in complex balanced networks is diffusion-independent. Here, complex balance is the nonequilibrium generalization of detailed balance. We also find that for a diffusion-included network with a Poisson-like steady-state distribution, the diffusion can be ignored at steady state. We then derive a necessary and sufficient condition for networks holding such steady- state distributions. Moreover, we show that for linear reaction networks the RDME reduces to the chemical master equation, which implies that the stochastic dynamics of networks is unaffected by diffusion at any arbitrary time. Our findings shed light on the fundamental question of when diffusion can be neglected, or (if nonnegligible) its effects on the stochastic dynamics of the reaction network. I. INTRODUCTION Diverse biological phenomena, such as cellular signal transductions and gene expression systems, are com- monly studied by stochastic reaction network modeling [1 -- 3]. These systems involve a set of reactant species which react through several channels. In most of the ex- isting studies, such systems are often assumed to be well mixed, meaning that the diffusion coefficients of the re- actants are infinitely large [4 -- 10]. However, experiments have shown that reactants in cells diffuse at considerably low rates [11], and that the smallest timescale of the sys- tem is a little larger than the timescale of molecular dif- fusion. In such cases, the well-mixed assumption cannot accurately obtain the stochastic dynamics of the system. For example, living cells continuously receive signals at their receptors, which are subsequently transmitted to the nucleus through biochemical reaction networks [12 -- 15]. This process is strongly influenced by extrinsic and intrinsic noise arising from fluctuations in the input and reactions. These effects induce unavoidable fluctuations in the biomolecule concentrations, which deteriorate the fidelity of information transfer [16, 17]. By accurately evaluating the fluctuations, we would better understand the mechanism underlying signal transmission in cells. In a numerical study of gene regulatory networks, Ref. [18] showed that the fluctuations are larger in the model with diffusion than in its well-mixed counterpart. Thus, how diffusion relates to the stochastic dynamics of reaction networks is a pertinent question. Recently, Ref. [19] has numerically studied the effects of diffusion on single-cell variability in multicellular organisms, and the limits of ∗ [email protected] † [email protected] slow and fast diffusion have been investigated. Two commonly used models for studying stochastic reaction-diffusion systems are the reaction-diffusion mas- ter equation (RDME) [20] and the Smoluchowski model [21]. The RDME, which is a mesoscopic model, is an extension of the nonspatial chemical master equation (CME) [20] and can be interpreted as an asymptotic ap- proximation to spatially continuous stochastic reaction- diffusion models [22]. The RDME has been successfully applied in studying many biological systems [19, 23 -- 25]. It is worth noting that the Langevin equation, which can be derived from an equivalent Fokker -- Planck equation or the Poisson representation, can handle continuum-limit diffusion in reaction networks [20, 26]. However, the Langevin equation is applicable to biochemical reactions occurring in infinite space with no physical boundary, which is unrealistic in biological cells. In the present work, we investigate the relations be- tween diffusion and the stochastic dynamics of reaction networks within a physical reflecting boundary. In this system, reactants diffuse within a closed 3-dimensional space without escaping. With the aid of the RDME, we find an intriguing law stating that diffusion does not affect the steady-state distribution of complex balanced networks, which have a Poisson-like distribution. Our proof reveals that if the network presents a steady-state distribution of product-of-Poissonians form, diffusion can be neglected. We then calculate the necessary and suf- ficient conditions for such steady-state network distri- butions. We also find another result, wherein steady state is not a requirement. Specifically, we prove that for linear reaction networks, one can derive the CME from the RDME, which indicates that diffusion can be ig- nored in this case. This result can be restated as follows: the stochastic dynamics of linear networks are diffusion- independent, which is consistent with the Smoluchowski model. In addition, we perform stochastic simulations on 2 also define a vector 1vi ∈ ZVN , in which the number of molecules of all species in all voxels is zero except for species Xi in voxel v (which is one), and a vector eVvj ∈ ZVN , in which all elements are zero except in voxel v (which holds Vj ). As the diffusion of each species into neighboring voxels can be modeled as a first-order reaction, the diffusion-included reaction network can be described in the following form: s1jX v 1 + · · · + sN jX v di−→ X v′ N kj−→ r1j X v 1 + · · · + rN jX v N , (4) both linear and nonlinear networks to verify our results. II. MODELS We consider a general reaction network consisting of N reactant species X1, . . . , XN and K reactions R1, . . . , RK. Assume that all reactions occur inside a cell with fixed volume Ω, and that reaction Rj (1 ≤ j ≤ K) is of the form s1jX1 + · · · + sN jXN kj−→ r1jX1 + · · · + rN jXN , (1) X v i i , ∀ 1 ≤ i ≤ N, v ∈ V, v′ ∈ Ne(v), defined analogously. IfPN where sij , rij ∈ N≥0 are the stoichiometric coefficients and kj ∈ R>0 is the macroscopic reaction rate. Here, N≥0 denotes the set of nonnegative integers. R>0 and R≥0 are i=1 sij ≤ 1 for all j = 1, . . . , K, then the reaction network is linear; otherwise, it is non- linear. The state of the system is fully determined by the molecule-number vector of all reactant species in the system, n = [n1, . . . , nN ]⊤, where ni ∈ N≥0 is the num- ber of molecules of species Xi. Assuming mass-action kinetics, the time evolution of a well-mixed system can be described by the following chemical master equation (CME): where X v i refers to species Xi in voxel v, di is the dif- fusion rate of species Xi, and Ne(v) is the set of voxels neighboring v. The stochastic dynamics of the system can then be described by the following RDME: ∂tP (n, t) =Xv∈V Xv′∈Ne(v) NXi=1(cid:0)E1vi−1 v′ i − 1(cid:1) dinviP (n, t) +Xv∈V KXj=1(cid:16)E− eVvj − 1(cid:17) fvj(n, ω)P (n, t), (5) where fvj(n, ω) is the propensity function, given by ∂tP (n, t) = KXj=1 (E−Vj − 1)fj(n, Ω)P (n, t), (2) fvj(n, ω) = kjω1−PN i=1 sij nvi! (nvi − sij)! . (6) NYi=1 where V = [rij − sij] ∈ ZN ×K is a stoichiometric ma- trix, Vj denotes the jth column of matrix V , and Ex is an operator that replaces n with n + x. P (n, t) is the probability of the system being in state n at time t, and the propensity function fj(n, Ω) of reaction Rj is given by fj(n, Ω) = kj Ω1−PN i=1 sij ni! (ni − sij)! . (3) NYi=1 To include diffusion in stochastic spatial dynamics, many researchers apply the RDME, in which space is parti- tioned discretely into many voxels. It is known that the RDME is accurate if an appropriate combination of the time- and length-scale is chosen [22, 27 -- 30]. We assume from now on that the volume of the system is optimally divided into small voxels and as such, the RDME yields a good description of the time evolution of the probabil- ity distribution. Diffusion then occurs among the vox- els, and the reaction can occur within the same voxel considered to be a well-mixed system. Assume that the volume Ω is divided into a set V of voxels labeled by integers v = 1, 2, . . . , V. Each voxel v occupies a con- stant volume ω and contains nvi molecules of reactant species Xi. The state vector of voxel v is denoted as nv = [nv1, . . . , nvN ]⊤. The state of the whole system is then described as the molecule-number vector n of each species in each voxel, namely, n = [n⊤ V]⊤. We 1 , . . . , n⊤ In the large-diffusion limit, the RDME converges to the CME [31]. tionsPN i=1 sijXi andPN Before stating our results, we describe several existing concepts and results of deterministic reaction networks. For each reaction Rj (1 ≤ j ≤ K), the linear combina- i=1 rij Xi of the species in Eq. (1) are called the complexes of the reaction. Defining C = {C1, C2, . . . , CM } as the set of complexes, with M = C, aii′−−→ Ci′ , where aii′ each reaction can be expressed as Ci denotes the reaction rate. For each 1 ≤ i, i′ ≤ M , aii′ = 0 if Ci → Ci′ is not present in the reaction network; oth- erwise, aii′ = kj for some j (1 ≤ j ≤ K). The matrix A ∈ RM×M , called the Kirchhoff matrix of the reaction network, is defined as follows: [A]ii′ =(−PM ai′i, j=1 aij, if i = i′ if i 6= i′ . (7) Let X = {X1, . . . , XN } be the set of species and R = Si,i′:aii′ >0{Ci → Ci′ } be the set of reactions in the network. The triple {X , C, R} then defines a reaction network. A reaction network {X , C, R} is called weakly reversible if for any reaction Ci → Ci′ ∈ R, there ex- ists a sequence of complexes Ci1 , . . . , Cip ∈ C such that Ci′ → Ci1 , Ci1 → Ci2 , . . . , Cip → Ci ∈ R. One can construct a directed graph G corresponding to a reaction network in the following manner. For each 1 ≤ i, i′ ≤ M , if and only if draw a directed edge from Ci to Ci′ Ci → Ci′ ∈ R. We denote by ℓ the number of con- nected components of the underlying undirected graph of G. The deficiency of a reaction network is an integer defined as δ = C − ℓ − rank(V ). According to Ref. [32], δ is always nonnegative. In a deterministic system, the vector of species concen- trations, c = [c1, c2, . . . , cN ]⊤ ∈ RN ≥0, temporally evolves as described by the following differential equations, which express the different form of rate equations: ∂tc = Y · A · Ψ(c). (8) Here, Y = [yij] ∈ NN ×M is the matrix of stoichiometric compositions of the complexes, i.e., yij is the stoichio- metric coefficient of Cj corresponding to species Xi, and Ψ : RN 7→ RM is a mapping given by ≥0 Ψj(c) = NYi=1 cyij i , j = 1, . . . , M. (9) A reaction network is called complex balanced at c ∈ RN >0 if A · Ψ(c) = 0. This condition means that for each com- ai′iΨi′ (c). In plex Ci ∈ C,PCi→Ci′ aii′ Ψi(c) =PCi′ →Ci this case, c is a positive equilibrium of the network. III. RESULTS The following states our first result. Theorem 1. If a reaction network is complex balanced, its steady-state distribution is unaffected by diffusion. Proof. As the network is complex balanced, there exists a positive equilibrium c = [c1, . . . , cN ]⊤ ∈ RN >0 such that A · Ψ(c) = 0. We note that the only requirement in our proof is the existence of some c such that A · Ψ(c) = 0. Here, c is not the steady-state concentration in the presence of diffusion. Let Γ ⊆ NN ≥0 be the state space of the network, which may depend on the initialization. First, we prove the following ansatz: that the steady- state distribution of the RDME is given by a product PΓ(n, t) of Poisson distributions: PΓ(n, t) =(NΓQv∈VQN 0, i=1 (ωci)nvi nvi! , if Pv∈V nv ∈ Γ if Pv∈V nv /∈ Γ , ≥0, we define P ∗ (10) where NΓ is the normalizing constant. For each nv ∈ NN i=1(ωci)nvi /nvi!. PΓ(n, t) Γ (nv, t). Now, we need to show that ∂tPΓ(n, t) = 0. Substituting PΓ(n, t) in Eq. (10) into Eq. (5), the first term of the right-hand side becomes can then be expressed as PΓ(n, t) = NΓQv∈V P ∗ Γ (nv, t) =QN Xv∈V Xv′∈Ne(v) NXi=1(cid:0)E1vi−1 v′ i − 1(cid:1) dinviPΓ(n, t) = 0. (11) 3 FIG. 1. Steady-state distributions (a) PΓ(bn1) of species X1 and (b) PΓ(bn2) of species X2 of nonlinear reaction net- work. Each panel shows the distributions of the 1-voxel sys- tem (green region), 100-voxel system (blue dots), and 225- voxel system (red line). The parameters are k1 = 4, k2 = 1, k3 = 2, Ω = 128. The diffusion rates of species X1, X2 are d1 = 1, d2 = 2 (100 voxels), and d1 = 2, d2 = 1 (225 voxels). Insets show the absolute probability differences P100−P1 (or- ange dots) and P225 − P1 (violet dots), where P1, P100 and P225 denote the probabilities in the 1-, 100-, and 225-voxel systems, respectively. The second term on the right-hand side becomes the sum of the following values over all voxels v ∈ V: KXj=1(cid:16)E− eVvj − 1(cid:17) fvj(n, ω)PΓ(n, t) = NΓ Yv′6=v Γ (nv′ , t) P ∗ KXj=1(cid:16)E−Vj − 1(cid:17) fj(nv, ω)P ∗ Γ (nv, t). Exploiting the condition A·Ψ(c) = 0, one can prove that [33] KXj=1(cid:16)E−Vj − 1(cid:17) fj(nv, ω)P ∗ Γ (nv, t) = 0. (12) Therefore, the second term also disappears and we obtain the total number of molecules of species Xi in the system. To complete our theorem, we compute the steady-state the desired result ∂tPΓ(n, t) = 0. Let bn = Pv∈V nv represent the number of molecules of all species, i.e.,bni is distribution PΓ(bn), and show its diffusion-independence. For bn /∈ Γ, obviously PΓ(bn) = 0. For bn ∈ Γ, the explicit form of PΓ(bn) is obtained as follows: PΓ(bn) = Xn:Pv As NΓ = (cid:16)Pbn∈ΓQN diffusion, the distribution PΓ(bn) is also independent of diffusion. The details of these derivations can be seen in Appendix A. bni! (cid:17)−1 does not depend on PΓ(n, t) = NΓ NYi=1 bni! (Ωci)bni . (13) nv=bn (Ωci)bni i=1 Our theoretical result is empirically verified in simula- tions of the following complex balanced network: ∅ k1−→ X1 + 2X2 k2−→ X2 k3−→ ∅. (14) We consider three cases with different numbers of vox- els in the system volume: 1 voxel (a well-mixed system), 100 voxels, and 225 voxels. The diffusion coefficients of the species in the 100-voxel system differ from those in the 225-voxel system. The steady-state distributions of species X1 and X2 are plotted in Fig. 1. As can be seen, the distributions of both species are consistent in all three cases. From these result, it is pertinent to ask which conditions define a complex balanced network. Refer- ence [32] proved that a weakly reversible reaction network with zero deficiency is a complex balanced network. This implies that in some cases, a complex balanced network can be identified by its network topology. In Ref. [34], complex balanced realizations of a given kinetic polyno- mial system were computed by a linear programming al- gorithm. Thus far, we show that the steady-state distribution of a complex balanced network is a product of Poisson distributions. A network with such a distribution im- plies that the system is diffusion-independent at steady state. Therefore, we desire to know the condition under which the system establishes a Poisson-like steady-state distribution. This condition is embodied in the following theorem. Theorem 2. The network possesses the steady-state dis- tribution defined in Eq. (10) in all state spaces Γ ⊆ NN ≥0 if and only if it is complex balanced. Proof. We use the Fock space representation [35] to de- scribe the molecule-number changes of each species in- side each voxel. A state vector ni with configuration n means that nvi molecules of species Xi exist in voxel v. Using the annihilation and creation operators avi, a† vi, i.e., avinvii = nvinvi − 1i, a† vinvii = nvi + 1i, we can map the probability distribution PΓ(n, t) to a state vec- tor ψ(t)iΓ, defined by ψ(t)iΓ =Xn PΓ(n, t)ni =Xn PΓ(n, t)(a†)n0i. (15) This expression sums over all possible configurations n weighted by their occurrence probabilities at time t. To establish the time evolution of this state vector, we apply the master equation to obtain the following Schrodinger equation: ∂tψ(t)iΓ = −H(a†, a)ψ(t)iΓ, (16) where H(a†, a) represents the Hamiltonian action on the Fock space, expressed as shown in Appendix B. In gen- eral, H(a†, a) is the sum of several sub-actions created by each reaction of the system, e.g., a reaction of the i yields a sub-action kjω1−PN i=1(avi)sij . The action H(a†, a) is considered to be normally ordered, i.e., a† vi is always to the left of avi. In a steady-state sys- tem, H(a†, a)ψ(t)iΓ = 0. Consequently, H(a†, a)ψ(t)i kj−→ PN vi)rij −QN i=1 sij(cid:0)QN vi)sij(cid:1)QN form PN i=1 sijX v i i=1 rij X v i=1(a† i=1(a† is also 0, where the state ψ(t)i is defined as follows: 4 ψ(t)i =XΓ ψ(t)iΓ NΓ Pv,i ωcia† = e vi 0i. = NYi=1 VYv=1 Xnvi≥0 vi)nvi (ωcia† nvi! 0i In other words, the following condition H(a†, a)e Pv,i ωcia† vi 0i = 0 (17) must hold. To derive a further condition with no in- volvement of a† and a, we consider the coherent states φvii and hφvi, defining the right and left eigenstates of avi and a† vi, respectively. Specifically, aviφvii = φviφvii and hφvia† vi = hφviφ∗ vi, with complex eigenvalue φvi ∈ C. Multiplying both sides of Eq. (17) by the left coherent state hφ, we obtain 0 = hφH(a†, a)e Pv,i ωcia† vi 0i ⇔ 0 = H(φ∗, ωec), (18) where ec ∈ RVN is defined asecvi = ci. As H(φ∗, ωec) is aii′−−→ Ci′ in voxel v contributes to H(φ∗, ωec) a polynomial of φ∗, this result is possible only when the coefficients of all monomials are zero. Each reaction of the form Ci v) − Ψi(φ∗ a quantity ωaii′ (Ψi′ (φ∗ collecting the coefficients of Ψi(φ∗ and v ∈ V, we obtain the following relation: v)) Ψi(c). Therefore, by v) for each i = 1, . . . , M H(φ∗, ωec) = 0, ∀φ ∈ CVN ⇔ A · Ψ(c) = 0, meaning that the network is complex balanced at c. The details of these calculations are shown in Appendix B. From these results, we conclude that the necessary and sufficient condition for a steady-state distribution (Eq. (10)) is that the network is complex balanced. (19) We note that the sufficient condition of Theorem 2 has been studied in Ref. [36] (i.e., if the network is complex balanced, then the steady-state distribution has a form as in Eq. (10)). Above we investigate the relation between diffusion and the distributions of the reactant species in steady state. We now present another result that holds under non-steady-state conditions. Theorem 3. When the reaction network is linear, the RDME can be reduced to the CME. Equivalently, the diffusion can be ignored in such case. number of molecules of species Xi in the system, we de- Proof. The system volume Ω is related to the voxel volume ω as Ω = Vω. Now, for each state vector ≥0 representing the number bn = [bn1,bn2, . . . ,bnN ]⊤ ∈ NN of molecules of the reactant species, i.e., bni is the total Pv∈V nv = bn(cid:9). Let fine the set S(bn) = (cid:8)n ∈ NVN P (bn, t) be the probability of the system being in state bn at time t. comes P (bn, t) = Pn∈S(bn) P (n, t). As Pbn P (bn, t) = Pn P (n, t) = 1, P (bn, t) is a probability distribution. In terms of P (n, t), this probability be- ≥0 To show that this probability distribution satisfies the CME given by Eq. (2), we calculate the time derivative 5 of P (bn, t) as follows: ∂tP (bn, t) = Xn∈S(bn) ∂tP (n, t). (20) FIG. 2. Probability distributions (a) P (bn1, t = 1), (b) P (bn2, t = 1), (c) P (bn1, t = 10), and (d) P (bn2, t = 10) of two species X1, X2 at times t = 1 (upper panels) and t = 10 (lower panels) of linear reaction network. Each panel shows the distributions of the 1-voxel system (green region), 100- voxel system (blue dots), and 225-voxel system (red line). The parameters are k1 = 1, k2 = 1, k3 = 2, k4 = 1, Ω = 128. The diffusion rates of species X1, X2 are d1 = 1, d2 = 2 (100 vox- els) and d1 = 2, d2 = 1 (225 voxels). Insets show the absolute probability differences P100 −P1 (orange dots) and P225 −P1 (violet dots), where P1, P100 and P225 indicate the probability in the 1-, 100-, and 225-voxel systems, respectively. does not affect the steady-state distribution of the sys- tem. We also showed that a diffusion-included reaction network has a Poisson-like steady-state distribution if and only if it is complex balanced, analogously to the well-mixed case described in [39]. Moreover, we demon- strated that the RDME can be reduced to the CME in the case of linear reaction networks. These results help to clarify the conditions under which diffusion is negligible. Under such conditions, the system can be described by the CME instead of the intractable RDME. In nonlinear networks that are not complex-balanced, how diffusion affects the stochastic system dynamics, or whether it can be ignored, requires further investigation. It appears that functional biological networks satisfy- ing the complex balanced condition are not widespread in real-world systems. Nevertheless, weakly reversible networks have been successfully applied in modeling sig- nal transduction pathways [40] and asymmetric stem-cell division [41]. Besides that, complex balanced networks whose Fano factor is equal to one can be used in ana- lyzing and approximating cascade networks or metabolic pathways, wherein noise is not propagated from upstream to downstream [42]. Although our results are obtained with the approximation of the RDME, it is expected that the derived diffusion-dynamics laws provide suggestive results for real physical systems. (21) Substituting Eq. (5) into the right-hand side of Eq. (20), we obtain an equation with both diffusion and reaction terms on the right. After some algebraic transforma- tions, the diffusion term disappears and only the reac- tion term remains (see Appendix C). As the reaction propensity function fvj(n, ω) must be one of two forms: fvj(n, ω) = kjnvi or kjω. Substituting the exact form of each propensity function into Eq. (20), we finally obtain network is linear, i.e., Pi sij ≤ 1 ∀ j = 1, . . . , K, the the following master equation for P (bn, t): ∂tP (bn, t) = KXj=1(cid:16)E−Vj − 1(cid:17) fj(bn, Ω)P (bn, t). Obviously, this differential equation is identical to the CME stated in Eq. (2), and contains no diffusion fac- tors. Therefore, it can be concluded that diffusion can be ignored in linear reaction networks. In Theorem 3, we demonstrate that the RDME reduces to the CME in the case of linear reaction networks, which implies that diffusion does not affect the stochastic dy- namics of the system at an arbitrary time. From the view of the Smoluchowski model, this statement appears to be obvious. By regarding the network as interact- ing many-particle system and introducing diffusion and reaction operators [37], the same result can be derived. However, it is not evident from the view of the RDME. The agreement of results in these different models serves as validation for the RDME. We numerically verify the result of Theorem 3 on a simple linear reaction network, namely, a coarse-grained model of enzymatic reactions and gene expressions. The network consists of two reactant species X1 and X2 and four reactions [6, 38]: ∅ k1 ⇄ k2 X1, X1 k3−→ X1 + X2, X2 k4−→ ∅. (22) Again, we divide the cell volume into 1, 100, and 225 vox- els with different diffusion coefficients of X1 and X2. The result is displayed in Fig. 2. As before, the distributions of each species at times t = 1 and t = 10 are identical in all three cases. These numerical results empirically validate Theorem 3. IV. CONCLUSIONS In summary, within the approximation of the RDME, we proved that diffusion in complex-balanced networks ACKNOWLEDGMENT This work was supported by MEXT KAKENHI Grant No. JP16K00325. Appendix A: Detailed calculations in Theorem 1 1. Detailed calculations of Equation (11) The detailed calculation of Eq. (11) is given below: v′ i − 1(cid:1) dinviPΓ(n, t) NXi=1(cid:0)E1vi−1 NXi=1(cid:18)di(nvi + 1)PΓ(n + 1vi − 1v′i, t) − dinviPΓ(n, t)(cid:19) 6 P ∗ Γ (nev, t) − nviPΓ(n, t)(cid:19) = Xv∈V Xv′∈Ne(v) =Xv∈V Xv′∈Ne(v) NXi=1 NXi=1 NXi=1 = = = 0. (ωci)nvi +1 (nvi + 1)! diXv∈V Xv′∈Ne(v)(cid:18)NΓ(nvi + 1) diXv∈V Xv′∈Ne(v)(cid:18)NΓ (nv′i − 1)! Yi′6=i diXv∈V Xv′∈Ne(v)(cid:18)nv′iPΓ(n, t) − nviPΓ(n, t)(cid:19) (ωci)nv′ i (ωci)nvi nvi! (ωci)nv′ i−1 (nv′i − 1)! Yi′6=i (ωci′ )nvi′ (ωci′ )nv′ i′ nvi′ !nv′i′ ! Yev6=v,v′ (ωci′ )nvi′ (ωci′ )nv′ i′ nvi′ !nv′i′ ! P ∗ Γ (nev, t) − nviPΓ(n, t)(cid:19) Yev6=v,v′ 2. Detailed calculations of Equation (12) We here reveal the details of Eq. (12). The equa- tion A · Ψ(c) = 0 means that PCi→Ci′ aii′ Ψi(c) − PCi′ →Ci ai′iΨi′ (c) = 0 for each complex Ci′ ∈ C. The left side of Eq. (12) can be transformed as Γ (nv, t) = (fj(nv − Vj, ω)P ∗ KXj=1(cid:16)E−Vj − 1(cid:17) fj(nv, ω)P ∗ KXj=1 = XCi→Ci′ = XCi→Ci′ k=1 yki NYk=1 k=1 yki NYk=1 aii′ ω1−PN aii′ ω1−PN Γ (nv − Vj, t) − fj(nv, ω)P ∗ Γ (nv, t)) (nvk + yki − yki′ )! (nvk − yki′ )! (ωck)nvk+yki−yki′ (nvk + yki − yki′ )! − (ωck)nvk+yki−yki′ (nvk − yki′ )! (ωck)nvk (nvk − yki)!! − NYk=1 nvk! (nvk − yki)! (ωck)nvk nvk! ! NYk=1 7 = ω XCi→Ci′ aii′  XCi→Ci′ = ω XCi′ ∈C NYk=1 = ω XCi′ ∈C Ψi′ (c) 1 = 0. Ψi(c) Ψi′ (c) ωnvk−yki′ cnvk (nvk − yki′ )! k NYk=1 − aii′ aii′ Ψi(c) Ψi′(c) ωnvk−yki′ cnvk (nvk − yki′ )! k NYk=1 ωnvk−yki′ cnvk (nvk − yki′ )!  XCi→Ci′ k k ωnvk−ykicnvk (nvk − yki)! ! NYk=1 (nvk − yki′ )! NYk=1 − XCi′ →Ci ai′iΨi′ (c) aii′ Ψi(c) − XCi′ →Ci ωnvk−yki′ cnvk ai′i k The same result (but omitting the details) is given in Ref. [33]. Appendix B: Detailed calculations in Theorem 2 Before presenting the calculations, we state several properties of the bosonic operators a† vi and avi. nvii = (a† vi)nvi 0i, 3. Detailed calculations of Equation (13) Finally, we compute the explicit form of the distribu- (avi)l(a† vi)knvii = (a† vi)k(avi)lnvii = (nvi + k − j)nvi + k − li, (nvi − j)nvi + k − li, l−1Yj=0 l−1Yj=0 tion PΓ(bn) in Eq. (13). We have PΓ(bn) = Xn:Pv PΓ(n, t) nv =bn = NΓ Xn1,...,nV∈NN (ωci)nvi nvi! (ωci)nvi ≥0 Pv nv=bn Pv∈V nvi=bni Yv∈V NYi=1( Xn1i,...,nVi≥0 NYi=1( (ωci)bni bni! NYi=1 bni! NYi=1 Yv∈V Xn1i,...,nVi≥0 NYi=1 Pv∈V nvi=bni Vbni = NΓ (ωci)bni nvi! ) Qv∈V nvi! ) (cid:0)Pv∈V nvi(cid:1)! bni! (Ωci)bni . = NΓ = NΓ = NΓ In transforming the fourth equation to the fifth one, we exploited the following equality: Xx1,...,xm≥0 i=1 xi=n Pm (x1 + · · · + xm)! x1! . . . xm! = mn, ∀m ∈ N>0, n ∈ N≥0. [avi, a† [a† vi, a† v′i′ − a† v′i′ ] = avia† v′i′ ] = [avi, av′i′ ] = 0. v′i′ avi = δvv′ δii′ , For a general configuration n, we define the correspond- ing state vector ni as ni = (a†)n0i = Yv∈V For convenience, we note that (a† vi)nvi 0i. NYi=1 vi) = f (a† ecavif (a† eca† vi + c)ecavi , vif (avi) = f (avi − c)eca† vi , (B1) (B2) (B3) where c ∈ C is a complex number and f is an arbitrary function. 1. Detailed calculations of Equation (16) We first derive the explicit form of the Hamiltonian action H(a†, a) in Eq. (16). Suppose that the network contains a set R of reactions Rj of the general form i=1 pj Pv∈VPN kj−→Pv∈VPN vi and qj vi are the stoichiometric coefficients. For each reac- tion Rj, we define a stoichiometric vector V j ∈ ZVN as V j vi. Starting from the master equation, we have i , where pj vi = qj vi − pj i=1 qj viX v viX v i 8 PΓ(n, t) (a†)n0i. Note that the two terms inside the bracket can be obtained using operators as follows: ∂tψ(t)iΓ =Xn ∂tPΓ(n, t)(a†)n0i =Xn XRj ∈R kj ω1−Pv,i pj vi)! (nvi + pj (nvi − qj vi − qj vi)! viYv,i PΓ(n − V j, t)(a†)n0i =Yv,i PΓ(n, t)(a†)n0i =Yv,i vi)! (nvi + pj vi)! vi − qj vi)! (nvi − qj Yv,i nvi! (nvi − pj Yv,i Using these equalities, ∂tψ(t)iΓ is calculated as follows: PΓ(n − V j, t) −Yv,i nvi! (nvi − pj vi)! (a† vi)qj vi (avi)pj vi PΓ(n − V j, t)(a†)n−V j 0i, (a† vi)pj vi (avi)pj vi PΓ(n, t)(a†)n0i. kjω1−Pv,i pj vi (avi)pj vi PΓ(n − V j, t)(a†)n−V j (a† vi)pj vi (avi)pj PΓ(n − V j, t)(a†)n−V j (a† vi)pj vi (avi)pj vi PΓ(n, t)(a†)n0i PΓ(n, t)(a†)n0i viXn ∂tψ(t)iΓ =Xn XRj ∈R = XRj ∈R = XRj ∈R kj ω1−Pv,i pj kj ω1−Pv,i pj vi (avi)pj (a† (a† vi)qj vi)qj viYv,i viYv,i viYv,i H(a†, a) = XRj ∈R vi)qj (a† vi (avi)pj viXn vi −Yv,i (a† vi)pj vi (avi)pj kjω1−Pv,i pj (a† viYv,i 0i −Yv,i 0i −Yv,i vi ψ(t)iΓ. vi −Yv,i vi)qj vi)pj (a† Thus, the general form of H is obtained as viYv,i (avi)pj vi . For a diffusion-included reaction network involving the following reactions s1jX v 1 + · · · + sN jX v di−→ X v′ N X v i i , ∀ 1 ≤ i ≤ N, v ∈ V, v′ ∈ Ne(v), kj−→ r1jX v 1 + · · · + rN jX v N , the Hamiltonian action H(a†, a) in Eq. (16) takes the following form H(a†, a) = KXj=1Xv∈V kjω1−PN i=1 sij" NYi=1 (a† vi)rij − NYi=1 (a† vi)sij# NYi=1 (avi)sij + NXi=1Xv∈V Xv′∈Ne(v) di(a† v′i − a† vi)avi. We note that this form of H is already normal-ordered. 2. Detailed calculations of Equation (18) Equation (18) is derived through the following steps: 0 = hφH(a†, a)e Pv,i ωcia† vi 0i (B6) Pv,i ωcia† ⇔ 0 = hφe ⇔ 0 = e ⇔ 0 = e ⇔ 0 = e Pv,i ωciφ∗ Pv,i ωciφ∗ Pv,i ωciφ∗ ⇔ 0 = H(φ∗, ωec). vi H(a†, a + ωec)0i vi hφH(a†, a + ωec)0i vi hφH(φ∗, ωec)0i vi H(φ∗, ωec)hφ0i (B4) (B5) (B7) (B8) (B9) (B10) (B11) In Eq. (B7), we use the property stated in Eq. (B3). In Eq. (B9), the operators avi (v ∈ V, i = 1, . . . , N ) are absorbed into 0i (∵ avi0i = 0), and the operators a† vi are replaced by φ∗ vi). The result Eq. (B11) is obtained by noting that vi (∵ hφvia† vi = hφviφ∗ Pv,i ωciφ∗ vi 6= 0, e hφ0i =Yv,i hφvi0i =Yv,i 2 φvi2 e− 1 6= 0. 9 3. Detailed calculations of Equation (19) Equation (19) is given by H(φ∗, ωec) = 0, ∀φ ∈ CVN ⇔ A · Ψ(c) = 0. The Hamiltonian action H(a†, a) is described by H(a†, a) = kjω1−PN KXj=1Xv∈V i=1 sij" NYi=1 The case H(φ∗, ωec) = 0 is equivalent to (a† vi)rij − (a† vi)sij# NYi=1 NYi=1 (avi)sij + NXi=1Xv∈V Xv′∈Ne(v) di(a† v′i − a† vi)avi. i=1 sij" NYi=1 (φ∗ vi)rij − (φ∗ NYi=1 ωaii′ [Ψi′ (φ∗ v) − Ψi(φ∗ v)] Ψi(c) + kjω1−PN KXj=1Xv∈V ⇔ XCi→Ci′Xv∈V Ψi(φ∗ ⇔ ωXv∈V XCi∈C ⇔ XCi′ →Ci v) XCi′ →Ci ai′iΨi′ (c) − XCi→Ci′ ⇔ A · Ψ(c) = 0. ai′iΨi′ (c) − XCi→Ci′ aii′ Ψi(c) = 0, ∀Ci ∈ C [(φ∗ NXi=1Xv∈V Xv′∈Ne(v) vi)sij# NYi=1 (ωecvi)sij + NXi=1 di Xv,v′∈V aii′ Ψi(c) = 0, ∀φ ∈ CVN v′∈Ne(v) v∈Ne(v′) v′ i − φ∗ di(φ∗ v′ i − φ∗ vi − φ∗ vi)ωecvi + (φ∗ vi)ωecvi = 0 v′i)ωecv′i] = 0 ∂tP (bn, t) = Xn∈S(bn) Appendix C: Detailed calculations in Theorem 3 The master equation of P (bn, t) is derived as follows: ∂tP (n, t) NXi=1(cid:18)di(nvi + 1)P (n + 1vi − 1v′i, t) − dinviP (n, t)(cid:19) (C1) = Xn∈S(bn)(cid:18)Xv∈V Xv′∈Ne(v) +Xv∈V KXj=1(cid:18)fvj(n − eVvj , ω)P (n − eVvj, t) − fvj(n, ω)P (n, t)(cid:19)(cid:19). 10 As n ∈ S(bn) → n + 1vi − 1v′i =en ∈ S(bn), the first term of the right-hand side in Eq. (C1) becomes Xn∈S(bn)Xv∈V Xv′∈Ne(v) = Xn∈S(bn)Xv∈V Xv′∈Ne(v) = Xen∈S(bn)Xv∈V Xv′∈Ne(v) = 0. NXi=1(cid:18)di(nvi + 1)P (n + 1vi − 1v′i, t) − dinviP (n, t)(cid:19) NXi=1 NXi=1 di(nvi + 1)P (n + 1vi − 1v′i, t) − Xn∈S(bn)Xv∈V Xv′∈Ne(v) dienviP (en, t) − Xn∈S(bn)Xv∈V Xv′∈Ne(v) dinviP (n, t) NXi=1 dinviP (n, t) NXi=1 As the reaction network is linear, the propensity function fvj(n, ω) takes one of two forms: fvj(n, ω) = kjnvi or fvj(n, ω) = kjω, where kj is the reaction rate and i is the index of some species. When fvj(n, ω) = kj nvi, the second term can be transformed as follows: Xn∈S(bn)Xv∈V(cid:18)fvj(n − eVvj, ω)P (n − eVvj, t) − fvj(n, ω)P (n, t)(cid:19) = kj Xn∈S(bn)Xv∈V(cid:18)(nvi − Vij )P (n − eVvj , t) − nviP (n, t)(cid:19) = kj(cid:18) Xn∈S(bn)Xv∈V nviP (n, t)(cid:19) (nvi − Vij )P (n − eVvj , t) − Xn∈S(bn)Xv∈V Vij P (n − eVvj, t) − Xn∈S(bn)bniP (n, t)(cid:19) = kj(cid:18) Xn∈S(bn)Xv∈V nviP (n − eVvj , t) − Xn∈S(bn)Xv∈V = kj(cid:18) Xen∈S(bn−Vj )Xv∈V P (n − eVvj, t) −bniP (bn, t)(cid:19) (envi + Vij)P (en, t) − VijXv∈V Xn∈S(bn) = kj(cid:18)(bni − Vij )P (bn − Vj , t) + Vij VP (bn − Vj, t) − VijVP (bn − Vj, t) −bniP (bn, t)(cid:19) = kj(bni − Vij )P (bn − Vj, t) − kjbniP (bn, t). When fvj(n, ω) = kjω, we similarly have Xn∈S(bn)Xv∈V(cid:18)fvj(n − eVvj , ω)P (n − eVvj , t) − fvj(n, ω)P (n, t)(cid:19) = kj Xn∈S(bn)Xv∈V(cid:18)ωP (n − eVvj, t) − ωP (n, t)(cid:19) = kj(cid:18) Xn∈S(bn)Xv∈V ωP (n, t)(cid:19) ωP (n − eVvj, t) − Xn∈S(bn)Xv∈V P (n, t)(cid:19) = kj(cid:18)Xv∈V P (n − eVvj, t) −Xv∈V ω Xn∈S(bn) ω Xn∈S(bn) = kj ΩP (bn − Vj , t) − kjΩP (bn, t). KXj=1 ∂tP (bn, t) = (fj(bn − Vj, Ω)P (bn − Vj, t) − fj(bn, Ω)P (bn, t)) . The master equation of P (bn, t) is then obtained as (C2) 11 [1] W. J. Blake, M. Kaern, C. R. Cantor, and J. J. Collins, Nature 422, 633 (2003). [2] Y. Lan and G. A. Papoian, Phys. Rev. Lett. 98, 228301 (2007). [3] L. S. Tsimring, Rep. Prog. Phys. 77, 026601 (2014). [4] J. Elf and E. Mans, Genome Res. 13, 2475 (2003). [5] C. A. G´omez-Uribe and G. C. Verghese, J. Chem. Phys. 126, 024109 (2007). R. Tostevin and P. [6] F. ten Wolde, Phys. Rev. E 81, 061917 (2010). [7] K. H. Kim, H. Qian, and H. M. Sauro, J. Chem. Phys. 139, 144108 (2013). [21] M. v. Smoluchowski, Z. Phys. Chem. 92, 129 (1917). [22] S. D. Isaacson, A. and Phys. Rev. E 80, 066106 (2009). A. Isaacson Howard and [23] M. D. Rutenberg, Phys. Rev. Lett. 90, 128102 (2003). [24] D. Fange and J. Elf, PLoS Comput. Biol. 2, 1 (2006). [25] M. J. Lawson, B. Drawert, M. Khammash, L. Petzold, and T.-M. Yi, PLoS Comput. Biol. 9, 1 (2013). [26] F. Benitez, C. Duclut, H. Chat´e, B. Delamotte, I. Dornic, and M. A. Munoz, Phys. Rev. Lett. 117, 100601 (2016). [27] J. Elf and M. Ehrenberg, IET Syst. Biol. 1, 230 (2004). [28] M. Dobrzy´nski, J. V. Rodr´ıguez, J. A. Kaandorp, and R. [8] K. Pilkiewicz and Phys. Rev. E 94, 032412 (2016). [9] H. S. Samanta, M. Hinczewski, Phys. Rev. E 96, 012406 (2017). M. L. Mayo, J. G. Blom, Bioinformatics 23, 1969 (2007). and D. Thirumalai, [29] S. A. Isaacson, SIAM J. Appl. Math. 70, 77 (2009). [30] R. Erban and S. J. Chapman, Phys. Biol. 6, 046001 (2009). [10] T. E. Ouldridge, C. C. Govern, and P. R. ten Wolde, Phys. Rev. X 7, 021004 (2017). [11] R. Ellis, Curr. Opin. Struc. Biol. 11, 114 (2001). [12] B. N. Kholodenko, Nat. Rev. Mol. Cell Biol. 7, 165 (2006). [13] R. Cheong, A. Rhee, C. J. Wang, I. Nemenman, and [31] S. Smith and R. Grima, Phys. Rev. E 93, 052135 (2016). [32] M. Feinberg, Arch. Ration. Mech. Anal. 132, 311 (1995). [33] D. F. Anderson, G. Craciun, and T. G. Kurtz, Bull. Math. Biol. 72, 1947 (2010). [34] G. Szederk´enyi and K. M. Hangos, A. Levchenko, Science 334, 354 (2011). J. Math. Chem. 49, 1163 (2011). [14] N. B. Becker, A. Mugler, and P. R. ten Wolde, Phys. Rev. Lett. 115, 258103 (2015). [15] Y. Hasegawa, Phys. Rev. E 97, 022401 (2018). [16] M. B. Elowitz, A. J. Levine, E. D. Siggia, and P. S. [35] L. Peliti, J. Phys. France 46, 1469 (1985). [36] D. K. Lubensky, Phys. Rev. E 81, 060102 (2010). [37] M. Doi, J. Phys. A: Math. Gen. 9, 1465 (1976). [38] S. Tanase-Nicola, P. B. Warren, and P. R. ten Wolde, Swain, Science 297, 1183 (2002). Phys. Rev. Lett. 97, 068102 (2006). [17] E. M. Ozbudak, M. Thattai, I. Kurtser, A. D. Grossman, [39] D. Cappelletti and C. Wiuf, and A. van Oudenaarden, Nat. Genet. 31, 69 (2002). SIAM J. Appl. Math. 76, 411 (2016). [18] J. S. van Zon, M. J. Morelli, S. Tanase-Nicola, and P. R. [40] E. Friedmann, R. Neumann, and R. Rannacher, ten Wolde, Biophys. J. 91, 4350 (2006). Commun. Math. Anal. 15, 76 (2013). [19] S. Smith and R. Grima, Nat. Commun. 9, 345 (2018). [20] C. Gardiner, Stochastic Methods (Springer-Verlag Berlin Heidelberg, 2009). [41] B. Mayer, G. Emery, D. Berdnik, F. Wirtz-Peitz, and J. A. Knoblich, Curr. Biol. 15, 1847 (2005). [42] E. Levine and T. Hwa, Proc. Natl. Acad. Sci. U.S.A. 104, 9224 (2007).
1201.4758	2	1201	2012-05-07T10:03:38	Directed Transport of Confined Brownian Particles with Torque	[ "physics.bio-ph", "cond-mat.stat-mech" ]	We investigate the influence of an external magnetic field (torque) on the motion of Brownian particles confined in a channel geometry with varying width. Furthermore, the particles are driven by random fluctuations modeled by the Ornstein-Uhlenbeck process (OUP) with given correlation time $\tau_c$. The latter is implemented as both a thermal and a nonthermal process. In contrast to the thermal OUP for the nonthermal process directed transport emerges, i.e. our setup now realizes a ratchet mechanism: Due to the assumed thermodynamic nonequilibrium situation random fluctuations are rectified. The transport quantities of the system are studied in detail with respect to the correlation time, the torque and the channel geometry. Eventually, the mechanism of the symmetry breaking is elucidated.	physics.bio-ph	physics	Directed Transport of Confined Brownian Particles with Torque Department of Physics, Humboldt-Universitat zu Berlin, Newtonstr. 15, 12489 Berlin, Germany Paul K. Radtke and Lutz Schimansky-Geier We investigate the influence of an additional torque on the motion of Brownian particles confined in a channel geometry with varying width. The particles are driven by random fluctuations modeled by an Ornstein-Uhlenbeck process (OUP) with given correlation time τc. The latter causes persistent motion and is implemented as (i) thermal noise in equilibrium and as (ii) noisy propulsion in nonequilibrium. In the nonthermal process a directed transport emerges, its properties are studied in detail with respect to the correlation time, the torque and the channel geometry. Eventually, the transport mechanism is traced back to a persistent sliding of particles along the even boundaries in contrast to scattered motion at uneven or rough ones. PACS numbers: 05.40.-a,87.17.Jj Introduction. Nowadays, ratchet models are em- ployed to explain a wide range of transport phenomena, in biophysics as well as in artificial devices [1, 2]. Gener- ally, the term ratchet refers to nonequilibrium phenom- ena that can arise provided one of the space-time sym- metries that inhibits directed motion is broken [3]. This can be caused either explicitly by an asymmetric, peri- odic structure or by an unbiased, but asymmetric drive [4]. In our case, the symmetry breaking is realized by a nonvanishing mean torque together with the confining geometry. The torque leads to a circular motion and an alter- ation of the diffusive properties [5]. Such motion appears in many real world phenomena. For example, it can be due to asymmetries in the propulsion of agents them- selves as occurring in the chemotaxis of sperm cells [6 -- 8] and nanorods [9, 10]. Also, a torque appears for Janus particles under laser irradiation [11] or due to an external magnetic field that is used to steer nanorods [12]. Furthermore, our particles are confined in an infinitely long channel with a periodically varying width, see Fig. 1. One channel boundary is introduced as a reflecting disk. Later on it will be interchanged by a reflecting tri- angle, thereby adding another symmetry breaking. Fi- nally, we also consider a 'rough' lower wall where elastic scattering is modeled by equidistributed reflection angles, regardless of the angle of incidence. Even though related studies of transport mechanisms possess some of the features, such as the channel layout in chaotic transport in Hamiltonian systems [13, 14], en- tropic particle transport [15] or for confined biological agents [8], its combination with a dissipative dynamics and a nonzero mean torque to realize directed transport is a novel feature. Hereby, the actual driving of the parti- cles is performed by the noise, which delivers a constant energy supply. Thermal Colored Noise. In our setting the par- ticle has unit mass m = 1, position r(t) = (x(t), y(t)) and velocity r(t) = v(t) = (vx(t), vy(t)) at time t, that are driven by correlated noise ǫ(t) = (ǫx(t), ǫy(t)). Such noise is called colored, its memory causes a persistent motion. We choose an OUP to drive our particles, which is a common implementation for it represents the most (a) 1 (b) 1 (c) 1 y 1 y y 1 1 L x W H x x FIG. 1. Illustration of the channel geometry, a box of unit size whose left and right side are identified with each other, real- izing periodic boundary conditions. At the top and bottom specular reflections take place. (a) Lower boundary formed by a disk of radius R = 1.2 that protrudes by L = 0.2 into the box. (b) Edged boundary parameterized by H = 0.5 and W = 2/3. (c) Flat lower boundary with equidistributed scat- tering angles due to a 'rough' surface. natural continuous valued colored noise [16]. In equi- librium a fluctuation dissipation relation (FDR) relates the autocorrelation function of the noise and the friction function hǫi(s)ǫj(s + t)i = Dξ τc e−t/τcδij = kT γ(t)δij, (1) with i, j ∈ {x, y}. Here the Einstein relation Dξ = γ0kT (γ0 -- friction constant, k -- Boltzmann constant, T -- tem- perature) is used to relate the noise intensity Dξ to the heat bath from which the noise derives. Assuming that a FDR holds, we are forced to relinquish the Markov prop- erty by introducing a dissipative memory kernel γ(t− s) to govern the friction. We then speak of a generalized Langevin equation [17, 18] for the velocity, v(t) = −Z t 0 γ(t − s)v(s)ds + ǫ(t) + ¯Ωv(t). (2) The mean torque is realized by multiplication of v with the rotation matrix ¯Ω, resulting in the force ¯Ωv = (Ωvy,−Ωvx). Without confinement, the stationary Fokker-Planck equation corresponding to this dynamics is easily solved. We then find that the velocity is Maxwell-Boltzmann dis- tributed. Numerically, we can confirm that this holds true in the channel geometry as well, both globally for Ω=0.3, τ Ω=1.0, τ Ω=1, τ c=3 c=3 c=10 2 v 0.5 0 -3 -2 -1 0 x 1 2 3 0.5 y 0 -0.5 FIG. 2. (Color online) Trajectories of confined (cf. Fig. 1 (a)) particles exposed to nonthermal colored noise with several torques and correlation times at constant γ0 = 0.2, Dξ = 0.04. The depicted time length is tl = 33, the velocity is indicated by the arrows, between two consecutive arrows ∆t = 1/4 has passed. the entire channel and in the vicinity of the boundaries. Hence we have a symmetrical velocity distribution and no directed transport occurs. The particles perform an undirected diffusive motion. As an estimate one can take the effective diffusion coef- ficient of this dynamics calculated for unbounded space, which reads [18] Deff = Dξ/(cid:0)γ 2 0 + Ω2(cid:1) . (3) Apparently, it does not depend on the correlation time and coincides with the expression for particles driven by white noise. Nonthermal Colored Noise. In the second case we consider noisy propulsion in nonequilibrium, hence no FDR holds [19]. Again equipped with an OUP-driving, our system is governed by the set of Langevin equations r(t) =v(t), ǫ(t) = − v(t) = −γ0v(t) + ¯Ωv(t) + ǫ(t), ǫ(t) + p2Dξ 1 τc ξ(t). τc (4) i.e. Here ξ(t) denotes white Gaussian noise of zero mean with uncorrelated components, hξi(t)i = 0 and hξi(s)ξj (s + t)i = δ(t)δij . In this way, we have effectively realized the motion of an active particle, it underlies a permanent corre- lated supply of energy from the OUP-noise which it dis- sipates during the persistent and curved motion due to Stokes friction. The persistence of this motion expresses nonequilibrium. With τc → 0 white noise follows realiz- ing again an equilibrium situation. By taking the Fourier transforms of eq. (4), the spec- trum of the velocity can be traced back to that of the noise Svivj (ω) = (ω2 + γ 2 0 + Ω2) (γ 2 0 + Ω2 − ω2)2 + 4ω2γ 2 0 Sǫiǫj (ω). (5) theorem and the Wiener-Khintchine With the Lorentzian noise spectrum of the OUP Sǫiǫj (ω) = 2Dξ (cid:0)1 + τ 2 δij, which follows from the left-hand side of eq. (1), we obtain a rather lengthy expression for hv(s)v(s + t)i. For t = 0 it simplifies to c ω2(cid:1)−1 hvivji = Dξ γ0 γ0τc + 1 (γ0τc + 1)2 + Ω2τ 2 c δij. (6) Obviously, the torque, the friction, and the correlation time dampen the particles' mean squared velocity. For a vanishing mean torque the result without external force field [19] is reproduced. We notice that the unconfined effective diffusion coef- ficient can be calculated from the Fourier transform of eq. (5) by integration over the time, which leads to the same result as for thermal colored noise (cf. eq. (3)). Directed Transport in Confined Geometries. Fig. 2 depicts some trajectories in the circular chan- nel for the nonthermal dynamics given by eq. (4). Rising correlation times lead to a reduction of the total spatial displacements as do increasing mean torques, although less severely. With both influences acting on the parti- cles together, another effect can be observed: The parti- cles stay in the vicinity of the reflecting boundaries for a longer time and perform a curly hopping motion that changes to a narrow creeping for large τc [20]. Also, we notice that while the red trajectory (empty arrow- heads) is largely unbiased, with bigger torques the parti- cles travel a longer distance along the flat wall than along the curved wall. The velocity distributions for the same situation are depicted in Fig. 3 [21]. While the velocity is Gaussian distributed for τc = 0, it remains symmetrical on an un- bounded plane or in the confined geometry without con- stant torque. In contrast, if all these influences act on the particles together, i.e. boundaries, finite correlation time and constant torque field, the velocity distribution loses its symmetry (green circles). For Ω = 1, the vx- distribution's left tail is lowered, while its right tail is raised. Thus, we have a net particle flux. Unlike for par- ticles in equilibrium, the velocity distributions are nar- rowing with growing Ω. Furthermore, we notice that the confinement particularly leads to a narrowing in P (vy), which now markedly differs from P (vx). Let us now address the stationary particle fluxes J that pass in the x-direction through our channel. Herein, J is normalized with respect to the number of particles. The numerical results for the net transport are shown in Fig. 4 as a function of the correlation time for several values of the mean torque. The flux J has a maximum at approximately τc ≈ 2 − 3 for all depicted Ω values. For vanishing correlation times, J converges to zero, which is consistent in view of the white noise case that follows 1.8 ) i v ( P 1.2 0.6 free, Ω = 1 Ω = 1 Ω = 0 τ c = 0 0.08 0.04 J 0 -0.04 0 -1.2 -0.8 -0.4 0 vi 0.4 0.8 1.2 -0.08 -6 -3 3 W = 1/3, H = 0.5 W = 2/3, H = 0.5 R = 1.2, L = 0.2 R = 1.2, L = 0.5 R = 1.2, L = 0.8 equidist. scattering 0 Ω 3 6 FIG. 3. (Color online) Velocity distribution (vx -- symbols, vy -- lines) for particles driven by nonthermal colored noise with τc = 1, γ0 = 0.2, Dξ = 0.04. Except for the red distribution (squares), particles are confined as in Fig. 1 (a). For τc = 0, the same Gaussian distribution emerges regardless of Ω. Ω = 0.1 Ω = 0.3 Ω = 1.0 Ω = 3.0 0.03 0.02 J 0.01 0 10-2 10-1 100 τ c 101 102 FIG. 4. (Color online) Net transport for particles driven by nonthermal colored noise for several mean torques at γ0 = 0.2, Dξ = 0.04. The channel geometry is specified in Fig. 1 (a). in this limit. There, the equilibrium condition forbids directed transport as the noise strength in eq. (4) van- ishes. For large correlation times the net transport also converges to zero. The effect of the mean torque is examined in further detail for several geometries in Fig. 5. Evidently, J van- ishes for Ω = 0 and also for strong torques regardless of the mean turning direction. The direction of the net transport changes with the orientation of Ω. Given the emergence of a directed transport, we can justify its qualitative form, i.e. the asymptotics in τc and Ω as well as the occurrence of a maximum, with regard to the behavior of the speed. As implied by eq. (6) it van- ishes both for large mean torques and correlation times. For transport to take place however, we need a torque to break the symmetry, and a nonzero correlation time for nonequilibrium. Hence we have no transport in either of the cases Ω → 0 or Ω → ∞ and τc → 0 or τc → ∞, leaving only a maximum value in between, which has to change its direction with the sign if the sign of the torque switches. We now turn to the effects of the differing geometries in more depth. As we see in Fig. 5, the antisymmetrical be- havior in Ω is conserved only as long as the lower bound- FIG. 5. (Color online) Net transport for various confining geometries (cf. Fig. 1). All curves are at γ0 = 0.2, Dξ = 0.04 and τc = 3. ary keeps as left-right symmetry (cf. Fig. 1 (a,c)). Oth- erwise, as for the edged boundaries (cf. Fig. 1 (b)), the heights of the positive and negative peaks differ, whereas the direction of the current coincides for W = 1/3 and W = 2/3 (i.e. is independent of the orientation of the sawteeth). Also, the position of the maximum moves to slightly smaller Ω values for the edged surface compared to the disk. If we narrow the channel by increasing the origin of the reflecting disk, the resulting transport curves also rise. However, their overall behavior remains; in particular the position of the maximum is barely altered and remains at Ωmax ≈ 1 − 1.5. From this we conclude that the position of the Ωmax cannot be traced to some resonant-like effect between the channel width and the cyclotron radius Rc = v/Ω. Rather, the position of Ωmax, as well as τ max , is a tradeoff between the necessary symmetry breaking and the reduction of the particles' speed. c Due to the unit scaling of the velocity, our net flux is identical to the mean velocity in the x-direction, J = hvxi. Other correlation ratchets [22, 23] display values for hvxi in the same order of magnitude. Elucidation of the Transport Mechanism. There are three possible types of trajectories in our channel: particles in the middle perform a more or less circular motion without net bias, whose regularity is enhanced by Ω and decreased by Dξ and γ0. Particles near the top eventually collide with the wall and are reflected. A counter clockwise mean turning rate then results in a forward motion and multiple further reflections. If we consider a particle that is driven to the wall by a noise pointing into that particular direction, the noise works largely as a drag force after reflection. It is still oriented in the same direction, whereas the particle has changed its own. As a result, the velocity normal to the bound- ary decreases (cf. Fig. 3). Once the noise orientation has changed (after about ∆t ≈ τc), the particle is driven away from the boundary, although this may be counter- acted by the particle's inertia together with the torque. Such situations are shown in Fig. 2. Clearly visible are the repeated reflections with ever decreasing radius and speed at the top. Without additional obstacle, particles near the bottom perform an antagonistic motion that cancels out the for- ward flux. The obstacle on the other hand modulates the reflection angles depending on the point of incidence. As seen in Fig. 2, the distance the particles have to travel until the next reflection varies depending on whether the surface tangential is increasing or decreasing. This con- stitutes a scattering effect that helps particles to escape from the vicinity of the boundary. Hence the backward flux is hindered and a positive net transport emerges. In this picture, we can also explain the augmentation of J with the narrowing of the channel (cf. Fig. 5), for the particles cannot move as far without interaction with the boundaries. Thusly, the proportion of intermediate unbiased motion is reduced and the im- pact of the symmetry breaking is enhanced. We confirm this rationale by considering a box with a flat but rough lower boundary, modeled by random equidistributed reflection angles (cf. Fig. 1 (c) and Fig. 5, black symbols). Obviously, this leads to a similar be- havior of J. Consequently, the flux is dominated by the sliding mo- tion along the flat boundary and affected by the per- sistence, the torque and the noise driven velocity. In contrast, the geometry of the confinement and the shape of the lower boundary appear to be less important and influence the total value of J rather than its form. To evaluate qualitatively the dynamical dependence of the current, we expect that J = J(τc, Ω,v). It has to vanish if any of these three parameters vanishes, which favors a product of them. We can assume proportionality to the 4 velocity, which we approximate by phvvi. Then, the de- pendence from the torque is met by an additional factor √Ω. The flux J rises linearly for small τc, in comparison to the numerical results this must be supplemented by an additional damping for large ones. Hence we propose J ∝ sgn(Ω)pΩhvviτc exp(−τc/4), which mimics the numerical results quite well. It yields a optimal torque at Ωmax = τ −1 c +γ, where the transport has a maximum of J max ∝ τc exp(−τc/4). This behavior can also be seen in Fig. 4, where Ωmax wanders to slightly smaller values with τc, and in the height of the corresponding peaks. It can be interpreted as a time scale merging between the period due to the torque, which is proportional to 1/Ω, and the two relaxation times τc and τinertia = 1/γ that characterize the persistence of the motion. In conclusion, we have constructed a setup in which un- biased correlated noise is rectified in nonequilibrium and leads to the emergence of directed transport. To that end, we needed a nonzero mean torque and an asym- metric confining channel. We have given a qualitative explanation of the mechanism that leads to this sym- metry breaking, namely the particle hopping along the boundaries disrupted by the scattering effect of the re- flecting disk or triangle. Despite the general settings of the propulsion, particles and geometry, this investigation can help in the qualitative understanding and might be the starting point for more precise application oriented modeling of noisy active particles. Acknowledgments. This paper was developed within the scope of the IRTG 1740 funded by the DFG. We acknowledge fruitful discussions with C. van den Broeck, T. Dittrich and B. Sonnenschein. [1] P. Hanggi and F. Marchesoni, Rev. Mod. Phys., 81, 387 wandte Chemie, 117, 754 (2005). (2009). [2] P. Reimann, Phys. Rep., 361, 57 (2002), ISSN 0370- 1573. [13] H. Schanz and M. Prusty, J. Phys. A, 38, 10085 (2005); M. Prusty and H. Schanz, Phys. Rev. Lett., 96, 130601 (2006), ISSN 1079-7114. [3] S. Flach, O. Yevtushenko, and Y. Zolotaryuk, Phys. Rev. [14] W. Acevedo and T. Dittrich, Prog. Theor. Phys. Supp., Lett., 84, 2358 (2000). 150, 313 (2003). [4] I. Goychuk and P. Hanggi, J. Phys. Chem. B, 105, 6642 [15] S. Martens, G. Schmid, L. Schimansky-Geier, and (2001). P. Hanggi, Phys. Rev. E, 83, 051135 (2011). [5] C. Weber, P. K. Radtke, L. Schimansky-Geier, and [16] P. Hanggi and P. Jung, Adv. Chem. Phys., 89, 239 P. Hanggi, Phys. Rev. E, 84, 011132 (2011). (1995). [6] B. M. Friedrich and F. Julicher, New J. Phys., 10, 123025 (2008); Phys. Rev. Lett., 103, 68102 (2009), ISSN 1079- 7114. [7] B. M. Friedrich, I. Riedel-Kruse, J. Howard, and F. Julicher, J. Exp. Biol., 213, 1226 (2010). [8] S. Van Teeffelen and H. Lowen, Phys. Rev. E, 78, 020101 (2008). [9] A. Sen, M. Ibele, Y. Hong, and D. Velegol, Faraday Discuss., 143, 15 (2009). [10] T. Mirkovic, N. Zacharia, G. Scholes, and G. Ozin, Small, 6, 159 (2010). [17] H. Mori, Prog. Theor. Phys., 34, 399 (1965). [18] F. N. C. Paraan, M. P. Solon, and J. P. Esguerra, Phys. Rev. E, 77, 022101 (2008). [19] L. H'walisz, P. Jung, P. Hanggi, P. Talkner, L. Schimansky-Geier, Z. Phys. B, 77, 471 (1989). and [20] A sliding along the channel walls has also been studied in [8] and observed experimentally, see P. Dhar, T. Fischer, Y. Wang, T. Mallouk, W. Paxton, and A. Sen, Nano Lett., 6, 66 (2006). [21] Errorbars are of symbol size and hence omitted. [22] R. Bartussek, P. Reimann, and P. Hanggi, Phys. Rev. [11] H. R. Jiang, N. Yoshinaga, and M. Sano, Phys. Rev. Lett., 76, 1166 (1996). Lett., 105, 268302 (2010). [23] B. Lindner, L. Schimansky-Geier, P. Reimann, P. Hanggi, [12] T. Kline, W. Paxton, T. Mallouk, and A. Sen, Ange- and M. Nagaoka, Phys. Rev. E, 59, 1417 (1999).
1006.3281	1	1006	2010-06-16T17:49:37	Periodic solutions and refractory periods in the soliton theory for nerves and the locust femoral nerve	[ "physics.bio-ph" ]	Close to melting transitions it is possible to propagate solitary electromechanical pulses which reflect many of the experimental features of the nerve pulse including mechanical dislocations and reversible heat production. Here we show that one also obtains the possibility of periodic pulse generation when the boundary condition for the nerve is the conservation of the overall length of the nerve. This condition generates an undershoot beneath the baseline (`hyperpolarization') and a `refractory period', i.e., a minimum distance between pulses. In this paper, we outline the theory for periodic solutions to the wave equation and compare these results to action potentials from the femoral nerve of the locust (locusta migratoria). In particular, we describe the frequently occurring minimum-distance doublet pulses seen in these neurons and compare them to the periodic pulse solutions.	physics.bio-ph	physics	Periodic solutions and refractory periods in the soliton theory for nerves and the locust femoral nerve Edgar Villagran Vargas1,3,∗, Andrei Ludu1,4, Reinhold Hustert5, Peter Gumrich5, Andrew D. Jackson2, and Thomas Heimburg1,∗ 1Membrane Biophysics Group, The Niels Bohr Institute, University of Copenhagen, Denmark 2Niels Bohr International Academy, The Niels Bohr Institute, University of Copenhagen, Denmark 3Departamento de F´ısica, Universidad Aut´onoma del Estado de M´exico, M´exico 4Department of Chemistry and Physics, Northwestern State University, Natchitoches, Louisiana 5Institut fur Zoologie und Anthropologie, Dept. Neurobiologie, University of Gottingen, Germany ABSTRACT Close to melting transitions it is possible to propagate solitary electromechanical pulses which reflect many of the experimental features of the nerve pulse including mechanical dislocations and reversible heat production. Here we show that one also obtains the possibility of periodic pulse generation when the boundary condition for the nerve is the conservation of the overall length of the nerve. This condition generates an undershoot beneath the baseline ('hyperpolarization') and a 'refractory period', i.e., a minimum distance between pulses. In this paper, we outline the theory for periodic solutions to the wave equation and compare these results to action potentials from the femoral nerve of the locust (locusta migratoria). In particular, we describe the frequently occurring minimum-distance doublet pulses seen in these neurons and compare them to the periodic pulse solutions. corresponding authors: E. Villagran Vargas ([email protected]) and T. Heimburg ([email protected]). keywords: solitons, action potential, grasshopper, Hodgkin-Huxley model abbreviations: DPPC: 1,2-dipalmitoyl-sn-glycero-3-phosphocholine; Introduction In several recent publications we have proposed that the ac- tion potential in nerves is an electromechanical solitary wave or soliton [1, 2, 3, 4]. Although the propagation of mechanical pulses in nerves has been discussed before [5, 6, 7, 8, 9], we have provided an quantitative formalism based on thermody- namics for how such pulses are generated and how they propa- gate. An important justification for the assumption of an elec- tromechanical process is the experimental observation of re- versible heat changes in phase with the action potential and a zero net heat release during the action potential (e.g., [10, 11, 12, 13, 14]). This finding is incompatible with an electrical pic- ture of the nerve pulse (Hodgkin-Huxley model [15]) in which currents (of ions) passing through resistors (i.e., ion channel proteins) play a dominant role. The latter model would result in a net heat release that is not found in experiments. The ini- tial heat release during the action potential was also found to be much larger than the free energy required to charge the mem- brane capacitor, which rules out the possibility that a purely electrical picture is able to explain the nerve signal. The pic- ture of the nerve pulse as an electromechanical phenomenon is further supported by the finding of mechanical changes of the nerve under the influence of the action potential, including changes in thickness and length (e.g., [6, 16, 14]). The dynamic equation underlying the soliton description is based on the simple wave equation for sound with the lateral density of the nerve membranes as a variable. When describing sound propagation in air, it is usually assumed that the sound velocity, c0, is constant. This leads to the simple differential equation ∂2∆ρ/∂t2 = c2 0∂2∆ρ/∂x2. Close to the melting transition in membranes, however, the speed of sound is a sensitive function of density [17, 18, 19, 20] and frequency [21, 17]. The simultaneous presence of non-linearity and dis- persion gives rise to the possibility of localized excitations or solitons. The resulting differential equation is somewhat more sophisticated but now displays analytical localized solutions [22, 2].1 Due to the fact that biomembranes carry charges, these solitons are of an electromechanical or piezoelectric nature. Thus, they are not only localized voltage pulses but also display reversible mechanical features and a reversible change in heat that are in agreement with experiment. Further, since our theory is of a thermodynamic nature, it implicitly has a role for all thermodynamic variables, e.g., temperature, pressure, electric potential and the chemical potentials of drugs and ions. This implies that the generation of the nerve pulse can be affected by pH [23], calcium concentration, and general or local anesthetics [2, 23]. While a reversible electromechanical picture of the nerve pulse is compelling and able to address many more physical features of the nervous impulse than traditional models, there remain issues that have not yet been addressed. In particular, this in- cludes refractory periods, hyperpolarization, and pulse trains. The refractory period is the minimum possible time between two nerve pulses. In the Hodgkin-Huxley picture it is a conse- quence of the complicated voltage and time dependence of the 1Here, we use the term 'soliton' as synonymous with 'solitary wave'. Gen- uine solitons can pass through each other without dissipation, which is not the case for our pulses [22]. Strictly speaking, the term solitary wave is thus more appropriate. 0 1 0 2 n u J 6 1 ] h p - o i b . s c i s y h p [ 1 v 1 8 2 3 . 6 0 0 1 : v i X r a 1 opening characteristics of ion channels. In simplified terms, the refractory period is due to the relaxation time of the ion chan- nel proteins after firing and is several milliseconds long, i.e., several pulse widths. Hyperpolarization is the undershoot of the voltage below the resting potential during one phase of the action potential. In the soliton model the action potential consists of a local re- gion of the nerve with high chain order and high local den- sity. An immediate consequence is that the excited nerve con- tracts. This has been observed in isolated nerves [6, 14]. How- ever, it is difficult to imagine that nerves also contract in the native situation when they are embedded in tissue where dis- tances are kept constant. Here, we demonstrate that one can obtain solutions for the electromechanical differential equation that maintain the length of the nerve. They consist of periodic pulses with an intrinsic minimal distance between pulses that is of the order of about 5-10 pulse widths. Using locust femoral nerves, we demonstrate that one finds minimal distances of ex- actly that order. In particular, we describe a phenomenon that we called pulse "doublets" that displays a constant distance be- tween pulses even if measured under different conditions. We have found such doublets in locust femoral nerves and also recently doublets and triplets in crayfish motor neurons. We show that the refractory period in the soliton model is a con- sequence of the conservation of mass and overall length of the nerve. This boundary condition also leads to periodic solutions or pulse trains and to hyperpolarization. Materials and Methods Locust nerve recordings: Adult male and female Locusta migratoria were reared in a laboratory colony at about 30◦ C and with a 12:12 hours dark-light cycle. Experiments were performed using amputated metathoracic legs which were mounted on a platform and fixed on its external lateral face with PlasticineTM (Modeling Clay KR 25, non-toxic, Becher Gottingen). The axis of rotation of the femoro-tibial joint was aligned with the origin, of a semicircular angular plane. Inside a Faraday cage and with the aid of an optical microscope, a small window of the postero-lateral cuticle was removed in order to expose the peripheral lateral internal nerve (properly called posterior tegumental femoral nerve n5B2c [24, 25], and the tibia was cut to a stump to allow oxygenation inside the leg via the trachea. See figure 3 (bottom). In order to record the electrical activity from the single antero-lateral (RDAL) and the two postero-lateral multipolar sensory afferents (RDPL and RVPL) (nomenclature according to [26]), monopolar platinum wire electrodes, 50 µm in diameter, were hooked on the lateral internal nerve n5b2c. With the aid of a pipette, a small drop of locust saline (NACl 140 mM, KCl 10 mM, Na H2PO4 5 mM, CaCl2 2 mM,Saccharose 90 mM; pH 6.8) was deposited on the wounds to avoid the coagulation of the hemolymphe and also used as a reference. A small amount of Vaseline (DAB 10, Roth GmbH, Karlsruhe, Germany) was added to isolate the nerve-electrode system as well as to avoid desiccation of the preparation. The femoro-tibial angle was extended and fixed with an ento- mological pin taking care to avoid lateral strain on the joint. Although primitive, the use of a pin to fix the angle is the most precise method, because the tibia tends always to the flexion and this counterforce allows a perfect static equilibrium at the desired angle of extension. All experiments were performed at room temperature of 22 -- 24◦ C. Detailed Procedure: Results presented in this report belong to 10 samples taken from different animals. In all cases, three electrodes were hooked on the lateral internal nerve n5b2c, and a small amount of Vaseline was added to isolate each electrode-nerve subsys- tem. The distance from the axis of rotation of the joint, O, to the electrodes was 10, 12 and 15 mm, respectively. We did not put the first electrode closer to the origin of the system for fear of damaging the receptors, and the last electrode could not be hooked beyond 17 mm because the nerve under study joins the nerve n5b2 at that point. In all experiments, the tibia was extended from 0 to 80◦ to verify that no activity could be recorded. Afterwards, at each angle from 80 to 150◦ in steps of 10◦, the tibia was fixed completely with an entomological pin and sessions of three minutes were recorded for each angle. We found that 150◦ is the limit of extension since beyond this angle the end of the tibia causes pressure against the distal dorsal edge of the femur and therefore the conditions are no longer physiologic [26]. All experiments were performed at room temperature of 22 -- 24◦ C. The action potentials were picked up externally by platinum wire hook electrodes. A preamplifier (four channels, 1000x amplification, no pass filter, 9V battery source, workshop of the Institute for Zoology, University of Gottingen) was used for primary signal amplification within the Faraday cage, and secondary amplifier (four channels, 10× amplification, 100 Hz high pass filter, 30 kH low pass filter, 15V transformer source) used to pre-filter the signals. The amplified action potentials were digitized via PC based data acquisition hardware (Packard Bell desktop computer, 8 channel RUN Technologies acquisi- tion port, PCI-DAS1200Jr board, Microsoft Windows XP Pro- fessional SP2, 2.66GHz Pentium(R) 4 CPU, 512 MB RAM) and recorded and analyzed with the DATAPAC 2K2 software (RUN Technologies, Mission Viejo, CA, USA). Analysis of the recordings: The smallest digital recording sample period achieved on the system was 49.6 µs; periods of 89.6 and 148.8 µs were cho- sen for longer recordings. No software-based secondary filter- ing was performed on the recorded channels. Pulse height was defined by the voltage distance between maximum depolariza- tion and maximum hyperpolarization in order to account for a 2 potential signal baseline slope during the recording. Pulse du- ration was defined as the time difference between maximum depolarization and maximum hyperpolarization during a pulse. In all pulse trains, the onset of the individual signals was de- fined by the maximum depolarization within the pulse, the off- set by the maximum hyperpolarization. Signal velocity along the nerve was estimated by subtracting the onset times of time- linked signals between the channels and multiplying the inverse by the measured distance between the corresponding hook elec- trodes. The instantaneous pulse period was defined as the pe- riod between the current onset and the onset of the next signal within a pulse train. In each region of interest, pulses (APs) were first identified and time localized via a simple threshold pass analysis. The thresh- old was established visually, positioned at roughly 80% of the maximum depolarization amplitude of the smallest unit of in- terest, then adjusted manually in such a way as to maximize the sensitivity and specificity regarding AP localization, ideally with no noise spikes falsely identified as APs. The onset of the marked pulses was then moved to the first peak depolarization, the offset to the first peak hyperpolarization. Theory In the following we want to derive the equations of motion in a cylindrical membrane along the coordinate x. The lateral density, ρA, of a membrane is larger in the gel phase of the membrane than in the fluid phase because the area per lipid molecule changes by about 24%. For this reason, one can in- duce a phase transition not only by adjusting the temperature but also by changes in lateral pressure or other thermodynamic variables. Let us consider the nerve axon as a one-dimensional cylinder with lateral density excitations moving along the coordinate x. As outlined in [1], sound propagation in the absence of disper- sion is governed by the equation (cid:18) (cid:19) (1) ∆ρA , ∂ ∂x (cid:113) c2 ∂ ∂x ∂2 ∂t2 ∆ρA = where ∆ρA = ρA − ρA 0 is the change in the area density of the membrane as a function of x and t and c = S ρA) is 1/(κA 0 is the den- the density-dependent velocity of sound. Here, ρA sity of the membrane at physiological conditions and is slightly above the melting transition. The above equation is related to the Euler equation in hydrodynamics and will not be derived here. To the extent that the compressibility is independent of the density and the amplitude of the propagating density wave, 0 , the sound velocity is approximately constant ∆ρA << ρA (c = c0). Eq.(1) reduces to ∂2 ∂t2 ∆ρA = c2 0 ∂2 ∂x2 ∆ρA . (2) 3 This is the well-known equation that governs sound propaga- tion in air. However, in several publications we showed that the compressibility, κA S , depends sensitively on temperature (and therefore also on density) close to the melting transition of lipid membranes [17, 18, 19, 20]. Therefore, the above simplifica- tion cannot be made. Both the liquid and gel phases are rel- atively incompressible. At densities near the phase transition where the two phases co-exist, a small increase in pressure can cause a significant increase in density by converting liquid to gel. Near this phase transition, the compression modulus is dramatically smaller. We thus approximate the sound velocity, c, as c2 = 1 ρAκA S = c2 0 + p∆ρA + q(∆ρA)2 (3) 0/ρA 0 and q = 79.5 c2 with p < 0 and q > 0. The sound velocity is also frequency dependent [21]. This means that the system is dispersive, which is a necessary requirement for soliton formation. At very low frequencies, ω ≈ 0, the adiabatic compressibility approaches the isothermal compressibility. In the following, we use the values for the isothermal com- pressibility as a low frequency approximation of the the adia- batic compressibility. For unilamellar DPPC vesicles we found c0 = 176.6 m/s, p = −16.6 c2 0 )2 0/(ρA 0 = 4.035× 10−3 g/m2, assuming a bulk temperature of with ρA T=45◦C [1]. Similar values are obtained for biological mem- branes at physiological temperature. The isentropic compressibility decreases at higher frequencies and leads to a propagation velocity which increases with in- creasing frequency. We approximate such dispersive effects by introducing a term, −h∂4∆ρA/∂x4 with h > 0, in eq. (1). As demonstrated in [1], this term implies a linear dependence of the pulse propagation velocity on frequency. This form of the dispersion term represents the leading term in a typical long- wavelength dispersion relation. In the absence of quantitative knowledge of the dispersion present in nerves, this term has the advantage of being relatively passive in that changes in h alter the spatial size of solitary waves but not their functional form. The inclusion of additional terms can lead to soliton instability and should not be considered without empirical justification for the form chosen. The equation governing sound propagation now becomes (cid:20)(cid:0)c2 0 + p∆ρA + q(∆ρA)2(cid:1) ∂ (cid:21) ∆ρA ∂x ∂2 ∂t2 ∆ρA = ∂ ∂x −h ∂4 ∂x4 ∆ρA . (4) This equation is closely related to the Boussinesq equation [27]. Since we are interested in finding solutions that propa- gate without changing their shape, we transform to a moving coordinate system with z = x − vt. The time- and space- dependent differential equation (eq. 4) can be transformed into a time-independent equation by using ∂∆ρ/∂x = (∂∆ρ/∂z) · (∂∆z/∂x) = ∂∆ρ/∂z and ∂∆ρ/∂t = (∂∆ρ/∂z)· (∂z/∂t) = −v∂∆ρ/∂z. Figure 1: The potential V (∆ρA) from eq. (8) for three different integration constants, C, and a velocity v = 0.8 · c0. The inserts show the potentials in a larger density range showing that they all possess two maxima and one minimum. The electromechanical pulses correspond to a movement in the potential between the limits given by the two dashed lines. Left: For C = 0 g/cm2 one finds a singular solution given in Fig. 2, right top. This solution displays a baseline of ∆ρA = 0. Center: For C = 2 g/s2 one obtains a periodic solution Fig. 2, right, 3rd trace from top. This solution displays a baseline of ∆ρA < 0. Right: For C = 5.72g/s2 one obtains a potential with two maxima of equal height, corresponding to the periodic solution given in Fig. 2, right bottom. This solution also displays a baseline of ∆ρA < 0, and the peak hits a maximum and broadens. (cid:20)(cid:0)c2 0 + p∆ρA + q(∆ρA)2(cid:1) ∂ ∂z Eq.(4) can now be rewritten as v2 ∂2 ∂z2 ∆ρA = ∂ ∂z → −h ∂4 ∂z4 ∆ρA . We now search for periodic solutions. We can now perform two integrals with respect to z from one minimum of the periodic solution to the next h 1 3 1 2 p(∆ρA)2 + 0 − v2)∆ρA + ∂2 q(∆ρA)3 + C . ∂z2 ∆ρA = (c2 (6) This can be easily checked, since the second derivative of eq. (6) yields eq. (5). C is an integration constant. The second integration constant disappears for reasons of symmetry. Now multiply both sides of eq. (6) by ∂(∆ρA)/∂z and integrate once more to yield (cid:18) ∂ (cid:19)2 h ∆ρA ∂z = (c2 0 − v2) (∆ρA)2 + q 6 p 3 (∆ρA)3 + (∆ρA)4 + C∆ρ + V0 . (7) Here, V0 is a further integration constant. In this equation there is one term that is proportional to (∂∆ρA/∂z)2 and a second term that explicitly only depends on ∆ρA. Since a mechanical analogy is useful, we call the first term a "kinetic term" and the second one a "potential term": (cid:18) ∂ h (cid:124) ∆ρA (cid:123)(cid:122) ∂z 'kinetic term' (cid:19)2 (cid:125) −(cid:104) (cid:124) (cid:21) ∆ρA → (5) or h (cid:18) ∂ ∂z (cid:123)(cid:122) (cid:19)2 0 − v2) (∆ρA)2 − p (c2 3 (∆ρA)3 − q 6 'potential term'≡V (∆ρA) (∆ρA)4 − C∆ρ (cid:105) (cid:125) = V0 . (9) ∆ρA + V (∆ρA) = V0 . This equation is formally equivalent to the equation of mo- 2 mv2 + V (x) = const. familiar from classical mechan- tion 1 ics where z plays the role of time and the density change ∆ρA serves as a spatial coordinate. The shape of the soliton is a re- sult of the motion in this potential, and all solutions of eq. (9) can be classified according to this potential energy surface [28]. It should be stressed that this analogy to classical mechanics must not be taken too literally. It is just an analogy because one of the terms depends on the square of the derivative of the density with respect to the coordinate, while the second term is an analytic function of the density. V0 is not relevant be- cause it moves the potential up or down without any influence on the solutions of the differential equation. The value for the integration constant C will depend on experimental boundary conditions. In the following we explore the solutions of eq (9). Results Solitary and periodic pulse solutions: − → (8) Boundary conditions are needed to determine the desired solu- tions. In the following we assume that the total length of the nerve is fixed. This implies that the mean area density of the nerve membranes stays constant (i.e.,(cid:10)∆ρA(cid:11) = 0), and that 4 Figure 2: Solutions to eq. (8) for different values of the velocity vand the integration constant C. Left: For C = 0g/s2 one obtains solitary pulses. Their peak amplitude depends on the velocity, v. Right: Solutions for v = 0.8 · c0 and various values of the integration constant C. For all C > 0g/s2 the solutions are periodic and one finds an undershoot. The pulse distance is a function of C. positive peaks must coexist with regions of negative density change. As shown below, this condition leads to periodic pulse solutions with an undershoot of the pulse (hyperpolarization) and refractory periods. We use the elastic parameters for the lipid DPPC given above. However, as shown in [1] values appropriate for biological membranes would work quite as well. The value for C will be varied. For instance, for a fixed propagation velocity v = 0.8c0 one obtains stable solutions in the range 0 g/s2 ≤ C ≤ 5.72 g/s2. The potential V (∆ρA) is a polynomial of order 4 with negative sign in the highest order term. Therefore, this potential can dis- play two maxima and one minimum. Stable solutions are con- fined to be between the two maxima; all other solutions would be unstable. For C = 0 g/s2 the potential is shown in Fig. 1, left. It displays one local maximum at ∆ρA = 0 and a second maximum at positive ∆ρA with higher value. The shape of the soliton is defined by the movement within the potential indi- cated by the arrow. It is defined by moving within the potential from the local maximum at ∆ρA = 0 to that positive value for ∆ρA for which the potential assumes the identical value. This defines the amplitude of the soliton. One obtains the iso- lated solitary solutions shown in Fig. 2, left. For each ampli- tude one obtains exactly one soliton with well-defined velocity somewhat smaller than the small-amplitude speed of sound, c0. Larger amplitude results in slower propagation velocity. There 5 0 − p2/6q ≈ 0.65 · c0 at which is a limiting velocity, vlimit = c2 the solitons assume maximum amplitude and minimum veloc- ity [1]. The solutions consist of a local change in density above the baseline. The neuron is considered being infinitely long. Therefore the mean density is still the baseline value even if it is locally different from zero. The solitary character is a con- sequence of the potential having zero slope at ∆ρA = 0. If the movement of the membrane starts from a value of ∆ρA > 0 one obtains periodic solutions. However, those solutions dis- play a mean density change larger than zero and are therefore no solutions are possible under the assumptions of constant nerve length. For 0 < C ≤ 5.72 g/s2, the first maximum of V is located at a negative value of ∆ρA. This is shown in Fig. 1, center. Un- der these conditions no solitary solution can exist because that would result in(cid:10)∆ρA(cid:11) < 0. However, there are periodic so- lutions that display a mean density change of zero. Such solu- tions are shown in Fig. 2 (right) a fixed velocity of v = 0.8 · c0. For each value of C and given velocity v pulse separation and amplitude are uniquely defined. The upper limit for the integration constant is C ≈ 5.72 g/s2. This case is shown in Fig. 1 (right). One recognizes that for this value of C the two maxima of the potential assume the same value. This influences the shape of the pulse such that it becomes flat on the top (see Fig. 2 (right), bottom trace). No stable solutions exist for C > 5.72 g/s2 and for C < 0 g/s2. In summary, one finds the following: 1. The integration constant C depends on the experimental conditions. There exist solitary and periodic solutions for the differential equation 9 and the boundary condition of constant mean density. 2. For each C and each velocity v there is exactly one solu- tion with a mean density change of zero. For C = 0 g/s2 this is a singular pulse (soliton) and for C > 0 g/s2 it is a periodic pulse. 3. There is a minimum distance between pulses dictated by the requirement that the mean density is constant, corre- sponding to a refractory period (Fig. 2). 4. For each pair of pulse amplitudes and pulse distances there is exactly one velocity. This implies that the velocity of the pulse train depends on am- plitude and frequency of the stimulation process. It further im- plies that any change of the amplitude resulting from a change in the conditions (e.g., the presence of anesthetics, changes in temperature or pressure, etc.) will have a predictable conse- quence for the propagation velocity and the spatial distance between pulses. This means that the individual pulses of the periodic solutions influence each other. One further result from Fig. 2 is that increasing pulse amplitude first leads to a narrow- ing of the pulses. When the pulses approach their maximum amplitude, they become broader and display a plateau at max- imum density. Pulse shapes of this kind are known in nerves, for example the prolonged action potentials of squid axons that occur under perfusion with aqueous media containing unusual salt composition [29]. The above considerations apply to an axon that is infinitely long. In such a system, there are no constant velocity, shape- preserving solutions involving a finite number of pulse peaks. Solutions in which a finite segment of a periodic solution is joined smoothly to the solution ∆ρA = 0 is expected to be long-lived since only the limited material at the interface can lead to its disintegration. Of course, real nerves have finite length and only a limited number of pulses can fit into a neuron at a time. It is therefore to be expected that the situation in the real nerve is close but not identical to the idealized situation described above. In particular, we suggest that the pulse dou- blets observed in the locust femoral nerve (described below) represent a limiting case of the periodic solutions considered here. The existence of periodic solutions, such as those shown in Fig. 2, is a very robust result. Arguments similar to those made here can be adopted to find periodic solutions for essen- tially any system exhibiting isolated solitons. This includes, e.g., the FitzHugh-Nagumo model [30, 31] which is closely related to the Hodgkin-Huxley model. The existence of (ap- proximate) periodic pulses is thus not likely to be useful in discriminating between models of the action potential. In this Figure 3: Representation of the locust metathoracic leg showing the main nerves and specifically the lateral nerve n5b2 we studied (image adapted from [38]). The femoral nerve extends along the whole leg. The angle of leg extension, α, influences firing frequencies (cf., Fig. 6). connection, we note that the model employed here was devel- oped to describe myelinated nerves. In [1] we suggested that myelination restricts the nerve to longitudinal density perturba- tions. This allowed us to adopt the one-dimensional represen- tation of the propagating pulse given in eq. 8 above. Without a myelin sheet we expect that the mechanical dislocations will contain out-of-plane components which would render the pulse significantly slower. So far, we have no mathematical model for this scenario. However, we expect such pulses to be based on physics similar to the longitudinal pulses described above. Below, we compare our theoretical results to pulse doublets in locust femoral nerves. Locust femoral nerve: Pulse trains and minimum distance doublets The femoral nerve of the locust contains three neurons (Fig. 3) from the extension of the femoro-tibial joint. The three neu- rons typically display periodic patterns with frequencies and amplitudes specific for each unit. The firing frequency also de- pends on the angle of extension, α (see Fig. 3). Therefore, in a recording one can distinguish the firing patterns of the three neurons and analyze them separately (Fig. 4). We have labeled them by signal amplitude and refer to them as neuron 1, 2 and 3. We have measured the pulse velocity by recording the action potentials at two positions of the femoral nerve separated by 5mm (Fig. 4). We found 1. v = 1.02 ± 0.01 m/s for neuron 1 (large amplitude) 2. v = 0.79 ± 0.01 m/s for neuron 2 (medium amplitude) 3. v = 0.73 ± 0.01 m/s for neuron 3 (small amplitude) Most of the pulses in the three neurons have time intervals of several times 10 ms, depending on the angle of extension of the leg. Fig. 5 shows recordings that display pulse doublets with identical amplitudes and separations. They occur in all three 6 not fit more than 2 or 3 pulses in the same neuron at the same time. All pulses with distances larger than 20 -- 30 ms must be regarded as uncorrelated, i.e., about 80-90% of all pulses. By uncorrelated we mean that the pulse shapes of those pulses can- not influence each other because they do not coexist in a nerve. However, the minimum distance pulse doublets are simultane- ously present in the neurons and can therefore display a mutual influence that may influence the separation and shape of the pulses. For this reason we will in the following consider the doublets as an approximation to the periodic solutions in the grasshopper nerve. It should be noted that we recently recorded similar pulse doublets and even triplets in the crayfish motor neuron 3 (data will be published elsewhere). We have recorded a histogram of pulse intervals of neuron 2 as a function of the angles of extension (Fig.6). One finds that for each of the three angles, 90◦, 120◦ and 150◦, the histogram displays 2 maxima. We fitted them to Gaussians (Fig. 6) and determined the mean peak positions and the percentage of doublet intervals (shown in Table 1). About 10% of all intervals in the recordings are related to doublets, corresponding to about 20% of all action potentials being doublet peaks. While we find that the longer interval is strongly affected by the angle of leg extension, the short interval corresponding to the doublet peak intervals is un- affected by the angle of extension within error. For this reason Figure 4: Recordings from the femoral nerve of the lo- cust: Three different pulse originating from three differ- ent neurons that are distinctly different in amplitude and shape. Indicated are the recordings from two electrodes that are separated by 5 mm. The time lags between the two electrodes yield the pulse propagation velocity which is of order 1 m/s for all three neurons (see text). neurons. These pulse doublets occurred quite frequently and account for some 10 -- 20% of all action potentials. The total length of the locust femoral nerve is about 2 cm. The pulse width is about 2 ms. The minimum distance of pulses we have recorded is about 11±2 ms (see below). In 11 ms the nerve pulse travels approximately 1 cm. This implies that one can- Figure 5: Pulse doublets occurring during the record- ings of the femoral nerve. Top: Recording from neuron 2. Center: Recording from neurons 1 and 2. Bottom: The two doublets marked in the top traces given in higher res- olution. Figure 6: Top: Histograms of pulse separation in neu- ron 2 for three different angles of extension: α =150◦, 120◦ and 90◦. Each histogram contains two peaks, one of which is dependent and the other one being independent of the angle of extension, α. Bottom: The small interval peak corresponds to a doublet spacing of about 11 ms. Details are given in table 1. 7 they are listed independently in table 1. The short interval is on average around 11 ms corresponding to the doublet spacing. The width/distance ratio for the doublet peaks is 0.1 − 0.2 de- pending on the somewhat arbitrary criterion for defining width. The closest distance of the theoretical periodic pulses (Fig. 2) corresponds to a width/distance ratio of about 0.125, i.e., it is of same order. Therefore we suggest that the doublets correspond to the closest possible pulses in our theory. Table 1: Mean intervals of action potentials in neuron 2 ex- tracted from the peaks of the distributions in Fig. 6 peak 1 position width [ms] [ms] 11.0 2.4 angle 150◦ 120◦ 90◦ all angles peak 2 position width [ms] 13.5 17.9 25.8 [ms] 39.7 82.5 147.5 % doublet intervals 8.5 9.2 13.9 Discussion this model The Hodgkin-Huxley model for the nerve pulse describes the nerve impulse as a voltage pulse generated by time- and voltage-dependent opening of ion channel proteins. As we have discussed before, is unable to explain a number of features of the nerve pulse including the genera- tion of reversible heat and the observed thickness and length changes [1, 2, 3, 4]. Reversible heat is a particular challenge to the HH-model because it implies that the physics of the nervous impulse is based on thermodynamically reversible processes while the HH-model is exclusively based on irre- versible processes due its reliance on the flow of ions along concentration gradients. In order to address these problems we have proposed that the nerve pulse instead consists of an electromechanical density pulse (soliton) that does not rely on ion channel proteins. In particular, we have shown that close to melting transitions in biomembranes the possibility of soliton propagation, i.e. of localized pulses, exists. This model automatically addresses the observed changes in length [6] and [14] and thickness [32, 16, 33, 34, 14, 35] and the reversible heat production [10, 11, 13, 14, 36]. The key feature of our theory is a localized pulse with con- stant entropy, i.e., without net heat exchange with the environ- ment. Since our theory is of thermodynamic nature, any change in a thermodynamic variable that has the potential to move the membrane through its transition is also able to generate a pulse. This includes pulse generation by changes in tempera- ture, voltage, lateral pressure, pH, calcium concentration, etc. In particular, we have shown that this theory implies a mech- 8 anism to explain anesthesia [2, 23]. So far, however, our the- ory has not included some of the features observed for the ner- vous impulse including hyperpolarization, refractory periods, and pulse trains. These features are addressed in the present paper. Our original approach to pulse propagation refers to an infinitely long axon that is quasi-one-dimensional, mean- ing that the thickness of the nerve is negligible compared to the length of the pulse, and that the density perturbation is of longitudinal nature. The pulse consists of a locally compressed region in which the membrane is transiently pushed through its melting transition while releasing and reabsorbing the latent heat of the transition. For an infinitely long nerve the mean area density in the presence of a singular pulse is unaffected. However, for a real nerve of finite length the overall length of the nerve should decrease while a nerve pulse is traveling along the axon. Such length changes have been observes in isolated nerves [6, 14]. However, it seems unlikely that a nerve within a body can undergo significant length changes because they are attached to muscles or other nerves with more or less fixed po- sitions. Therefore, we must distinguish between nerves with constant length and nerves removed from bodies without any external constraint on length. Here, we have shown that the constraint of fixed overall length leads to the emergence of periodic solutions. These solutions share the property that their average area density is constant, leading to an undershoot under the baseline that is enforced by mass conservation. For each pair of pulse distances and am- plitudes there is a well-defined velocity that is smaller than the speed of sound in such membranes. Further, the model implies that there is a minimum distance between pulses or a maximum frequency in a pulse train. Here, we have provided recordings from locusts (locusta migratoria) that display periodic pulses with long intervals, but with significant likelihood also min- imum distance pulse doublets. In contrast to other pulses in this nerve, the doublet spacing in time is unaffected by changes of the angle of extension of the locust leg, and no pulses of closer distance were found in the individual neurons. Since the femoral nerve is only about 2 cm long and the pulse propagation velocity is about 1 m/s (see results section) one can conclude that action potentials with a distance larger than 20 ms cannot be found simultaneously in a neuron. Therefore, they cannot influence each other and must be considered uncorrelated. In the femoral nerve we found to classes of pulses. One class with large time intervals indicating separations too large for temporal coexistence, and pulse doublets with spatial distances smaller than the overall length of the nerves. Such pulses can influence each other, and we consider them here as an approxi- mation to a periodic solution for a nerve that has a length that is not larger than two pulse distances. The distance of the pulses in the femoral nerve is about 11 ms corresponding to about 1 cm in space. The individual pulse width is 1-2 mm (or 1 -- 2 ms). Thus, the minimum pulse distance in the doublets is about 5-10 pulse widths. The minimum distance in the electromechanical model is about 8 pulse widths. It should be stressed that the the- ory does not allow for stable doublet solutions. However, we consider the doublets as an approximation to such solutions for the case of a nerve with a length only slightly larger than 2 -- 3 pulse intervals. It should be added here that in our electrome- chanical description a simple linear relation between density and voltage has been assumed. Therefore, the integral of volt- age changes over time (or space) should depend in a simple manner on the integrated density changes. Another feature of our theory is that it attributes a meaning to myelination. We have chosen parameters leading to pulse velocities of approximately 100 m/s, comparable to the pulse velocity of myelinated nerves and much larger than those found in the locust femoral nerve. We have noted before that our electromechanical theory assumes longitudinal compression [1], and we do not consider transverse excitations normal to the nerve membrane. However, it is well-known that longitu- dinal wave propagation is significantly faster than transverse propagation. E.g., the speed of sound in bulk water is 1500 m/s while a surface wave displays a typical velocity of about 1 m/s. Therefore, it is to be expected that surface excitations in nerves involving changes in membrane curvature are consid- erably slower than longitudinal pulses. Thus, if one prevents out-of-plane waves by spatial confinement, one expects faster pulse propagation. This is the role of myelinization in the soliton theory. It has already been noted by Abbott et al. [10] that the magnitude of the reversible heat changes suggest that the myelinated regions of the nerve play an active role in the pulse propagation process. This is not the case in the Hodgkin-Huxley model in which myelinization plays the role of an electrical insulator preventing an active amplification of the nerve signal. Therefore, they assigned a special role to the 'nodes of Ranvier' which are the only parts of the myelinated nerve that could amplify the nerve signal. In the soliton model, the 'nodes of Ranvier' do not play an active role different from the rest of the nerve. The construction of a theory for transverse pulse propagation is a far more demanding task. However, such a model would offer a better description of non-myelinated nerves. Thus, although the present results are suggestive, their relevance to the non-myelinated locust femoral nerve has not been demonstrated. A further notable feature of the electromechanical soliton theory is that it does not require ion channel proteins. Such proteins are, however, indispensable in the Hodgkin-Huxley model. In the present model the most important requirement for obtaining localized pulses is the presence of a cooperative melting transition in the biomembrane. This transition is not only responsible for the reversible heat but also for the localiza- tion of the pulse. Interestingly, we and others have shown that in this transition the lipid membrane displays quantized ion conduction events that are indistinguishable from the channel events typically attributed to protein ion channels, even in the complete absence of proteins. In particular, one finds a straight-forward explanation of these quantized current events in inevitable thermodynamic fluctuations. (For a review, see [37]). Thus, during the pulse propagation, one expects that the lipid membrane should become permeable for ions and that quantized conduction events should occur. Similarly, any agency that inhibits pulse formation in the soliton model would also inhibit quantized currents since the two phenomena are unavoidably coupled thermodynamically. Summary We have shown here that the soliton theory for electromechani- cal action potentials in nerves possesses the following features: • It generates localized voltage pulses while displaying a reversible heat and mechanical changes including short- ening and swelling • When the overall length of the nerve is constant, one ob- tains periodic solutions. • One finds an undershoot or hyperpolarization. • The periodic pulses display a minimum distance of about 5-10 pulse widths as a consequence of mass conserva- tion. • This pulse separation is approximately the same as that found in the femoral nerve of the locust. • During the pulse one finds quantized ion conduction events through the membrane resembling those usually attributed to ion channel proteins. All features of the nerve pulses and the generation process dis- play unavoidable thermodynamic couplings with a predictable influence of changes in temperature, lateral pressure, pH, cal- cium concentration or anesthetics. Thus, we have shown that the soliton theory for nerves is able to explain most known fea- tures of nerves, the effect of anesthetics [23] and the emergence of ion channel phenomena from the thermodynamics of the membrane without employing molecular features of the mem- brane components. Acknowledgments: Experiments were partially carried out at the Zoology Dept. of the University of Gottingen, Germany, during a guest visit of E. Villagran Vargas. He acknowledges funding from the mex- ican National Council for Science and Technology (ConaCyt). A. Ludu was a guest of the Niels Bohr International Academy during a visit in 2009. References [1] Heimburg, T., and A. D. Jackson. 2005. On soliton propaga- tion in biomembranes and nerves. Proc. Natl. Acad. Sci. USA 102:9790 -- 9795. 9 [2] Heimburg, T., and A. D. Jackson. 2007. On the action poten- tial as a propagating density pulse and the role of anesthetics. Biophys. Rev. Lett. 2:57 -- 78. [3] Heimburg, T., and A. D. Jackson, 2008. Thermodynamics of the In Structure and Dynamics of Membranous nervous impulse. Interfaces. (Nag, K., editor). Wiley, 317 -- 339. [4] Andersen, S. S. L., A. D. Jackson, and T. Heimburg. 2009. Towards a thermodynamic theory of nerve pulse propagation. Progr. Neurobiol. 88:104 -- 113. [5] Wilke, E. 1913. Das Problem der Reizleitung im Nerven vom Standpunkte der Wellenlehre aus betrachtet. Pflugers Arch. 144:35 -- 38. [6] Wilke, E., and E. Atzler. 1912. Experimentelle Beitrage zum Problem der Reizleitung im Nerven. Pflugers Arch. 146:430 -- 446. [7] Cole, K. S., and H. J. Curtis. 1939. Electrical impedance of the squid giant axon during activity. J. Gen. Physiol. 220:649 -- 670. [8] Hodgkin, A. L., and A. F. Huxley. 1945. Resting and action potentials in single nerve fibres. J. Physiol. London 104:176 -- 195. [9] Kaufmann, K., 1989. Lipid membrane. http://membranes. nbi.dk/Kaufmann/pdf/Kaufmann book5 org. pdf, Caruaru. [10] Abbott, B. C., A. V. Hill, and J. V. Howarth. 1958. The positive and negative heat production associated with a nerve impulse. Proc. R. Soc. London. B 148:149 -- 187. [11] Howarth, J. V., R. Keynes, and J. M. Ritchie. 1968. The origin of the initial heat associated with a single impulse in mammalian non-myelinated nerve fibres. J. Physiol. 194:745 -- 793. [20] Schrader, W., H. Ebel, P. Grabitz, E. Hanke, T. Heimburg, M. Hoeckel, M. Kahle, F. Wente, and U. Kaatze. 2002. Com- pressibility of lipid mixtures studied by calorimetry and ultra- sonic velocity measurements. J. Phys. Chem. B 106:6581 -- 6586. [21] Mitaku, S., and T. Date. 1982. Anomalies of nanosecond ultra- sonic relaxation in the lipid bilayer transition. Biochim. Biophys. Acta 688:411 -- 421. [22] Lautrup, B., A. D. Jackson, and T. Heimburg. 2005. The stability of solitons in biomembranes & nerves. arXiv:physics/0510106 . [23] Heimburg, T., and A. D. Jackson. 2007. The thermodynamics of general anesthesia. Biophys. J. 92:3159 -- 3165. [24] Mucke, A. 1991. Innervation pattern and sensory supli of the midleg of schistocerca gregaria (insecta orthopteroidea). Zoomorph. 110:175 -- 187. [25] Hustert, R., E. Lodde, and W. Gnatzy. 1999. Mechanosensory pegs constitute stridulatory files of grasshoppers. J. Comp. Neu- rol. 410:444 -- 456. [26] Coillot, J. P., and J. Boistel. 1969. ´Etude de l'activit´e electrique propag´ee de r´ecepteurs `a l'´etirement de la patte m´etathoracique du criquet, schistocerca gregaria. J. Insect. Physiol. 15:1449 -- 1470. [27] Remoissenet, M., 1999. Waves Called Solitons. Springer, Berlin. [28] Eichmann, U. A., A. Ludu, and J. P. Draayer. 2002. Analysis and classification of nonlinear dispersive evolution equations in the potential representation. J. Phys. A - Math. Gen. 35:6075 -- 6090. [29] Kobatake, Y., I. Tasaki, and A. Watanabe. 1971. Phase transition in membrane with reference to nerve excitation. Adv. Biophys. 208:1 -- 31. [12] Howarth, J. 1975. Heat production in non-myelinated nerves. [30] FitzHugh, R. 1961. Impulses and physiological states in theo- Phil. Trans. Royal Soc. Lond. 270:425 -- 432. retical models of nerve membrane. Biophys. J. 445 -- 466. [13] Ritchie, J. M., and R. D. Keynes. 1985. The production and absorption of heat associated with electrical activity in nerve and electric organ. Quart. Rev. Biophys. 392:451 -- 476. [31] Nagumo, J., S. Arimoto, and S. Yoshizawa. 1962. An ac- tive pulse transmission line simulating nerve axon. Proc. IRE 50:2061 -- 2070. [14] Tasaki, I., K. Kusano, and M. Byrne. 1989. Rapid mechani- cal and thermal changes in the garfish olfactory nerve associated with a propagated impulse. Biophys. J. 55:1033 -- 1040. [32] Iwasa, K., and I. Tasaki. 1980. Mechanical changes in squid giant-axons associated with production of action poten- tials. Biochem. Biophys. Research Comm. 95:1328 -- 1331. [15] Hodgkin, A. L., and A. F. Huxley. 1952. A quantitative descrip- tion of membrane current and its application to conduction and excitation in nerve. J. Physiol. 117:500 -- 544. [33] Tasaki, I., and P. M. Byrne. 1982. Tetanic contraction of the crab nerve evoked by repetitive stimulation. Biochem. Biophys. Research Comm. 106:1435 -- 1440. [16] Iwasa, K., I. Tasaki, and R. C. Gibbons. 1980. Swelling of nerve fibres associated with action potentials. Science 210. [17] Heimburg, T. 1998. Mechanical aspects of membrane ther- modynamics. Estimation of the mechanical properties of lipid membranes close to the chain melting transition from calorime- try. Biochim. Biophys. Acta 1415:147 -- 162. [18] Halstenberg, S., T. Heimburg, T. Hianik, U. Kaatze, and R. Kri- vanek. 1998. Cholesterol-induced variations in the volume and enthalpy fluctuations of lipid bilayers. Biophys. J. 75:264 -- 271. [19] Ebel, H., P. Grabitz, and T. Heimburg. 2001. Enthalpy and vol- ume changes in lipid membranes. i. the proportionality of heat and volume changes in the lipid melting transition and its im- plication for the elastic constants. J. Phys. Chem. B 105:7353 -- 7360. [34] Tasaki, I., and K. Iwasa. 1982. Further studies of rapid mechan- ical changes in squid giant axon associated with action potential production. Jap. J. Physiol. 32:505 -- 518. [35] Tasaki, I., and M. Byrne. 1990. Volume expansion of non- myelinated nerve fibers during impulse conduction. Biophys. J. 57:633 -- 635. [36] Tasaki, I., and P. M. Byrne. 1992. Heat production associated with a propagated impulse in bullfrog myelinated nerve fibers. Jpn. J. Physiol. 42:805 -- 813. [37] Heimburg, T. 2010. Lipid ion channels. Biophys. Chem., in print. http://arxiv.org/pdf/1001.2524. [38] Heitler, W. J., and M. Burrows. 1977. The locust jump. II. Neural circuits of the motor programme. J. Exp. Biol. 66:221 -- 241. -- -- -- -- -- -- 10
1710.06913	1	1710	2017-08-30T12:46:33	pH sensing properties of flexible, bias-free graphene microelectrodes in complex fluids: from phosphate buffer solution to human serum	[ "physics.bio-ph", "physics.app-ph", "physics.med-ph" ]	Advances in techniques for monitoring pH in complex fluids could have significant impact on analytical and biomedical applications ranging from water quality assessment to in vivo diagnostics. We developed flexible graphene microelectrodes (GEs) for rapid (< 5 seconds), very low power (femtowatt) detection of the pH of complex biofluids. The method is based on real-time measurement of Faradaic charge transfer between the GE and a solution at zero electrical bias. For an idealized sample of phosphate buffer solution (PBS), the Faradaic current varied monotonically and systematically with the pH with resolution of ~0.2 pH unit. The current-pH dependence was well described by a hybrid analytical-computational model where the electric double layer derives from an intrinsic, pH-independent (positive) charge associated with the graphene-water interface and ionizable (negative) charged groups described by the Langmuir-Freundlich adsorption isotherm. We also tested the GEs in more complex bio-solutions. In the case of a ferritin solution, the relative Faradaic current, defined as the difference between the measured current response and a baseline response due to PBS, showed a strong signal associated with the disassembly of the ferritin and the release of ferric ions at pH ~ 2.0. For samples of human serum, the Faradaic current showed a reproducible rapid (<20s) response to pH. By combining the Faradaic current and real time current variation, the methodology is potentially suitable for use to detect tumor-induced changes in extracellular pH.	physics.bio-ph	physics	Article type: Full Paper Title pH sensing properties of flexible, bias-free graphene microelectrodes in complex fluids: from phosphate buffer solution to human serum Jinglei Ping†, Jacquelyn E. Blum‡, Ramya Vishnubhotla†, Amey Vrudhula§, Carl H. Naylor†, Zhaoli Gao†, Jeffery G. Saven‡, & A. T. Charlie Johnson*,† †Department of Physics and Astronomy, University of Pennsylvania, Philadelphia 19104, United States Email: [email protected] ‡Department of Chemistry, University of Pennsylvania, Philadelphia 19104, United States §Department of Bioengineering, University of Pennsylvania, Philadelphia 19104, United States KEYWORDS: pH, graphene, flexible, microelectrode, tumor ABSTRACT: Advances in techniques for monitoring pH in complex fluids could have significant impact on analytical and biomedical applications ranging from water quality assessment to in vivo diagnostics. We developed flexible graphene microelectrodes (GEs) 1 for rapid (< 5 seconds), very low power (femtowatt) detection of the pH of complex biofluids. The method is based on real-time measurement of Faradaic charge transfer between the GE and a solution at zero electrical bias. For an idealized sample of phosphate buffer solution (PBS), the Faradaic current varied monotonically and systematically with the pH with resolution of ~0.2 pH unit. The current-pH dependence was well described by a hybrid analytical-computational model where the electric double layer derives from an intrinsic, pH-independent (positive) charge associated with the graphene-water interface and ionizable (negative) charged groups described by the Langmuir-Freundlich adsorption isotherm. We also tested the GEs in more complex bio- solutions. In the case of a ferritin solution, the relative Faradaic current, defined as the difference between the measured current response and a baseline response due to PBS, showed a strong signal associated with the disassembly of the ferritin and the release of ferric ions at pH ~ 2.0. For samples of human serum, the Faradaic current showed a reproducible rapid (<20s) response to pH. By combining the Faradaic current and real time current variation, the methodology is potentially suitable for use to detect tumor- induced changes in extracellular pH. 2 1. Introduction In vivo monitoring of pH is important in investigations of tissue metabolism, neurophysiology, and diagnostics1. Extracellular pH-sensing, though of great interest for cancer diagnosis and medical treatment1-4, is currently based mainly on relatively slow fluorescent techniques such as fluorogenic pH probes5, 6 and fluorophore-decorated micelles7. Moreover, although optical methods hold promise for in vivo applications, improvement in detection platforms is still needed.8 Other methods for in vivo measurement of tumor pH, including positron emission tomography (PET) radiotracers, magnetic resonance (MR) spectroscopy, and magnetic resonance imaging (MRI), are limited in sensitivity and require expensive instrumentation and exogenous and even radioactive indicators.8 Electrical or electrochemical devices have the potential to be developed for in vivo pH monitoring but they are typically based on metal and glass, making them fragile and bulky. Existing approaches have additional disadvantages including the need for frequent recalibration, excessive power consumption, and lack of biocompatibility1. Flexible field-effect transistors (FETs) based on graphene, a biocompatible9, chemically inert, and scalable10 two-dimensional material with high quality pH-sensing properties11- 19, are promising for monitoring pH changes in biological systems. One important application is in cancer research and diagnostics since tumors demonstrate substantial reduction in extracellular pH2, 20, 21 by 1.5 pH unit (from ~7.5 for healthy tissue to ~6.0 for tumor) but only moderate fluctuations in sodium concentration (~ 7%)22 with respect to normal tissue. However, graphene FETs are commonly operated with ~ 100 mV source-drain bias and ~ 400 mV liquid-gate voltage. The application of these 3 potentials/biases may complicate device fabrication, scaling, and stability; perturb the system under investigation; and set a power (and thus size) constraint on the device. Since each gate-sweep measurement requires ~100 seconds to identify the charge neutrality point that characterizes the pH value, the pH measurement process with a FET is also relatively slow and may not be suitable for real-time monitoring. Here we demonstrate the use of flexible graphene microelectrodes (GEs) 23 for rapid, bias-free pH measurement in phosphate buffer solution (PBS), ferritin solution in PBS (0.1 µM), and human serum. The GE fabrication process is based on scalable photolithographic approaches, and the measurements are conducted without using an external bias voltage 23, so the methodology is intrinsically low-power and minimally perturbative. We find that the spontaneous Faradaic charge transfer between the GE and PBS is modulated by the pH. The Faradaic current extracted from 5 seconds of charge measurement (20 times faster than graphene FETs11-19) varies systematically with the pH of PBS and is very insensitive to moderate fluctuations of the extracellular ionic strength that would be induced by a tumor (~7%). The GE response to pH is well described by a hybrid analytical/computational model where the electric double layer derives from an intrinsic, pH-independent (positive) charge associated with the graphene-water interface and ionizable (negative) charged groups described by the Langmuir-Freundlich adsorption isotherm. For the ferritin solution, we focus on the relative Faradaic current obtained by subtracting the baseline Faradaic current for PBS from that for the ferritin solution. The relative Faradaic current shows a very strong feature that we associate with the disassembly of the ferritin cage and the associated release of ferric ions into the solution. For human serum, the GE reaches equilibrium with the solution in short time 4 (~20 s) and also demonstrates remarkable performance: the Faradaic current responds systematically to pH in the range from 6.0 to 7.6 with high resolution (<0.2 pH unit); the differential current with respect to the pH flips sign and changes by ~ 150% as the pH decreases from 7.1 to 6.4. Together these findings suggest the suitability of the GE for both monitoring of biomolecular activity or protein disassembly in solution and for measurement of pH reduction expected for tumor extracellular fluid (1.5 pH unit)2, 20, 21 in vitro or in vivo. 2. Results and Discussion 2.1. Device Fabrication and Setup Inch-size graphene sheets for scalable electrode fabrication24 were synthesized via low- pressure chemical vapor deposition on copper25, and then transferred onto a flexible Kapton polyimide film with a pre-fabricated array of gold contacts. An Al2O3 sacrificial layer was deposited onto the sample by e-beam evaporation, and then the GE structures were defined using photolithography and oxygen plasma etching. This was followed by spin-coating of a 7-µm thick SU-8 2007 (Microchem) passivation layer, which was patterned to define 100 µm × 100 µm wells over the graphene electrodes. (See Experimental Section for further details of the device fabrication) An example of as- fabricated flexible devices is shown in Figure 1a. The sub-pA Faradaic current between the GEs and the solution under test was measured using an electrometer (Keithley 6517a) with high resolution (~ fC) and low noise (0.75 fC/s peak-to-peak), as shown in Figure 1b. The noninverting input of the electrometer was initially grounded. The GE was exposed to fluid samples with various pH values. To 5 conduct the measurement, the graphene electrode was connected to the inverting input of the operational amplifier of the electrometer, and the charge transferred from the solution to graphene accumulated on the feedback capacitor Cf to provide the readout of the electrometer. 2.2. Modulation of Faradaic Current through pH Variation First, we monitored the Faradaic charge transfer as a function of time for PBS (ionic strength 150 mM) as the pH was decreased from 11.2 to 2.2 and then increased back to 7.1 (Figure 2a). For each pH value, the charge transferred from the solution to the graphene increased linearly with time, with the slope used to determine the Faradaic current 𝑖. In contrast to gate-sweep measurements for graphene FETs, where several minutes might be needed to determine the shift in Dirac voltage that indicates the pH, the Faradaic current measurement described here was completed in less than 5 sec. The Faradaic current decreases monotonically with increasing pH, with excellent reproducibility, and minimal hysteresis (Figure 2b). For pH > 3, the Faradaic current is negative, i.e., electrons are transferred from the solution to the graphene. At low pH (<3), the Faradaic current is positive indicating that the proton concentration in the solution is large enough to reverse the direction of the current. The dependence of the Faradaic current on pH is approximately (but not exactly) linear, with a sensitivity of ~0.12 ± 0.01 pA/pH for pH in the range 2.2 – 11.2. We get an excellent fit to the data (red curve in Figure 1b) using a model that incorporates the electric double layer and ionizable defect groups on graphene, as described in the following paragraphs. 6 The equivalent circuit model26 describing the graphene-solution interface is shown in the inset of Figure 2b. The interfacial capacitance 𝐶# (~ µF cm-2)27 of the graphene/solution interface can be ignored in the DC measurement used here, so the Faradaic current 𝑖 is determined by the electrostatic potential 𝜓% of the Stern plane due to adsorbed charges near the graphene surface and the charge transfer resistance 𝑅'( (~ 6.7 MΩ cm2)23 between the graphene and the ionic solution: 𝑖= 𝜓% 𝑅'(. The measured sensitivity of the GE, 0.12 ± 0.01 pA/pH, is equivalent to 6.8 ± 0.7 mV/pH at the Stern plane, in good agreement with the value16 of ~ 6 mV/pH reported by others for experiments on graphene FETs in ionic aqueous solution using an electrolytic gate. The current-pH dependence can be fit quantitatively using a model where the Grahame equation28 is used to quantify the potential at the Stern plane associated with a surface proton concentration through the the Langmuir-Freundlich isotherm13, 14: density and a set of ionizable defect sites in the graphene whose charge state varies with charge density, 𝜎%, with two components: a constant (i.e., pH-independent) offset charge 𝑖= 𝜓% 𝑅'( (1) 𝜓%=-./01 sinh67 89 :;;<./0' (2) 7=7>? pKa@pH exp6A1B9 ./0+ σoff (3) 𝜎%= In Equation 2, 𝑘E is the Boltzmann constant, 𝑇 the absolute temperature, 𝑒 the electronic charge, 𝜖 (𝜖>) the relative (vacuum) permittivity, and 𝑐 = 150 mM the ionic strength of graphene defects), pKa is the dissociation constant, and 𝑛 the degree of heterogeneity. the solution. In Equation 3, σmax is the areal charge density of ionizable groups (i.e., σmax 7 The parameter σoff allows for the presence of a surface charge density that is independent of pH. Combining Equation 1-3, we obtain an excellent fit to the measured current-pH response, where 𝜎max, 𝑛, pKa, and σoff are the fit parameters (solid line in Figure 2b). The best fit value for 𝜎max is - 0.077 ± 0.005 C m-2, consistent with earlier reports for following paragraph. The best fit values for 𝑛 = 0.24 ± 0.03 and pKa = 6.5 ± 0.1 show reasonable agreement with values of 𝑛 = 0.3 and pKa = 7.6 found by others for single- 2. The best fit value for σoff is 0.007 ± 0.002 C m-2, which we discuss further in the graphene and carbon nanotubes19, 29, 30 with values in the range from - 0.01 to - 0.08 C m- wall carbon nanotubes (SWCNs) in KCl solution30. Our value for pKa is also in the range 4.3 – 9.8 from earlier reports for ionizable groups on graphene31. The pH-independent areal charge density σoff characterizes an intrinsic electric double layer at the graphene-water interface. To provide a molecular basis for this quantity, we first simulated the distribution of water molecules associated with defect-free graphene in contact with pure water with molecular dynamics. The charge density obtained from the simulation was then used to calculate the potential difference as a function of distance from the graphene (Figure 2c). (See Simulation Section for details.) At the graphene surface, a potential of Φ = +360 mV is calculated relative to bulk water. Excess hydrogen density compared to oxygen density close to the graphene surface (z < 0.3 nm) leads to the positive potential. Considering the double-layer capacitance at the graphene-solution interface, ~ 1.3 µF cm-2 (assuming the hydrogen-graphene distance of 0.12 nm), 32 the corresponding charge density is approximately 0.005 C m-2, in good agreement with the value of σoff inferred from the experiment (0.007 ± 0.002 C m-2). 8 2.3. pH Response of Graphene Electrode to Complex Biofluids Building on this understanding of GE operation in an idealized PBS sample, we conducted experiments to explore the use of GEs in more complex biological solutions. As a first step, we used a GE to measure the Faradaic current as a function of pH in a 0.1 µM equine spleen ferritin (Sigma Aldrich F4503) solution in PBS. Ferritin is a globular protein complex of 24 subunits found in most tissues and in serum (pH ~ 7.0) that stores iron oxide and releases it in a controlled fashion. Ferritin is known to disassemble and release the stored iron ions for pH below ~ 2.0 33, with partial disassembly beginning to occur for pH below 4.2 34. Since kapton degrades at low pH below 2.0 35, the GEs for this experiment were fabricated on oxidized silicon substrates. First we measured the Faradaic current for the ferritin solution as a function of pH over the range 1.0 – 7.0, and then we conducted the same measurement for a pure PBS solution to determine a baseline response (data not shown). The pH of all solutions was adjusted in steps of ~ 1.0 pH unit by adding 150 mM hydrochloride acid solution. In order to observe the signature of ferritin disassembly, we focused on the relative Faradaic current (Figure 3), obtained by subtracting the baseline Faradaic current for PBS from that for the 5 µM ferritin solution. The relative Faradaic current increases abruptly at pH near 2.0, exactly in the range where ferritin disassembles and positively charged iron ions enclosed in the intact globular ferritin 24-mer are released. Furthermore, there is a noticeable increase in the relative Faradaic current in the pH range 2.0 - 4.0, in agreement with the expectation that partial disassembly of horse spleen ferritin occurs in this range.34 Thus the pH dependence of the relative Faradaic current for the ferritin 9 solution, although not analytically interpretable, is a sensitive probe of biomolecular processes that substantially change the electrostatic environment of the GE. To test the GE performance in a complex human bio-fluid, we investigated its response to pH changes in a sample of human serum (ThermoFisher) diluted with PBS to bring it to physiological ionic strength ~ 150 mM. The pH range tested was 6.0 (typical extracellular pH for a tumor) to 7.6 (typical for normal tissue), which covers the range of pH variation that can be induced by non-metastatic/metastatic tumor.2, 20, 21 The GE Faradaic current was measured over the same pH range in PBS at ionic strengths of 139.5mM, 150 mM, and 160.5 mM (Figure 4a), corresponding to the variation ionic strength expected in extracellular fluid (~7%)22. Over the relevant pH range, the Faradaic current varied by nearly 0.3 pA (~ 45%), with an estimated pH resolution < 0.2 pH unit and sensitivity of 0.144 ± 0.0098 pA/pH. For fixed pH, the variation of Faradaic current over the range of ionic strengths tested was only 0.01 – 0.02 pA, more than an order of magnitude smaller. For human serum, the Faradaic charge transfer (Figure 4b) had a more gradual time dependence than that for PBS (Figure 2a). The variation of the GE Faradaic current with time in serum was well described by simple relaxation with a single time constant 𝜏 to a 7.60 (Figure 4b,c), the time constant was 𝜏 = 3.81 ± 0.09 s, and over the range of pH constant value that we term the steady-state Faradaic current (Figure 4b). At a pH of tested, this time constant varied by ± 0.5 s. This relaxation time is presumed to reflect equilibration processes such as non-specific adsorption of organic and inorganic components in human serum36 onto the graphene surface, in rough agreement with earlier 10 reports of the saturation time scale for non-specific adsorption of protein onto graphene (~ 30 s) measured with graphene FETs12. The magnitude of the Faradaic current measured in serum (Figure 4d) is smaller by 0.1 – 0.4 pA over the whole pH range than that for PBS. The reduced current magnitude is ascribed to the inhibition of electronic communication to the GE by biomolecules adsorbed onto its surface 37, 38. The differential current with respect to the pH ∆𝐼/∆[𝑝𝐻] can be derived from the current-pH response (Figure 4a), with results shown in Figure 4d. The differential current response shows two different behaviors over the tested range: it is positive for pH 6.1 to 6.6 (saturating at ~ 0.47 pA/pH), and it is negative for pH 6.6 to 7.6 (saturating at ~ - 0.23pA/pH) with an abrupt transition at pH ~ 6.6. Since tumor development almost exclusively leads to acidosis with very rare exceptions,39 tumor-induced pH decrease will result in either Faradaic current reduction in the pH range of 6.1 to 6.6 or increase in the range of 6.6 to 7.6 (Figure 4e). Thus for tumor diagnosis, the range of the pH can be determined from the current variation and further the pH can be identified based on the current-pH response. 3. Conclusion In summary, we have demonstrated the use of flexible graphene microelectrodes for monitoring of pH in idealized and complex bio-solutions, specifically PBS, 5 µM ferritin solution, and human serum. The measurement signal is the zero-bias, sub-pA Faradaic current between the GE and the solution, making this a low-power, minimally 11 perturbative approach. For PBS, the variation of the current with pH can be understood quantitatively in a model where the current reflects the potential of the Stern layer, which is derived from an intrinsic (positive) charge associated with the graphene-water interface and ionizable (negative) charged groups whose density is described by a Langmuir- Freundlich adsorption isotherm. The charge density intrinsic to the graphene-water interface derived from the model is in excellent agreement with that found via molecular dynamics simulation. For the ferritin solution, the relative Faradaic current, compared to a PBS baseline, shows a strong feature at pH ~ 2.0, reflecting the disassembly of the ferritin cage and release of iron atoms. For human serum, the microelectrode rapidly (~ 20 s) reaches equilibrium with the solution. The Faradaic current and the current variation together can be used for identifying pH changes on the scale of that induced by a tumor. This electrode-based technique is therefore potentially suitable for use as a miniature portable or implantable pH-sensor for early-stage cancer diagnosis. 4. Experimental Section Graphene growth Copper foil (99.8% purity) was loaded into a four-inch quartz tube furnace and annealed for 30 minutes at 1050 °C in ultra-high-purity (99.999%) hydrogen atmosphere (flow rate 80 sccm; pressure of 850 mT at the tube outlet) for removal of oxide residues. Graphene was then deposited by low-pressure chemical vapor deposition (CVD) using methane as a precursor (flow rate 45 sccm, growth time of 60 min). Graphene device fabrication Contacts (5 nm/40 nm Cr/Au) were pre-fabricated by photolithography on the device substrate (either a Kapton film or an oxidized silicon wafer). The graphene/copper growth substrate was coated with a 500 nm layer of 12 poly(methyl methacrylate) (PMMA; MicroChem). The PMMA/graphene/copper trilayer was immersed in a 0.1 M NaOH solution and connected to the cathode of a power supply so that the PMMA/graphene bilayer was peeled off the copper substrate by hydrogen bubbles generated between the bilayer and the copper. After thoroughly rinsing with DI- water, the PMMA/graphene bilayer was transferred onto the contact array on the device substrate. After the removal of PMMA with acetone, the film was annealed on a hot plate at 200 °C for 1 hour under ambient conditions. A 5 nm Al2O3 sacrificial layer40 was deposited onto the graphene by e-beam evaporation. Photolithography (AZ 5214 E, Microchem) was then used to define 100 µm x 100 µm graphene electrodes; the AZ MIF developer also removes the Al2O3 sacrificial layer, so unwanted graphene material could then be etched with an oxygen plasma. Next a passivation layer of 7-𝜇m thick photoresist (SU-8; MicroChem) was spin-coated onto the sample, and windows were defined at the locations of the graphene electrodes. Finally, the Al2O3 sacrificial layer graphene electrodes was removed with AZ 422 MIF and the whole as-fabricated film was hard-baked at 200 °C on a hotplate for 2 hours. Biosample preparation Equine spleen ferritin (Sigma Aldrich F4503) samples were prepared at 0.1 µM concentration in full PBS solution (ionic strength 150 mM). Delipidated and dialyzed human serum (ThermoFisher 31876) was diluted by 1.73 times in DI-water, resulting in ionic strength of ~ 150 mM. The pH for solutions of ferritin or human serum was adjusted by adding diluted chloride acid or sodium hydroxide solution. 5. Simulation Section 13 The simulations consisted of two sheets of graphene, generated by the Nanotube Builder plug-in of the Visual Molecular Dynamics software (VMD)41, separated by 20 Å each in contact with atomistic water molecules. Periodic boundary conditions were used and the graphene sheets were positioned parallel to the x-y plane. The x-y dimensions of a periodic rectangular box were selected such that each sheet and its images formed a defect-free, continuous sheet of graphene. Each sheet had dimensions of 50.348 Å by 45.376 Å. The pair of parallel sheets was centered within the box in the z-dimension, and the distance between the two sheets was 20.000 Å. Each sheet contained 924 carbon atoms, the positions of which were constrained throughout the simulation. Water molecules were added with VMD's Solvate plug-in. Water was present above and below the sheets with a vacuum between them. The initial dimensions of the box were 51.577 Å by 46.794 Å by 119.942 Å and the system contained a total of 7153 water molecules. The simulations were carried out in the NPT ensemble at 300 K and 1.0 atm. The area of the periodic box in the x-y plane was held constant, and the cell length in the z direction was allowed to vary. The CHARMM3642-44 force field parameters were used with the NAMD software package45. The water model was the three-site TIPS3P model46-48. The charge on each hydrogen atom is + 0.417e and - 0.834e on each oxygen atom, where e is the elementary charge. Bond distances in water molecules were constrained using the SHAKE algorithm49. Temperature was controlled with a Langevin thermostat with a damping coefficient of 1.0 ps-1. Pressure was controlled with a Langevin piston barostat50, 51 with a period of 200 fs and a damping time of 100 fs. The particle mesh Ewald method was used to calculate long-range electrostatics beyond 14.0 Å, with a grid spacing of 1.0 Å. A 2 fs time step was used. The system was minimized for 20,000 steps, then heated 14 incrementally to 300 K in steps of 5 K and 50 K over 160 ps. The system was equilibrated for 200 ps, then run for 10 ns, with configurations sampled every 1 ps. For each configuration, the charge density of the water molecules was calculated as a function of distance from the plane containing the carbon atoms of graphene. The electric potential (ϕ) as a function of distance from graphene (z) was calculated from the charge density (ρ) using the Poisson equation52, 53: 𝑑-𝜙(𝑧) 𝑑𝑧- =−𝜌(𝑧)𝜀> where ε0 is the vacuum permittivity, 8.854 × 10-12 C m-1 V-1 (F/m) or 5.526 × 10-5 e nm-1 mV-1. This equation is integrated twice under the boundary conditions that electric field and potential are zero in bulk, to give 𝜌𝑧′′ 𝑑𝑧′′ ^ _< 𝜙𝑧 −𝜙(𝑧>)= −1𝜀> 𝑑𝑧′ _ _< =−1𝜀> 𝜌𝑧'𝑑𝑧′ 𝑧−𝑧' _ _< where the final expression is obtained using integration by parts. The bulk, reference value of z0 used was 4.4 nm. Acknowledgement This work was supported by the Defense Advanced Research Projects Agency (DARPA) and the U. S. Army Research Office under grant number W911NF1010093. J.E.B. and J.G.S. acknowledge partial support from NSF CHE 1412496 and from the Penn 15 Laboratory for Research on the Structure of Matter (NSF DMR 1120901). This work used the Extreme Science and Engineering Discovery Environment (XSEDE), which is supported by NSF grant ACI-1053575, under grant TG-CHE 110041. R.V. and C.H.N. acknowledge support from NSF EFRI 2-DARE 1542879. A.V. is grateful for support through Penn's Rachleff Scholars Program. Z.G. acknowledges support through grant 1P30 ES013508 from the National Institute of Environmental Health Sciences, NIH. The contents of this paper are solely the responsibility of the authors and do not necessarily represent the official views of NIEHS, NIH. Author Contributions A.T.C.J. directed the research. J.P. proposed and designed the experiment, fabricated the graphene microelectrodes, and carried out the graphene microelectrode measurements. J.E.B. conducted the all-atom molecular dynamics simulations under the supervision of J.G.S. R.V and A.V prepared the graphene samples, and R.V., A.V., C.H.N., and Z.G. assisted with device fabrication. J.P., J.E.B., J.G.S., and A.T.C.J. wrote the manuscript, with input and approval from all the authors. Reference Korostynska, O.; Arshak, K.; Gill, E.; Arshak, A. IEEE Sensors J. 2008, 8, (1), 20-8. Vinnakota, K.; Kemp, M. L.; Kushmerick, M. J. Biophys J. 2006, 91, (4), 1264-87. Bizzarri, R.; Arcangeli, C.; Arosio, D.; Ricci, F.; Faraci, P.; Cardarelli, F.; Beltram, F. Biophys J. 2006, 90, (9), 1. 2. 3. 3300-14. 4. 5. 6. S. A. 2016, 113, (29), 8177-8181. 7. 136, (31), 11085-11092. 8. 9. Tannock, I. F.; Rotin, D. Cancer Res. 1989, 49, (16), 4373-84. Tung, C.-H.; Qi, J.; Hu, L.; Han, M. S.; Kim, Y. Theranostics 2015, 5, (10), 1166-1174. Anderson, M.; Moshinikova, A.; Engelman, D. M.; Reshetnyak, Y. K.; Andreev, O. A. Proc. Natl. Acad. Sci. U. Ma, X.; Wang, Y.; Zhao, T.; Li, Y.; Su, L.-C.; Wang, Z.; Huang, G.; Sumer, B. D.; Gao, J. J. Am. Chem. Soc. 2014, Zhang, X.; Liu, Y.; Gillies, R. J. J. Nucl. Med. 2010, 51, 1167-1170. Chung, C.; Kim, Y.-K.; Shin, D.; Ryoo, S.-R.; Hong, B.; Min, D.-H. Acc. Chem. Res. 2013, 46, (10), 2211-2224. 16 Ang, P. K.; Chen, W.; Wee, A. T. S.; Loh, K. P. J. Am. Chem. Soc. 2008, 130, (44), 14392-3. Lee, M. H.; Kim, B. J.; Lee, K. H.; SHIN, I.-S.; Huh, W.; Cho, J. H.; Kang, M. S. Nanoscale 2015, 7, (17), 7540- Lerner, M. B.; Matsunaga, F.; Han, G. H.; Hong, S. J.; Xi, J.; Crook, A.; Perez-Aguilar, J. M.; Park, Y. W.; Saven, Sohn, I.-Y.; Kim, D.-J.; Jung, J.-H.; Yoon, O. J.; Thanh, T. N.; Quang, T. T.; Lee, N.-E. Biosensors and Li, X.; Cai, W.; An, J.; Kim, S.; Nah, J.; Yang, D.; Piner, R.; Velamakanni, A.; Jung, I.; Tutuc, E.; Banerjee, S. K.; Kuzum, D.; Takano, H.; Shim, E.; Reed, J. C.; Juul, H.; Richardson, A. G.; de Vries, J.; Bink, H.; Dichter, M. A.; Fu, W.; Nef, C.; Knopfmacher, O.; Tarasov, A.; Weiss, M.; Calame, M.; Schonenberger, C. Nano Lett. 2011, Cheng, Z.; Li, Q.; Li, Z.; Zhou, Q.; Fang, Y. Nano Lett. 2010, 10, (102), 1864-1868. Ristein, J.; Zhang, W.; Speck, F.; Ostler, M.; Ley, L.; Seyller, T. J. Phys. D: Appl. Phys. 2010, 43, (34), 345303. Heller, I.; Chatoor, S.; Männik, J.; Zevenbergen, M.; Dekker, C.; Lemay, S. J. Am. Chem. Soc. 2010, 132, (48), Ohno, Y.; Maehashi, K.; Yamashiro, Y.; Matsumoto, K. Nano Lett. 2009, 9, (9), 3318-3322. Mailly-Giacchetti; Hsu, A.; Wang, H.; Vinciguerra, V.; Pappalardo, F.; Occhipinti, L.; Guidetti, E.; Coffa, S.; Webb, B. A.; Chimenti, M.; Jacobson, M. P.; Barber, D. L. Nature Reviews 2011, 11, (9), 671-677. Madelin, G.; Kline, R.; Walvick, R.; Regatte, R. R. Scientific Reports 2014, 4, (4763), 1-7. Ping, J.; Johnson, A. T. C. Appl. Phys. Lett. 2016, 109, (1), 013103. Lerner, M.; Matsunaga, F.; Han, G.; Hong, S.; Xi, J.; Crook, A.; Perez-Aguilar, J.; Park, Y.; Saven, J.; Liu, R. 10. J. G.; Liu, R.; Johnson, A. T. C. Nano Lett. 2014, 14, (5), 2709-2714. 11. Bioelectronics 2013, 45, (15), 70-6. 12. 13. Kong, J.; Palacios, T. J. Appl. Phys. 2013, 114, (8), 084505. 14. 15. 7544. 16. 11, (9), 3597-3600. 17. 18. 19. 17149-17156. 20. Estrella, V.; Chen, T.; Lloyd, M.; Wojtkowiak, J.; Cornnell, H. H.; Ibrahim-Hashim, A.; Bailey, K.; Balagurunathan, Y.; Rothberg, J. M.; Sloane, B. F.; Johnson, J.; Gatenby, R. A.; Gillies, R. J. Cancer Res. 2012, 73, (5), 1524-1538. 21. 22. 23. 24. Nano Lett. 2014. 25. Colombo, L.; Ruoff, R. S. Science 2009, 324, 1312. 26. Lucas, T. H.; Coulter, D. A.; Cubukcu, E.; Brain, L. Nature Communications 2014, 5, (5259), 1-10. Ping, J.; Xi, J.; Saven, J. G.; Liu, R.; Johnson, A. T. C. Biosensors and Bioelectronics 2015. 27. Israelachvili, J., Intermolecular and surface forces: Revised third edition. Acadmeic Press: 2011. 28. Zuccaro, L.; Krieg, J.; Desideri, A.; Kern, K.; Balasubramanian, K. Scientific Reports 2015, 5, (11794), 1-13. 29. Back, J. H.; Shim, M. The Journal of Physicsl Chemistry B 2006, 110, (47), 23736-23741. 30. Yoon, J.-C.; Thiyagarajan, P.; Ahn, H.-J.; Jang, J.-H. RSC Advances 2015, 5, (77), 62772-62777. 31. Ping, J.; Xi, J.; Saven, J. G.; Liu, R.; Johnson, A. T. C. Biosensors and Bioelectronics 2016, in press. 32. Crichton, R. R.; Bryce, C. F. Biochem. J. 1973, 133, (2), 289-299. 33. Kim, M.; Rho, Y.; Jin, K. S.; Ahn, B.; Jung, S.; Kim, H.; Ree, M. Biomacromolecules 2011, 12, (5), 1629-1640. 34. DeIasi, R.; Russell, J. J. Appl. Polym. Sci. 1971, 15, 2965-2974. 35. Krebs, H. A. Biochemistry 1950, 19, 409-430. 36. Wang, C. H.; Yang, C.; Song, Y. Y.; Gao, W.; Xia, X. H. Advanced Funcional Materials 2005, 15, (8), 1267-1275. 37. Kim, B. S.; Hayes, R. A.; Ralston, J. Carbon 1995, 33, (1), 25-34. 38. Chong, W. H.; Molinolo, A. A.; Chen, C. C.; Collins, M. T. Endocrine-related Cancer 2011, 18, R53-R77. 39. Hsu, A.; Wang, H.; Kim, K. K.; Kong, J.; Palacios, T. IEEE Electron Device Lett. 2011, 32, (8), 1008-1010. 40. Humphrey, W.; Dalke, A.; Schulten, K. Journal of Molecular Graphics 1996, 14, 33-38. 41. Best, R. B.; Zhu, X.; Shim, J.; Lopes, P. E.; Mittal, J.; Feig, M.; D., M. J. A. J. Chem. Theory Comput. 2012, 8, 42. 3257-3273. MacKerellJr., A. D.; Feig, M.; Brooks, C. L. J. Am. Chem. Soc. 2004, 126, 698-699. 43. 44. MacKerell, J., A. D.; Bashford, D.; Bellott, M.; Dunbrack Jr., R. L.; Evanseck, J. D.; Field, M. J.; Fischer, S.; Gao, J.; Guo, H.; Ha, S.; Joseph-McCarthy, D.; Kunchnir, L.; Kuczera, K.; Lau, F. T. K.; Mattos, C.; Michnick, S.; Ngo, T.; Nguyen, D. T.; Prodhom, B.; Reiher, I., W. E.; Roux, B.; Schlenkrich, M.; Smith, J. C.; Stote, R.; Staub, J.; Watanabe, M.; Wiorkiewicz-Kuczera, J.; Yin, D.; Karplus, M. J. Phys. Chem. B 1998, 102, 3586-3616. 45. Schulten, K. Journal of Computational Chemsitry 2005, 26, 1781-1802. 46. Computational Chemsitry 1983, 4, (2), 187-217. 47. 935. 48. 49. Phillips, J. C.; Braun, R.; Wang, W.; Gumbart, J.; Tajkhorshid, E.; Villa, E.; Chipot, C.; Skeel, R. D.; Kale, L.; Brooks, B. R.; Bruccoleir, R. E.; Olafson, B. D.; Sttes, D. J.; Swaminathan, S.; Karplus, M. Journal of Jorgensen, W. L.; Chandrasekhar, J.; Madura, J. D.; Impey, R. W.; Klein, M. L. J. Chem. Phys. 1983, 79, 926- Neria, E.; Fischer, S.; Karplus, M. J. Chem. Phys. 1996, 105, (5), 1902-1921. Ryckaert, J. P.; Ciccotti, G.; Berendsen, H. J. C. J. Comput. Phys. 1977, 23, 317-341. 17 Martyna, G. J.; Tobia, D. J.; Klein, M. L. J. Chem. Phys. 1994, 101, (5), 4177-4189. Feller, S. E.; Zhang, Y.; Pastor, R. W.; Brooks, B. R. J. Chem. Phys. 1995, 103, (11), 4613-4621. Gurtovenko, A. A.; Vattulainen, I. J. Chem. Phys. 2009, 130, (21), 215107. Tieleman, D. P.; Berendsen, H. J. C. J. Chem. Phys. 1996, 105, (11), 4871-4880. 50. 51. 52. 53. 18 Figure 1. a Graphene electrode devices on a flexible polyimide substrate. b Schematic of the device and the measurement configuration. 19 Figure 2. a Real-time Faradaic charge transfer for phosphate buffer solution of various pH values. The solid lines are linear fits to the data. b The Faradaic current extracted from a as a function of the pH. The starting pH was 11.2. The Faradaic current was measured as the pH was decreased to 2.2 (solid symbols) and then increased to 7.1 (open symbols). The solid curve is a fit to an equation derived from Equations 1-3 in the main text. Inset: Equivalent circuit for the graphene-solution interface. c Molecular simulations were used to calculate electrostatic potential Φ(z) (black) and densities of water hydrogen atoms (blue) and water oxygen atoms (red) as functions of z, the distance from the plane containing the graphene carbon nuclei. The densities of oxygen 𝜌O and hydrogen 𝜌H are presented relative to the bulk values for these quantities, 𝜌O,bulk and 𝜌H.bulk. Superimposed on the figure are space-filling representations of graphene (green) and representative configurations of water molecules at three different orientations and distances relative to the graphene surface; graphene and water molecules are rendered on the scale of the abscissa (z). 20 Figure 3. pH-dependence of the relative current for the ferritin solution, defined as the difference between the Faradaic current for the ferritin solution and the baseline current for PBS. 21 Figure 4. a Faradaic current for human serum sample diluted to ionic strength of 150.0 mM (solid circles) and for phosphate buffer solution (PBS) as a function of pH in the physiological range. PBS measurements were made at ionic strength values of 139.5 mM (hollow squares), 150.0 mM (solid squares), and 160.5 mM (hollow triangles). The red curve is a linear fit to the PBS data. b, c Time-dependence of the Faradaic charge transfer (panel b) and Faradaic current (panel c) for human serum at pH = 7.60. The red curves in b and c are fits to a model where the Faradaic current is described by a single relaxation time. d Differential current with respect to pH calculated based on the current response to serum in a. 22
1112.3245	1	1112	2011-12-14T15:19:23	Reconciling cyanobacterial fixed-nitrogen distributions and transport experiments with quantitative modelling	[ "physics.bio-ph", "q-bio.CB" ]	Filamentous cyanobacteria growing in media with insufficient fixed nitrogen differentiate some cells into heterocysts, which fix nitrogen for the remaining vegetative cells. Transport studies have shown both periplasmic and cytoplasmic connections between cells that could transport fixed-nitrogen along the filament. Two experiments have imaged fixed-nitrogen distributions along filaments. In 1974,Wolk et al found a peaked concentration of fixed-nitrogen at heterocysts using autoradiographic techniques. In contrast, in 2007, Popa et al used nanoSIMS to show large dips at the location of heterocysts, with a variable but approximately level distribution between them. With an integrated model of fixed-nitrogen transport and cell growth, we recover the results of both Wolk et al and of Popa et al using the same model parameters. To do this, we account for immobile incorporated fixed-nitrogen and for the differing durations of labeled nitrogen fixation that occurred in the two experiments. The variations seen by Popa et al are consistent with the effects of cell-by-cell variations of growth rates, and mask diffusive gradients. We are unable to rule out a significant amount of periplasmic fN transport.	physics.bio-ph	physics	Reconciling cyanobacterial fixed-nitrogen distributions and transport experiments with quantitative modelling Aidan I Brown and Andrew D Rutenberg Department of Physics and Atmospheric Science, Dalhousie University, Halifax, Nova Scotia, Canada B3H 4R2 E-mail: [email protected], [email protected] Abstract. Filamentous cyanobacteria growing in media with insufficient fixed nitrogen differentiate some cells into heterocysts, which fix nitrogen for the remaining vegetative cells. Transport studies have shown both periplasmic and cytoplasmic connections between cells that could transport fixed-nitrogen along the filament. Two experiments have imaged fixed-nitrogen distributions along filaments. In 1974, Wolk et al found a peaked concentration of fixed-nitrogen at heterocysts using autoradiographic techniques. In contrast, in 2007, Popa et al used nanoSIMS to show large dips at the location of heterocysts, with a variable but approximately level distribution between them. With an integrated model of fixed-nitrogen transport and cell growth, we recover the results of both Wolk et al and of Popa et al using the same model parameters. To do this, we account for immobile incorporated fixed-nitrogen and for the differing durations of labeled nitrogen fixation that occurred in the two experiments. The variations seen by Popa et al are consistent with the effects of cell-by-cell variations of growth rates, and mask diffusive gradients. We are unable to rule out a significant amount of periplasmic fN transport. 1 1 0 2 c e D 4 1 ] h p - o i b . s c i s y h p [ 1 v 5 4 2 3 . 2 1 1 1 : v i X r a Reconciling cyanobacterial metabolite and transport experiments with modelling 2 1. Introduction Filamentous cyanobacteria are model multicellular organisms that form unbranched filaments of clonal cells. In conditions of low exogenous fixed-nitrogen, terminally- differentiated heterocyst cells develop along the filament separated by clusters of photosynthetic vegetative cells [1, 2]. Heterocysts fix dinitrogen [3] and provide the fixed- nitrogen (fN) to vegetative cells where it accommodates ongoing growth. Vegetative cells cannot themselves fix nitrogen due to oxygen produced by photosynthesis [4]. Because fN transport is required for ongoing growth, and because local fN deprivation is a key early signal for heterocyst development, it is important to understand how fN is transported within the cyanobacterial filament. Since fN is a metabolite many convenient genetic visualization techniques are not directly applicable. Nevertheless, radio-isotopes of nitrogen were used to show that fN is rapidly incorporated into glutamine (molecular mass 146 Da) within heterocysts [5, 6]. Fixed-nitrogen is likely transported to vegetative cells as glutamine or other amino acids [1, 4]. Various imaging approaches have indicated cytoplasm-to-cytoplasm connections between adjacent cells and a shared contiguous periplasmic space, either of which might transport fN between cells [7]. Early indications for direct cytoplasmic connections between adjacent vegetative cells within cyanobacterial filaments were electron-microscopy images of small ( ≈ 50A diameter) holes in the septal cell membrane called microplasmodesmata [7, 8, 9]. Larger pores have also been observed connecting heterocysts to adjoining vegetative cells [10]. Recently, direct cytoplasmic transport of the small exogenous fluorophore calcein (molecular mass 623 Da) was observed in Anabaena sp. strain PCC7120 (hereafter PCC7120) using fluorescence recovery after photobleaching (FRAP) [11]. After photobleaching of a cell, fluorescence recovered on a timescale of tens of seconds -- allowing transport constants to be estimated. Since calcein is not native to cyanobacteria, it is thought that direct cytoplasmic transport is nonspecific and could be used to transport fN. The protein SepJ (FraG) appears to play a key role in direct cytoplasmic transport. SepJ includes a coiled-coil domain that prevents filament fragmentation and a permease domain required for diazotrophic growth [12]. SepJ also appears to be needed for microplasmodesmata formation [13]. There are no indications of ATPase activity of SepJ, or any other energy requirements for transport between cells -- indicating that direct cytoplasmic transport of small molecules such as fN between cells is likely to be by passive diffusion. In addition to direct cell-to-cell connections, electron microscopy images indicate that the periplasmic space of (Gram negative) cyanobacterial filaments is contiguous (see Flores et al [14]). Supporting this, transport of periplasmic GFP (molecular mass ≈ 27kDa) has been recently observed in FRAP studies [15], though this is somewhat controversial [2, 7, 16]. Nevertheless, the outer membrane has been shown to provide a permeability barrier to fN compounds [17]. Furthermore, knocking out amino- acid permeases (transporters) that act between the periplasm and cytoplasm leads to Reconciling cyanobacterial metabolite and transport experiments with modelling 3 impaired diazotrophic growth [18, 19, 20]. Together these imply that there may be some amount of periplasmic fN present during diazotrophic growth, which would then passively diffuse between adjacent cells via the contiguous periplasmic space. However, it is not clear how large diffusive fluxes are in the periplasm compared to those between cytoplasmic compartments. There have been two studies of nitrogen transport along cyanobacterial filaments [3, 21], but neither had the transverse spatial resolution to distinguish cytoplasm from periplasm. The first study was by Wolk et al in 1974 [3] (hereafter simply "Wolk"). Diazotrophically growing filaments of Anabaena cylindrica were provided with dinitrogen gas composed of 13N (half-life of less than 10 minutes) for 2 minutes before being imaged. Radioactive decay tracks from 13N were used to approximate the distribution of labeled fN in the filament. Wolk found that the fN was peaked at the location of heterocysts -- consistent with diffusive transport away from a heterocystous source. A more recent study by Popa et al in 2007 [21] (hereafter simply "Popa") studied diazotrophically growing Anabaena oscillarioides filaments with a nanoSIMS (nanometer-scale secondary ion mass spectrometry) technique after they were provided with stable dinitrogen gas composed of 15N. After four hours, they found dips of the fN distribution at the locations of heterocysts, and a noisy plateau between them (see red curve in Fig. 3c, below) -- with no evidence of a concentration gradient consistent with diffusive transport away from heterocysts. It has not been immediately clear how to reconcile the Wolk and Popa fN distributions. It is tempting, but not necessary, to attribute qualitative differences between the Wolk and Popa distributions to distinct species-specific physiology. Instead, we hypothesize that fN transport is qualitatively similar in all heterocystous filamentous cyanobacterial species and investigate the quantitative fN patterns expected from modelling the two experimental approaches with a single model. Wolk et al [3] developed a quantitative model of fN transport and incorporation along the cyanobacterial filament, including diffusive transport and fN consumption in vegetative cells. However, the Wolk model did not distinguish periplasmic from cytoplasmic transport, used a homogeneous continuum model of the filaments, did not include stochasticity, and assumed that cytoplasmic fN was always rate-limiting for growth -- i.e. growth vs local fN concentration was strictly linear. We found (see Appendix A) that with this last assumption the Wolk model was unable to reproduce the distribution of Popa after prolonged production of fN at heterocysts. More recently, a stochastic cellular computational model of cyanobacterial growth and fN transport was developed by Allard et al [22]. However, they assumed that only periplasmic fN transport was significant and they did not investigate the possibility of local fN starvation. We adapt their model, with both cytoplasmic and periplasmic transport but stripped of heterocyst development and supplemented by vegetative growth that is independent of non-zero fN concentrations, to the study of fN distributions. Using this model, we reconcile the two qualitatively distinct fN distributions of Wolk and of Popa. Reconciling cyanobacterial metabolite and transport experiments with modelling 4 Figure 1. Schematic of fixed-nitrogen dynamics as represented by Eqns. 1 and 2. Fixed-nitrogen amounts are indexed to each cell by i, with NC (i) the cytoplasmic fixed-nitrogen and NP (i) the periplasmic fixed-nitrogen of each cell. Each shaded green region represents the cytoplasms of two adjacent cells. DC describes the cytoplasm- cytoplasm transport, DP the periplasm-periplasm transport, DI the import from the periplasm to the cytoplasm, DE the export from the cytoplasm to the periplasm and DL the loss from the periplasm to outside of the filament. We then use the qualitative Popa results to constrain possible periplasmic fN transport. We find periplasmic fN transport is not necessary to explain the current data but cannot be ruled out as a significant contribution. 2. Model 2.1. Fixed-nitrogen transport Our fixed-nitrogen (fN) transport and incorporation model is similar to that of Allard et al [22], and tracks the total amount of cytoplasmic fN, NC(i, t), in each cell i vs. time t: d dt NC(i, t) = ΦC(i) + DINP (i, t) − DENC(i, t) + Gi, (1) where ΦC(i) ≡ ΦRC (i−1) + ΦLC(i+ 1)−ΦLC(i)−ΦRC(i) is the net diffusive cytoplasmic flux into cell i and is equal to the sum of the fluxes into the cell from the left and right minus the fluxes out of the cell to the left and the right. Similarly we track the corresponding periplasmic fixed-nitrogen NP (i, t): d dt NP (i, t) = ΦP (i) − (DI + DL)NP (i, t) + DENC(i, t), (2) where, similarly, ΦP (i) ≡ ΦRP (i − 1) + ΦLP (i + 1)− ΦLP (i) − ΦRP (i) is the net diffusive periplasmic flux into the periplasm of cell i. In addition to fluxes along the filament, DI is the coefficient for periplasm-to-cytoplasm transport, DE is the coefficient for cytoplasm- to-periplasm transport, and DL is the coefficient for losses from the periplasm to the external medium. These transport processes are shown schematically in Fig. 1. Note that any fN transferred between cytoplasm and periplasm, with DI or DE, shows up with opposite sign in Eqns. 1 and 2 -- so that the total amount of free fN is unchanged by the exchange. In addition to these transport terms, there is a source/sink term Gi in Reconciling cyanobacterial metabolite and transport experiments with modelling 5 Eqn. 1 that describes fN production and consumption in the cytoplasms of heterocysts and vegetative cells, respectively. This G term is discussed more extensively below in Sec. 2.2. Diffusive transport between cells is described by the flux terms Φ. Following Allard et al [22] and Mullineaux et al [11] we assume that they are limited by the junctions between cells. Supporting this, there is no evidence in the images of fN by Popa et al [21] or of calcein by Mullineaux et al [11] of subcellular gradients. Fick's law states that net diffusive fluxes between two cells will be proportional to the density difference between the cells. Each cell will have two outgoing fluxes, one to the left neighbour ΦL and one to the right neighbour ΦR, each proportional to the local density N(i, t)/L(i, t) and to transport coefficients DC or DP for cytoplasmic or periplasmic fluxes, respectively. Each cell will also have two incoming fluxes from its neighbouring cells. The value of the transport coefficients DC or DP will be proportional to the microscopic diffusivity of fN, but will also depend on the size and shape of cytoplasmic or periplasmic connections between cells -- which are not well characterized. Nevertheless, we can take the cytoplasm-to-cytoplasm transport coefficients for calcein [11] and simply scale them to glutamine under the assumption that contributions due to connection geometry are unchanged. Calcein transport coefficients reported by Mullineaux et [11] assumed a constant cell length. Multiplying their E in units of s−1 by a al length of 3.38 µm, gives our transport coefficients in units of µm s−1. We cell then scale from calcein to glutamine with a factor of the cube-root of the ratio of molecular masses, assuming Stokes-Einstein diffusivity that inversely depends upon radius. We obtain DC=1.54 µm s−1 between two vegetative cells and DC=0.19 µm s−1 between a vegetative cell and a heterocyst -- we keep these values fixed in this paper. Note that with uniform transport coefficients, such as among clusters of vegetative cells, the flux terms in in Eqn. 1 or 2 look like a discrete Laplacian, ∇2 d(N/L) ≡ N(i + 1, t)/L(i + 1, t) + N(i− 1, t)/L(i− 1, t)− 2N(i, t)/L(i, t)) and lead to the more familiar diffusion equation when coarse-grained (see Eqn. 5 and Appendix A, below). We can estimate a lower bound of periplasmic fN transport with the GFP-FRAP data of Mariscal et al [15]. They observed a fluorescence recovery after photobleaching timescale of approximately 150s [15]. This corresponds to DP gf p ≈ L/150s ≈ 0.02µm s−1, using a cell length L ≈ 3µm. Scaling the transport of GFP (27 kDa) to glutamine (146 Da) by the cube-root of the molecular masses [23], i.e. by a factor of (27 kDa/146 Da)1/3 ≈ 5.7, and taking this as a lower bound, then gives DP & 0.1 µm/s. However, the relatively large GFP may diffuse particularly slowly through periplasmic connections compared to glutamine. FRAP experiments with GFP in Anabaena PCC 7120 showed recovery in part of a single cell's periplasm after around 60s [16] and after around 5s [15]. This faster time of 5s is similar to the 4s recovery time observed in the periplasm of E. coli [24], which also measured an associated diffusivity of Dcoli ≈ 2.6µm2/s. We use this diffusivity to determine an upper bound of DP by scaling it to glutamine and dividing by cell length to obtain DP = Dcoli × 5.7/3.38µm ≈ 4.4µm/s. The broad range Reconciling cyanobacterial metabolite and transport experiments with modelling 6 we consider for periplasmic transport is then 0.1 µm s−1 . DP . 10 µm s−1. When we consider periplasmic transport, we must also consider exchange between the cytoplasm and the periplasm -- as well as loss from the periplasm. We expect that uptake of fN from the periplasmic to the cytoplasmic compartment, controlled by DI in Eqns. 1 and 2, is an active process, controlled by e.g. ABC transporters [20]. Nevertheless, we expect that the rate of import is proportional to the periplasmic density, NP /L, and to the number of transporters, which will themselves be proportional to the cell length ∼ L. Similarly, leakage of fN from the cytoplasmic to the periplasmic compartment, controlled by DE, will be proportional to the cytoplasmic density NC/L but also to the amount of membrane or of (leaky) ABC transporters (∼ L) [25]. The result is transport terms DI and DE that are independent of length L and have units of s−1. Using the transport data of Pernil et al [20] for uptake of extracellular glutamine, we estimate (see Appendix B) DI = 4.98 s−1 and DE = 0.498 s−1. We also consider the possibility of stronger transport, with DI = 49.8 s−1 with DE = 4.98 s−1, because the outer membrane permeability barrier [17] may decrease extracellular uptake with respect to the periplasmic uptake DI. There is evidence that fixed-nitrogen is lost from the filament to the external medium [26, 27] and we include a loss term for the periplasm with coefficient DL in the Eq. 2. We take the loss rate as very small compared to the reuptake rate, using 1% of the lower DI value i.e. DL = 0.0498 s−1 -- qualitatively corresponding to the observation of a permeability barrier in the outer membrane [17]. Our results are qualitatively unchanged in the range of 0 ≤ DL/DI . 0.1. 2.2. Cell Growth and Division We take PCC7120 cells to have a minimum size of Lmin = 2.25µm and a maximum size of Lmax = 2Lmin [1, 2]. When a cell reaches Lmax it is divided into two cells of equal length, each of which is randomly assigned a new growth rate and half of the fixed nitrogen in the parent cell. We found that average concentration profiles were largely insensitive to stochastic effects. Accordingly, most of our results came from a deterministic model where all cells were initialized with the same length L0 = 2.8µm and the same doubling time TD = 20h [19, 28]. For our stochastic results, presented in Figs. 3(b) and (c), we initialized lengths randomly from an analytical steady state distribution of cell lengths ranging between Lmin and Lmax [29]. We define a minimum doubling time Tmin = TD − ∆ and a maximum doubling time Tmax = TD + ∆. These then define a minimum growth rate Rmin = Lmin/Tmax and a maximum growth rate Rmax = Lmin/Tmin and we randomly and uniformly select optimal growth rates Ropt from the range Ri ∈ [Rmin, Rmax]. The standard deviation of the maximal growth rate is then σR = p2/3∆Lmin/[(TD + ∆)(TD − ∆)]. We expect that the coefficient of variation of the growth rate σR/Ravg is of similar magnitude to that seen in a study of mutant Anabaena [22], which was σR/Ravg ≈ 0.165. In Eqn. 1, for vegetative cells Gi = Gveg is a growth term determining the rate of freely diffusing cytoplasmic fN removed to support growth of the cell. Gveg depends on i Reconciling cyanobacterial metabolite and transport experiments with modelling 7 the actual growth rate R of the cell (in µm/s), which in turn depends upon the locally available cytoplasmic fN: Gveg = −gR(Ropt i , NC(i, t)), (3) where g is the amount of fixed nitrogen a cell needs to grow per µm -- which we now estimate. Dunn and Wolk [30] measure a dry mass for one cell of A. cylindrica of 1.65×10−11g. Cobb et al [31] measured the nitrogen content of A. cylindrica to be 5-10% of dry mass (see also [32]). We take their typical value of 10% [31] and scale it by volume from A. cylindrica to PCC7120. A. cylindrica appears to be 1.5× as wide and 1.5× as long as PCC7120 [1], giving ≈ 2.07×1010 N atoms per average cell of PCC7120. According to Powell's steady state length distribution of growing bacteria [29] the average cell size is 1.44× as long as the smallest cell, so that ≈ 1.4×1010 N atoms are needed for a newly born cell to double in length. This implies that g = 1.4 × 1010/Lmin ≃ 6.2 × 109µm−1. We assume that cells increase their length at their optimal growth rate Ropt as long as there is available cytoplasmic fixed-nitrogen (see discussion in Sec. 4). Otherwise, they can only grow using the fixed-nitrogen flux into the cell from neighbouring cells and the periplasm: i R =( Ropt i , min(Φin/g, Ropt i ) if NC(i, t) > 0 if NC(i, t) = 0 (4) Note that cells with NC = 0 may still grow, but will be limited by the incoming fluxes of fixed-nitrogen from adjoining cells and the periplasm, Gveg = Φin = ΦRC(i − 1) + ΦLC(i + 1) + DINP (i). When the incoming flux exceeds the requirement of maximal growth, then the cytoplasmic nitrogen concentration will rapidly become non-zero since we always have R ≤ Ropt In Eqn. 1, for heterocyst cells Gi = Ghet. The heterocyst fN production rate Ghet is chosen to supply the growth of approximately 20 vegetative cells and is 3.15×106 s−1 unless otherwise stated. . i 2.3. Some numerical details Filaments were initiated with two heterocysts, separated on either side by vegetative cells, on a periodic loop. Periodic boundary conditions were used to minimize end effects. Vegetative cells were then allowed to grow and divide, subject to available fixed- nitrogen. Because of strong spatial gradients of free (unincorporated) fixed-nitrogen between heterocysts, we group and report our data with respect to the number of vegetative cells between two heterocysts -- typically we took a separation of 20 cells. This is approximately twice the typical heterocyst separation [1, 2], which allows us to investigate fixed-nitrogen depletion at the midpoint -- corresponding to the location of an intercalating heterocyst. We distinguish free from incorporated fixed-nitrogen (fN). Free fN is freely diffusing in the cytoplasm, and is simply given by NC(i, t). We report this as a linear density Reconciling cyanobacterial metabolite and transport experiments with modelling 8 ρF ≡ NC(i)/Li. Incorporated fN is the fixed-nitrogen that has been incorporated by cellular growth through the growth term Gveg. We report the incorporated concentration ρI ≡ R Gvegdt/Li, where the growth is integrated over a fixed duration. During cell division, half of any previously incorporated fN is assigned to either daughter cell. Note that since we are interested in isotopically labeled fixed-nitrogen [3, 21], we set all NC(i) = 0 at the start of our data gathering (t = 0), corresponding to the introduction of isotopically labeled dinitrogen to the filament that is then fixed by the heterocysts and subsequently supplied to the filament as free fixed-nitrogen via Ghet. The total fN, for e.g. Fig. 4, in a cell is ρT ≡ ρF + NP /L + ρI, where we include periplasmic contributions as well. 3. Results and Discussion 3.1. Short-time fixed-nitrogen distributions We first examined systems with only cytoplasmic transport (i.e. DI = DE = DP = 0) at short times comparable to those of Wolk et al [3]. In Fig. 2 we show fixed-nitrogen (fN) distributions along the filament, with respect to distance in cells from the nearest heterocyst, i. The cytoplasmic transport parameter was held constant at DC = 1.54 µm/s, estimated from calcein transport values in Section 2.1. We varied the time-interval between the start of labeled nitrogen fixation (t = 0) and the times (t = 10s, 30s, 60s, 120s, and 180s -- see legend in panel (b)) at which the fN concentration profile along the filament was considered. As shown in (a), the incorporated fN, ρI , increases steadily over time for all vegetative cells reached by free fN, while incorporated fN remains at zero in the non-growing heterocyst ("H"). As shown in (b), the free fN, ρF , spikes at the heterocyst, where it is produced, and progressively decreases with distance away from the heterocyst. At later times, the free fN approaches a steady-state -- the distribution does not change significantly with time. We show in (c) the total fN, ρT = ρI + ρF . At very short times the free fN dominates the total, resulting in the spike in the distribution at the heterocyst. At longer times the incorporated fN dominates the total -- which then loses its spike at the heterocyst and instead begins to form a dip at the heterocyst (see below). The radiographic technique of Wolk et al [3] does not distinguish between free and incorporated fN, so it is appropriate to compare it with the total fN ρT from Fig. 2(c). In (d) we show the average of the two distributions determined by Wolk et al [3], with filled red circles (left scale). We superimpose a qualitatively similar ρT curve at t = 30s with the blue curve (right scale). We have shifted the ρT = 0 axis to agree with the experimental background far from the heterocyst, and have adjusted the scale so that the curves agree at the heterocyst. The agreement is not bad, and captures the qualitative feature of the Wolk data -- namely the distinctive spike at the heterocyst. This spike is a feature of short exposure to labeled fN. Indeed, while Wolk et al took data at t = 120s we find our best qualitative agreement at t = 30s. In our model heterocysts immediately Reconciling cyanobacterial metabolite and transport experiments with modelling 9 Figure 2. Model fixed-nitrogen (fN) concentration vs. cell index i (the number of cells from a heterocyst at i = 0), for short intervals of labeled nitrogen fixation ranging from 10s to 180s as indicated by the legend in (b). The incorporated fN ρI is shown in (a), the free fN ρF in (b), and the total fN ρT in (c). In (d) the red circles indicate the experimental data digitized from Wolk et al [3], and the blue squares are the t = 30s ρT data from (c) (using the right axis, scaled and shifted to best agree with experimental data). Our model data is from our deterministic model with no periplasmic transport, i.e. DC = 1.54 µm s−1 and DI = DE = DP = 0. provide fN able to be transported. This is not the case in real heterocysts: glutamine levels in Anabaena cylindrica linearly increase during the first minute of exposure to N2, and other fixed nitrogen products appear after two minutes [6]. This one to two minute lag between exposure to labeled fN and its transport and incorporation is consistent with the 90s difference between the illustrative curves in Fig. 2(d). 3.2. Steady-state fixed-nitrogen distributions As seen in the last section, for t & 120s the free fN distribution has reached an approximate steady-state. We can obtain an analytic expression for this steady-state distribution of ρF by approximating the filament as a continuous diffusive medium. The Reconciling cyanobacterial metabolite and transport experiments with modelling diffusion and consumption of free fN within a continuous filament is ∂ρF /∂t = D∇2ρF − c 10 (5) with ρF as the free fN concentration; D is the diffusivity and c is the fN consumed by growth per unit length. For the steady state (∂ρF /∂t = 0) the general solution is quadratic (parabolic) vs. distance x from a nearby heterocyst. In steady-state, the flux on either side of the heterocyst at x = 0 equals half of the fN production, i.e. Ghet/2 = −D∂ρF /∂x. This flux, Ghet/2, must also equal the total fN consumed by the growing vegetative cells on either side of it, i.e. cx0 where x0 is where growth halts and ρF (x0) = 0. Applying ρF (Ghet/(2c)) = 0, and −D∂ρF /∂x(0) = Ghet/2 gives us the steady-state profile ρF (x) = 1 2 c D x2 − Ghet 2D x + G2 het 8cD , (6) for x ≤ x0 and ρF (x) = 0 for x ≥ x0. This is in excellent agreement with the steady- state ρF from our numerical model with non-stochastic cytoplasmic-only transport, as shown in Figure 3(a). In our non-stochastic model all cells with free fN levels above zero grow at the same rate and so incorporate fN at the same rate. This produces a constant plateau in ρI for vegetative cells between the heterocyst at x = 0 and the cell where the free fN vanishes, i.e. for 0 < x < x0. This is seen in our discrete cellular model, as shown with ρT in Fig. 3(a). We note that the cells at i = 9 and i = 11 grow at a reduced rate despite having ρF = 0, since some free fN is provided from their neighbours -- i.e. Φin > 0 in Eq. 4. The plateau height depends upon how much time has elapsed since labeled fN was introduced. In Fig. 3(a) we show results for t = 4h, to be able to directly compare with the data of Popa [21]. The predicted plateau height (<Ropt>gt/<L>) from our previous continuum model is shown with a black dashed line, and the agreement is excellent. We also see that the free fixed-nitrogen ρF (scale on right side) is approximately two orders of magnitude smaller than the total fixed-nitrogen ρT (scale on left side), which is dominated by the incorporated fixed-nitrogen ρI. We see excellent qualitative agreement with the plateau-like distribution seen by Popa. The expected gradient of free fixed- nitrogen is too small to be easily resolved by their technique, compared to the much larger plateau of incorporated fixed-nitrogen. 3.3. Stochasticity of Long Time Distributions Fig. 3(b) shows the total fN distribution using our stochastic model that has been averaged over 1000 vegetative segments between two heterocysts. Qualitatively it looks like our deterministic model results. In (c) we show both the experimental results of Popa et al [21] (red curve, left axis) and one single sample distribution from our stochastic simulation (blue curve, right axis, with an arbitrary factor to separate the curves). Qualitatively the distributions are very similar, but strikingly rougher than the smooth curves in (a) and (b). The ρF in (c) is smooth and relatively small in magnitude, indicating that the variation is almost entirely due to incorporated fN. Reconciling cyanobacterial metabolite and transport experiments with modelling 11 Figure 3. (a) non-stochastic concentrations after t = 4h of labeled fixed-nitrogen (fN) exposure vs. the cell index i. Heterocysts are indicated by ("H"). Shown are the total fN (ρT , blue solid line with circles using the left axis) and the free fN (ρF , red solid line with squares using the right axis). Note the dramatic difference in scale. The steady- state analytic result for the free fN from a continuum approximation, Eqn. 6, is shown as a green line; while the corresponding plateau in ρT due to incorporated fN is shown by the black dashed line. (b) averaged stochastic ρT using doubling-time variation ∆ = 4.5h, with error bars too small to be seen outside data points. The average is over 1000 distributions. (c) the experimental fN concentrations of Popa et al [21] in red circles (scale on left axis) and from a single stochastic result from our model using ∆ = 4.5h and Ghet = 4.25×106 s−1 with ρT in blue squares and 25ρF in green diamonds (scale on right axis, offset for clarity). Also shown as black lines are the vegetative cell averages. (d) the coefficient of variation in vegetative cell fN levels, σT /∆ρT , vs. the coefficient of variation in the growth rate of the vegetative cells, σR/Ravg. The error bars indicate statistical errors of the average variation measured over 100 samples. The grey shaded region indicates the size of the standard deviation, which corresponds to the expected error for a single sample. The red dashed line indicates the variation in the experimental data of Popa et al [21], with a single sample. For (a)-(d), we have used only cytoplasmic transport with DC = 1.54 µm s−1 and DI = DE = DP = 0. Reconciling cyanobacterial metabolite and transport experiments with modelling 12 In our stochastic simulation, the variation of ρT shown in (c) is due both to the varying growth rates of cells along the filament and to their varying lengths -- which leads to corresponding differences in the density of incorporated fN after a fixed time. This raises the possibility that the variability seen by Popa et al [21] is not due to instrumental or sampling noise, but simply reflects growth rate variability in the filament. In (d) we plot the coefficient of variation of the total-fixed nitrogen, σT /∆ρT , vs the coefficient of growth rate variation, σR/Ravg. Here σR is the standard deviation of the growth rate distribution, and Ravg is the average growth rate. Similarly σT is the standard deviation of ρT in vegetative cells, while ∆ρT is the difference between the average ρT in vegetative cells and in heterocysts. While σR/Ravg was calculated exactly, we measured σT /∆ρT over 100 segments of cells, and used this to determine the average coefficient of variance, the statistical errors, and the standard deviation around the average -- as shown in (d). With the dashed red line, we also show σT /∆ρT from the Popa experimental data in (c). The grey region shows one standard deviation, which approximates the single interval error for the Popa data. We see that a broad range of growth rate stochasticities are consistent with the Popa's observed variation, including the coefficient of variation σR/Ravg = 0.165 reported for a mutant of PCC7120 [22]. We conclude that the variability seen by Popa et al is probably determined by the stochasticity of the growth rate. After extended isotopic dinitrogen exposure, much longer than typical cell-doubling times, we expect a nearly constant distribution of ρT -- corresponding to the nearly uniform concentration of fN in a cyanobacterial filament. Notably, the variation within the plateau, evident in the data of Popa et al [21] and modelled in Fig. 3(d), will also vanish at late times -- since it arose due to differential growth rates leading to variations in labeled vs. unlabelled ρI. Only old heterocysts, which retain some unlabelled fN, will present local dips in the concentration profile. On top of this nearly constant ρI, will be much smaller steady-state free fN gradients coming from heterocysts -- as described by Eq. 6 and illustrated in Fig. 3(a-c). Imaging fN distributions in filaments that have been grown for weeks in labeled dinitrogen may allow small gradients of ρF to be discerned with NanoSIMS techniques [21]. 3.4. Long Time Nitrogen Distributions With Periplasmic Transport [3] and the flat distribution of Popa et al We have established that the qualitative characteristics of both the spiked of Wolk et [21] can be reproduced using a single al quantitative model for the cyanobacterial filament using only cytoplasmic transport. In this section, we explore the question of how much additional periplasmic transport is possible while retaining the qualitative phenomenology. The strongest effect is seen after a Popa-like plateau is formed in ρT . In Figs. 4(a) and (b) import and export between the cytoplasm and periplasm has DI = 4.98 s−1 and DE = 0.498 s−1 (see Appendix B). The only parameter that varies is the value of the periplasmic diffusivity DP , as indicated by the legend in (b). We Reconciling cyanobacterial metabolite and transport experiments with modelling 13 Figure 4. Fixed-nitrogen distributions for systems with additional periplasmic transport, as discussed in Section 3.4. In all four plots the periplasmic transport rate DP (in units of µm s−1) is varied according to the legend in (b). (a) the total fixed-nitrogen distribution, ρT , for transport between the cytoplasm and periplasm with DI = 4.98 s−1 and DE = 0.498 s−1. The data of (b) corresponds to the curves of the same colour in (a) and shows the fraction of the total fixed-nitrogen flux along the filament that is periplasmic vs. the cell index i. (c) and (d) are similar to (a) and (b), respectively, except that import and export between the cytoplasm and periplasm are tenfold stronger, with DI = 49.8 s−1 and DE = 4.98 s−1. As before, DC =1.54 µm s−1. We use DL = 0.0498s−1 and Ghet = 3.67×106 s−1. vary DP up to the magnitude expected for unimpeded diffusive transport of glutamine along the filament, as discussed in Sec. 2.1. The characteristic dip in the centre of the total fixed-nitrogen distribution, ρT , in (a) fills in at larger values of DP -- but maintains its qualitative features throughout the range. The mid-segment dip fills in because periplasmic transport allows free fixed-nitrogen to move past the first starving vegetative cell with NC = 0. To evaluate how significant the periplasmic transport is, we show in (b) the fraction of the flux along the periplasm to the total flux along the cytoplasm and periplasm. The flux is mostly cytoplasmic next to the heterocysts, and mostly periplasmic next to starving cells. In between, up to 50% of the flux can be periplasmic. For the larger DI and DE values of Figs. 4(c) and (d) the results are Reconciling cyanobacterial metabolite and transport experiments with modelling 14 very similar. We conclude that significant, though probably not dominant, periplasmic transport is consistent with the qualitative fixed-nitrogen distribution reported by Popa et al [21]. 4. Summary We have presented a quantitative model for fixed-nitrogen (fN) transport and dynamics in a cyanobacterial filament, including growth of vegetative cells and production of fN by heterocysts. Our model reproduces the qualitative fN distribution near heterocysts (Fig. 2) seen after short exposure to isotopically labeled dinitrogen by Wolk et al. [3] ("Wolk"); but also reproduces the qualitative distribution between heterocysts (Fig. 3) seen with nanoSIMS techniques after much longer exposure by Popa et al. [21] ("Popa"). The apparent lack of a fN gradient in the Popa distribution is explained by the much larger amount of incorporated fixed-nitrogen, which gives the characteristic noisy plateau between heterocysts. Although much smaller in magnitude, the modelled free fN distribution does show a smooth gradient away from heterocysts (Fig. 3(a)) reflecting diffusive transport. Qualitatively, our results indicate the importance of the large, noisy, plateau- like concentration profile of incorporated fixed-nitrogen. The level of the plateau will increase linearly in time after the free fN gradient away from heterocysts reaches its steady-state, which takes minutes. The free fN, ρF , is soon much smaller than the incorporated fN, ρI. The spike in the Wolk distribution, seen in the first few minutes of exposure to isotopically labeled dinitrogen, was when the free fN still dominated the total ρT . In contrast, the plateau seen in the Popa distribution was after hours of exposure, when the total fN was dominated by the incorporated component and the gradient in ρF was relatively small even with respect to variations of ρI within the plateau (see Fig. 3(c)). At these intermediate times, long after the steady-state ρF but much shorter than typical division times, the heterocysts present dips of labeled fixed- nitrogen -- as seen by Popa [21]. Stochasticity, through initial cell lengths and cellular growth rates, adds noise to the incorporated fN distribution, ρI . As we show in Fig. 3 (d), the modelled cell-to-cell variation is due to differences in growth rate which manifest themselves as different amounts of labeled fN incorporated by growth. We believe this explains the observed variation in the experimental nitrogen distribution of Popa et al [21]. The possibility of periplasmic fixed-nitrogen transport raised by a contiguous periplasm [14], transport of GFP [15], and an outer membrane permeability barrier [17], led us to explore the possible impact of periplasmic transport in addition to cytoplasmic cell-to-cell connections [11]. As illustrated in Fig. 4, we found that a significant proportion (up to 50%) of fN transport along the filament can be observed without qualitatively changing our fN distributions. We cannot rule out a role for the periplasm in fN transport, and larger proportions may be achievable with more extensive parameter searches. Indeed, the fact that knocking out amino-acid permeases leads to Reconciling cyanobacterial metabolite and transport experiments with modelling 15 impaired diazotrophic growth [18, 19, 20] does indicate that significant leakage from the cytoplasm does occur (i.e. DE > 0) -- which is a necessary part of periplasmic transport. Nevertheless, our results are also consistent with no significant periplasmic transport. We have used a simple model, Eq. 4, to limit growth in response to local fixed- nitrogen starvation: cells grow at a fixed rate Ropt if there is sufficient free fixed-nitrogen available locally (NC > 0), and at the maximal rate allowed by local input fluxes of free fixed-nitrogen if not. This model reflects the absolute limitation on growth placed by available fixed-nitrogen, but also recovers the characteristic plateau of ρT seen in the Popa distribution. As we illustrate in Appendix A, "graded" growth rates, that smoothly depend on NC and vanish at NC = 0, respond to the graded distribution of free fixed-nitrogen away from heterocysts, Eq. A.2, and produce graded growth patterns and graded distributions of ρI and ρT (see Fig. A1) -- unlike those observed by Popa [21]. Graded growth models, such as the Monod model [33], are typically designed to describe response to external metabolite concentrations rather than cytoplasmic concentrations. Like our growth model, Salmonella typhimurium [34] and Escherichia coli [35, 36] both show non-graded growth in response to cytoplasmic fixed-nitrogen starvation. However, there is little direct evidence of how growth of vegetative cells in cyanobacterial filaments depends on the cytoplasmic freely available fixed-nitrogen, ρF . The experimental work was done on different species of filamentous cyanobacteria, namely Anabaena cylindrica by Wolk et al [3] and Anabaena oscillarioides by Popa et al [21]. In contrast, our models were parameterized by consideration of transport studies done in Anabaena sp. strain PCC7120 (see Sec. 2.1). While we expect parameter values to differ between species, so that precise numerical agreement with the Wolk and Popa results is not to be expected, our qualitative agreement indicates that similar fixed-nitrogen transport and dynamics could apply in these different model organisms. We expect the qualitative features of the Wolk and Popa fixed-nitrogen distributions to apply to the early and late-time fixed-nitrogen distributions, respectively, of all filamentous cyanobacteria with heterocysts that are growing in media lacking fixed- nitrogen and that have significant cytoplasmic connections. Nevertheless, quantitative details will depend, via the model parameterization, on the cyanobacterial strain and the experimental conditions. For example, sufficiently large periplasmic transport (via DP ) would broaden Popa's dip in incorporated fixed-nitrogen between heterocysts due to transport and growth past the first starving cell. Conversely, a larger cytoplasmic diffusivity (via DC) would alter the magnitude of Wolk's free fixed-nitrogen (via Eqn. 6) but not the incorporated fixed-nitrogen. The average doubling time determines the characteristic time at which we expect the later time distribution to emerge. Cell-to- cell variations in the doubling time, in turn, determine the amount of variation expected within the Popa-like plateau of incorporated fixed-nitrogen near heterocysts for the first few generations of labelled growth. Our results reconcile the metabolite patterns shown by the autoradiographic It is [3] with the nanoSIMS technique of Popa et al technique of Wolk et al [21]. Reconciling cyanobacterial metabolite and transport experiments with modelling 16 not obvious which would currently have the best resolution, and it would be useful to have either or both applied to Anabaena PCC7120 -- the most popular current model for filamentous cyanobacteria and for which recent transport studies exist [11, 15]. We can make four qualitative but testable predictions from our work. First, as the duration of labelled fN is increased the total fN distribution away from heterocysts will evolve from a Wolk-like peak, dominated by free fN, to a Popa-like plateau, dominated by incorporated fN. As indicated by Fig. 2, the crossover takes place after only a few minutes of labeled dinitrogen fixation. Second, on a similar timescale the free fN will reach a steady-state smooth parabolic shape even in a single sample. This free fN may be directly measurable, for example, by fluorescence resonance energy transfer (FRET)- based metabolite nanosensors [37]. The parabolic gradient of ρF could also be uncovered by averaging the results of many experiments measuring ρT . Third, the variations seen by Popa et al [21] will decrease in relative magnitude after at least one doubling-time of exposure to labeled fN. Fourth is more of an assumption, but we expect that growth of vegetative cells will slow abruptly when the freely available fN vanishes -- according to Eqn. 4 rather than in a graded fashion according to Eqn. A.3. Finally, with more precise parameterization of our transport parameters, sources, and sinks we can make quantitatively testable predictions. Prediction of the variance of cell-to-cell fN, σI, requires only the observed growth rate variance σR and the duration of exposure to labeled dinitrogen. Acknowledgments We thank the Natural Science and Engineering Research Council (NSERC) for support, and the Atlantic Computational Excellence Network (ACEnet) for computational resources. AIB also thanks NSERC, ACEnet, and the Killam Trusts for fellowship support. Reconciling cyanobacterial metabolite and transport experiments with modelling 17 Appendix A. Graded growth models In this Appendix we address the effects of a graded growth model, where the growth rate of a cell depends continuously on the cytoplasmic free fixed-nitrogen concentration ρF = NC/L, and vanishes when NC = 0. Examples of such graded growth models include the Monod model (normally applied to concentrations of extracellular metabolites) [33] and the growth model used by Wolk et al [3] for modelling the response of cyanobacterial filaments to short exposures to labeled dinitrogen. Here we consider the model of Wolk et al [3], though we expect similar results with any graded growth model. The graded-growth model of Wolk et al [3] has growth linearly proportional to free fixed-nitrogen concentration, ρF . The dynamics of ρF are then ∂ρF /∂t = D∇2ρF − kρF , (A.1) where the first term on the right represents diffusion along the filament and the second term is the local sink due to vegetative growth. (Note the difference with our coarse- grained growth model in Eq. 5.) We ignore the discrete cellular structure of the filament in order to obtain an analytic solution. A heterocyst at x = 0 will impose a flux J = −D∂ρF /∂x = Ghet/2 there, leading to a steady-state solution ρF = 1 2√kD e−√k/Dx. (A.2) The incorporated fixed-nitrogen is proportional to this, and so ρT will exponentially distribution of Popa et al [21] -- since the plateau shown in Fig. 3 is inconsistent with a decay away from the heterocyst with a length-scale pD/k. This is not observed in the small pD/k while the sharp dip seen between existing heterocysts is inconsistent with a large pD/k. Figure A1. The numerical total fixed-nitrogen distributions vs. cell index i using graded growth proportional to local ρF as Eq. A.3. The growth-rate proportionality constant k (in units of µm s−1) is varied, as indicated by the legend. This is a deterministic cellular model with cytoplasmic transport: DC = 1.54 µm s−1 and DI = DE = DP = 0. Reconciling cyanobacterial metabolite and transport experiments with modelling 18 With our discrete deterministic model, we have modelled the Wolk-style growth Rgraded(Nc) = kNc/Li. (A.3) The results are shown in Fig. A1 for various values of k. The discrete nature of the model modifies the observed pattern near the bounding heterocysts compared to Eq. A.2 -- but either a plateau but no mid-segment dip is recovered (for small k) or a wide mid- segment trough with no plateau is recovered (for large k). We are unable to recover the experimental distribution seen by Popa et al [21] with such a graded growth model. Appendix B. Parameter estimation of DI and DE To estimate the value of DI (from Eqns. 1 and 2) we begin with the measurement of Pernil et al [20] that 1.65 nmol (mg protein)−1 is imported into a PCC7120 filament in 10 minutes from a medium containing 1 µM glutamine. From our discussion in Section 2.2, the dry mass of a PCC7120 cell is 4.89×10−12g and the mass of protein, approximated at 55% of the dry mass of the cell [38], is 2.69×10−12g. This implies that there is 4.44×10−9 nmol of glutamine, or NG = 2.67×106 glutamine molecules, imported in 10 minutes into each cell. The term in Eqn. 1 describing import from the periplasm to the cytoplasm is ∂NC/∂t=DINP . If we assume that the periplasmic concentration is equal to the glutamine concentration in the external medium, then NP can be replaced by the product of the glutamine concentration in the periplasm, ρP , the cross sectional area of the periplasm A (for a 0.1µm thick periplasm surrounding a cytoplasm 1µm in radius A = 0.21πµm2), and the cell length L giving NG = DIρP LAτ , with τ = 10 minutes. Solving this for DI using L = Lmin = 2.25µm yields DI = 4.98 s−1. ABC transporters, known to transport amino acids [19], are asymmetric, and we assume that export is 10× weaker than import, and that DE = 0.498 s−1. It is also known the the outer cell membrane provides a permeability barrier [17], implying that the periplasmic glutamine concentration may not be as high as the external glutamine concentration, and so we also consider a DE and DI pair that is ten times larger with DI = 49.8 s−1 and DE = 4.98 s−1. Reconciling cyanobacterial metabolite and transport experiments with modelling 19 References [1] Flores E and Herrero A. Compartmentalized function through cell differentiation in filamentous cyanobacteria. Nat Rev Microbiol, 8:39 -- 50, 2010. [2] Kumar K, Mella-Herrera R A, and Golden J W. Cyanobacterial heterocysts. Cold Spring Harb Perspect Biol, 2:a000315, 2009. [3] Wolk C P, Austin S M, Bortins J, and Galonsky A. Autoradiographic localization of 13N after fixation of 13N-labeled nitrogen gas by a heterocyst-forming blue-green alga. J Cell Biol, 61:440 -- 453, 1974. [4] Meeks J C and Elhai J. Regulation of cellular differentiation in filamentous cyanobacteria in free-living and plant-associated symbiotic growth states. Microbiol Mol Biol Rev, 66:94 -- 121, 2002. [5] Thomas J, Meeks J C, Wolk C P, Shaffer P W, Austin S M, and Chien W S. Formation of glutamine from 13N ammonia, 13N dinitrogen, and 14C glutamate by heterocysts isolated from Anabaena cylindrica. J Bacteriol, 129:1545 -- 1555, 1977. [6] Wolk C P, Thomas J, and Shaffer P W. Pathway of nitrogen metabolism after fixation of 13N- labeled nitrogen gas by the cyanobacterium, Anabaena cylindrica. J Biol Chem, 251:5027 -- 5034, 1976. [7] Haselkorn R. Cell-cell communication in filamentous cyanobacteria. Molecular Microbiology, 70:783 -- 785, 2008. [8] Giddings T H and Staehelin L A. Plasma membrane architecture of Anabaena cylindrica: occurrence of microplasmodesmata and changes associated with heterocyst development and the cell cycle. Cytobiologie, 16:235 -- 249, 1978. [9] Giddings T H and Staehelin L A. Observation of microplasmodesmata in both heterocyst-forming and non-heterocyst forming filamentous cyanobacteria by freeze-fracture electron microscopy. Arch Microbiol, 129:295 -- 298, 1981. [10] Lang N J and Fay P. The heterocysts of blue-green algae. I. Ultrastructural integrity after isolation. Proc R Soc Lond B, 178:193 -- 203, 1971. [11] Mullineaux C W, Mariscal V, Nenninger A, Khanum H, Herrero A, Flores E, and Adams D G. Mechanism of intercellular molecular exchange in heterocyst-forming cyanobacteria. EMBO Journal, 27:1299 -- 1308, 2008. [12] Mariscal V, Herrero A, Nenninger A, Mullineaux C W, and Flores E. Functional dissection of the three-domain SepJ protein joining the cells in cyanobacterial trichomes. Mol Micro, 79:1077 -- 1088, 2011. [13] Flores E, Pernil R, Muro-Pastor A M, Mariscal V, Maldener I, Lechno-Yossef S, Fan Q, Wolk C P, and Herrero A. Septum-localized protein required for filament integrity and diazotrophy in the heterocyst-forming cyanobacterium Anabaena sp. strain PCC 7120. J Bacteriol, 189:3884 -- 3890, 2007. [14] Flores E, Herrero A, Wolk C P, and Maldener I. Is the periplasm continuous in filamentous multicellular cyanobacteria? Trends Microbiol, 14:439 -- 443, 2006. [15] Mariscal V, Herrero A, and Flores E. Continuous periplasm in a filamentous, heterocyst-forming cyanobacterium. Mol Microbiol, 65:1139 -- 1145, 2007. [16] Zhang L-C, Chen Y-F, Chen W-L, and Zhang C-C. Existence of periplasmic barriers preventing green fluorescent protein diffusion from cell to cell in the cyanobacterium Anabaena sp. strain PCC 7120. Mol Micro, 70:814 -- 823, 2008. [17] Nicolaisen K, Mariscal V, Bredemeier R, Pernil R, Moslavac S, L´opez-Igual R, Maldener I, Herrero A, Schlieff E, and Flores E. The outer membrane of a heterocyst-forming cyanobacterium is a permeability barrier for uptake of metabolites that are exchanged between cells. Mol Micro, 74:58 -- 70, 2009. [18] Montesinos M L, Herrero A, and Flores E. Amino acid transport systems required for diazotrophic growth in the cyanobacterium Anabaena sp. strain PCC 7120. J Bacteriol, 177:3150 -- 3157, 1995. Reconciling cyanobacterial metabolite and transport experiments with modelling 20 [19] Picossi S, Montesinos M L, Pernil R, Lichtle C, Herrero A, and Flores E. ABC-type neutral amino acid permease N-I is required for optimal diazotrophic growth and is repressed in the heterocysts of Anabaena sp. strain PCC 7120. Mol Microbiol, 57:1582 -- 1592, 2005. [20] Pernil R, Picossi S, Mariscal V, Herrero A, and Flores E. ABC-type amino acid uptake transporters Bgt and N-II of Anabaena sp strain PCC 7120 share an ATPase subunit and are expressed in vegetative cells and heterocysts. Mol Micro, 67:1067 -- 1080, 2008. [21] Popa R, Weber P K, Pett-Ridge J, Finzi J A, Fallon S J, Hutcheon I D, Nealson K H, and Capone D G. Carbon and nitrogen fixation and metabolite exchange in and between individual cells of Anabaena oscillarioides. ISME Journal, 1:354 -- 360, 2007. [22] Allard J F, Hill A L, and Rutenberg A D. Heterocyst patterns without patterning proteins in cyanobacterial filaments. Dev Biol, 312:427 -- 434, 2007. [23] Nenninger A, Mastroianni G, and Mullineaux C W. Size dependence of protein diffusion in the cytoplasm of Escherichia coli. J Bacteriol, 192:4535 -- 4540, 2010. [24] Mullineaux C W, Nenninger A, Ray N, and Robinson C. Diffusion of green fluorescent protein in three cell environments in Escherichia Coli. J Bacteriol, 188:3442 -- 3448, 2006. [25] Rees D C, Johnson E, and Lewinson O. ABC transporters: the power to change. Nat Rev Mol Cell Biol, 10:218 -- 227, 2009. [26] Paerl H W. Role of heterotrophic bacteria in promoting N2 fixation by Anabaena in aquatic habitats. Microb Ecol, 4:215 -- 231, 1978. [27] Thiel T. Protein turnover and heterocyst differentiation in the cyanobacterium Anabaena variabilis. J Phycol, 26:50 -- 54, 1990. [28] Herrero A and Flores E. Transport of basic amino acids by the dinitrogen-fixing cyanobacterium Anabaena PCC 7120. J Biol Chem, 265:3931 -- 3935, 1990. [29] Powell E O. Growth rate and generation time of bacteria, with special reference to continuous culture. J Gen Microbiol, 15:492 -- 511, 1956. [30] Dunn J H and C P Wolk. Composition of the cellular envelopes of Anabaena cylindrica. J Bacteriol, 103:153 -- 158, 1970. [31] Cobb H D and Myers J. Comparative studies of nitrogen fixation and photosynthesis in Anabaena cylindrica. Amer Jour Bot, 51:753 -- 762, 1964. [32] Flores E and Herrero A. Assimilatory nitrogen metabolism and its regulation. In Bryant D A, editor, The molecular biology of cyanobacteria, pages 487 -- 517. Kluwer Academic Publishers, 1994. [33] Kovarova-Kovar D and Egli T. Growth kinetics of suspended microbial cell: from single substrate- controlled growth to mixed-substrate kinetics. Microbiol Mol Biol Rev, 62:646 -- 666, 1998. [34] Ikeda T P, Shauger A E, and Kustu S. Salmonella typhimurium apparently perceives external nitrogen limitation as internal glutamine limitation. J Mol Biol, 259:589 -- 607, 1996. [35] Brauer M J, Yuan J, Bennett B D, Lu W, Kimball E, Botstein D, and Rabinowitz J D. Conservation of the metabolomic response to starvation across two divergent microbes. P Natl Acad Sci (USA), 103:19302 -- 19307, 2006. [36] Hart Y, Madar D, Yuan J, Bren A, Mayo A E, Rabinowitz J D, and Alon U. Robust control of nitrogen assimilation by a bifunctional enzyme in E. coli. Mol Cell, 41:117 -- 127, 2011. [37] Fehr M, Ehrhardt D W, Lalonde S, and Frommer W B. Minimally invasive dynamic imaging of ions and metabolites in living cells. Curr Opin Plant Biol, 7:345 -- 351, 2004. [38] Philips R, Kondev J, and Theriot J. Physical Biology of the Cell. Garland Science, New York, 2009.
1706.01285	3	1706	2018-04-05T12:26:17	Gaussian fluctuation of the diffusion exponent of virus capsid in a living cell nucleus	[ "physics.bio-ph", "cond-mat.stat-mech" ]	In their work [Proc. Natl. Acad. Sci. USA 112 (2015) E5725], Bosse et al. experimentally showed that virus capsid exhibits not only normal diffusion but also anomalous diffusion in nucleus of a living cell. There, it was found that the distribution of fluctuations of the diffusion exponent characterizing them takes the Gaussian form, which is, quite remarkably, the same form for two different types of the virus. This suggests high robustness of such fluctuations. Here, the statistical property of local fluctuations of the diffusion exponent of the virus capsid in the nucleus is studied. A maximum-entropy-principle approach (originally proposed for a different virus in a different cell) is applied for obtaining the fluctuation distribution of the exponent. Largeness of the number of blocks identified with local areas of interchromatin corrals is also examined based on the experimental data. It is shown that the Gaussian distribution of the local fluctuations can be derived, in accordance with the above form. In addition, it is quantified how the fluctuation distribution on a long time scale is different from the Gaussian distribution.	physics.bio-ph	physics	Gaussian fluctuation of the diffusion exponent of virus capsid in a living cell nucleus Yuichi Itto Science Division, Center for General Education, Aichi Institute of Technology, Aichi 470-0392, Japan Abstract. In their work [Proc. Natl. Acad. Sci. USA 112 (2015) E5725], Bosse et al. experimentally showed that virus capsid exhibits not only normal diffusion but also anomalous diffusion in nucleus of a living cell. There, it was found that the distribution of fluctuations of the diffusion exponent characterizing them takes the Gaussian form, which is, quite remarkably, the same form for two different types of the virus. This suggests high robustness of such fluctuations. Here, the statistical property of local fluctuations of the diffusion exponent of the virus capsid in the nucleus is studied. A maximum-entropy-principle approach (originally proposed for a different virus in a different cell) is applied for obtaining the fluctuation distribution of the exponent. Largeness of the number of blocks identified with local areas of interchromatin corrals is also examined based on the experimental data. It is shown that the Gaussian distribution of the local fluctuations can be derived, in accordance with the above form. In addition, it is quantified how the fluctuation distribution on a long time scale is different from the Gaussian distribution. Keywords: Diffusion-exponent fluctuations; Gaussian distribution; Maximum entropy principle 1 1. Introduction There is growing interest in viruses from the viewpoint of physics. Examples include the mechanical property (such as elasticity) of viruses, self-assembly of viruses as a thermodynamic process (phase transition), the use of viruses in nanotechnology, etc. (see Refs. [1-3], for example). They reveal intriguing aspects of the physical properties of viruses and related phenomena. In their recent work [4] (see also Ref. [5]), Bosse et al. have experimentally studied the diffusion property of herpesviruses in nuclei of living PtK2 cells by making use of the technique of single particle tracking. (Here, the herpesvirus is an enveloped virus particle, whereas the PtK2 cell is a kidney epithelial cell.) This virus contains a protein shell, which is referred to as capsid. In the experiment, the cells were infected with pseudorabies virus (i.e., suid herpesvirus 1) or herpes simplex virus 1, the capsid of which is labeled with a fluorescent protein. The nucleus is organized by chromatin (i.e., chromosomal substance), by which interchromatin compartments called corrals are made. Such an organization offers a "sponge-like" architecture of the nucleus, in which these compartments represent for example enclosed pores. Due to this, large macromolecules are trapped in the corrals. In fact, the fluorescent virus capsids have been found to diffuse inside the corrals in the following situation. During virus infection, chromatin structure becomes more porous, making the corral size increase, and simultaneously, viral replication compartments (i.e., sites of viral DNA replication) form in the nucleus. In order to characterize the diffusion property, the mean square displacement of the virus capsid, which is expressed here by ,2x has been evaluated by analyzing tracks of the capsids in the corrals. Then, it has been found to scale as 2 x ~2 αt (1) for elapsed time, .t The experimental result shows not only normal diffusion but also anomalous diffusion: in the former, ,1=α whereas 1≠α in the latter. A remarkable observation there is that ,α which is termed here the diffusion exponent, fluctuates for a wide range from subdiffusion, i.e., 0 < α < ,1 to superdiffusion, i.e., .1>α Since the virus capsids are uniformly distributed over the nucleus [4], this fluctuation seems to occur in different corrals on a large spatial scale. Then, the corresponding distribution of fluctuations has been presented. Surprisingly, as can be seen in Fig. 2 (C) in Ref. [4], this distribution takes the Gaussian form (see also Ref. [5]). It is here worth briefly mentioning recent advances in experimental studies of anomalous diffusion based on the technique of single particle tracking (see Refs. [6,7] for its theory and foundations, for example). It has been shown in Ref. [8] that lipid granules in living cells exhibit subdiffusion. In Ref. [9], it has been observed that a particle such as vesicle in a pathogen shows superdiffusion. In addition, the impressiveness of the technique with molecular simulation has been discussed in Ref. [10]. It should be also noted that the phenomenon of exponent fluctuations itself is equally nontrivial as well as anomalous diffusion [11,12] widely discussed in the literature. We describe the Gaussian distribution observed in the experiment as follows: αf ( ~) − αα ( 0 2 σ 2 2 )     , (2) exp     − 3 where 0 =α 85.0 and 24.0=σ are the mean value and the standard deviation of ,α respectively [4]. In particular, it is seen to behave as .0~)0(f Here is the important feature concerning the distribution in Eq. (2). That is, the Gaussian distribution is robust in the sense [4,5] that it takes the same form for two different herpesviruses (i.e., the pseudorabies virus and the herpes simplex virus 1). Thus, the experimental result presents an evidence for high robustness of the observed exponent fluctuations. Now, for a different virus in cytoplasm of a different cell, over which the diffusion exponent fluctuates depending on localized areas [13], a theoretical approach has been developed in recent works [14-16] for describing the statistical property of the local fluctuations of the diffusion exponent. There, a statistical distribution of the local fluctuations has been proposed based on experimental data. Then, it has been shown that the proposed distribution can be derived by the maximum entropy principle for such exponent fluctuations. As mentioned earlier, exponent fluctuations for the virus capsids seem to occur in different corrals over the nucleus. This indicates that the diffusion exponent, ,α in Eq. (1) fluctuates depending on the corrals: that is, the local fluctuations of .α Therefore, this naturally motivates us to examine if the above approach can offer the Gaussian distribution in Eq. (2). This issue has an obvious importance for understanding the robustness of the Gaussian fluctuation (in the above-mentioned sense) deeper. In addition, such an understanding sheds new light on structure of the nucleus as a medium for diffusion of the virus capsid. In fact, it has been shown, in recent experimental studies (e.g., [17-24]) of diffusion of proteins concerning other viruses and other 4 proteins in cells, that knowledge about local fluctuations yields nontrivial feature of cellular structure such as its heterogeneity (as a medium for diffusion), although fluctuating quantity discussed there is not the diffusion exponent but the diffusion coefficient. In the present paper, we study the statistical property of the local fluctuations of the diffusion exponent of a herpesvirus capsid in nucleus of a living PtK2 cell. We regard the region of corrals over the nucleus as a medium for diffusion of the capsids of both pseudorabies virus and herpes simplex virus 1. Then, imaginarily dividing it into many small blocks, we introduce entropy associated with such fluctuations. Based on the experimental data of subdiffusion, the case of which is an explicitly tractable one in the data, we also examine how large the number of these blocks is. We show that, in accordance with the form in Eq. (2), the Gaussian distribution of the fluctuations can be derived by the maximum-entropy-principle approach [14-16], in which the statistical behavior of the exponent at its second order is considered to play an informative role. In addition, we quantitatively show how the fluctuation distribution after long duration of time is different from the Gaussian distribution in Eq. (2). 2. Entropy associated with the fluctuations of the diffusion exponent and largeness of the number of blocks We start our discussion with considering diffusion of the virus capsid in the nucleus. As a first step, taking the presence of chromatin into account, we regard the region of the corrals over the nucleus as a medium for diffusion of the capsids of both the pseudorabies virus and the herpes simplex virus 1. Then, we imaginarily divide this medium into many small blocks, in which the virus capsid exhibits anomalous diffusion 5 as well as normal diffusion. Therefore, the diffusion exponent, ,α in Eq. (1) fluctuates depending on these local blocks. It is noted [4] that 2x in Eq. (1) is evaluated for the elapsed time smaller than that taken for estimation of the size of corral. Accordingly, a local block is identified with a certain area of the corral. (We will quantitatively discuss this point later.) Although the diffusion exponent may slowly vary over a period of time, we assume that α is approximately constant, here. We introduce the entropy associated with exponent fluctuations. Our situation is that no information is available about how the exponent locally fluctuates over the region of the corrals. So, we consider a collection formed by constructing the local blocks, which is equivalent to the medium at the statistical level of the fluctuations. As we shall see below, the medium is, in fact, expected to contain many blocks, and accordingly, there are a large number of distinct collections in terms of the local fluctuations. Suppose that the medium consists of the blocks with a set of different values of discrete exponents, iα For this medium, let N and { .} i nα be the number of total blocks and the i number of blocks with the i th exponent, ,iα respectively. It is clear that the relation, ∑ α n i i = ,N is fulfilled. Then, for average taken in the analysis of ,2x by which the exponent in a given area of the corral is determined, it is considered that tracks in other areas of the corral as well as other corrals are not included in it. This naturally suggests that the blocks are independent each other with respect to the exponent. Therefore, the α With it, we !. i NG ∏= /! n i total number of distinct collections, ,G is given by then introduce the entropy as follows: 6 S = ln G N . (3) Next, we present a discussion about largeness of the number of blocks by evaluating the volume of the region of the corrals. Observation turns out to indicate that S in Eq. (3) takes the form of the Shannon entropy. In Ref. [5], the volume of the nucleus is assumed to be given by about 570 3 .mµ Accordingly, we consider that the volume of the region of the corrals is smaller than this volume, since chromatin is present in the nucleus as mentioned earlier. We further stress the following points. As mentioned in the Introduction, during virus infection, it has been observed that the corral size increases since chromatin structure becomes more porous. Simultaneously, it has also been found that viral replication compartments form in the nucleus. These compartments may occupy the volume of the nucleus less compared to the corrals for a period of time, over which the Gaussian fluctuation in Eq. (2) is realized [4]. Therefore, in the present paper, we suppose that the volume of the region of the corrals is not so small but still large enough. Now, for a given local block in the medium, we treat it as a cubic block, the side of which has the length of the value of 2x at large elapsed time. This treatment (originally performed for a localized area of cytoplasm in Ref. [25]) enables us to estimate the volume of the cubic block. As the value of α in Eq. (1), we take, as explicitly tractable cases, the following two values presented in Fig. 2 (B) in Ref. [4]: .0=α 961 for the pseudorabies virus and .0=α 918 for the herpes simplex virus 1. Regarding these exponents, it should be noted [4] that each exponent is obtained by average of the mean square displacement for all tracks over the nucleus. Then, the 7 diffusion coefficient, which is necessary for estimating 2x in our discussion, presented there is also obtained in this way. Such a diffusion coefficient as well as the diffusion exponent may be different from those to be used in 2x in Eq. (1), but they are sufficient in the present discussion since the difference is expected to be small [4]. Therefore, for the elapsed time s36.0 [4], the volumes of the cubic block are estimated as follows: .0 022 3µm for the pseudorabies virus and .0 013 3µm for the herpes simplex virus 1. Recalling that the local block is identified with a certain area of the corral, these volumes are, in fact, smaller than that of the corral [5], as expected. In each case, if the total number of blocks with the above-estimated volume in the medium is, for example, 100, then it follows that n = 100 in the former, while n = 100 .0 918 .0 961 in the latter, for each of which >N ,100 provided that no additional blocks with the corresponding exponent are supposed to be present. So, in this example, the total volumes of the blocks with each of the above volumes are calculated to be 3mµ2.2 and ,mµ3.1 3 respectively. Thus, the volume of the region of the corrals is seen to be large enough compared to these volumes. From this observation, it is therefore implied that .0n 961 and .0n 918 are actually larger than the above values and the numbers of blocks with other exponents are also large, since the volume of the medium should be equal to that of the region of the corrals. 3. Derivation of the Gaussian fluctuation by the maximum entropy principle From the above discussion about the blocks, it seems that the entropy in Eq. (3) can be evaluated by the use of the Stirling approximation [i.e., ln ( M )! ≅ MMM ln − for large M ]. This leads to the following form of the Shannon entropy: 8 S ∑−≅ i f α i ln f α i , where f α = i /α n i N is the probability of finding iα in a given local block of the medium. Therefore, we write its continuum limit as fS [ ] ∫−= αα d ( f ln) f α ( ), (4) where we are using the same symbol (αf ) for the probability density as that in Eq. (2), but it may not lead to confusion. Let us show that the Gaussian distribution can be derived by the maximum entropy principle with the Shannon entropy in Eq. (4). Recall that there is no information about how the exponent locally fluctuates over the region of the corrals. It is natural in such a situation to impose the constraint on the average of α in the principle: ∫ ααα d ) ( f = . m (5) Then, as stressed earlier, viral replication compartments form during virus infection. The fluctuation distribution in the vicinity of the above average may slightly change due to these compartments, but we consider the fluctuation distribution that is stable under such a formation over a period of time. Accordingly, additional constraint seems to be required in order to take into account this stability, since a class of the fluctuation distributions to be realized is limited by it with keeping the above average. To do so as simple as possible, we impose the constraint on the statistical behavior of α at its second order, i.e., the size of the fluctuations of .2α Therefore, if additional 9 information is available on such a statistical behavior, then the average of ,2α i.e., the second moment, should be further constrained: ∫ ααα d ) ( f 2 2 = s . (6) It is noticed from Eqs. (5) and (6) that the variance of α is automatically fixed, reflecting the above-mentioned stability. Thus, together with another constraint on the normalization of (αf ), the corresponding maximum entropy principle reads where { δ f fS [ ] − ( ∫ αακ ) d ( f − 1 ) + ( ∫ αααλ d ) ( f − 2 αααµ d ) ( f ( ∫ )m − − 2 s ) } = ,0 (7) ,{ µλκ } , is the set of the Lagrange multipliers associated with the constraints, and fδ stands for the variation with respect to (αf ). Note that we have imposed the following two conditions in Eq. (7): lim α ∞→ f α ) ( = ,0 f )0( ∗ < αf ( ), (8) where ∗α is a certain fixed positive value of .α The limit, ,∞→α should be interpreted as the large-α limit, i.e., the limit α going to a large but finite value. The first condition in Eq. (8) is nothing but the requirement of physically plausible behavior 10 of (αf ) due to the fact that the region of the corrals is finite. Then, the second one comes from the following natural considerations. The virus capsid tends to reach the membrane of the nucleus for egress (i.e., a budding process), implying the existence of many blocks with a typical large value of .α This tendency might seem to indicate that such a value could be very large. This is however seen to lead to contradiction regarding the above first condition, and therefore we take the value as .∗α Due to the time dependence in Eq. (1), the tendency then may become strong in the case when the number of blocks with the exponent ∗= αα is larger than that of blocks with the exponent ,0=α which yields the second condition. It turns out that the conditions in Eq. (8) require λ and µ to be positive Lagrange multipliers. Consequently, the stationary solution of Eq. (7) is given by ( f α ) ∝ exp ( 2αµαλ − ). (9) Clearly, this distribution can be written in the following form, ( f α ) ∝ exp { − 2µλαµ ])2(/ − [ }, and accordingly, becomes identical to the distribution in Eq. (2) after the following identifications: λ = α 0 σ ,2 µ = 1 2σ 2 . (10) Thus, we see that the maximum-entropy-principle approach can give rise to the Gaussian distribution of the local fluctuations, which is precisely (αf ) in Eq. (2). It is anticipated from the present results that the average of 2α in Eq. (6) can be 11 informative in the approach. 4. Comment on the difference between the fluctuation distributions on a long time scale So far, we have discussed the situation that the virus capsids are uniformly distributed over the nucleus, in which the Gaussian fluctuation in Eq. (2) is realized. In contrast to this situation, after long duration of time, it has been observed in the experiment [4] that the virus capsids are marginalized around the membrane of the nucleus in the case of the pseudorabies virus, which is nontrivial [5]. Accordingly, the fluctuation distribution to be observed on such a long time scale seems to be different from Eq. (2). So, it is of interest to quantify the difference between the Gaussian distribution in Eq. (2) and such a fluctuation distribution, since it may be possible to measure how the marginalization makes the Gaussian distribution approach the fluctuation distribution to be observed. Therefore, we wish to make a comment on this issue. Suppose that Gaussian form similar to Eq. (2) still holds for the fluctuation distribution to be observed, ~ αf ), ( but the associated mean value and standard deviation can differ from those in Eq. (2) and are denoted by ~α and 0 ,~σ respectively. (This is supported by the raw data of the experiment [4], although the data is not shown here.) Then, we quantify the difference between (αf ) and ~ αf ) ( by the distance between them. Such a distance may be supplied by the symmetric Kullback-Leibler divergence: ~ fD [ , f ] = ~ fK [ f ] + fK [ ~ f ]. (11) 12 Here, ~ fK [ f ] describes the Kullback-Leibler relative entropy [26] and is given in the present case by ~ fK [ f ] ∫= ~ f ln) αα d ( ~ f f α ( ) α ) ( , (12) which is positive semidefinite and vanishes if and only if ~ f = f . Thus, we obtain ~ fK [ f ] = 1 σ 2 2 { fK [ ~ f ] = 1 ~2 σ 2 { ~( αα 0 − 0 ~( αα 0 − 0 2 ) ~ σ 2 + } − ln ~ σ σ − 1 2 , (13) 2 ) 2 + σ } + ln ~ σ σ − 1 2 , (14) ~ fD [ , f ] = 1 2    1 1 ~ 2 2 σσ +    ~( αα 0 − 0 2 ) + ~ σ σ    1 2 − 2 σ ~ σ    . (15) leading to As an example of interest, we consider the case when ~ σσ = . In this case, we have ~ fD [ , =f ] 1 2 σ ~( αα 0 − 0 2 ,) (16) showing that ~ αf ) ( deviates from (αf ) in a monotonic way with respect to .~ 0α This fact also illustrates that if the fluctuation distribution in the course of the marginalization remains the Gaussian form, then the distribution approaches ~ αf ) ( in 13 this way. Thus, we quantitatively see how the fluctuation distribution to be observed on a long time scale differs from the Gaussian distribution in Eq. (2). 5. Concluding remarks We have studied the statistical property of the local fluctuations of the diffusion exponent of a herpesvirus capsid over the region of corrals in nucleus of a living PtK2 cell. Regarding the region as a medium for diffusion of the capsids of both pseudorabies virus and herpes simplex virus 1, and imaginarily dividing it into many small blocks, which are identified with local areas of the corrals, we have introduced the entropy associated with the fluctuations. Treating the local block as the cubic one, we have also examined largeness of the number of blocks. Based on the maximum entropy principle, in which the second moment of the exponent plays an informative role, we have derived the Gaussian form of the fluctuation distribution in consistent with the experimental observations. This highlights a possibility of applying the present approach to the diffusion-exponent fluctuations observed in the experiment. In addition, we have quantitatively discussed how the fluctuation distribution to be observed on a long time scale is different from the Gaussian distribution in Eq. (2). In the present study, we have focused our attention on the fluctuations of the diffusion exponent. It may be natural to suppose that the diffusion coefficient of the virus capsid also fluctuates in the nucleus. So, it could be expected by the entropic approach that, if the diffusion coefficient satisfies similar constraints and conditions discussed for the diffusion exponent, then its statistical distribution is of the Gaussian form. In addition, in a recent work [27], a unimodal distribution of the diffusion-exponent fluctuations of a viral protein in a cell nucleus has experimentally been observed. 14 Therefore, it is of interest to examine application of the present approach for such a distribution. Acknowledgements The author would like to thank J. B. Bosse for providing him with the raw data of the experiment in Ref. [4] and for valuable discussions. He also would like to acknowledge the warm hospitality of Heinrich Pette Institute, Leibniz Institute for Experimental Virology. Note added. In a recent paper [28], an issue regarding the continuum limit of the entropy [14-16] has been carefully examined. The discussion given there is seen to be applicable to the entropy in Eq. (4). References [1] P.G. Stockley, R. Twarock (Eds.), Emerging Topics in Physical Virology, Imperial College Press, London, 2010. [2] W.H. Roos, R. Bruinsma, G.J.L. Wuite, Nat. Phys. 6 (2010) 733. [3] M.G. Mateu (Ed.), Structure and Physics of Viruses, Springer, Dordrecht, 2013. 15 [4] J.B. Bosse, I.B. Hogue, M. Feric, S.Y. Thiberge, B. Sodeik, C.P. Brangwynne, L.W. Enquist, Proc. Natl. Acad. Sci. USA 112 (2015) E5725. [5] J.B. Bosse, L.W. Enquist, Nucleus 7 (2016) 13. [6] F. Höfling, T. Franosch, Rep. Prog. Phys. 76 (2013) 046602. [7] C. Manzo, M.F. Garcia-Parajo, Rep. Prog. Phys. 78 (2015) 124601. [8] J.-H. Jeon, V. Tejedor, S. Burov, E. Barkai, C. Selhuber-Unkel, K. Berg-Sørensen, L. Oddershede, R. Metzler, Phys. Rev. Lett. 106 (2011) 048103. [9] J.F. Reverey, J.-H. Jeon, H. Bao, M. Leippe, R. Metzler, C. Selhuber-Unkel, Sci. Rep. 5 (2015) 11690. [10] J.-H. Jeon, M. Javanainen, H. Martinez-Seara, R. Metzler, I. Vattulainen, Phys. Rev. X 6 (2016) 021006. [11] J.-P. Bouchaud, A. Georges, Phys. Rep. 195 (1990) 127. [12] R. Metzler, J.-H. Jeon, A.G. Cherstvy, E. Barkai, Phys. Chem. Chem. Phys. 16 (2014) 24128. 16 [13] G. Seisenberger, M.U. Ried, T. Endress, H. Büning, M. Hallek, C. Bräuchle, Science 294 (2001) 1929. [14] Y. Itto, J. Biol. Phys. 38 (2012) 673. [15] Y. Itto, Phys. Lett. A 378 (2014) 3037. [16] Y. Itto, in: Atta-ur-Rahman, M. Iqbal Choudhary (Eds.), Frontiers in Anti-Infective Drug Discovery, Vol. 5, Bentham Science Publishers, Sharjah, 2017. [17] S. Manley, J.M. Gillette, G.H. Patterson, H. Shroff, H.F. Hess, E. Betzig, J. Lippincott-Schwartz, Nat. Methods 5 (2008) 155. [18] N. Dross, C. Spriet, M. Zwerger, G. Müller, W. Waldeck, J. Langowski, PLoS ONE 4 (2009) e5041. [19] T. Kühn, T.O. Ihalainen, J. Hyväluoma, N. Dross, S.F. Willman, J. Langowski, M. Vihinen-Ranta, J. Timonen, PLoS ONE 6 (2011) e22962. [20] S. Bakshi, B.P. Bratton, J.C. Weisshaar, Biophys. J. 101 (2011) 2535. [21] E. Adu-Gyamfi, M.A. Digman, E. Gratton, R.V. Stahelin, Biophys. J. 102 (2012) 2517. 17 [22] N. Hoze, D. Nair, E. Hosy, C. Sieben, S. Manley, A. Herrmann, J.-B. Sibarita, D. Choquet, D. Holcman, Proc. Natl. Acad. Sci. USA 109 (2012) 17052. [23] B.-C. Chen et al., Science 346 (2014) 1257998. [24] H. Anton, N. Taha, E. Boutant, L. Richert, H. Khatter, B. Klaholz, P. Rondé, E. Réal, H. de Rocquigny, Y. Mély, PLoS ONE 10 (2015) e0116921. [25] Y. Itto, Physica A 462 (2016) 522. [26] S. Kullback, Information Theory and Statistics, Dover, New York, 1997. [27] B.L. Rice, R.J. Kaddis, M.S. Stake, T.L. Lochmann, L.J. Parent, Front. Microbiol. 6 (2015) 925. [28] Y. Itto, Open Conf. Proc. J. 9 (2018) 1. 18
1605.03626	1	1605	2016-05-11T21:47:29	Predictability and hierarchy in Drosophila behavior	[ "physics.bio-ph", "cs.IT", "cs.IT", "q-bio.NC", "stat.AP" ]	Even the simplest of animals exhibit behavioral sequences with complex temporal dynamics. Prominent amongst the proposed organizing principles for these dynamics has been the idea of a hierarchy, wherein the movements an animal makes can be understood as a set of nested sub-clusters. Although this type of organization holds potential advantages in terms of motion control and neural circuitry, measurements demonstrating this for an animal's entire behavioral repertoire have been limited in scope and temporal complexity. Here, we use a recently developed unsupervised technique to discover and track the occurrence of all stereotyped behaviors performed by fruit flies moving in a shallow arena. Calculating the optimally predictive representation of the fly's future behaviors, we show that fly behavior exhibits multiple time scales and is organized into a hierarchical structure that is indicative of its underlying behavioral programs and its changing internal states.	physics.bio-ph	physics	Predictability and hierarchy in Drosophila behavior Gordon J. Berman,∗ William Bialek, and Joshua W. Shaevitz Joseph Henry Laboratories of Physics and Lewis -- Sigler Institute for Integrative Genomics, Princeton University, Princeton, NJ 08544 (Dated: September 17, 2018) Even the simplest of animals exhibit behavioral sequences with complex temporal dynamics. Prominent amongst the proposed organizing principles for these dynamics has been the idea of a hierarchy, wherein the movements an animal makes can be understood as a set of nested sub-clusters. Although this type of organization holds potential advantages in terms of motion control and neural circuitry, measurements demonstrating this for an animal's entire behavioral repertoire have been limited in scope and temporal complexity. Here, we use a recently developed unsupervised technique to discover and track the occurrence of all stereotyped behaviors performed by fruit flies moving in a shallow arena. Calculating the optimally predictive representation of the fly's future behaviors, we show that fly behavior exhibits multiple time scales and is organized into a hierarchical structure that is indicative of its underlying behavioral programs and its changing internal states. I. INTRODUCTION Animals perform a vast array of behaviors as they go about their daily lives, often in what appear to be re- peated and non-random patterns. These sequences of actions, some innate and some learned, have dramatic consequences with respect to survival and reproductive function -- from feeding, grooming, and locomotion to mat- ing, child rearing, and the establishment of social struc- tures. Moreover, these patterns of movement can be viewed as the final output of the complicated interactions between an organism's genes, metabolism, and neural sig- naling. As a result, understanding the principles behind how an animal generates behavioral sequences can pro- vide a window into the biological mechanisms underlying the animal's movements, appetites, and interactions with its environment, as well as broader insights into how be- haviors evolve. The prevailing theory for the temporal organization of behavior, rooted in work from neuroscience, psychology, and evolution, is that the pattern of actions performed by animals is hierarchical [1 -- 3]. In such a framework, ac- tions are nested into modules on many scales, from simple motion primitives to complex behaviors to sequences of actions. Neural architectures related to behavior, such as the motor cortex, are anatomically hierarchical, sup- porting the idea that animals use a hierarchical repre- sentation of behavior in the brain [4 -- 7]. Additionally, hierarchical organization is a hallmark of human design, from the layout of cities to the wiring of the internet, and its potential use in various biological contexts has been proposed as an organizing principle [2]. Despite the theoretical attractiveness of behavioral hi- erarchy, measurements showing that a particular ani- mal's behavioral repertoire is organized in this manner often are limited in their applicability and scope. Typi- ∗ E-mail: [email protected]. Current address: Depart- ment of Biology, Emory University, Atlanta, GA 30322 cally, observations of hierarchy in the ordering of move- ment have considered a single behavioral type, such as grooming, ignoring relationships between more varied be- havioral motifs [8 -- 13]. Perhaps more problematic is that most analyses of behavior make use of methods, such as hierarchical clustering, that implicitly or explicitly im- pose a hierarchical structure onto the data without show- ing that such a representation is accurate. Lastly, to our knowledge, all measurements of a hierarchical organiza- tion of behavior limit their analysis to behavioral dynam- ics at a single time scale. This scale is often given by the results of fitting a Markov model, where the next step in a behavioral pattern only depends on the animal's current state. Even in the simplest of animals, however, there are many internal states such as hunger, reproductive drive, etc., and sequences of behaviors possess an effec- tive memory of an animal's behavioral state that persists well into the future, a result noted in a wide variety of systems [14 -- 17]. In this paper, we study the behavioral repertoire of fruit flies (Drosophila melanogaster ), attempting to char- acterize the temporal organization of their movements over the course of an hour. Decomposing the flies' move- ments into a set of stereotyped behaviors without mak- ing any a priori behavioral definitions [18], we find that their behavior exhibits long time scale,s far beyond what would be predicted from a Markovian model. Applying methods from information theory, we show that a hier- archical representation of actions optimally predicts the future behavioral state of the fly. These results show that the best way to understand how future actions fol- low from the current behavioral state is to group these current behaviors in a nested manner, with fine grained partitions being useful in predicting the near future, and coarser partitions being sufficient for predicting the rela- tively distant future. These results show that these ani- mals control their movement via a hierarchy of behaviors at varying time scales, affirming and making precise a key concept in ethology. II. EXPERIMENTS AND BEHAVIORAL STATES As a testbed for probing questions of behavioral or- ganization and hierarchy, we sought to measure the en- tire behavioral repertoire of a population of Drosophila melanogaster in a specific environmental context. We probed the behavioral repertoire of individual, ground- based fruit flies in a largely featureless circular arena for one hour using a 100Hz camera. Under these conditions, flies display many complex behaviors, including locomo- tion and grooming, that involve multiple parts of their bodies interacting at varying time scales. We recorded videos of 59 male flies using a custom-built tracking setup, producing more than 21 million images [18]. These data were used to generate a two -- dimensional map of fly behavior based on an unsupervised approach that automatically identifies stereotyped actions (Fig. 1A, for full details see [18]). Briefly, this approach takes a set of translationally and rotationally aligned images of the flies and decomposes the dynamics of the observed pixel values into a low -- dimensional basis set describing the flies' posture. Time series are produced by project- ing the original pixel values onto this basis set and the local spectrogram of these trajectories is then embedded into two dimensions [19]. Each position in the behavioral map corresponds to a unique set of postural dynamics; although this was not required by the analysis, nearby points represent similar motions, i.e. those involving re- lated body parts executing similar temporal patterns. In the resulting behavioral space, z, we estimate the probability distribution function P (z) and find that it contains a set of peaks corresponding to short segments of movement that are revisited multiple times by mul- tiple individuals (Figure 1A). Pauses in the trajecto- ries through this space, z(t), are interspersed with quick movements between the peaks. These pauses in z(t) at a particular peak correspond to the fly performing one of a large set of distinct, stereotyped behaviors such as right wing grooming, proboscis extension, or alter- nating tripod locomotion [18]. In all, we identify 117 unique stereotyped actions, with similar behaviors, i.e. those that utilize similar body parts at similar frequen- cies, located near each other in the behavioral map. A watershed algorithm is used to separate the peaks and, combined with a threshold on dz(t)/dt, to segment each movie into a sequence of discreet, stereotyped behaviors. In this paper, we treat pauses at these peaks to be our states, the lowest level of description of behavioral organization, and investigate the pattern of behavioral transitions among these states over time. We count time in units of the transitions between states, so we have a description of behavior as a discrete variable S(n) that can take on N = 117 different values at each discrete time n. Note that since we count time in units of transitions, we always have S(n + 1) (cid:54)= S(n). Combining data from all 59 flies, we observe ≈ 6.4× 105 behavioral transitions, or ≈ 104 per experiment. 2 FIG. 1. Transition probabilities and behavioral modular- ity. (A) Behavioral space probability density function (PDF). Here, each peak in the distribution corresponds to a dis- tinct stereotyped movement. (B) One-step Markov transition probability matrix T(τ = 1). The 117 behavioral states are grouped by applying the predictive information bottleneck calculation and allowing 6 clusters (Eq. 4). Black lines de- note the cluster boundaries. (C) Transitions rates plotted on the behavioral map. Each red point represents the maximum of the local PDF, and the black lines represent the transi- tion probabilities between the regions. Line thicknesses are proportional to the corresponding value of T(τ = 1)ij, and right -- handed curvature marks the direction of the transition. For clarity, all lines representing transition probabilities of less than .05 are omitted. (D) The clusters found using the information bottleneck approach (colored regions) are con- tiguous in the behavioral space. Behavioral labels associated with each partitioned graph cluster from B are shown. Black lines thickness represents the conditional transition probabil- ities between clusters. All transition probabilities less than .05 are omitted. III. TRANSITION MATRICES AND NON-MARKOVIAN TIME SCALES To investigate the temporal pattern of behaviors, we first calculated the behavioral transition matrix over dif- ferent time scales, [T(τ )]i,j ≡ p(S(n + τ ) = iS(n) = j), (1) which describes the probability that the animal will go from state j to state i after τ transition steps. We expect that this distribution becomes less and less structured as τ increases because we lose the ability to make predic- tions of the future state as the horizon of our predictions extends further. In addition, it will be useful to think about these matrices in terms of their eigendecomposi- 0PDF2 x 10-4CD= 1= .5LocomotionSide Leg2020A0PDF2 x 10-4B0.15Initial StateFinal StateIdleSlowSide LegAnteriorPosteriorLocomotionIdleSlowPosteriorAnteriorT(1)i,j3 FIG. 2. Long time scale transition matrices and non -- Markovian dynamics. (A) Markov model transition matrix for τ = 100, TM (100), from Eq (3). (B and C) Transition matrices for τ = 100 and τ = 1, 000, respectively, from Eq (1). (D) Absolute values of the leading eigenvalues of the transition matrices T(τ ) as a function of τ . The curves represent the average over all flies, and thicknesses represent the standard error of the mean. Dashed lines are the predictions for the Markov model TM (τ ). The black line is a noise floor, corresponding to the typical value of the second largest eigenvalue in a transition matrix calculated from random temporal shuffling of our finite data set. (E) Eigenmode decay rates, rµ(τ ) ≡ − log λµ(τ )/τ , as a function of the number of transitions. Line colors represent the same modes as in (D) and the black line again corresponds to a "noise floor," in this case the largest decay rate that we can resolve above the random structures present in our finite sample. tions: (cid:88) µ [T(τ )]i,j = λµ(τ )uµ i (τ )vµ j (τ ), of motion including locomotion, behaviors involving an- terior parts of the body etc. (Fig. 1D). (2) i } and vµ ≡ {vµ where uµ ≡ {uµ i } are the left and right eigenvectors, respectively, and λµ(τ ) is the eigenvalue with the µth largest modulus. Because probability is conserved in the transitions, the largest eigenvalue will always be equal to one, λ1(τ ) = 1, and v1 i (τ ) describes the stationary distribution over states at long times. All the other eigenvalues have magnitudes less than one, λµ(cid:54)=1(τ ) < 1, and describe the loss of predictability over time, as shown in more detail below. The matrix T(τ = 1) describes the probability of tran- sitions from one state to the next, the most elementary steps of behavior (Fig. 1B). To the eye, this transition matrix appears modular, with most transitions out of any given state only going to one of a handful of other states. By appropriately organizing the states in Figure 1B, T(τ = 1) takes on a nearly block-diagonal structure, which can be broken up into modular clusters using the information bottleneck formalism (see below). Plotting this matrix on the behavioral map itself (Fig. 1C), we see that the transitions are largely localized, with nearly all large probability transitions occurring between nearby behaviors. Furthermore, the transition clusters are con- tiguous in the behavioral space, defining gross categories It is important to note that T(τ = 1) does not di- rectly contain information about the location of behav- ioral states in the two dimensional map, and hence any relationship we observe between the transition structure and the patterning of behaviors in the map is a con- sequence of the animal's behavior and not the way we construct the analysis. We thus conclude that behavioral transitions are mostly restricted to occur between similar actions -- e.g., grooming behaviors are typically followed by other grooming behaviors of close-by body parts and animals transition between locomotion gates systemati- cally by changing gate speed and velocity. These observa- tions are consistent with classical ideas of postural facili- tation and previous observations that transitions largely occur between similar behaviors [9, 20 -- 22]. We begin to see the necessity of looking at longer time scales as we measure the transition matrices for τ (cid:29) 1. If the observed dynamics are purely Markovian, then the transitions from one state to the next do not depend on the history of behavior, and T(τ = 1) provides a complete characterization of the system. In particular, if the behavior is Markovian then we can calculate the transition matrix after τ state just by iterating the matrix AB100103101102Number of Transitions01Dλ μ = 2 μ = 3 μ = 4 μ = 5 μ = 6Random100103101102Number of Transitions10-310110-1EDecay Rate (transitions-1)TM(100)i,jT(100)i,j0.05CT(1000)i,j0.050.05Initial StateFinal StateInitial StateFinal StateInitial StateFinal Statefrom one step: TM (τ ) ≡ [T(1)]τ = (cid:88) µ [λµ(1)]τ uµ(1)vµ(1). (3) Because λµ(1) < 1 for all but the leading eigenvalue, the contributions from the µ > 1 terms decay to zero exponentially as τ → ∞. For very long times, there- fore, TM (τ ) loses all information about the current state and instead reflects the average probabilities of perform- ing any particular behavior. Thus, in a Markovian system, the slowest time scale in the system is deter- mined by λ2(1), resulting in a characteristic decay time t2 = −1/ log λ2(1). Calculating these eigenvalues for each fly and averaging, we find (cid:104)λ2(1)(cid:105) = 0.953 ± 0.004, or (cid:104)t2(cid:105) = 29 ± 2 transitions. Thus, any memory that extends beyond ≈ 30 transitions into the future is di- rect evidence for hidden states that carry a memory over longer times and modulate behavior. Initial evidence for long-time structure in T(τ ) comes by comparing the lack of structure within TM (100) to that within T(τ ) for τ = 100 and τ = 1, 000 (Fig 2A- C). After 100 transitions, (≈ 3(cid:104)t2(cid:105)), the Markov model retains essentially no information, as demonstrated by the similarity between all of the rows, implying that all transitions have been randomized. Conversely, although some of the block -- diagonal structure from Fig. 1B has dissipated, we see that T(100) and T(1000) retrain a great deal of non-randomness. This observation can be made more precise by looking at the eigenvalue spectra of the transition matrices. In Figure 2D, we plot λµ(τ ) as a function of τ for µ = 2 through 6 (solid color lines) in addition to the predictions from the Markov model of Eq (3) based on T(1) (colored dashed lines). In a Markovian system, it would be more natural to plot these results with a logarithmic axis for λ, but here we see that structure extends over such a wide range of time scales that we need a logarithmic axis for τ . We can make this difference more obvious by mea- suring the apparent decay rate, rµ(τ ) = − log λµ(τ )/τ , which should be constant for a Markovian system. For the leading mode, the apparent decay rate falls by nearly two orders of magnitude before the corresponding eigen- value is lost in the noise (Figure 2E). Similar patterns appear in higher modes, but we have more limited dy- namic range for observing them. These results are direct evidence that many time scales are required to model behavioral sequences, even in this simple context where no external stimuli are provided. Accordingly, we can infer that the organism must have in- ternal states that we do not directly observe, even though we are making rather thorough measurements of the mo- tor output. Roughly speaking, the appearance of decay rates ≈ 10−3 means that the internal states must hold memory across at least ≈ 103 behavioral transitions, or approximately 20 minutes -- much longer than any time scale apparent in the Markov model. 4 IV. PREDICTABILITY AND HIERARCHY The modular structure of the flies' transition matrix, combined with the observed long time scales of behav- ioral sequences, suggests that we might be able to group the behavioral states into clusters that preserve much of the information that the current behavioral state pro- vides about future actions (predictive information [23]). Furthermore, we should be able to probe whether this results in a hierarchical organization: if the states are grouped into a hierarchy, then increasing the number of clusters will largely subdivide existing clusters rather than mix behaviors from two different clusters. To make this idea more precise, we hope to map the behaviors into groups, S(n) → Z, that compress our description in a way that preserves information about a state τ transitions in the future, S(n + τ ). Mathe- matically, this means that we should maximize the in- formation about the future, I(Z; S(n + τ )), while hold- ing fixed the information that we keep about the past, I(Z; S(n)). Introducing a Lagrange multiplier to hold I(Z; S(n)) fixed, we wish to maximize F = I(Z; S(n + τ )) − βI(Z; S(n)). (4) At β = 0 we retain the full complexity of the 117 be- havioral states, and as we increase β, we are forced to tighten our description into a more and more compressed form, thus losing predictive power. This is an example of the information bottleneck problem [24]. If the com- pressed description Z involves a fixed number of clusters, then we find solutions that range from soft clustering, where behaviors can be assigned to more than one cluster probabilistically, to hard clustering, where each behav- ior belongs to only one cluster, as β increases; changing the number of clusters allows us to move along a curve that trades complexity of description against predictive power, as shown in Fig 3 (see §VI C for details). FIG. 3. Optimal trade-off curves for lags from τ = 1 to τ = 5000. For each time lag τ , number of clusters, and β, we optimize Equation 4 and plot the resulting complexity of the partitioning, I(Z; S(n)), versus the predictive informa- tion, I(Z; S(n + τ )). I(X;Z) (bits)I(Y;Z) (bits)0321210log10 n0123τI(Z;S(n))(bits)I(Z;S(n+τ))(bits)5 FIG. 4. Information bottleneck partitioning of behavioral space for τ = 67 (approximately twice the longest time scale in the Markov model). Borders from the previous partitions are shown in black. For 25 clusters (bottom right), the partitions, still contiguous, are denoted by dashed lines. As expected, the optimal curves move downward as the time lag increases, implying that the ability to predict the behavioral state of the animal decreases as we look further into the future. We also observe a relatively rapid decrease in the height of these curves for small τ , followed by increasingly-closely spaced optimal curves as the lag length increases. It this slowing that is indicative of the long time-scales in behavior. Along each of these trade-off curves lie partitions of the behavioral space that contain an increasing number of clusters. We can make several observations about these data. First, in agreement with our investigation of the single-step transition matrix, we find that the clusters are spatially contiguous in the behavioral map as exem- plified in Figure 4 for τ = 67. Thus, even when we add in the long time-scale dynamics, we find that transitions predominantly occur between similar behaviors. Second, these spatially-contiguous clusters separate hierarchically as we increase the number of clusters, i.e. new clusters largely result from subdividing existing clusters instead of emerging from multiple existing clusters. One example of this can be seen in Figure 5, where the probability flow between partitions of increasing size subdivide in a tree- like manner. It is important to note that these results are not built in to the information bottleneck algorithm: we can solve the bottleneck problem for different num- bers of clusters independently, and hence (in contrast to hierarchical clustering) this method could have found non-hierarchical evolution with new clusters comprised of behaviors from many other clusters, That this does not happen is strong evidence that fly behavior is organized hierarchically. We can go beyond this qualitative description, how- ever, by quantifying the degree of hierarchy in our rep- resentation as the number of clusters increases using a "treeness" metric, T (Fig. 6). The idea behind this met- ric, which is similar to the one introduced by Corominas -- Murta et al [25], is that if our representation is perfectly hierarchical, then each cluster has precisely one "parent" in a partitioning with a smaller number of clusters. Thus, the better our ability to distinguish the lineage of a clus- ter as it splits through increasingly complex partitionings implies a higher value of T . More precisely, the treeness index is given by the relative reduction in entropy going FIG. 5. Hierarchical organization for optimal solutions with lag τ = 100 ranging from 1 cluster to 25. The displayed clusterings are those that have the largest value of I(Z; S(n + τ )) for that number of clusters. The length of the vertical bars are proportional to the percentage of time a fly spends in each of the clusters, and the lines flowing horizontally from left to right are proportional in thickness to the flux from the clustering on the left to the clustering on the right. Fluxes less than .01 are suppressed for clarity. # of Clusters = 1# of Clusters = 2# of Clusters = 3# of Clusters = 4# of Clusters = 5# of Clusters = 6# of Clusters = 7# of Clusters = 25IdleSlowLocomotionAnteriorPosteriorSide Legs6 the hierarchy was modeled, and/or relied on hierarchi- cal clustering and other types of analyses that only yield hierarchical outputs. The type of organization we observe is reminiscent of the functional clustering seen in mouse and primate mo- tor cortex, where groupings of neurons from millimeter scales down to single cells have been found to exhibit increasing temporal correlation as the distance between them decreases [4, 6]. Although no such pattern has been specifically found in Drosophila, our results suggest that such neuronal patterns may exist. As circuits for differ- ent behavioral modules are uncovered, our results suggest that such hierarchical neuroanatomical organization will also be found in the fly, serving as a general principle that may apply across organisms to provide insight to- wards how the brain controls behavior and adapts to a complex environment. ACKNOWLEDGMENTS We thank Ugne Klibaite, David Schwab, and Thibaud Taillefumier for discussions and suggestions. JWS and GJB also acknowledge the Aspen Center for Physics, where many ideas for this work were formulated. This work was funded through awards from the National In- stitutes of Health (GM098090, GM071508), The National Science Foundation (PHY-1305525, PHY-1451171, CCF- 0939370), the Swartz Foundation, and the Simons Foun- dation. VI. METHODS A. Experiments We imaged 59 individual male flies (D. melanogaster, Oregon-R strain) for an hour each, following the pro- tocols originally described in [18]. All flies were within the first two weeks post-eclosion during the filming ses- sion. Flies were placed into the arena via aspiration and were subsequently allowed 5 minutes for adaptation be- fore data collection. All recording occurred between the hours of 9:00 AM and 1:00 PM. The temperature during all recordings was 25o ± 1oC. B. Generating Markovian Models Markovian model data sets were generated by first ran- domly selecting a state, and then finding another, ran- domly chosen, instance in the measured data set where the fly was performing that behavior. The behavior per- formed immediately after that behavior is chosen, and the process is iterated until the generated sequence is equivalent in size to the original data set, similar to the first-order alphabets generated in Shannon's original work on information theory [26]. FIG. 6. Partitionings are tree-like over all measured time scales. (A) Definition of the treeness metric, T ; Methods for details. (B) T as a function of the number of transitions in the future and the number of clusters in the most fine-grained partition. Colored lines represent values of T for partitions at varying times in the future, and black lines are values for randomized graphs generated from partitionings that were as- signed randomly. backwards rather than forwards through the tree, T = Hf − Hb Hf , (5) where Hf and Hb are the entropies over all possible paths going forward and backwards, respectively. This metric is bounded between zero and one, 0 ≤ T ≤ 1, and T = 1 implies a perfect hierarchy. We find that the partitionings derived from the in- formation bottleneck algorithm are much more tree-like than random partitions of the behavioral space (Fig. 6B). This is true even when we attempt to optimally predict behavioral states thousands of transitions into the fu- ture. Thus, by finding optimally-predictive representa- tions that best explain the relationship between states over long time scales, we have uncovered a hierarchical ordering of actions, supporting decades-old theory with- out relying on hierarchical clustering, Markov models, or limiting the measured behavioral repertoire. V. CONCLUSIONS We have measured the behavioral repertoires for dozens of fruit flies, paying particular attention to the structure of their behavioral transitions. We find that these transitions exhibit multiple time scales and pos- sess memory that persists thousands of transitions into the future, indicative of internal states that carry mem- ory across thousands of observable behavioral transitions. Using an information bottleneck approach to find the compressed representations that optimally predict our observed dynamics, we find that behaviors are orga- nized in a hierarchical fashion, with fine grained repre- sentations being able to predict short -- time structure and coarser representations being sufficient to predict the fly's actions that are further removed in time. This is funda- mentally different from previous measurements of hier- archy in behavior, which were more limited in the types of behaviors they measured, the time scales over which Number of Clusters025510152001.2.4.6.8100103102101Number of TransitionsHbHfT=Hf−HbHfABP(2)1,1P(1)1,1Q(2)3,2Q(1)1,1TC. Predictive Information Bottleneck The solution to the information bottleneck problem, Eq (4), obeys a set of self -- consistent equations that can be iterated in a manner equivalent to the Blahut-Arimoto algorithm in rate -- distortion theory [24, 27]. For a given Z = K and inverse temperature β, a random initial condition for p(zx) is chosen, and the following self -- consistent equations are iterated until the convergence criterion ((Ft − Ft+1)/Ft < 10−6) is met: (cid:104) (cid:16) exp − βDKL p(yx)p(yz) p(z) Z(β, x) (cid:88) p(zx) = p(z) = p(zx)p(x) x p(yz) = p(yx)p(zx)p(x), (cid:17)(cid:105) , (6) (7) (8) where x ∈ S(n), y ∈ S(n + τ ), z ∈ Z, DKL is the Kullback-Leibler divergence between two probability dis- tributions, and Z(β, x) is a normalizing function. Because this study focuses on hard clusterings of the behavioral space, we find solutions by starting at β = 0.1 and annealing with 40 exponentially-spaced values up to β = 500. After starting from a random initial condi- tion at the initial value of β, the optimization is per- formed at that value until the convergence criterion is met, and that solution is used as the initial condition for the next value of β. All intermediate solutions, p(n) (zx) are stored so they can potentially be included in the found Pareto front. In addition, we perform 24 replicates of this process with different random initial conditions for K = 2, . . . , 25 and for 81 time lag values between n = 1 and n = 5, 000. (cid:96) Given the set of solutions for a given lag, we first take the deterministic limit of each clustering (p(zx) = δz,arg maxz(cid:48) p(z(cid:48)x)) and recalculate I(Z; S(n)) and I(Z; S(n + τ )) accordingly. We then defined the Pareto front, ξ(n), as the set of all solutions, p(n) (zx), such that no other solution for that given lag results in a smaller value for I(Z; S(n)) and a larger value for (cid:96) 7 I(Z; S(n+τ )). Between 150 and 350 solutions were found for all of the fronts. We choosing a clustering for a fixed number of clusters, here, we always pick the representa- tion along the optimal front that has the highest value of I(Z; S(n + τ )). D. Treeness Index To calculate the treeness index, T , we construct a di- rected, acyclic forward graph that connects the partitions as the number of clusters increases for a given time lag with values P ((cid:96)) ij . These values are the probability that a state contained in one cluster, i, in the partitioning with (cid:96) clusters also belongs to cluster j in the partitioning with (cid:96) + 1 clusters. Similarly, we can create the backwards graph, Q((cid:96)) ij , that links clusters in the opposite direction; Q((cid:96)) is the probability that a state in cluster i in the par- ij titioning with (cid:96) + 1 clusters also belongs to the cluster j in the partitioning containing (cid:96) clusters. v ) = (cid:81)N−1 Given these two graphs, we can calculate the entropy of picking a path, π(f ) in the forward direction ver- sus the entropy of picking a path, π(b) in the back- wards direction. These probabilities can be calculated via p(π(f ) v(cid:96)+1,v(cid:96), with v being a chosen sequence of clusters. Thus, we define the forward and backwards entropies as follows: v ) = (cid:81)N−1 v(cid:96),v(cid:96)+1 and p(π(b) (cid:96)=1 Q((cid:96)) (cid:96)=1 P ((cid:96)) (cid:88) (cid:88) v∈V Hf = − p(π(f ) v ) log p(π(f ) v ) Hb = (cid:104)− w∈Wr p(π(b) w ) log p(π(b) w )(cid:105)r, (9) (10) where V is the set of all possible paths and Wr is the set of all paths ending at cluster r in the most fine-grained (cid:104)···(cid:105)r denotes an average over each end partitioning. state. T is then calculated as the relative reduction in entropy between backwards and forwards path probabil- ity distributions, as given by Equation 5. [1] N. Tinbergen, The Study of Instinct (Oxford University Press, Oxford, U. K., 1951). [2] R. Dawkins, in in Growing points in ethology, edited by P. Bateson and R. Hinde (Cambridge Univ. Press, Cam- bridge, U.K., 1976) pp. 7 -- 54. zon, T. L. Jernigan, L. T. Eyler, C. Fennema-Notestine, A. J. Jak, M. C. Neale, C. E. Franz, M. J. Lyons, M. D. Grant, B. Fischl, L. J. Seidman, M. T. Tsuang, W. S. Kremen, and A. M. Dale, Science 335, 1634 (2012). [8] W. J. Davis, G. J. Mpitsos, M. V. Siegler, J. M. Pinneo, [3] H. A. Simon, in Hierarchy Theory, edited by H. H. Pattee and K. B. Davis, American Zoologist 14, 1037 (1974). (Braziller, New York, NY, 1973). [9] R. Dawkins and M. S. Dawkins, Animal Behaviour 739- [4] M. S. A. Graziano and T. N. Aflalo, Neuron 56, 239 755, 973 (1976). (2007). [5] D. S. Bassett, E. Bullmore, B. A. Verchinski, V. S. Mat- and A. Meyer-Lindenberg, J. tay, D. R. Weinberger, Neuroscience 28, 9239 (2008). [6] D. A. Dombeck, M. S. Graziano, and D. W. Tank, Jour- nal of Neuroscience 29, 13751 (2009). [7] C. H. Chen, E. D. Gutierrez, W. Thompson, M. S. Paniz- [10] L. Lefebvre, Animal Behaviour 29, 973 (1981). [11] L. Lefebvre, Behavioural Processes 7, 93 (1982). [12] L. Lefebvre and R. Joly, Animal Behaviour 30, 1020 (1982). [13] A. M. Seeds, P. Ravbar, P. Chung, S. Hampel, F. M. Midgley, Jr, B. D. Mensh, and J. H. Simpson, eLife 3, e02951 (2014). 8 [14] G. A. Miller, Trends in Cognitive Sciences 7, 141 (2003). [15] W. Heiligenberg, Animal Behaviour 21, 169 (1973). [16] D. Z. Jin and A. A. Kozhevnikov, PLoS Computational [22] B. Hopkins and L. Ronnqvist, Developmental psychobi- ology 40, 168 (2002). [23] W. Bialek, I. Nemenman, and N. Tishby, Neural com- Biology 7, e1001108 (2011). putation 13, 2409 (2001). [17] R. Dawkins and M. Dawkins, Behaviour , 83 (1973). [18] G. J. Berman, D. M. Choi, W. Bialek, and J. W. Shae- vitz, J. Royal Soc. Interface 11, 20140672 (2014). [19] L. van der Maaten and G. Hinton, J. Mach. Learning Research 9, 85 (2008). [24] N. Tishby, F. C. Pereira, and W. Bialek, in Proceedings of the 37th Annual Allerton Conference on Communica- tion, Control and Computing (University of Illinois Press, Urbana-Champaign, IL, 1999) pp. 368 -- 377. [25] B. Corominas-Murtra, C. Rodr´ıguez-Caso, J. Goni, and [20] M. Takahata, M. Yoshino, and M. Hisada, The Journal R. Sol´e, Chaos 21, 016108 (2011). of experimental biology (1981). [26] C. E. Shannon, Bell Systems Technical Journal 27, 379 [21] H. Ackermann, E. Scholz, W. Koehler, and J. Dich- gans, Electroencephalography and Clinical Neurophysi- ology/Evoked Potentials Section 81, 71 (1991). (1948). [27] R. E. Blahut, IEEE Trans. Info. Theory IT-18, 460 (1972).
1704.07413	1	1704	2017-04-24T18:45:29	RT-TDDFT study of hole oscillations in B-DNA monomers and dimers	[ "physics.bio-ph", "physics.chem-ph" ]	We employ Real-Time Time-Dependent Density Functional Theory to study hole oscillations within a B-DNA monomer (one base pair) or dimer (two base pairs). Placing the hole initially at any of the bases which make up a base pair, results in THz oscillations, albeit of negligible amplitude. Placing the hole initially at any of the base pairs which make up a dimer is more interesting: For dimers made of identical monomers, we predict oscillations with frequencies in the range $f \approx$ 20-80 THz, with a maximum transfer percentage close to 1. For dimers made of different monomers, $f \approx$ 80-400 THz, but with very small or small maximum transfer percentage. We compare our results with those obtained recently via our Tight-Binding approaches and find that they are in good agreement.	physics.bio-ph	physics	RT-TDDFT study of hole oscillations in B-DNA monomers and dimers M. Tassi,∗ A. Morphis, K. Lambropoulos, and C. Simserides National and Kapodistrian University of Athens, Department of Physics, Panepistimiopolis, Zografos, GR-15784, Athens, Greece We employ Real-Time Time-Dependent Density Functional Theory to study hole oscillations within a B-DNA monomer (one base pair) or dimer (two base pairs). Placing the hole initially at any of the bases which make up a base pair, results in THz oscillations, albeit of negligible amplitude. Placing the hole initially at any of the base pairs which make up a dimer is more interesting: For dimers made of identical monomers, we predict oscillations with frequencies in the range f ≈ 20-80 THz, with a maximum transfer percentage close to 1. For dimers made of different monomers, f ≈ 80-400 THz, but with very small or small maximum transfer percentage. We compare our results with those obtained recently via our Tight-Binding approaches and find that they are in good agreement. I. INTRODUCTION Carriers can be inserted in DNA via electrodes or generated by UV irradiation or by oxidation and reduction. We had better discriminate between the words transport (implying the use of bias between electrodes), transfer (a carrier moves without bias) and migration (a transfer over rather long distances). Charge transport, transfer and migration in DNA are affected by many external factors such as chemical environment, interaction with the substrate or quality of contacts [1], as well as by many intrinsic factors like the sequence of bases [2]. Charge transfer in DNA is important for DNA damage and repair [3–6] and it may be used as an indicator for the separation between cancer tumor and healthy tissue [7]. Although (unbiased) charge transfer in DNA nearly vanishes after 10 to 20 nm [8–10], DNA is still a promising component for (biased) charge transport in molecular electronics, e.g., as a short molecular wire [11]. It may also be exploited in nanotechnology for nanosensors [12] and molecular wires [13, 14]. The scope of this article is to study the effect of base sequence on charge transfer, in all possible monomers (i.e. within one base pair) and dimers (i.e. within two base pairs), using Real-Time Time-Dependent Density Functional Theory (RT-TDDFT) [15, 16]. In previous articles, we studied small and long B-DNA segments [8–10, 17, 18] with Tight Binding (TB) approaches, specifically with a wire model and an extended ladder model. We determined the spatio-temporal evolution of an extra carrier (hole or electron) along a N base-pair DNA segment and showed that the carrier movement has frequency content mainly in the THz domain, i.e., depending on the case, in the NIR, MIR and FIR range [19]. This part of the electromagnetic (EM) spectrum is significant because it can be used to extract complementary to traditional spectroscopic information e.g. about hydrogen bond stretching, bond torsion in liquids and gases, low-frequency bond vibrations and so on, and because it is relatively non-invasive compared to higher-frequency regions of the EM spectrum [20]. In this article we study B-DNA monomers and dimers with RT-TDDFT, which is one of the few computationally viable techniques to model carrier dynamics in molecular systems [21–24]. RT-TDDFT can simulate density fluctua- tions of the order of attoseconds to femtoseconds and can model the effects of an ultrafast, intense, short-pulse laser field, including the full (nonlinear) response of the electron density [25]. RT-TDDFT also provides the full frequency dependence of properties of interest via Fourier transform from the time domain to the frequency domain. In addition, the determination of many-particle eigenstates is avoided, and the calculation is reduced to an independent particle problem carried out in real time [26]. Recently, RT-TDDFT was applied [27] to study the charge transfer between the oxidized or reduced a-sulfur monomers and the neighbouring neutral ones. Another recent work used RT-TDDFT to study electron transfer through oligo-p-phenylenevinylene (OPV) and carbon bridged OPV (COPV) [28]. The rest of this article is organized as follows. In Section II we sketch the basic formalism of RT-TDDFT. In Section III we give some computational details and explain our notation. In Section IV we present our results. Finally, in Section V we state our conclusions. ∗Corresponding author: [email protected] 7 1 0 2 r p A 4 2 ] h p - o i b . s c i s y h p [ 1 v 3 1 4 7 0 . 4 0 7 1 : v i X r a II. RT-TDDFT FORMALISM 2 Density Functional Theory (DFT) [29, 30] is an efficient method for treating ground state properties of many electron systems e.g. molecules or solids. Years after its development, it was extended [31] to time dependent systems (TDDFT). Specifically, the Time-Dependent Kohn-Sham (TDKS) equations with an effective potential energy υKS(r, t), uniquely described by the time-dependent charge density, ρ(r, t), are, in atomic units, i ∂ ∂t Ψj(r, t) = (cid:20) − ∇2 + υKS(r, t)(cid:21)Ψj(r, t) = ∇2 + υext(r, t) + υH(r, t) + υxc[ρ](r, t)(cid:21)Ψj(r, t). 1 2 (cid:20) − 1 2 The charge density is the sum over all occupied orbitals j = 1, 2, . . . Nocc, i.e., ρ(r, t) = Nocc Xj=1 Ψj(r, t)2. (1) (2) υext(r, t) includes external fields and nuclear potentials, υH(r, t) is the Hartree potential energy. Exchange and correlation effects are included in υxc[ρ](r, t). RT-TDDFT is based on a direct numerical integration of Eq. 1. This differs from the traditional linear-response approach, which is not actually a time-resolved method but instead solves Eq. 1 in the frequency domain for the excitation energies of a system subject to a small perturbation [23]. Within RT-TDDFT, we solve the TDKS equations and obtain the electron density at each time step. The electron density is then used for the calculation of the Hamiltonian in the next cycle of the self-consistent process. III. COMPUTATIONAL DETAILS AND NOTATIONS With the notation YX we mean that the bases Y and X of two successive base pairs Y-Ycompl and X-Xcompl are located at the same strand in the direction 5′-3′. Xcompl (Ycompl) is the complementary base of X (Y). In other words, the notation YX means that Y-Ycompl is the one base pair and X-Xcompl is the other base pair, separated by 3.4 A and twisted clockwise by 36o relatively to the first base pair, along the growth axis of the nucleotide chain. For example, the notation AC denotes that one strand contains A and C in the direction 5′-3′ and the complementary strand contains T and G in the direction 3′-5′. The notation (10) means that the hole is initially placed at the 1st base pair of the dimer and the notation (01) means that it is initially placed at the 2nd base pair of the dimer. For our DFT and RT-TDDFT calculations we used the NWChem open-source computational package [32]. Ad- ditionally, we performed Fourier analysis, implemented with MATLAB, to determine the frequency content of the charge oscillations. The geometry of monomers and all dimers was produced via the BIOVIA Discovery Studio [33]. The backbone was removed and H atoms were added at the corresponding atomic sites. The range-separated functional CAM-B3LYP [34], which is appropriate for the correct estimation of the exchange energy, both at short and long ranges, was used for all monomers and dimers. The calculations were performed using the 6-31++G [35], 3-21++G [36] and aug-cc-pVDZ [37] basis sets, which include diffuse functions, for all systems. Within a small number of time steps, we noticed that our results were similar (cf. Appendix). Hence, we chose to perform our longer simulations with the 6-31++G basis set [35], that is, the least computationally expensive between the converging sets. In a Gaussian basis set, it is most natural to use the single particle reduced density matrix, whose time evolution is governed by the von Neumann equation. The Magnus propagator is used in NWChem's RT-TDDFT implementation, which is both stable and conserves the density matrix idempotency [23]. At the end of each time step the fragments' charge is calculated with an appropriate population analysis method, along with the total dipole moment. For monomers, the initial state (i.e. hole localized at a specific base) was produced by CDFT, where the charge constraint was calculated with Lowdin population analysis that is also used in the subsequent RT-TDDFT simulation. For dimers, the initial state (i.e. charge localization) was produced by the following procedure: first, a ground state DFT calculation of the charged and neutral isolated monomers was performed, yielding the corresponding monomers' eigenstates. Then, the dimer's eigenstates were approximated in a non self-consistent manner (NOSCF) by the orthogonalized monomers' eigenstates. Since the monomers comprising each dimer are 3.4 A apart, this is a fairly good approximation that also helps circumvent CDFT convergence problems. 3 Lowdin [38] population analysis was integrated into RT-TDDFT module of NWChem for the calculation of each monomer's charge at each time step. Lowdin population analysis is known to be much less basis-set dependent and also it does not suffer from the ultra-fast charge oscillations that Mulliken analysis (which is the default scheme in NWChem's RT-TDDFT) artificially introduces in RT-TDDFT charge simulations. As a result, Lowdin population analysis gives a more clear picture of charge oscillations convergence towards the basis-set limit, cf. Figs. 7-8. The main frequencies of charge oscillations are extracted from the results via Fourier analysis. In order to increase the resolution, zero-padding with appropriate signal attenuation was used where necessary. IV. RESULTS Before discussing our RT-TDDFT results, we give a brief summary of the TB picture. In TB, charge transfer between the two bases of a monomer or between the two base pairs of a dimer can be easily analytically solved [8, 17, 18]. In both cases, it is a movement between the two sites i, j of a two-site system and it is mathematically similar to Rabi oscillations in Quantum Optics [39]. The maximum transfer percentage p, i.e., the maximum probability to find the carrier at the site where it was not initially placed is p = (2tij)2 (2tij )2 + (∆ij )2 , (3) where ∆ij is the difference between the two sites' on-site energies and tij is the hopping parameter between the two sites. Since between the two bases of a monomer, ∆ij is much larger than tij, p is negligible. For movements between the two monomers of a dimer things are different: For dimers made of identical monomers, since ∆ij = 0, it follows that p = 1. For dimers made of different monomers p < 1; in fact, it is usually negligible with one exception, a hole moving between the two monomers of a GA dimer, where p is of the order of 0.25. In TB, the frequency and period of charge oscillations is given by f = 1 T = q(2tij)2 + ∆2 ij h , (4) hence, increasing tij and / or ∆ij results in higher frequencies. In what follows, this simple picture is qualitatively obeyed by our RT-TDDFT results. Hole oscillations between the two bases of a base pair are in the THz regime, but have negligible maximum transfer percentage, in agreement with the TB results [18]. Hole oscillations between the two base pairs of a dimer, with the 6-31++G basis set, are shown in Fig 1. We place the hole initially at one of the monomers which make up the dimer and call p the maximum transfer percentage to the other monomer. For dimers made of identical monomers (AT, TA, GC, CG) the hole is transferred almost completely to the other monomer (p ≈ 0.9). For dimers made of different monomers (AC, CA, GA, AG), p is close to zero with the exception of the GA dimer, where p ≈ 0.3. Our calculations for the cases AT(01), TA(01), GC(01), and CG(01) are not included in Fig.1, since they are identical with the cases AT(10), TA(10), GC(10), CG(10), respectively, as expected from base-pair symmetry and found also in Refs. [8, 17, 18]. In Fig. 2 we present the mean probabilities to find the extra hole at each monomer, having placed it initially either at the 1st monomer (10) or at the 2nd monomer (01). Let us call Pi the mean probability at the monomer where the hole is initially placed, and Pf the mean probability at the other monomer. We observe that for dimers made of identical monomers, where p ≈ 1, we have Pi ≈ Pf ≈ 0.5, while for dimers made of different monomers Pi ≈ 1 and Pf ≈ 0, with the exception of the GA dimer. This, again, agrees qualitatively with the TB picture [8, 17, 18]. In Fig. 3 we depict the maximum transfer percentage p as well as the electronic coupling energy (also called the electron transfer matrix element) VRP between the reactant and product states R and P [40–43], which reflects the magnitude of the interaction between the two monomers. In our case reactant state corresponds to the hole at the initial placement and product state corresponds to the hole at the other monomer. We also show the energy difference δE between reactant and product states. In analogy with a Rabi oscillation, we expect charge transfer to increase increasing VRP and to decrease decreasing δE. Additionally, for δE ≈ 0, we expect p to be close to 1. For the AT, TA, GC and CG dimers, δE ≈ 0, hence p is indeed close to 1. For dimers made of different monomers, we observe that AC, CA and AG have large δE ≈ 0.45-0.75 eV and small VRP ≈ 0-1 eV, therefore their p is insignificant. An exception is the dimer GA which not only has a large δE ≈ 0.65 eV, but also the largest VRP of all dimers ≈ 5 eV which makes charge transfer significant, with p ≈ 0.3. In Fig. 4 we present the frequencies of hole oscillations obtained by Fourier analysis. For dimers made of identical monomers (upper panel) the frequencies f ≈ 20-80 THz and their amplitudes ≈ 0.2-0.5. For dimers made of different AT TA GC CG 1.0 0.8 4 20 t (fs) AC(01) AG(01) 30 40 50 CA(10) GA(10) CA(01) GA(01) 0.6 0.4 e g r a h c 0.2 0.0 0 10 AC(10) AG(10) 1.0 0.8 0.6 0.4 0.2 0.0 e g r a h c 0 10 20 t (fs) 30 40 50 FIG. 1: Net charge versus time for the monomer where the hole was initially placed: (upper panel) dimers made of identical monomers, (lower panel) dimers made of different monomers. The notation (10) means that the hole is initially placed at the 1st base pair of the dimer, while the notation (01) means that the hole is initially placed at the 2nd base pair of the dimer. monomers (lower panel) f ≈ 80-400 THz and the amplitudes ≈ 0.003-0.02, except for the two GA cases which have amplitudes ≈ 0.1-0.2. Obviously, larger amplitudes reflect larger transfer. As expected by TB, for dimers made of different monomers we generally get higher frequencies than for dimers made of identical monomers. Additionally, in Fig. 5 we observe that for dimers made of identical monomers the frequencies follow VRP, which does not hold for dimers made of different monomers. This, again, agrees with the TB prediction (Eq. 4). Finally, in Fig. 6 we compare our RT-TDDFT results with those of TB [8, 18] for two different sets of TB parameters: (a) used in Ref. [18] (calculated in Ref. [44]) and (b) used in Ref. [8]. We observe that the maximum transfer percentages obtained by RT-TDDFT are in good agreement with those obtained by TB. For the periods, the results, both for RT-TDDFT and TB are quite close especially when we compare them with parametrization (b) used in Ref. [8]. 1.0 0.8 0.6 0.4 0.2 0.0 1st base pair 2nd base pair ) 0 1 ( A T ) 0 1 ( T A ) 0 1 ( G C ) 0 1 ( C G ) 0 1 ( G T A C , ) 1 0 ( G T A C , ) 0 1 ( T G C A , ) 1 0 ( T G C A , ) 0 1 ( T C G A , ) 1 0 ( T C G A , ) 0 1 ( C T A G , ) 1 0 ( C T A G , 5 1.0 0.8 0.6 0.4 0.2 0.0 s e i t i l i b a b o r p n a e m FIG. 2: The mean probabilities to find the extra hole, at each base pair of a dimer, having placed it initially at the 1st (10) or at the 2nd (01) base pair. 1,0 0,9 0,8 0,7 0,6 0,5 0,4 0,3 0,2 0,1 0,0 r e f s n a r t x a m max transfer VRP (eV) Delta (eV) 6 5 4 ) 3 V e ( P R V 2 1 0 ) 0 1 ( A T ) 0 1 ( T A ) 0 1 ( G C ) 0 1 ( C G ) 0 1 ( G T A C , ) 1 0 ( G T A C , ) 0 1 ( T G C A , ) 1 0 ( T G C A , ) 0 1 ( T C G A , ) 1 0 ( T C G A , ) 0 1 ( C T A G , ) 1 0 ( C T A G , base pairs 0,8 0,7 0,6 0,5 0,4 ) V e ( 0,3 E 0,2 0,1 0,0 FIG. 3: Maximum transfer, i.e., oscillation amplitude p, electron transfer coupling energy VRP and energy difference δE between reactant and product states. Initially, the hole is placed at the 1st monomer (10) or at the 2nd monomer (01). 6 e d u t i l p m a Dimers TA AT GC CG 100 10-1 10-2 10-3 100 100 101 f (THz) 102 103 e d u t i l p m a 10-1 10-2 Dimers CA,TG(1,0) CA,TG(0,1) AC,GT(1,0) AC,GT(0,1) AG,CT(1,0) AG,CT(1,0) GA,TC(1,0) GA,TC(0,1) 10-3 100 101 102 f (THz) 103 FIG. 4: Frequencies of hole oscillations in dimers made of identical monomers (upper panel) and different monomers (lower panel), obtained via Fourier analysis. 400 360 320 280 240 200 160 120 80 40 0 ) z H T ( f HOMO dimers frequency (THz) VRP (eV) ) 0 1 ( A T ) 0 1 ( T A ) 0 1 ( G C ) 0 1 ( C G ) 0 1 ( G T A C , ) 1 0 ( G T A C , ) 0 1 ( T G C A , ) 1 0 ( T G C A , ) 0 1 ( T C G A , ) 1 0 ( T C G A , ) 0 1 ( C T A G , ) 1 0 ( C T A G , - - 6 5 4 3 2 1 0 ) V e ( P R V FIG. 5: Extra hole oscillation frequency versus VRP for all dimers. 1.0 0.8 0.6 0.4 0.2 0.0 p Max transfer RT-TDDFT TB(a) TB(b) ) 0 1 ( A T ) 0 1 ( T A ) 0 1 ( G C ) 0 1 ( C G 1.0 0.8 0.6 0.4 0.2 0.0 ) 0 1 ( G T A C , ) 1 0 ( G T A C , ) 0 1 ( T G C A , ) 1 0 ( T G C A , ) 0 1 ( T C G A , ) 1 0 ( T C G A , ) 0 1 ( C T A G , ) 1 0 ( C T A G , - - 104 103 ) s f ( T 102 101 100 7 Period RT-TDDFT TB(a) TB(b) ) 0 1 ( A T ) 0 1 ( T A ) 0 1 ( G C ) 0 1 ( C G ) 0 1 ( G T A C , ) 1 0 ( G T A C , ) 0 1 ( T G C A , ) 1 0 ( T G C A , ) 0 1 ( T C G A , ) 1 0 ( T C G A , ) 0 1 ( C T A G , ) 1 0 ( C T A G , - - FIG. 6: Comparison of RT-TDDFT (this work) and TB from (a) Ref. [18] and (b) Ref. [8]: maximum transfer percentage (left panel) and oscillation periods (right panel). V. CONCLUSION 8 We studied how base sequence affects hole transfer in small DNA segments (between the two bases of a monomer and between the two base pairs of a dimer) with the RT-TDDFT method. We calculated the maximum transfer percentage, i.e., the oscillation amplitude, p, the mean probability at the site where the hole is initially placed, Pi, and the mean probability at the other site, Pf , as well as the frequency content of the oscillations. Hole oscillations between the two bases of a base pair are in the THz regime, but have negligible maximum transfer percentage. For dimers made of identical monomers, the maximum transfer percentage is almost 1. For dimers made of different monomers, we observed negligible hole transfer except for the GA dimer where p ≈ 0.3. For all dimers, we found frequency content in the THz domain and noticed that Fourier amplitudes reflect p and Pf . For dimers made of different monomers we generally get higher frequencies than for dimers made of identical monomers. We compared the RT-TDDFT results with those obtained by TB in Refs. [8, 17, 18] and we found that they are in good agreement. In Figs. 7-8 we show charge oscillations calculated with different basis sets: 6-31++G[35], 3-21++G [36] and aug-cc-pVDZ [37] and we see that the results are similar (cf. Sec. III). Appendix 1.0 0.8 0.6 0.4 0.2 0.0 1.0 0.8 0.6 0.4 0.2 0.0 e g r a h c e g r a h c AT(10) 6-31++G aug-cc-pVDZ 3-21++G e g r a h c 0 10 20 t (fs) 30 40 50 TA(10) 6-31++G aug-cc-pVDZ 3-21++G e g r a h c 0 10 20 t (fs) 30 40 50 1.0 0.8 0.6 0.4 0.2 0.0 1.0 0.8 0.6 0.4 0.2 0.0 9 GC(10) 6-31++G aug-cc-pVDZ 3-21++G 0 20 40 t (fs) 60 80 100 CG(10) 6-31++G aug-cc-pVDZ 3-21++G 0 10 20 t (fs) 30 40 50 FIG. 7: Time evolution of hole transfer in dimers made of identical monomers, for the three different basis sets. e g r a h c e g r a h c e g r a h c e g r a h c 1.0 0.8 0.6 0.4 0.2 0.0 1.0 0.8 0.6 0.4 0.2 0.0 1.0 0.8 0.6 0.4 0.2 0.0 1.0 0.8 0.6 0.4 0.2 0.0 0 10 20 t (fs) e g r a h c AC,GT(10) 6-31++G aug-cc-pVDZ 3-21++G 30 40 50 AC,GT(01) 6-31++G aug-cc-pVDZ 3-21++G e g r a h c 30 40 50 0 10 20 t (fs) 0 10 20 t (fs) e g r a h c AG,CT(10) 6-31++G aug-cc-pVDZ 3-21++G 30 40 50 AG,CT(01) 6-31++G aug-cc-pVDZ 3-21++G e g r a h c 0 10 20 t (fs) 30 40 50 1.0 0.8 0.6 0.4 0.2 0.0 1.0 0.8 0.6 0.4 0.2 0.0 1.0 0.8 0.6 0.4 0.2 0.0 1.0 0.8 0.6 0.4 0.2 0.0 10 0 10 20 t (fs) CA,TG(10) 6-31++G aug-cc-pVDZ 3-21++G 30 40 50 CA,TG(01) 6-31++G aug-cc-pVDZ 3-21++G 0 10 20 t (fs) 30 40 50 0 10 20 t (fs) GA,TG(10) 6-31++G aug-cc-pVDZ 3-21++G 30 40 50 GA,TG(01) 6-31++G aug-cc-pVDZ 3-21++G 0 10 20 t (fs) 30 40 50 FIG. 8: Time evolution of hole transfer in dimers made of different monomers, for the three different basis sets. Acknowledgments 11 This work was supported by computational time granted from the Greek Research & Technology Network (GRNET) in the National HPC facility - ARIS - under project ID pr002008 - CODNA. A. Morphis thanks the State Scholarships Foundation-IKY for a Ph.D. research scholarship via 'IKY Fellowships of Excellence', Hellenic Republic-Siemens Settlement Agreement. http://users.uoa.gr/~csimseri/physics_of_nanostructures_and_biomaterials.html Related work at [1] Aperiodic Stuctures in Condensed Matter: Fundamentals and Applications, edited by E. Macia, (Taylor and Francis CRC, Boca Raton FL, USA 2009). [2] G. Cuniberti, E. Macia, A. Rodriguez, R.A. Romer, Tight-Binding Modeling of Charge Migration in DNA Devices, in Charge Migration in DNA: Perspectives from Physics, Chemistry, and Biology edited by T. Chakraborty (Springer, Berlin Heidelberg New York 2007). [3] P.J. Dandliker, R.E. Holmlin, J.K. Barton, Oxidative thymine dimer repair in the DNA helix, Science 275, 1465 (1997). [4] S.R. Rajski, B.A. Jackson, J.K. Barton, DNA repair: Models for damage and mismatch recognition, Mutation Research 447, 49 (2000). [5] B. Giese, Long-distance electron transfer through DNA, Annual Review of Biochemistry 71, 51 (2002). [6] B. Giese, Electron transfer through DNA and peptides, Bioorganic & Medicinal Chemistry 14, 6139 (2006). [7] C.-T. Shih, Y.-Y. Cheng, S.A. Wells, C.-L. Hsu, R.A. Romer, Charge transport in cancer-related genes and early carcino- genesis, Computer Physics Communications 182, 36 (2011). [8] C. Simserides, A systematic study of electron or hole transfer along DNA dimers, trimers and polymers, Chemical Physics 440, 31 (2014). [9] K. Lambropoulos, M. Chatzieleftheriou, A. Morphis, K. Kaklamanis, M. Theodorakou, and C. Simserides, Unbiased charge oscillations in B-DNA: Monomer polymers and dimer polymers, Physical Review E 92, 032725 (2015). [10] K. Lambropoulos, M. Chatzieleftheriou, A. Morphis, K. Kaklamanis, R. Lopp, M. Theodorakou, M. Tassi and C. Simserides, Electronic structure and carrier transfer in B-DNA monomer polymers and dimer polymers: Stationary and time-dependent aspects of a wire model versus an extended ladder model, Physical Review E 94, 062403 (2016). [11] C.H. Wohlgamuth, M.A. McWilliams, and J.D. Slinker, DNA as a molecular wire: Distance and sequence dependence, Anal. Chem. 85, 8634 (2013). [12] M. Huang, H. Li, H. He, X. Zhang and S. Wang, An electrochemical impedance sensor for simple and specific recognition of G–G mismatches in DNA, Analytical Methods 8, 7413 (2016). [13] J.C. Genereux and J.K. Barton, Mechanisms for DNA Charge Transport, Chemical Reviews 110, 1642 (2010). [14] K. Kawai and T. Majima, Increasing the Hole Transfer Rate Through DNA by Chemical Modification, in Chemical Science of π -Electron Systems, edited by T. Akasaka, A. O. S. Fukuzumi, H. K. Y. Aso, (Springer Springer, Tokyo Heidelberg New York Dordrecht London 2015). [15] J. Theilhaber, Ab initio simulations of sodium using time-dependent density-functional theory, Phys. Rev. B 46, 12990 (1992). [16] K. Yabana and G.F. Bertsch, Time-dependent local-density approximation in real time, Phys. Rev. B 54, 4484 (1996). [17] K. Lambropoulos, K. Kaklamanis, G. Georgiadis and C. Simserides, THz and above THz electron or hole oscillations in DNA dimers and trimers, Ann. Phys. (Berlin) 526, 249 (2014). [18] K. Lambropoulos, K. Kaklamanis, A. Morphis, M. Tassi, R. Lopp, G. Georgiadis, M. Theodorakou, M. Chatzieleftheriou and C. Simserides, Wire and extended ladder model predict THz oscillations in DNA monomers, dimers and trimers, J. Phys.: Condens. Matter 28, 495101 (2016). [19] ISO 20473 specifies: Near-Infrared (NIR) 0.78 - 3 µm, Mid-Infrared (MIR) 3-50 µm, Far-Infrared (FIR) 50-1000 µm. [20] Terahertz Sources and Detectors, in Terahertz Imaging for Biomedical Applications: Pattern Recognition and Tomo- graphic Reconstruction edited by X. Yin, B.W.-H Ng, and D. Abbott, Springer Science+Business Media, LLC 2012, http://www.springer.com/978-1-4614-1820-7, ISBN 978-1-4614-1820-7 e-ISBN 978-1-4614-1821-4, (Springer, New York, Dordrecht, Heidelberg, London, 2012). [21] Time-Dependent Density Functional Theory edited by M. Marques (Springer, Berlin, 2006). [22] W. Liang, C.T. Chapman, X. Li, J. Chem. Efficient first-principles electronic dynamics, Phys. 134, 184102 (2011). [23] K. Lopata and N. Govind, Modeling Fast Electron Dynamics with Real-Time Time-Dependent Density Functional Theory: Application to Small Molecules and Chromophores, J. Chem. Theory Comput. 7, 1344 (2011). [24] Time-Dependent Density-Functional Theory: Concepts and Applications edited by C. Ullrich (Oxford University Press: Oxford; New York, 2012). 12 [25] M.R. Provorse and C.M. Isborn, Electron Dynamics with Real-Time Time-Dependent Density Functional Theory, Int. J. Quantum Chem. 116, 739 (2016). [26] Y. Takimoto, F.D. Vila, and J.J. Rehr, Real-time time-dependent density functional theory approach for frequency- dependent nonlinear optical response in photonic molecules, J. Chem. Phys. 127, 154114 (2007). [27] P. Partovi-Azar and P. Kaghazchi, Time-Dependent Density Functional Theory Study on Direction-Dependent Electron and Hole Transfer Processes in Molecular Systems, J. Comput. Chem. 38 698 (2017). [28] O. Granas, G. Kolesov, and E. Kaxiras, Impact of Vibrations and Electronic Coherence on Electron Transfer in Flat Molecular Wires, MRS Advances (2017), pp. 1–6, doi:10.1557/adv.2017.157 [29] P. Hohenberg and W. Kohn, Inhomogeneous electron gas, Phys. Rev. 136, B864 (1964). [30] W. Kohn and L.J. Sham, Self-consistent equations including exchange and correlation effects, Phys. Rev. 140, A1133 (1965). [31] E. Runge and E.K.U. Gross, Density-Functional Theory for Time-Dependent Systems, Phys. Rev. Lett. 52, 997 (1984). [32] M. Valiev, E.J. Bylaska, N. Govind, K. Kowalski, T.P. Straatsma, H.J.J. Van Dam, D. Wang, J. Nieplocha, E. Apra, T.L. Windus, W.A. de Jong, NWChem: A comprehensive and scalable open-source solution for large scale molecular simulations, Comput. Phys. Commun. 181, 1477 (2010). [33] http://accelrys.com/resource-center/downloads/freeware/index.html [34] T. Yanai, D.P. Tew, N.C. Handy, A new hybrid exchangecorrelation functional using the Coulomb-attenuating method (CAM-B3LYP), Chem. Phys. Lett. 393, 51 (2004). [35] W.J. Hehre, R. Ditchfield and J.A. Pople, Self –Consistent Molecular Orbital Methods. XII. Further Extensions of Gaus- sian–Type Basis Sets for Use in Molecular Orbital Studies of Organic Molecules, J. Chem. Phys. 56, 2257 (1972). [36] J.S. Binkley, J.A. Pople and W.J. Hehre, Self-consistent molecular orbital methods. 21. Small split-valence basis sets for first-row elements, J. Am. Chem. Soc 102, 939 (1980). [37] T.H. Dunning, Gaussian basis sets for use in correlated molecular calculations. I. The atoms boron through neon and hydrogen, J. Chem. Phys. 90, 1007 (1989). [38] P.-O Lowdin, On the Non-Orthogonality Problem Connected with the Use of Atomic Wave Functions in the Theory of Molecules and Crystals, J. Chem. Phys. 18 (1950) 365. [39] C. Simserides, Link. 2016, (http://hdl.handle.net/11419/2108) Quantum Optics and braries - http://repository.kallipos.gr/handle/11419/2108?locale=en Derivatives. URI: No Lasers. License: Athens: Attribution Hellenic - http://hdl.handle.net/11419/2108 ISBN: Li- Academic Non-Commercial 978-960-603-073-4 [40] M.D. Newton and N. Sutin, Electron Transfer Reactions in Condensed Phases, Ann. Rev. Phys. Chem. 35, 437 (1984). [41] R.A. Marcus and N. Sutin, Electron transfers in chemistry and biology, Biochim. Biophys. Acta 811, 265 (1985). [42] Electron Transfer in Inorganic, Organic and Biological Systems, edited by J.R. Bolton, N. Mataga, and G. McLendon, (American Chemical Society, Washington, D.C., 1991). [43] A. Farazdel, M. Dupuis, E. Clementi, and A. Aviram, Electric field induced intramolecular electron transfer in spiro π- electron systems and their suitability as molecular electronic devices. A theoretical study, J. Am. Chem. Soc. 112, 4206 (1990). [44] L.G.D. Hawke, G. Kalosakas and C. Simserides, Electronic parameters for charge transfer along DNA, Eur. Phys. J. E 32 (2010) 291.
1502.07044	1	1502	2015-02-25T03:57:10	Mapping the energy and diffusion landscapes of membrane proteins at the cell surface using high-density single-molecule imaging and Bayesian inference: application to the multi-scale dynamics of glycine receptors in the neuronal membrane	[ "physics.bio-ph", "q-bio.QM" ]	Protein mobility is conventionally analyzed in terms of an effective diffusion. Yet, this description often fails to properly distinguish and evaluate the physical parameters (such as the membrane friction) and the biochemical interactions governing the motion. Here, we present a method combining high-density single-molecule imaging and statistical inference to separately map the diffusion and energy landscapes of membrane proteins across the cell surface at ~100 nm resolution (with acquisition of a few minutes). When applying these analytical tools to glycine neurotransmitter receptors (GlyRs) at inhibitory synapses, we find that gephyrin scaffolds act as shallow energy traps (~3 kBT) for GlyRs, with a depth modulated by the biochemical properties of the receptor-gephyrin interaction loop. In turn, the inferred maps can be used to simulate the dynamics of proteins in the membrane, from the level of individual receptors to that of the population, and thereby, to model the stochastic fluctuations of physiological parameters (such as the number of receptors at synapses). Overall, our approach provides a powerful and comprehensive framework with which to analyze biochemical interactions in living cells and to decipher the multi-scale dynamics of biomolecules in complex cellular environments.	physics.bio-ph	physics	Mapping energy and diffusion landscapes 2 Mapping the energy and diffusion landscapes of membrane proteins at the cell surface using high-density single-molecule imaging and Bayesian inference: application to the multi-scale dynamics of glycine receptors in the neuronal membrane. Jean-Baptiste Masson1 Physics of Biological System, Pasteur Institute, Paris , France, CNRS UMR 3525, Paris, France. Patrice Dionne* Laboratoire Kastler Brossel, CNRS UMR 8552, Ecole Normale Superieure, Paris , France, Centre de recherche Universit Laval Robert-Giffard, Quebec, Canada. Charlotte Salvatico* Biologie Cellulaire de la Synapse, Institut National de la Sante et de la Recherche Medicale U1024, Institut de Biologie de l'Ecole Normale Superieure (IBENS), 75005 Paris, France. Marianne Renner Biologie Cellulaire de la Synapse, Institut National de la Sante et de la Recherche Medicale U1024, Institut de Biologie de l'Ecole Normale Superieure (IBENS), 75005 Paris, France. Christian G. Specht Biologie Cellulaire de la Synapse, Institut National de la Sante et de la Recherche Medicale U1024, Institut de Biologie de l'Ecole Normale Superieure (IBENS), 75005 Paris, France. Mapping energy and diffusion landscapes 3 Antoine Triller 2 Biologie Cellulaire de la Synapse, Institut National de la Sante et de la Recherche Medicale U1024, Institut de Biologie de l'Ecole Normale Superieure (IBENS), 75005 Paris, France. Maxime Dahan 3 Laboratoire Physico-Chimie, Institut Curie, CNRS UMR 168, Universit Pierre et Marie Curie-Paris 6 Paris France 1Corresponding author. email: [email protected]. Address: Institut Pasteur, Physics of Biological Systems, 28 rue du Dr Roux, CNRS, UMR 3525 75724, Paris Cedex 15, France. Tel.: +33 (0)1 40 61 39 23. ∗ These authors contributed equally 2Corresponding author, email: [email protected], Address: Ecole Normale Superieure, Biologie Cellulaire de la Synapse Institut National de la Sante et de la Recherche Medicale U1024 Institut de Biologie de l'Ecole Normale Superieure (IBENS), 45 rue d'Ulm, Tel.: +33 (0)1 44 32 35 47. 3Corresponding author, email: [email protected], Laboratoire Physico- Chimie, Institut Curie, CNRS UMR 168, 26 Rue d'Ulm, 75005, Paris France, Tel.: +33 (0)1 56 24 67 54. Abstract Protein mobility is conventionally analyzed in terms of an effective diffu- sion. Yet, this description often fails to properly distinguish and evaluate the physical parameters (such as the membrane friction) and the biochemi- cal interactions governing the motion. Here, we present a method combining high-density single-molecule imaging and statistical inference to separately map the diffusion and energy landscapes of membrane proteins across the cell surface at ∼ 100 nm resolution (with acquisition of a few minutes). When applying these analytical tools to glycine neurotransmitter receptors (GlyRs) at inhibitory synapses, we find that gephyrin scaffolds act as shallow energy traps (∼3 kBT ) for GlyRs, with a depth modulated by the biochemi- cal properties of the receptor-gephyrin interaction loop. In turn, the inferred maps can be used to simulate the dynamics of proteins in the membrane, from the level of individual receptors to that of the population, and thereby, to model the stochastic fluctuations of physiological parameters (such as the number of receptors at synapses). Overall, our approach provides a powerful and comprehensive framework with which to analyze biochemical interac- tions in living cells and to decipher the multi-scale dynamics of biomolecules in complex cellular environments. Key words: Single Molecule, Neurons, Statistical Physics Mapping energy and diffusion landscapes 2 Introduction Determining the parameters that regulate the mobility of proteins in cells is key for many cellular functions. The motion of proteins depends on a variety of factors, including the local viscosity, their intermittent binding to other proteins, the molecular crowding and the dimensionality of the accessible space (1). Since all these factors are difficult or impossible to reconstitute in vitro using purified constituents, there is a compelling need for analytical tools that bypass in vitro assays and directly access the properties of macro- molecular assemblies and the kinetics of their interactions in their native cellular environment. Thanks to single-molecule (SM) imaging tools, it is now possible to record trajectories of individual proteins in a variety of cellular systems. An important challenge is to extract relevant biochemical and biophysical information from these trajectories. This is commonly done by computing the mean square displacement (MSD) along the trajectories and estimating the effective diffusion coefficient of the molecule. By associating the diffu- sional states to the functional states of the biomolecules, one can identify molecular behaviors (1) and evaluate the transition kinetics between them (2). Although this approach has often proved useful, it is conceptually in- appropriate in many biological situations. Measuring a diffusion coefficient places emphasis on the friction encountered by the protein and assumes that the movement is characterized by a MSD scaling linearly with time. Yet, the primary factor controlling the motion of a protein is often not the fric- tion but, rather, its interactions with molecular or macromolecular partners leading to transient stabilization or transport. In this case, the relevant in- formation is not the diffusion coefficient but the binding energies between the protein of interest and its interacting partners. Furthermore, regulatory processes are often mediated by changes in these binding energies, which should ideally be evaluated with in situ measurements. Methods that go beyond the computation of the MSD generally aim to identify deviations from Brownian movement within single molecule trajec- tories, due for instance to trapping or transport (3 -- 5). However these meth- ods essentially remain ad hoc tools and do not constitute a comprehensive framework to describe the parameters underlying the motion. Furthermore, biological media are often spatially inhomogeneous and this heterogeneity is poorly conveyed by measuring a few, sparse trajectories. A conceptu- ally different approach using Bayesian inference methods has been recently proposed to analyze the motion of molecules (7, 8). It assumes that the membrane environment is characterized by two spatially-varying quantities: Mapping energy and diffusion landscapes 3 (i) the diffusivity D (r) = kBT /γ (r) (where γ (r) is the local viscosity), (ii) the potential energy V (r) that reflects the biochemical interactions of the molecule. In this framework, the protein is a random walker with a motion governed by the Langevin equation (6): dr dt = − D (r) ∇V (r) kBT +p2D (r)ξ (t) (1) (where ξ (t) is a rapidly varying Gaussian noise with zero mean). From a general standpoint, a knowledge of D (r) and V (r), which are protein- specific, can not only reveal how fast the protein moves in the membrane but also identify areas where it can be stabilized (energy traps) or from which it is excluded (energy barriers). However, in the few cases where D (r) and V (r) have been experimentally determined (9, 11), the analysis has been limited to movements confined in local regions (< 1 µm2), falling short of providing a complete description of the heterogeneous diffusivity and energy landscapes in the cell membrane. Here, we introduce a novel and generic approach, combining high density single molecule imaging and computational tools, that enables the mapping of the environment of membrane receptors across the entire cell surface and at ∼ 100 nm resolution. This approach allows the mapping of the membrane over regions of several hundred µm2 in a few minutes of data acquisition. Furthermore, the inferred maps are used to numerically generate massive number of trajectories. These simulated trajectories, whose characteristics match those of the experimental ones, enable a complete analysis of the dynamics in the complex membrane environment by means of various sta- tistical estimators. To illustrate the relevance and benefits of our approach, we applied it to the neuronal membrane, a cellular system in which the spatial organization is critical for the detection and processing of external information. In past years, tracking experiments have underlined the role of membrane dynamics in ensuring rapid exchange of receptors (e.g. glu- tamate, glycine or GABA receptors) between extrasynaptic and synaptic localizations (12). Therefore, the number of receptors at synapses depends on the motion of receptors at the cell surface and their stabilization at synap- tic loci, the latter being regulated by the number of scaffolding molecules and the affinity of the receptor-scaffold interactions (13). A quantitative analysis of the protein mobilities and of their regulatory mechanisms is thus paramount for characterizing and modeling the variability of the synaptic response and the plasticity of the nervous system (involved in higher brain functions such as learning and memory or during pathological processes). Mapping energy and diffusion landscapes 4 Results and discussion Mapping the diffusion and energy landscapes with Bayesian infer- ence. Our approach for the large-scale mapping of D (r) and V (r) builds on Bayesian statistical tools recently developed to analyze the motion of individual particles (7, 8). The principle of the method is as follows (see de- tails in the Supplementary materials). We first acquire high-density single- molecule data (14, 15), with a number of individual translocations on the order of 1000 − 10000 /µm2. Next, the surface of the cell is meshed with sub-domains Si,j (labeled with the index (i, j) along the x and y axis) with a size proportional by a factor δ ∼ 2-3 to the average step size of a translo- cation, such that consecutive positions of the molecules are either in the same or in adjacent domains (Fig. 1a). From the information contained in the massive number of individual translocations, we determine Di,j and ∇Vi,j in each sub-domain (i, j) using Bayesian inference techniques adapted from (7). In brief, we compute the global posterior distribution P of the pa- rameters {Di,j}(i,j) and {∇Vi,j}(i,j) given the observed trajectories {Tk}(k). Since all the sub-domains are independent, P is the product of the posterior distributions inside each of them: P (cid:16){∇Vi,j}(i,j), {Di,j}(i,j){Tk}(k)(cid:17) = Y(i,j) P (cid:16)∇Vi,j, Di,j{Tk}(k)(cid:17) exp − (rk µ+1 −rk µ −Di,j∇Vi,j∆t/kBT)2 4(cid:16)Di,j+ σ2 ∆t(cid:17)∆t µ of the kth trajectory where µ designates the index for which the points rk are in Si,j, σ is the experimental localization accuracy (∼30 nm), ∆t the acquisition time and P (∇V, D) the prior information on the potential and the diffusivities. In the second line of equation (2) we display the prior we commonly used, Jefferey's prior, that is discussed in the supplementary Materials. The estimators (DMAP ) of the local diffusivity and force are the Maximum a Posteriori (MAP) of the posterior distribution P (16, 17). Finally, we solve the inverse problem to determine in each sub- domain the potential field Vi,j associated to the force. The estimation of Vi,j , ∇V MAP i,j i,j ∝ Y(i,j) Yk Yµ:rk µ ∈Si,j   4π(cid:16)Di,j + σ2 ∆t(cid:17) ∆t ! ×  × P (∇V, D) (2)   (Di,j∆t + σ2)2 D2 i,j (3) Mapping energy and diffusion landscapes 5 dr dt = D(r) V (r) k B T + 2D(r) (t) j i Di,j Δ Vi,j D Us interaction transmembrane extracellular loop domain pHluorin a b c wt-TM β S403D-TM β β−-TM d anti-GFP antibody coupled to a red fluorescent dye Figure 1: General scheme of the assay. a) Principle of the Bayesian in- ference method. Left: High-density single-molecule data (red dots) are recorded at the cell surface. Right: In a mesh domain, multiple translo- cations (top) are used to infer the local diffusivity and force (gradient of the potential) that underlie the motion (bottom). b) GlyRs (blue) diffuse in the membrane and are in dynamic equilibrium between synaptic and ex- trasynaptic domains in the neuronal membrane. At synapses, GlyRs are stabilized by their interactions with gephyrin clusters (orange), which can be modeled as trapping potential (with depth US). c) Expression constructs of transmembrane (TM) proteins with an extracellular pHluorin tag and an intracellular interaction loop derived from the GlyR β-subunit. d) Principle of high-density single-molecule uPaint imaging (15). (cid:1)2 +β (δ)X(i,j) (∇Vi,j)2 (i,j) Mapping energy and diffusion landscapes 6 is performed by minimizing ξ ({Vi,j}), defined as: ξ ({Vi,j} (i, j) ∈ {N (i, j)} 6= 0) = X(i,j)(cid:0)∇Vi,j − ∇V M AP i,j i,j (4) with N (i, j) the number of neighboring occupied mesh domains, β (δ) a con- stant (optimized on numerically generated trajectories) depending on δ (see Supplementary Information). Eventually, the set of quantities {DMAP , ∇Vi,j} constitute the diffusivity and potential energy maps. Glycine receptors and their interactions with scaffolding proteins. We applied our inference-based mapping method to investigate the dynam- ics of GlyRs in the neuronal membrane as well as their stabilization at inhibitory synapses (18). This stabilization is achieved through the binding of the receptors to the scaffold protein gephyrin (Fig. 1b) via an intracellular loop (the β-loop) present in the two β-subunits of the pentameric GlyR com- plex. The high affinity component of the β-loop-gephyrin interaction is in the nanomolar range (KD ∼ 20 nM), as determined by isothermal titration calorimetry (19). To characterize the GlyR-gephyrin interaction in living neurons, we used recombinant membrane proteins consisting of a trans- membrane domain (TM), a C-terminal pHluorin tag (a pH-sensitive GFP mutant that is quenched in intracellular acidic vesicular compartments), that were fused N-terminally to the intracellular GlyR β-loop (Fig. 1c). This βWT-TM-pHluorin construct recapitulates the interactions of the en- dogenous GlyR complexes with the gephyrin scaffold proteins, with the im- portant benefit that individual elements of the receptor-scaffold interaction can be manipulated independently (19). It also overcomes the difficulty of defining the sub-unit composition of oligomeric receptors where transfected sub-units compete with endogenous ones. As a control, we used β−-TM- pHluorin, a construct with a mutated β-loop that does not interact with gephyrin. High-density single-molecule imaging of TM proteins. We acquired a high-density of individual trajectories using uPaint, a single-molecule tech- nique in which cells are imaged at an oblique illumination in a buffer con- taining dye-labeled primary antibodies (15). As antibodies (in our case, anti-GFP antibodies coupled to Atto647N dyes) continuously bind to their membrane targets, they can be tracked until they either dissociate or photo- bleach (Fig. 1d and Supplementary Videos 1 and 2). Hence, the entire field of view is constantly replenished with new fluorescent labels and a large number of individual trajectories covering a field of view of ∼ 500-1000 Mapping energy and diffusion landscapes 7 µm2 can be recorded. Experiments were performed on cultured rat hip- pocampal neurons co-transfected with mRFP-tagged gephyrin and with the pHluorin-tagged transmembrane constructs (Fig. 1c). In typical measure- ments, movies were recorded for ∼5-15 minutes with an acquisition time ∆t = 50 ms (Supplementary Videos 1 and 2), yielding up to hundreds of thou- sands of individual translocations per field of view, with an average of 30 points per mesh domain (size ∼100x100 nm2). On this time scale, the cells and synaptic sites remained relatively stable, meaning that the diffusivity and energy landscapes could be considered constant. Diffusion and energy maps of TM proteins. Figures 2a-f show ex- amples of the diffusivity and energy maps for the two constructs βWT-TM- pHluorin and β−-TM-pHluorin. In both cases, the diffusion map exhibits fluctuations at short scale (∼1 µm or less), with local peaks and valleys and a characteristic diffusivity in the range of 0.05-0.2 µm2.s−1 (Fig. 2b and 2e). More striking differences were observed between the energy landscapes. For βWT-TM, the landscape is characterized by the existence of small regions (< 0.5 µm2) corresponding to local energy minima (Fig. 2c). Importantly, gephyrin clusters coincide with energy minima, consistent with the stabiliza- tion of the transmembrane proteins at synaptic sites. Yet, we also observed that some other minima did not colocalize with gephyrin clusters, suggest- ing that βWT-TM-pHluorin might interact with other partners outside of synapses (such as the cytoskeleton or lipid domains). It is possible that these extrasynaptic interactions are still mediated by gephyrin (present in number too small to be detected), since gephyrin is known to associate with GlyRs both inside and outside of synapses (20). In contrast, the energy map for β−-TM (Fig. 2f) shows variations at a longer length-scale, without correlation to gephyrin clusters. To more quantitatively compare the heterogeneous properties of the neu- ronal membrane for βWT-TM and β−-TM, we computed two quantities (av- eraged over 7 cells in each case): (i) the distribution of diffusion coefficients in the maps (Fig. 2g), and (ii) the rugosity of the energy landscape, defined as the standard deviation of the potential in circular region of defined radius averaged over the complete surface of the cell over circular region of the fluc- tuations of the potential averaged over circular regions of increasing radius (Fig. 2h and Supplementary Information). These parameters revealed that the interacting β-loop led to a lower average diffusivity (0.06 µm2.s−1 and 0.13 µm2.s−1 for βWT-TM and β−-TM, respectively) and a larger rugosity of the potential. This is consistent with the notion that moving TM proteins, Mapping energy and diffusion landscapes 8 Figure 2: Diffusion and energy maps in live neurons. a) Fluorescence im- ages of cultured neurons expressing mRFP-gephyrin and βWT-TM-pHluorin. Scale bar: 10 µm. b-c) Diffusion and energy maps. d-f) Equivalent set of images and maps for β−-TM-pHluorin. g) Distribution of diffusion coef- ficients for the membrane constructs βWT-TM (black), βS403D-TM (blue) and β−-TM (red). The vertical bars on the x-axis indicate the mean values of the respective distributions. Insert: distribution in a lin-log scale. h) Rugosity of the membrane potential as a function of the region radius. Mapping energy and diffusion landscapes 9 when bound to intracellular scaffolding proteins, encounter more obstacles that increase the viscosity of their environment. Also they are more likely to interact with membrane or sub-membrane structures that contribute to the energy landscapes. Synaptic scaffolds as crowded energy traps. Given the pronounced differences between the energy landscapes of the βWT-TM and β−-TM con- structs, we examined the behavior of βWT-TM at gephyrin clusters in closer details. An example of the energy profile of βWT-TM proteins at a synaptic cluster (identified by the presence of mRFP-gephyrin fluorescence) is shown in Fig 3a. The profile reinforces the view that clusters of scaffolding proteins act as energy traps for membrane receptors (11, 12, 18). The average trap depth was 3.6 ± 0.4 kBT (± S.E.M., n = 69 clusters), a relatively shal- low potential from which receptors can escape rapidly. Yet, about 15% of clusters had stabilization energies greater than 6 kBT , corresponding to a much more stable anchoring of receptors (Fig. 3b). This reflects the hetero- geneity of the synaptic domains in the neuronal membrane and underlines the need for measurements at the single synapse level. Of note, the binding energies between βWT-TM and gephyrin seem to be significantly lower than the stabilization energy of AMPA receptors at synaptic sites, for which 25 % of the wells had a depth larger than 8 KBT (11). The method used in (11), also based on a combination of high-density single molecule imaging and statistical inference, evaluates the diffusion and drift by computing the maximal likelihood estimation in a mesh square as described in (7). The confining potentials were subsequently evaluated by L2 minimization of a parabolic shaped potential from the force (drift) fields. In (11) the authors do not discuss the role of known biases with confining potentials (see (8 -- 10)) or the effect of the positioning noise, and do not provide information on the posterior distribution of the parameters. It is thus delicate to precisely com- pare their experimental results with ours. Yet, given that the diffusivity of AMPARs at excitatory synapses appears to be higher than the diffusivity of GlyRs at inhibitory synapses (gephyrin clusters), higher confining potentials may be necessary to stabilize the AMPARs. In addition, we noticed that the average diffusivity of βWT-TM (∼0.01 µm2s−1) inside gephyrin clusters was reduced by a factor ∼6 compared to extrasynaptic regions (Fig. 3c), probably due to the combined effect of membrane crowding within synaptic sites and the binding to scaffolding elements. In comparison, the diffusivity of β−-TM proteins inside gephyrin clusters, which we expect to be predom- inantly influenced by molecular crowding (21), was 0.07 µm2.s−1 (Fig. 3c), only a factor ∼2 lower than in extrasynaptic domains. In other words, the Mapping energy and diffusion landscapes 10 a b 1 0.75 f . d c . 0.5 0.25 0 0 2 y g r e n e l a i t n e t o P 17.6 x [μm] 17.4 17.2 33.1 33.5 33.3 y [μm] 4 3 2 1 0 [ k B T ] y g r e n e n a e m 5 4 3 2 1 0 wt-TM β β S403D-TM 8 6 4 10 12 Trapping energy [kBT] 14 c ] 1 - s . 2 m μ [ y t i v i s u ff D i 0.08 0.07 0.06 0.05 0.04 0.03 0.02 0.01 0 wt-TM β β S403D-TM --TM β Figure 3: Analysis of the synaptic gephyrin scaffold. a) Example of a gephyrin cluster (indicated by a box) acting as a local trap in the energy landscape. Scale bar: 5 µm b) Cumulative distribution function (c.d.f) of trapping energy for the constructs βWT-TM (black) and βS403D-TM (blue). Insert: mean values of the distribution. Error bars indicate the S.E.M. c) Mean diffusivity for βWT-TM (black), βS403D-TM (blue) and β−-TM (red). Error bars indicate the S.E.M. synaptic scaffold stabilizes the receptor by simultaneously diminishing the diffusivity of the receptor and by acting as a trapping potential. Modulation of the β-loop gephyrin binding affinity. Since the compu- tation of the energy landscape allows the unambiguous distinction between interacting membrane constructs and those lacking interaction domains, we tested the sensitivity of our approach with the phosphomimetic construct βS403D-TM, a mutated β-loop known to have a lower gephyrin binding affin- ity in vitro (KD ∼ 0.9 µM (19)) (Fig. 1c). As a result, βS403D-TM displayed increased membrane diffusion and reduced synaptic accumulation compared to βWT-TM. The phosphorylation of amino acid residue S403 of the GlyRβ subunit by protein kinase C thus contributes to the regulation of GlyR levels at inhibitory synapses (19). The diffusion and energy landscapes of βS403D- Mapping energy and diffusion landscapes 11 TM (computed over 6 different cells) yielded a diffusivity (average value 0.11 µm2.s−1) and an energetic rugosity precisely intermediate between those of the wild-type and of the binding-deficient constructs (Fig. 2g-h). Com- pared to βWT-TM, the average trap depth of βS403D-TM at synaptic sites was reduced to 2.4 ± 0.4 kBT (n = 58 clusters), with less than 5% of the traps above 6 kBT (Fig. 3b). Inside clusters, the average diffusivity (0.015 µm2.s−1) was slightly higher than for the wild-type (Fig. 3c). Importantly, the binding energy reported here corresponds to TM pro- teins moving in a two-dimensional membrane and interacting with macro- molecular gephyrin scaffolds that are believed to be two-dimensional as well (22, 23). This is in contrast with measurement of the equilibrium con- stant KD by isothermal calorimetry, which reports on the individual inter- action between the β-loop and the scaffolding protein in an isotropic, three- dimensional measurement of the β-loop-scaffold interaction. Obtaining the stabilization energy thus constitutes a first and important step to bridge the gap between in vitro and in situ biochemical measurements. When further complemented with data on the ultra-structure and stoichiometry of synap- tic scaffolds (that are now accessible with single molecule imaging techniques (23, 24)), we expect our approach to enable a true determination of the two- dimensional affinity of the membrane proteins for the synaptic scaffolds (25). Connecting the landscapes and the global mobility of proteins. An important question for the dynamics of proteins is how the variability of their diffusion and energy landscapes at short scale (∼ 100 nm) affects their long-distance mobility and, thereby, the kinetics of many intermolecular re- actions. Reaching a multi-scale description of the motion in the membrane has long been a challenge in single-molecule experiments. High-density sam- pling is usually achieved with poorly stable probes, yielding numerous but short trajectories (14, 15). In contrast, long trajectories obtained with more stable markers such as quantum dots (26) only provide a sparse sampling of the cell surface. Furthermore, the nature of the motion, such as sub- diffusion, may prevent efficient space sampling with single long trajectories. Here, we adopted a different strategy and used the inferred maps as phe- nomenological templates to simulate the motion of proteins. Practically, we used the Gillespie scheme (27) to generate individual trajectories lasting up to 500s (see Methods and Supplementary Information). From a large number of simulated trajectories, we could compute ensemble- averaged quantities. We first evaluated the propagator Π(d, t), namely the probability density function of moving a distance d in a time t, which is the fundamental estimator characterizing the random motion in a complex Mapping energy and diffusion landscapes 12 Figure 4: Analysis of simulated trajectories in the inferred maps. Unless otherwise mentioned, the results correspond to the constructs βWT-TM (black), βS403D-TM (blue) and β−-TM (red). a) Ensemble-averaged propa- gator Π(d, t), defined as the probability density function (pdf) to move by a given distance in t=10 s. b) Propagator Π(d, t) for the construct βWT-TM, computed at different times t. The plain lines are adjustments with the gaus- sian curves exp(−d2/2χ2(t))/2πχ2(t). c) Curves χ(t). Inset: Propagators for the construct βWT-TM as a function of the rescaled variable ρ = d/χ(t). d) Mean-squared displacement (MSD) as a function of time. The straight lines indicate the sub-diffusive behavior at short time-scales. e) Time-course of the number of receptors at a single synapse. Inset: distribution of the minimum (in red) and maximum (in blue) number of receptors computed over traces of 300 s for all the gephyrin clusters. f) Autocorrelation func- tions (in grey) for the time traces of number of receptors at gephyrin clusters (computed over 300 s). The red line indicates the average autocorrelation function. Mapping energy and diffusion landscapes 13 environment (28). Although the difference in the average trapping energy at gephyrin clusters was only ∼1 kBT between βWT-TM and βS403D-TM, it led to significant changes in the mobility, reducing the probability of moving over long distances with increasing strength of the β-loop-gephyrin interac- tion (Fig. 4a). To more carefully examine the nature of the movement of βWT-TM, we plotted Π(d, t) at different times t. The curves could be approx- imated by gaussian curves exp(−d2/2χ2(t)) with χ(t) ∝ tα and α = 0.33, less than 0.5 the value expected for a standard Brownian motion (Fig. 4b-c). In fact, these results are consistent with a subdiffusive movement resulting from a fractional brownian motion due to heteregoneities in the diffusion and energy landscapes (28). Similar results were obtained for βS403D-TM and β−-TM, with α increasing to 0.39 and 0.41, respectively (Fig. 4c). The subdiffusive nature of the motion could be further illustrated by computing the ensemble-averaged MSD for the three transmembrane constructs (Fig. 4d). On the time scales 0.05-5 s, all the MSDs increased sublinearly, with an anomalous exponent α equal to 0.75, 0.82 and 0.89 for βWT-TM, βS403D-TM and β−-TM, respectively. The MSD anomalous exponents are slightly larger than 2α, likely due to boundary effects associated to the size and geometry of the neurons. Finally, we examined the implications of the local properties of the mobil- ity of individual GlyRs on their global distribution in the membrane and on the receptor occupancy at synapses. To do so, we simulated the membrane dynamics of a population of receptors, using surface densities derived from prior experimental reports (Supplementary information). We computed in particular the time course of the number of receptors at individual synaptic clusters, which we expect to fluctuate due to the exit and entry fluxes of receptors (Fig. 4e and Supplementary Information). The exit kinetics at a given synapse is determined by the shape and amplitude of the trapping potential combined with the reduced diffusivity in the cluster. In contrast, the entry kinetics depends on the motion of all the receptors over the entire cell surface and need to be computed using the diffusion and energy maps. The number of receptors varied significantly over times, as illustrated by the distribution of their minimal and maximal numbers at individual synapses (Fig. 4e). Furthermore, the time scale of these fluctuations, analyzed by computing the autocorrelation function, is comprised between ∼ 1 s and a few tens of seconds, showing a large heterogeneity among gephyrin clusters (Fig. 4f). These observations may account for the dynamic range of receptor numbers at synapses and for the variability of synaptic transmission (31). The receptor fluctuations, that are equivalent to a noise, may also favor the transition from one steady state to another during synaptic plasticity (32). Mapping energy and diffusion landscapes 14 Conclusion The motion of proteins in the plasma membrane is influenced by both a viscous landscape, γ (r), and an interaction potential, V (r). We have in- troduced a method to map the interaction energy and diffusion landscapes in the cellular membrane with ∼ 100 nm resolution over surfaces of several hundred µm2. The possibility of simulating trajectories in the inferred maps offers many possibilities to address the multiscale dynamics of membrane proteins. In particular, it bridges the gap between the information obtained from numerous, dense but short - trajectories acquired using uPaint or sptPALM techniques, and that from the much longer, but usually sparse, trajectories extracted through the tracking of proteins labeled with photo- stable fluorophores (Qdots, nanoparticles). These trajectories can be used to accurately evaluate various statistical estimators, thus enabling the analysis of the dynamics of biomolecules in complex media. We anticipate that our method will be instrumental to identify the factors governing the mobility of specific molecules (such as friction, molecular interactions and geometry of the cell) and thereby to model and analyze reaction-diffusion processes in biological media. As illustrated in the case of GlyR-gephyrin binding, it also paves the way to in situ biochemical measurements, which is key for a quantitative analysis of the regulation of molecular interactions in a cellular environment. Our approach should also be helpful to describe the molec- ular noise that results from variability of protein concentrations across the cell surface and may play an important role in information processing at the single cell level (31). Beyond the case of receptor-scaffold interactions, our analytical tools can be applied to other biological questions, such as the stability of macromolecular assemblies in the cytoplasm or the nucleus, or to the sequence-dependent movement of proteins along DNA (33). Materials and methods Antibody coupling. Rat anti-GFP monoclonal antibody (Roche) was la- beled with Atto-647 dye using standard conjugation methods. In brief, 40 µL of antibodies at 0.4 mg/ml in PBS were mixed with 4 µL of 1 M sodium bicarbonate buffer at pH 8.5. This solution was incubated with 10-fold molar excess of Atto-647-NHS-ester (Sigma) diluted at 1 mg/mL in anhy- drous DMSO. After 1 h of incubation at room temperature, the solution was filtered with a Microspin G50 column (GE Healthcare) to remove unconju- gated dye. The overall coupling efficiency of the dye, estimated by UV-Vis Mapping energy and diffusion landscapes 15 absorption, was about 12%. The labeled antibodies were washed with PBS and concentrated using three rounds of centrifugation with a vivaspin500 10 kDa cut-off PES membrane filter (GE Healthcare). The concentrated antibody solution was stored at 4 C and used for up to one week. Cell culture and plasmid transfection. Hippocampal neurons from Sprague Dawley rats at embryonic day 18 were cultured at a density of 6 x 104 cells/cm2 on 18 mm coverslips precoated with 80 mg/ml poly-D,L- ornithine (Sigma) and 5% fetal calf serum (Invitrogen) as described pre- viously (19). Cultures were maintained in serum-free Neurobasal medium supplemented with 1x B27 and 2 mM glutamine (Invitrogen). Cells were transfected after 6 to 8 days in vitro using Lipofectamine 2000 (Invitrogen), and imaged 1 to 2 days after transfection. All coverslips were co-transfected with mRFP-tagged gephyrin and pHluorin-tagged TM constructs, using 0.4 µg of each plasmid per coverslip. The expression constructs βWT-TM- pHluorin, βS403D-TM-pHluorin and β−-TM-pHluorin are all described in In brief, βS403D corresponds to the mutation of serine S403 of the (19). GlyRβ subunit that mimics the phosphorylation of the residue by protein kinase C. β−-TM corresponds to the double mutation F398A and I400A of the wild-type GlyR β-loop that abolishes binding to gephyrin. Cell labeling. Prior to imaging, we prepared a stock solution of diluted an- tibodies using casein (Vector laboratories) as a blocking reagent. We added 2 µl of Atto-647 conjugated anti-GFP antibodies and 10 µl of 10 mg/ml casein to 40 µl of PBS, resulting in an antibody solution of 0.1-0.2 µM. We also prepared a stock of Tetraspeck fluorescent microbeads (Invitrogen) by mixing 1 µl of 0.1 µM microbeads with 400 µl of imaging solution. These multi-color fluorescent beads were used as a reference to align the different imaging channels and to correct for x/y drifts of the stage and the coverslip. The coverslip was mounted in an imaging chamber and incubated with 20 µl of warmed microbead solution for 10 seconds. After rinsing, the chamber was filled with 600 µl of warmed imaging solution (MEMair: phenol red- free MEM, glucose 33 mM, HEPES 20 mM, glutamine 2 mM, Na-pyruvate 1 mM, and B27 1x) and placed on the microscope. To avoid saturating the cell membrane with fluorescent antibodies, we first selected a transfected neuron and added the fluorescent antibodies at a final concentration of 0.3- 0.6 nM directly before the start of the acquisition. Imaging. Measurements were performed on an inverted epi-fluorescence microscope (Olympus IX70) equipped with a 100x 1.45NA oil objective and Mapping energy and diffusion landscapes 16 a back-illuminated electron-multiplying CCD camera (Quantem, Roper). We imaged the neurons at 37 C in MEMair recording medium using a heated stage. For each neuron, we first recorded images of the pHluorin signal of the TM constructs and of mRFP-gephyrin fluorescence, using a UV lamp (Uvico, Rapp Optoelectronic) and standard sets of filters for GFP (excitation 475AF40, dichroic 515DRLP and emission 535AF45) and RFP (excitation 580DF30, dichroic 600DRLP and emission 620DF30). Next we acquired a uPaint movie of the transmembrane proteins labeled with Atto- 647-coupled anti-GFP antibodies (20 000 images at 20 frames/s). Atto-647 dyes were excited with a 640 nm laser and their fluorescence was collected through using a 650DRLP dichroic and a 690DF40 emission filter. The laser was tightly focused on the back focal plane of the objective. The angle of incidence of the beam on the coverslip, controlled by laterally moving the focused spot, was just under the limit of total internal reflection, such that the laser beam in the sample was almost parallel to the glass surface. This angle was slightly adjusted in each experiment to maximize the signal/noise ratio of the single fluorescent spots diffusing in the membrane. Data analysis. Tracking analysis of the movies was carried out using an adapted version of the multiple target tracking algorithm (34). In brief, fluorescence spots corresponding to the point-spread function of single emit- ting fluorophores were fitted with a two-dimensional Gaussian. The centre of the fit yielded the position of single molecules with localization accuracy ∼ 30 nm. Trajectories were then computed from individual detections with a nearest-neighbor algorithm. Simulations in the landscapes. The maps of the diffusion and energy landscapes, D (r) and V (r), can be used to simulate the behavior of the molecules at different time and space scales. In each mesh sub-domain (i, j) a diffusivity Di,j is associated with a potential energy value Vi,j. The dynamics of the molecules are described by the Fokker-Planck equation: ∂P (r, tr0, t) ∂t = −∇.(cid:18)− ∇V (r) P (r, tr0, t) γ (r) − ∇ (D (r) P (r, tr0, t))(cid:19)(5) where P (r, tr0, t) is the conditional transition probability from (r0, t0) to (r, t). Fokker-Planck equations can always be approximated by Master equa- tions: dP(i,j) (t) dt = X(i′,j ′)∈N (i,j) W(i,j),(i′,j ′)P(i′,j ′) − X(i′,j ′)∈N (i,j) W(i′,j ′),(i,j)P(i,j) (6) Mapping energy and diffusion landscapes with in our case W(i,j),(i′,j ′) = D (i′, j′) ∆x2 exp − ∆xF x (i,j),(i′,j ′) 2γ (i′, j′) D(i′,j ′)! 17 (7) if the transition is in the x direction and a similar formula in the y direction and with W(i,j),(i′,j ′) the transition rate from the (i′, j′) site to the (i, j), ∆x ((∆y)) the mesh size in the x (y) direction, and F x (i,j),(i′,j ′) the poten- tial gradient acting on the random walker in the x direction when moving from (i′, j′) to (i, j). The motion of the molecule following equation 6 was simulated using the Gillespie scheme (27). When the molecule was at the site (i, j), the transitions rates, rewritten aν to match Gillespie formalism,ν taking values from 1 to 4, were evaluated on all neighboring sites. We define log(cid:16) 1 ν=0 aν ≤ r2a0 ≤ Pk is extracted from an exponential probability density function of rate a0, so that τ = 1 a0 a0 = Pν aν. The time, τ , to move from the site (i, j) to a neighboring site r1(cid:17) with r1 a random number in [0, 1]. The destination site, k, is chosen to satisfy : Pk−1 ν=0 aν with r2 a random number in [0, 1]. Limits of the neuronal cells and unvisited sites are defined as inaccessible sites. Note that the trajectory generation process leads to trajectories with non-constant time steps. In order to evaluate the different estimators, trajectories were regularized to obtain the molecule position at regular time lags by imposing that as long as each τ was not reached the molecule did not move. Acknowledgments We are grateful to Paul de Koninck for his support and discussion. We also thank Diego Krapf for his critical reading of the paper and multiple sugges- tions. This work was funded by CNRS, Inserm, C'Nano Ile de France, the program "Prise de Risque" from CNRS, Agence Nationale pour la Recherche PiriBio, the grant Synaptune from the Agence Nationale pour la Recherche, the program ANR-10-IDEX-0001-02 PSL and the state program Investisse- ments d'avenir managed by Agence Nationale de la Recherche (Grant ANR- 10-BINF-05 Pherotaxis). References 1. Saxton, M. J. & K. Jacobson. (1997) Single-particle tracking: applica- tions to membrane dynamics. Annu Rev Biophys Biomol Struct 26, 373- 399. Mapping energy and diffusion landscapes 18 2. Persson, F. et al. (2013). Extracting intracellular diffusive states and transition rates from single-molecule tracking data. Nat Methods 10, 265- 9. 3. Simson, R., E.D. Sheets & K. Jacobson. Detection of temporary lateral confinement of membrane proteins using single-particle tracking analysis. Biophys. J. 69, 989-993 (1995). 4. Huet, S. et al. Analysis of transient behavior in complex trajectories: application to secretory vesicle dynamics. Biophys. J. 91, 3542-3559 (2006). 5. Bouzigues, C. & M. Dahan. Transient directed motions of GABA(A) receptors in growth cones detected by a speed correlation index. Biophys. J. 92, 654-660 (2007). 6. Risken, H. (1997) The Fokker-Planck Equation: Methods of Solutions and Applications. (Springer) 7. Masson, J. B. et al. . (2009) Inferring maps of forces inside cell membrane microdomains. Phys Rev Lett 102, 048103 . 8. Turkcan, S., A. Alexandrou & J.-B Masson. (2012) Bayesian inference scheme to extract diffusivity and potential fields from confined single- molecule trajectories. Biophys J 102, 2288-2298. 9. Turkcan, S. et al. (2013) Probing membrane protein interactions with their lipid raft environment using single-molecule tracking and Bayesian inference analysis. Plos One 8, e53073. 10. Turkcan, S., A. Aelxadrou & J.-B Masson (2010) Quantifying biomolecule diffusivity using an optimal Bayesian method. Biophys J. 98,596-605. 11. Hoze, N. et al. (2012) Heterogeneity of AMPA receptor trafficking and molecular interactions revealed by superresolution analysis of live cell imaging. Proc Natl Acad Sci U S A 109, 17052-7. 12. Triller, A. & D. Choquet (2008) New concepts in synaptic biology de- rived from single-molecule imaging. Neuron 59, 359-374 . 13. Renner, M., L. Cognet, B. Lounis, A . Triller & D. Choquet. (2009) The excitatory postsynaptic density is a size exclusion diffusion environment. Neuropharmacology 56, 30-36. Mapping energy and diffusion landscapes 19 14. Manley, S. et al. (2008) High-density mapping of single-molecule trajec- tories with photoactivated localization microscopy. Nat Methods 5, 155- 157. 15. Giannone, G. et al. (2010) Dynamic superresolution imaging of endoge- nous proteins on living cells at ultra-high density. Biophys J 99, 1303-1310. 16. Mac Kay, D. J. C (2003) Information Theory, Inference, and Learning Algorithms. (Cambridge University Press) 17. von Toussaint, U. (2011) Bayesian inference in physics. Reviews of Mod- ern Physics 83, 943-999. 18. Dahan, M. et al (2003) Diffusion dynamics of glycine receptors revealed by single-quantum dot tracking. Science 302, 442-445. 19. Specht, C. G. et al. (2011) Regulation of glycine receptor diffusion prop- erties and gephyrin interactions by protein kinase C. EMBO J 30, 3842- 3853. 20. Ehrensperger, M. V.,C. Hanus, C. Vannier, A. Triller & M. Dahan (2007) Multiple association states between glycine receptors and gephyrin identi- fied by SPT analysis. Biophys J 92, 3706-3718. 21. Renner, M., D. Choquet & A. Triller. (2009) Control of the postsynaptic membrane viscosity. J Neurosci 29, 2926-2937. 22. Fritschy, J. M., R. J. Harvey & G. Schwarz. (2008) Gephyrin where do we stand, where do we go?. Trends Neurosci 31, 257-264c. 23. Specht, C.G. et al. (2013) Quantitative nanoscopy of inhibitory synapses: counting gephyrin molecules and receptor binding sites. Neu- ron, 79(2), 308-321 24. Lord, S. J., H. L. Lee & W. E. Moerner. (2010) Single-molecule spec- troscopy and imaging of biomolecules in living cells. Anal Chem 82, 2192- 2203. 25. Wu, Y., J. Vendome, L. Shapiro, A. Ben-Shaul & B. Honig. (2011) Transforming binding affinities from three dimensions to two with appli- cation to cadherin clustering. Nature 475, 510-513. 26. Pinaud, F., S. Clarke, A. Sittner, & M. Dahan (2010) Probing cellular events, one quantum dot at a time. Nat Methods 7, 275-85. Mapping energy and diffusion landscapes 20 27. Gillespie, D. (1977) Exact Stochastic Simulation of Coupled Chemical Reactions. Journal of Physical Chemistry 81, 2340-2361. 28. Metzler R. & J. Klafter (2004). The restaurant at the end of the random walk: recent developments in fractional dynamics descriptions of anoma- lous dynamical processes, J. Phys. A 37, R161. 29. Benichou, O., C. Chevalier, J. Klafter, B. Meyer & R. Voituriez (2010) Geometry-controlled kinetics. Nat Chem 2, 472-477. 30. Condamin, S., O. Benichou, V. Tejedor, R. Voituriez & J. Klafter. (2007) First-passage times in complex scale-invariant media. Nature 450, 77-80. 31. Ribrault, C., K. Sekimoto & A. Triller. (2011) From the stochasticity of molecular processes to the variability of synaptic transmission. Nat Rev Neurosci 12, 375-387. 32. Sekimoto, K. & A. Triller (2009) Compatibility between itinerant synap- tic receptors and stable postsynaptic structure. Phys Rev E Stat Nonlin Soft Matter Phys 79, 031905. 33. Leith J.S. et al. (2012) Sequence-dependent sliding kinetics of p53. Proc Natl Acad Sci U S A. 109, 16552-7. 34. Serge, A., N. Bertaux, H. Rigneault & D. Marguet (2008) Dynamic multiple-target tracing to probe spatiotemporal cartography of cell mem- branes. Nat Methods 5, 687-694.
1507.04581	1	1507	2015-07-16T14:03:08	Feeding ducks, bacterial chemotaxis, and the Gini index	[ "physics.bio-ph", "cond-mat.soft", "q-bio.CB", "q-bio.PE" ]	Classic experiments on the distribution of ducks around separated food sources found consistency with the `ideal free' distribution in which the local population is proportional to the local supply rate. Motivated by this experiment and others, we examine the analogous problem in the microbial world: the distribution of chemotactic bacteria around multiple nearby food sources. In contrast to the optimization of uptake rate that may hold at the level of a single cell in a spatially varying nutrient field, nutrient consumption by a population of chemotactic cells will modify the nutrient field, and the uptake rate will generally vary throughout the population. Through a simple model we study the distribution of resource uptake in the presence of chemotaxis, consumption, and diffusion of both bacteria and nutrients. Borrowing from the field of theoretical economics, we explore how the Gini index can be used as a means to quantify the inequalities of uptake. The redistributive effect of chemotaxis can lead to a phenomenon we term `chemotactic levelling', and the influence of these results on population fitness are briefly considered.	physics.bio-ph	physics	Feeding ducks, bacterial chemotaxis, and the Gini index Fran¸cois J. Peaudecerf and Raymond E. Goldstein Department of Applied Mathematics and Theoretical Physics, Centre for Mathematical Sciences, University of Cambridge, Wilberforce Road, Cambridge CB3 0WA, United Kingdom (Dated: August 3, 2021) Classic experiments on the distribution of ducks around separated food sources found consistency with the 'ideal free' distribution in which the local population is proportional to the local supply rate. Motivated by this experiment and others, we examine the analogous problem in the microbial world: the distribution of chemotactic bacteria around multiple nearby food sources. In contrast to the optimization of uptake rate that may hold at the level of a single cell in a spatially varying nutrient field, nutrient consumption by a population of chemotactic cells will modify the nutrient field, and the uptake rate will generally vary throughout the population. Through a simple model we study the distribution of resource uptake in the presence of chemotaxis, consumption, and diffusion of both bacteria and nutrients. Borrowing from the field of theoretical economics, we explore how the Gini index can be used as a means to quantify the inequalities of uptake. The redistributive effect of chemotaxis can lead to a phenomenon we term 'chemotactic levelling', and the influence of these results on population fitness are briefly considered. PACS numbers: 87.18.Gh, 87.17.Jj, 87.23.-n, 89.65.-s INTRODUCTION In one of the more amusing, yet influential experiments on animal behavior, Harper [1] studied the distribution of mallards around two separated sources of standard- ized pieces of bread. After an induction period on the order of a minute, the average number of ducks clus- tered tightly around each station stabilized. The distri- bution he observed was simple: the number of ducks at each source was proportional to the flux of bread there (pieces/minute). This constituted the first experimental observation of the so-called ideal free distribution previ- ously introduced in theoretical ecology. Using the termi- nology of Fretwell and Lucas [2], 'ideal' means that ducks can identify the source where their uptake is maximized, and 'free' implies unfettered ability to access the source of choice. This distribution, resulting from individual ratio- nal behaviors, achieves a population-wide uniformization of the probability of uptake, and can be understood as a Nash equilibrium [3]. These works impacted not only or- nithology, but ecology [4, 5], evolutionary biology [6] and the study of human behavior [7 -- 9], all areas involving resource acquisition in a heterogeneous environment. Here we take motivation from Harper's experiment, and others discussed below, to examine resource acqui- sition in a heterogeneous microbial world [10] where swimming microorganisms respond to nutrient sources through concentration fields determined by molecular dif- fusion and microbial uptake. For the specific case of per- itrichously flagellated bacteria such as E. coli and B. sub- tilis, cells move in a run-and-tumble random walk biased by concentration gradients, resulting in drift of the popu- lation up these gradients [11]. Because chemotaxis [12] is quite different from the visually-based searching of higher animals, and because of the diffusive behavior of nutri- ents and the cell populations the microbial problem is distinct in character. This feature motivates the present investigation of the consequences of a collection of in- dividual chemotactic responses on the population-scale distribution of resources. While chemotaxis is generally thought to optimize uptake at the single-cell level [13], even the mere presence of translational diffusion in a pop- ulation would lead to a distribution of uptake rates. And the interplay of chemotaxis and consumption will mod- ify the distribution of resources and cells, with further potential impact on the uptake rate distribution. In this paper we focus on three key questions in this area: What is the distribution of bacteria around spatially distinct nutrient sources and their associated impact on the re- source field? What is the distribution of resource uptake rates within that population? What are the consequences of such distributions for cellular fitness? A historically important experiment on spatially- varying resources is Engelmann's 1883 determination of the action spectrum of photosynthesis [14]. Having dis- covered bacteria that are attracted to the oxygen pro- duced by photosynthesis [15], he imaged the solar spec- trum onto a linear algal cell in an air-tight chamber con- taining such bacteria. They clustered around the alga in proportion to the local oxygen production, revealing with greater precision than the available techniques of the time the peaks of photosynthetic activity for blue and red wavelengths. How reliably the local bacterial accumulation reflects the oxygen production rate, in the face of both bacterial and oxygen diffusion, remains an open question in the spirit of the present investigation. Moreover, this work demonstrates how micro-domains re- leasing a limited quantity of attractive nutrients in a con- tinuous fashion can arise in the microbial world. Understanding bacterial organization and uptake around algal resources also finds an important biologi- cal context in the case of bacterial-algal symbiosis. The 5 1 0 2 l u J 6 1 ] h p - o i b . s c i s y h p [ 1 v 1 8 5 4 0 . 7 0 5 1 : v i X r a recent discovery [16] that many algae dependent on vita- min B12 obtain it through symbiotic relationships with bacteria raises questions about spatio-temporal aspects of symbiosis: how the two species find each other and arrange themselves to achieve the symbiosis. These are examples of a more general problem of mi- croorganisms responding to the 'patchy' nature of nu- trients in ecosystems [17], such as the prosaically named 'marine snow' [18]. We emphasize the fundamental differ- ence between live microbial sources, such as Engelmann's alga, and inert sources, such as lysis events; the former continuously release nutrients at low rates. They are thus stable in time but can be significantly impacted by bacte- rial uptake. These characteristics makes them the natu- ral microbial equivalent to the limited continuous sources considered by Harper in his animal experiments. To make concrete the interplay between production, consumption, diffusion, and chemotaxis we consider a generalized Keller -- Segel (KS) model [19]. While the KS model is a well-established model that has found frequent application in the study of spatially extended microor- ganism populations [20], the biological setting of local- ized, low-intensity sources with a steady release of nutri- ents is little-studied, and the overarching issue of resource uptake distribution is essentially unexplored. In what fol- lows the steady-state distributions arising from multiple localized sources are analyzed to understand the conse- quences of chemotaxis on the distribution of uptake in the bacterial population, the total uptake being fixed. Bor- rowing from theoretical economics, we next propose that the Gini index [22], a number originally used to char- acterize wealth inequality, can be used to quantify the individual uptake distribution. Varying the model pa- rameters and the dimensionality of space we show that chemotaxis can switch from redistributing the resource to generating greater inequalities. Finally, we explore the potential biological consequences of uptake redistribution through an example of growth at low nutrient levels. THE MODEL Consider bacteria with mean concentration b0 and lo- cal concentration b(r, t) in a d-dimensional volume Ld. Within the volume are nutrient sources with fluxes {φi} of typical value φ0 leading to a concentration field c(r, t). b and c obey KS equations [19], 2 FIG. 1. (color online). Numerical results in one dimension. (a) Sketch of the setup. (b) Example of a steady-state dis- tribution of bacteria in the rescaled KS model (4) & (5). (c) Corresponding distribution of nutrient at steady state. In (b) & (c) the parameters values are λ = 3 for the domain size parameter and s = 1/8 for the relative strength of the left source. The importance of chemotaxis against bacterial diffusion increases with the chemotactic parameter α. is expected to behave at low c as f (c) ∼ kc, with satu- ration at high c: f (c) ∼ kmax. When the nutrient is es- sential for life, such as oxygen for obligate aerobes, f (c) may vanish below some c∗ [24, 25]. As in Harper's study and Engelmann's experiment, the interesting regime has the resource limiting, with b0 and L such that the to- tal nutrient flux can be consumed and is thus below the maximum value kmaxb0Ld. Otherwise, steady state can- not be attained and c increases indefinitely. For small c, we identify the length scale (cid:96)k = (Dc/kb0)1/2 for concentration gradients due to uptake. For the a run-and-tumble chemotaxis mechanism to op- erate and the continuum model to be relevant, (cid:96)k should be large compared to the run length (cid:96)run = vτ , where v is a typical swimming speed and τ the time between tumbles. That is, we require the Knudsen-like number lrun/lk (cid:28) 1. For E. coli, with v ∼ 20 µm/s and τ ∼ 1 s, (cid:96)run ∼ 20 µm, but it can be considerably longer for other bacteria [28]. Still at low c, with χ(c) = χ0 [21], interesting behavior occurs when the chemotactic flux ∼ χ0b0φ0/Dc dominates the diffusive flux ∼ Dbb0/(cid:96)k, with φ0/Dc a typical concentration gradient. Thus, the P´eclet-like number α = χ0φ0 DbDc(cid:96)k (3) exceeds unity. = Dc∇2c − bf (c) , = Db∇2b − ∇ · [χ(c)b∇c] , ∂c ∂t ∂b ∂t A one-dimensional version of Harper's experiment (Fig. 1(a)) has fluxes φl and φr at the left and right domain boundaries, and Φ = φl + φr. Nondimensional- izing as above we obtain in the low-c regime (1) (2) with Dc and Db the diffusion coefficients of nutrients and bacteria respectively, and χ(c) the chemotactic response coefficient. The nutrient uptake rate f (c) per bacterium ∂c ∂t δ−1 ∂b ∂t = ∇2c − bc , = ∇2b − α∇ · [b∇c] , (4) (5) where δ = Db/Dc, with the following nondimensional boundary conditions: n · ∇c0 = − s n · ∇cλ = 1 − s n · (∇b − αb∇c)0 = 0 n · (∇b − αb∇c)λ = 0 , (6) (7) where n is the outward unit normal to the domain. Two new parameters drive the evolution of the sys- tem: λ = L/(cid:96)k, the domain size relative to the screening length, and the relative strength s = φl/Φ of the left source. Before investigating the steady-state solutions to (4) and (5), we point out that by construction the total uptake U over the population is equal at steady-state to the total nutrient flux into the chamber, (cid:90) λ (cid:90) λ U = dx bc = 0 0 dx∇2C = n · ∇cλ 0 = 1 , (8) independent of the choice of parameters and strength of chemotaxis. Thus overall uptake optimization is not part of the present study: our interest resides instead in how the individual behaviors result in a distribution of finite resource among population members, just as in Harper's experiment. One of the natural questions to ask is whether there is a variational structure to the KS equations (4) and (5). The only related result of which we are aware concerns the case when the nutrient consumption rate f (c) takes the aforementioned high-c form of a constant. After suit- able rescaling that dynamics is ∂c ∂t δ−1 ∂b ∂t = ∇2c + νb , = ∇2b − α∇ · [b∇c] , (10) where ν = −1. As shown recently [26], a dynamics of a closely related form is variational. If we introduce the energy functional (9) E[b, c] = dx b log b − bc + ∇c2 (11) (cid:90) (cid:26) 1 α (cid:27) 1 2 (cid:21) (cid:20) b∇ δE δb then the variational relations = − δE δc ∂c ∂t , δ−1 ∂b ∂t = α∇ · 3 here, with f (c) ∼ c also appears not to possess a vari- ational structure and thus it is not possible to conclude that the steady-state solutions are in any way minimizers or maximizers of some energy-like functional. It follows that the distribution of uptake rates is a nontrivial fea- ture of the underlying diffusion-consumption-chemotaxis dynamics. Figures 1(b) and 1(c) show steady state distributions of b and c for α = 3, λ = 3, δ = 0.6 and s = 1/8, which correspond to physical values of the dimensional param- eters (Dc = 5×10−6 cm2 s−1, Db = 3×10−6 cm2 s−1, k = 2 × 10−7 cell −1 s−1 cm3, χ0 = 2.7 × 10−3 cm2 s−1 mM−1, b0 = 106 cells cm−3, φ0 = 3.4 × 10−6 mM cm s−1 and L = 150 µm [29, 30]), along with the non-chemotactic case and the case α = 7. Bacteria accumulate on both sides, with more closer to the stronger source, as one might expect. Intriguingly, this leads to what we term chemotactic levelling of the nutrient: a more uniform concentration field than without chemotaxis. In par- ticular, we notice that the maximum uptake rate of a bacterium in this population, obtained closest to the strongest source, is decreased by chemotaxis. Diffusion of b and c precludes the ideal free distribution. QUANTIFYING INEQUALITIES WITH THE GINI INDEX Whereas studies on biological consequences of chemo- taxis usually measure the increased uptake over the whole or part of the population with respect to the non- chemotactic case (e.g. [31, 32]), as emphasized above, the present study is focused on the case for which the mean uptake rate is independent of the chemotactic behavior. As our interest then resides in how equally resources are spread among the population, comparison of the chemo- tactic results with the non-chemotactic reference distri- butions thus requires a measure of the proximity with the ideal free distribution. Among the many possible mea- sures of inequality [33], we consider here the Gini index G ∈ [0, 1] [22], which for a distribution P (w) of wealth w in a population can be expressed as (cid:82)(cid:82) dudvP (u)P (v)u − v 2(cid:82) du uP (u) . (12) G = . (13) yield (9) and (10), but with ν = +1. This case corre- sponds to the well-studied situation in which the bac- teria are sources for the chemoattractant, rather than sinks. Because the same term (−bc) in E yields both the production/consumption term in (9) and the chemotac- tic term in (10), the case ν = −1 appears not to have any variational structure of this type [27]. But since the con- sumption rate per bacterium is constant in this regime, the distribution of resource acquisition rates is trivial and a variational structure would provide no new infor- mation. More importantly, the case under consideration The ideal free distribution, in which every individual has the same wealth u0, is P = δ(u − u0) and thus G = 0, while larger values hold for more unequal distributions. G can be used with any notion of 'wealth' [34], such as biodiversity [35]. We should empasize that in using the Gini index to quantify uptake inequalities in the present study we do not imply any preferred status to G as a metric for resource acquisition distributions. There are many that could be explored, the Gini index presenting the advantage of being easily translated in the context of the continuous distribution and thus enabling analytical 4 rameter: the ratio of diffusion coefficients δ impacts tran- sient dynamics but does not modify steady state solu- tions. From numerical solutions in the phase space de- limited by s ∈ [0, 0.5] (from one source on the right to equal sources), α ∈ [0, 15] (strength of chemotaxis) and λ ∈ [0.5, 15] (domain size), we obtain first the intuitive result that for given chemotactic and domain size pa- rameters, the more equal the sources are, the more equal the uptake is among the population, and the lower is G. More balanced sources indeed create smaller nutri- ent gradients, thus a lesser range of uptakes and weaker chemotaxis. We also find that G increases with the size parameter λ (Fig. 2 (a) in the case s = 0), for larger λ corresponds to stronger variations of the concentration field and higher variations of uptake. The question of whether chemotactic levelling of the nutrient concentration field can make the distribution of individual uptake more ideal is addressed by varying α. Its impact is subtle: chemotaxis levels nutrients across the domain, but accumulation of cells near sources im- proves uptake of some to the detriment of others. We find that G actually decreases with α (Fig. 2(a) for a single source), which reveals chemotactic levelling of the uptake rate among the population. In Fig. 1, G decreases from (cid:39) 0.3 with no chemotaxis to (cid:39) 0.25 when chemotaxis (α = 3) is allowed. This decrease is best understood for a single source on one side of the domain. Because c decreases monoton- ically from the source, the bacterial population can be split into quantiles of uptake that are ordered in space. Chemotaxis lowers the nutrient concentration over the whole domain, but mostly close to the source, and it shifts the center of mass and mean uptake level of each quantile. Fig. 3a show that bacteria with the higher up- take, which are also closer to the source, are transferred to lower uptake levels. For the lower uptake quantiles, higher levels of uptake compared to the non-chemotactic case are attained due to chemotaxis toward the source. Together, these effects bring uptake levels closer to the average, lowering G: the bacterial system moves closer to the ideal free distribution. The generality of this result can be established in the limit of weak chemotaxis (α (cid:28) 1) still with a single source, where a series solution G (cid:39) G0(λ) + α G1(λ) +··· of (4) and (5) yields G0(λ) = 1 − 2 G1(λ) = − 2 3λ cosh λ − 1 , λ sinh λ cosh λ − 1 λ2 sinh λ + (cid:18) 1 + λ cosh λ 3 sinh(λ) (cid:19) (15) , (16) where G1(λ) < 0, so G indeed decreases with chemotaxis. A cumbersome analysis (not shown) for the general case s ∈ [0, 0.5], that is G (cid:39) G0(λ, s) + α G1(λ, s) + ··· , also shows that G1(λ, s) < 0, thus extending this result to any ratio of source stengths. FIG. 2. (color online). Variations of Gini index for bacterial uptake, with s = 0 (single source on the right side). (a) G at steady-state with respect to the chemotactic parameter α for various domain sizes λ. (b) Ratio of G in the chemotactic case to that without, for various chemotactic strengths α, showing optimal domain size parameter for relative decrease of G. work on its variations. Using the individual nutrient up- take rate as wealth, we can transform the integrals in the uptake rate levels into integrals in space. This makes the bacterial density appear as the equivalent in space of the frequency distribution of uptake levels. Normalizing this new density function, thus making the integral for the total number of bacterial cells appear at the denomina- tor, we finally re-express (13) for time-dependent spatial distributions in a domain Ω: Ω dxdy b(x, t)b(y, t)f (c(x, t)) − f (c(y, t)) 2(cid:82) Ω dx b(x, t)f (c(x, t))(cid:82) Ω dx b(x, t) . (cid:82) (cid:82) G(t) = Ω (14) With this measure of inequality, we investigate how the uptake distribution at steady-state depends on the parameters of the KS model (4)-(5). Before proceeding we should make clear that the Gini index values calcu- lated within the present approach and the corresponding inequalities in the uptake levels are instantaneous: phys- ically, bacteria would swim along a biased random walk inside the steady-state nutrient distribution, thus sam- pling different concentration levels. Over a time long in comparison with the typical time of bacterial diffusion at the scale of the experimental chamber this motility would tend to level the integrated uptake within the popula- tion and yield lower values of G. The fundamental issue is then whether the time scale for this smoothing-out of inequalities is large or small compared to the timescale τint for a relevant internal biological process based on nutrient uptake. The present approach is thus in the limit τintDb/L2 (cid:28) 1, thus most relevant to large system sizes and short internal times. As an example, it takes approximately 17h for a typical run-and-tumble bacteria (Db = 4.10−6 cm2.s−1) to explore the space between two sources separated by 5 mm, a time that is much bigger than the typical scale of key cellular processes such as division (approx. 30 minutes). We now go back to investigating the role of each pa- 1510500.00.40.8Gini index G1510500.51.0G(α,λ)/G(α,0)chemotactic parameter αdomain size parameter λ(a)(b)λαλ=235α=137where ω and β1 are determined through (cid:16) ω (cid:17) (cid:105) ω tan β1 (−β1 + λ) 2 = αs = α(1 − s) . ω tan (cid:104) ω 2 5 (19) (20) (cid:26) The constant β2 depends on ω and the model parameters. Neglecting terms in 1/α2 and higher order, it can be written as β2 (cid:39) − 2 log α + 2 + 2 log ω (cid:21)(cid:27) 1 (cid:20) (1 − s)2 1 − 2s + 4s2 + s(1 − s)log − 2 [s log s + (1 − s) log(1 − s)] − ω2λ/α + ··· . (21) This solution enables us to obtain an analytical expres- sion for the Gini index in the limit of strong chemotaxis: +··· , G (cid:39) λ (22) to leading order in 1/α, for s ∈ [0, 0.5]. This establishes further the generality of its decrease with α together with its increase with λ: in 1D, chemotaxis levels the uptake throughout the population. Moreover, in this range of high α, inequalities of uptake initially increase when the system changes from a single source to more balanced sources. s2 α As this levelling of the uptake distribution appears as the microbial equivalent of the more uniform uptake dis- played by ducks, it is natural to ask if, as in Harper's experiments, the bacterial population reaches the ideal free distribution as α → ∞ and splits into two localized sub-populations proportional to the source strengths. If we consider that the position x0 of minimum bacterial concentration separates a left population BL associated to the left source and its right-hand-side equivalent BR, our analytical solution for α (cid:29) 1 directly yields dx b(x) = BL (cid:39) λ s . (23) 0 √ Thus, the population associated to one source is directly proportional, to leading order, to the flux of this source, as in the ideal free distribution. Moreover, analysis of (18) shows that in the limit α → ∞, b(x) is localized in regions of width ∼ 1/ α at both x = 0 and x = λ, with peak values ∼ α at these positions. In the limit of α → ∞, we thus get a localisation of the number of cells proportional to the source at the source: this is the (unphysical) limit of a microbial ideal free distribution. When the domain size parameter λ (cid:29) 1, the central portion of the domain has a steady-state concentration c ∼ 0, with very small gradients. Bacteria there are screened from the sources, unable to feel sufficient gradi- ents to move chemotactically closer to them. The relative redistributive effect of chemotaxis compared to its ab- sence must then diminish with distance. Considering the relative Gini index G(α)/G0 in the approximation (15)- (16), we indeed find an optimal domain size, λG (cid:39) 3.12, (cid:90) β1 (color online). FIG. 3. Interpretation of steady-state Gini index variation in different dimensions d. (a) d = 1 with do- main size parameter λ = 3, s = 0 (single source on right side), chemotactic strength parameter α = 3. (b) d = 2 with a cir- cular source of size l = 0.5 in a domain of radius L = 4.5, and chemotactic strength parameter α = 3. Shown are center of mass and mean uptake level of the sextiles of uptake dis- tribution (colored circles), overlaid on nutrient concentration, without and with chemotaxis. Dashed line shows concentra- tion for average uptake. Inset: G vs. chemotactic strength parameter α, with blue circles corresponding to displayed so- lutions. In the limit of strong chemotaxis (α (cid:29) 1), where bac- terial diffusion becomes irrelevant, we may expect to re- cover the ideal free distribution. Analytical progress on this steady-state problem is achieved by integrating twice (5) to obtain b(c), substituting into (4) and then expand- ing in powers of 1/α. One obtains b(x) = c(x) = 1 α 1 λ λω2 2 cos2 [ω (x − β1) /2] + 1 α {β2 − 2 log [cos (ω(x − β1)/2)]} , (18) (17) 32100.00.40.8position xnutrient concentration cα51000.00.5G(a)1.52.01.00.5position x0.00.5nutrient concentration c(b)α5100G0.250.35chemotaxisno chemotaxischemotaxisno chemotaxisfor which the redistributive effect of chemotaxis is the strongest. An optimal size is also found in simulations beyond the linear regime in α (Fig. 2 (b)) and with in- flux from both sides, with λG (cid:39) 3. The decrease of this relative change for high values of λ embodies the afore- mentioned screening, while the behavior at low λ results from a nearly uniform concentration over the domain, with only weak gradients for a chemotactic response. Does the uptake levelling found in d = 1 hold in higher dimensions? To answer this, we solve (4) and (5) for a single spherical source of radius l in a closed spherical do- main of radius L, both measured in units of lk. We find that in d = 2 and 3 the effects of chemotaxis, for a given size of source and domain, are much weaker. Moreover, for certain parameter values, chemotaxis can actually in- crease G (inset in Fig. 3 (b)). Analysis of the quantiles (Fig. 3b for a two-dimensional example) shows that in these cases, even though bacteria closest to the source have a lowered uptake, a majority of the bacteria that are already above the average uptake in the non-chemotactic case gain access to even higher uptake. Bacteria furthest from the source, and below the average uptake level in the non-chemotactic case, see their mean uptake decrease even further. Overall this corresponds to an increase of G: in higher dimensions, chemotaxis can bring the bacte- ria further away from the ideal free distribution, that is, it increases the inequalities among the population. The increase or decrease of inequalities of uptake, as revealed by the positive or negative change of G, may thus depend in detail on the system characteristics embodied in l, L and α. IMPLICATIONS FOR FITNESS What would be the biological consequences of chemo- tactic levelling of resources and of uptake rates? Uptake of nutrients governs a wide range of bacteriological pro- cesses, among which are cell growth and division. In par- ticular, the yield of biomass per unit nutrient taken up is an increasing function of the available nutrient concen- tration, a feature which has been suggested as selective for the response characteristics of chemotaxis [13]. Here we provide a brief discussion of how the chemotactically- driven redistribution of resources throughout a popula- tion can impact on the average growth rate of the popu- lation which we consider a measure of fitness. Continuing the point of view taken in the introduction, we show that while at the single cell level in a defined resource field chemotaxis may increase fitness, this is not necessarily true at the population level. We compute an average growth rate ¯µ over the pop- ulation from the steady-state distributions of (5), in a model in which the local growth rate µ(x) is proportional to the local uptake rate through a yield function y(c): 6 µ(x) = y(c) c and (cid:90) λ 0 ¯µ = 1 λ dx y(c)cb . (24) Whereas the mean uptake is fixed by the boundary conditions in our problem, this average growth coefficient will depend on the distribution of the resource among the bacteria and thus will be modified by chemotaxis. In order to capture the increase of y with c, we consider that y = 0 below of threshold concentration cmin, and adopt a Michaelis-Menten form above it [37]: y(c) = y0 (c − cmin) (c − cmin) + K for c > cmin (25) with K a saturation constant of the yield (Fig. 4a). The threshold concentration cmin can be considered as the limit below which all the uptake is directed toward maintenance costs [36]. The relative change of the aver- age growth rate due to chemotaxis (¯µchemo − ¯µnonchemo) is shown in Fig. 4b. We observe that for lower values of the threshold concentration cmin the redistributive ef- fect increases the number of cells that reach the growth threshold: the population fitness becomes higher with chemotaxis than without. However, for higher values of the threshold concentration cmin, the redistribution of uptake throughout the population leaves more cells be- low the growth threshold: we get the counter-intuitive result that chemotaxis effectively lowers population fit- ness with respect to the non-chemotactic case. This re- sult, which stems from the competition of the bacteria for the same resource, shows that in the context of contin- uous sources, an homogeneously chemotactic population could be selected against due to the dilution of a scarce resource resulting from the chemotactic behavior. CONCLUSIONS We have shown that the organization of bacteria around localized nutrient sources is fundamentally dif- ferent from that of higher animals due to diffusion of re- sources and feeders. Yet, there are still common charac- teristics. First, what might be termed 'foraging' behavior decreases the maximum uptake rate through competition for the resource; in the bacterial case this corresponds to a decrease of the maximum of the concentration field with chemotaxis. Second, foraging generates, quite evidently, localization and accumulation of the population closer to the resources. But whereas the conjunction of these two phenomena brings Harper's ducks to the ideal free dis- tribution, it may fail in the microbial world: it brings the system closer or further from this ideal distribution depending on the spatial dimensionality and parameters capturing the strength of chemotaxis, the size of the re- sources and the distance between them. The redistribu- tion of uptake is not without consequences: when the 7 [2] S. D. Fretwell and H. L. Lucas Jr., On territorial behavior and other factors influencing habitat distribution in birds, Acta Biotheor. 19, 16 (1969). [3] P. Glimcher, M. Dorris and H. Bayer, Physiological utility theory and the neuroeconomics of choice, Games Econ. Behav. 52(2), 213 (2005). [4] M. Kennedy and R. D. Gray, Can ecological theory predict the distribution of foraging animals? A critical analysis of experiments on the Ideal Free Distribution, Oikos 68, 158 (1993). [5] M. L. Rosenzweig, A Theory of Habitat Selection, Ecol- ogy 62, 327 (1981). [6] J. K. Parrish and L. Edelstein-Keshet, Complexity, Pat- tern, and Evolutionary Trade-Offs in Animal Aggrega- tion, Science 284, 99 (1999). [7] G. J. Madden, B. F. Peden, and T. Yamaguchi, Hu- man group choice: discrete-trial and free-operant tests of the ideal free distribution, J. Exp. Anal. Behavior 78, 1 (2002). [8] K. E. Abernethy, E. H. Allison, P. P. Molloy and I. M. Cot´e, Why do fishers fish where they fish? Using the ideal free distribution to understand the behaviour of artisanal reef fishers, Can. J. Fish. Aquat. Sci. 64, 1595 (2007). [9] J. R. Kraft, W. M. Baum and M. J. Burge, Group choice and individual choices: modeling human social behavior with the Ideal Free Distribution, Behav. Process. 57, 227 (2002). [10] F. Azam and F. Malfatti, Microbial structuring of marine ecosystems, Nat. Rev. Microbiol. 5, 782 (2007). [11] H. C. Berg, Random Walks in Biology, 2nd ed. (Prince- ton, Princeton University Press, 1993). [12] H. C. Berg an E. M. Purcell, Physics of chemoreception, Biophys. J. 20, 193 (1977). [13] A. Celani and M. Vergassola, Bacterial strategies for chemotaxis response, P. Natl. Acad. Sci. USA 107, 1391 (2010). [14] T.W. Engelmann, Untersuchungen uber die quantitativen Beziehungen zwischen Absorption des Lichtes und Assim- ilation in Pflanzenzelle, Bot. Zeit. 42, 81 (1884). [15] T. W. Engelmann, Neue Methode zur Untersuchung der Sauerstoffausscheidung pflanzlicher und thierischer Or- ganismen, Pflug. Arch. Ges. Phys. 25 285 (1881). [16] M. T. Croft, A. D. Lawrence, E. Raux-Deery, M. J. War- ren, and A. G. Smith, Algae acquire vitamin B12 through a symbiotic relationship with bacteria, Nature 438, 90 (2005). [17] N. Blackburn, T. Fenchel, and J. Mitchell, Microscale nutrient patches in planktonic habitats shown by chemo- tactic bacteria, Science 282, 2254 (1998). [18] T. Kiørboe and G. A. Jackson, Marine snow, Organic Solute Plumes, and Optimal Chemosensory Behavior of Bacteria, Limnol. Oceanogr. 46, 1309 (2001). [19] E. F. Keller and L. A. Segel, Traveling bands of chemo- tactic bacteria: a theoretical analysis, J. Theor. Biol. 30, 235 (1971). [20] J. D. Murray, Mathematical Biology II Spatial Models and Biomedical Applications, 3rd ed. (Springer-Verlag New York, 2003). [21] M. Tindall, P. Maini, S. Porter, and J. Armitage, Overview of Mathematical Approaches Used to Model Bacterial Chemotaxis II: Bacterial Populations, Bull. Math. Biol. 70, 1570 (2008). [22] C. Gini, Variabilit`a e mutabilit`a (Bologna, C. Cuppini, FIG. 4. (color online). Consequences of uptake levelling on population growth rate. Using α = 4, λ = 3 and s = 0.125 in d = 1. (a) Yield as a function of nutrient concentration, for cmin = 0.2, K = 0.05 and y0 = 1. (b) Change of growth rate as a consequence of chemotaxis for K = 0.05 and varying cmin: if chemotaxis can increase growth for low yield thresh- old, the uptake leveling makes it less advantageous for higher threshold by diluting the resource. resource is scarce in comparison to the metabolic needs, chemotaxis effectively dilutes it and reduce the average population fitness. The issues addressed here suggest experimental stud- ies of model systems in physical ecology for which in situ measurements of local metabolic activity and nutrient concentration fields are possible. Optically-based quan- titative measures of photosynthetic activity [38], probes of local oxygen concentration [39], and local mass spec- trometry [40] are examples of relevant techniques. Micro- bial communities in biofilms, sediments [41], and algae sustaining a motile population of bacteria around them by releasing oxygen [42] represent interesting systems in which to study the distribution of uptake rates. ACKNOWLEDGMENTS We are grateful to Wolfram Schultz for bringing Ref. 1 to our attention, and thank O. Croze, T.J. Pedley, W. Poon, A. Smith, G. Peng and R. Watteaux for discus- sions. This work was supported by a Raymond and Bev- erley Sackler Scholarship (FJP), MinesParisTech (FJP), and the European Research Council Advanced Investiga- tor Grant 247333 (REG). [1] D. G. C. Harper, Competitive foraging in mallards: 'ideal free' ducks, Anim. Behav. 30, 575 (1982). 1912). [23] M. O. Lavrentovich, J. H. Koschwanez, and D. R. Nelson, Nutrient Shielding in Clusters of Cells, Phys. Rev. E 87, 062703 (2013). [24] A. J. Hillesdon, T. J. Pedley and J. O. Kessler, The de- velopment of concentration gradients in a suspension of chemotactic bacteria, B. Math. Biol. 57, 299 (1995). [25] I. Tuval, L. Cisneros, C. Dombrowski, C. W. Wolgemuth, J. O. Kessler, and R. E. Goldstein, Bacterial swimming and oxygen transport near contact lines, P. Natl. Acad. Sci. USA 102, 2277 (2005). [26] See, for instance, A. Blanchet, J. A. Carrillo, D. Kinder- lehrer, M. Kowalczyk, P. Laurencot and S. Lisini, A hy- brid variational principle for the Keller-Segel System in R2, arixiv:1407.5562. [27] See also G. Wolansky, Multi-components chemotactic sys- tem in the absence of conflicts, Euro. J. App. Math. 13, 641 (2002); G. Wolansky, A critical parabolic estimate and applications to nonlocal equations arising in Chemo- taxis, App. Anal. 66, 291 (1997). [28] J. G. Mitchell, M. M. Martinez-Alonso, J. Lalucat, I. Esteve and S. Brown, Velocity Changes, Long Runs, and Reversals in the Chromatium minus Swimming Response, J. Bacteriol. 173, 997 (1991). [29] A. Natarajan and F. Srienc, Dynamics of Glucose Up- take by Single Escherichia coli Cells, Metab. Eng. 1, 14 (1999). [30] T. Ahmed and R. Stocker, Experimental Verification of the Behavioral Foundation of Bacterial Transport Param- eters Using Microfluidics, Biophys. J. 95 13 (2008). [31] R. Stocker, J. Seymour, A. Samadani, D. Hunt, and M. Polz, Rapid chemotactic response enables marine bacteria 8 to exploit ephemeral microscale nutrient patches, P. Natl. Acad. Sci. USA 105, 4209 (2008). [32] J. Taylor, and R. Stocker, Trade-offs of chemotactic for- aging in turbulent water, Science 338, 675 (2012). [33] F. Cowell, Measuring inequality (Oxford University Press, 2011). [34] C. Gini, Measurement of Inequality of Incomes, Econ. J. 31, 124 (1921). [35] L. Wittebolle et al., Initial community evenness favours functionality under selective stress, Nature 458, 623 (2009). [36] D. Tempest, The biochemical significance of microbial growth yields: A reassessment, Trends Biochem. Sci., 3, 180 (1978). [37] H.L. Smith and P. Waltman (1995) The Theory of the Chemostat: Dynamics of Microbial Competition, Cam- bridge Studies in Mathematical Biology Vol 13, (Cam- bridge Univ Press, Cambridge, UK, 1995). [38] N. R. Baker, Chlorophyll fluorescence: a probe of photo- synthesis in vivo, Annu. Rev. Plant Biol. 59, 89 (2008). [39] C. Douarche, A. Buguin, H. Salman, and A. Libchaber, E. Coli and Oxygen: A Motility Transition, Phys. Rev. Lett. 102, 198101 (2009). [40] J. D. Watrous and P. C. Dorrestein, Imaging mass spec- trometry in microbiology, Nat. Rev. Microbiol. 9, 683 (2011). [41] R. N. Glud et al., An in situ instrument for planar O2 optode measurements at benthic interfaces, Limnol. Oceanogr. 46, 2073 (2001). [42] J. Adler and B. Templeton, The effect of environmental conditions on the motility of Escherichia coli, J. Gen. Microbiol. 46, 175 (1967).
1303.0985	1	1303	2013-03-05T11:14:29	A highly specific gold nanoprobe for live-cell single-molecule imaging	[ "physics.bio-ph" ]	Single molecule tracking in live cells is the ultimate tool to study subcellular protein dynamics, but it is often limited by the probe size and photostability. Due to these issues, long-term tracking of proteins in confined and crowded environments, such as intracellular spaces, remains challenging. We have developed a novel optical probe consisting of 5-nm gold nanoparticles functionalized with a small fragment of camelid antibodies that recognize widely used GFPs with a very high affinity, which we call GFP-nanobodies. These small gold nanoparticles can be detected and tracked using photothermal imaging for arbitrarily long periods of time. Surface and intracellular GFP-proteins were effectively labeled even in very crowded environments such as adhesion sites and cytoskeletal structures both in vitro and in live cell cultures. These nanobody-coated gold nanoparticles are probes with unparalleled capabilities; small size, perfect photostability, high specificity, and versatility afforded by combination with the vast existing library of GFP-tagged proteins.	physics.bio-ph	physics	A highly specific gold nanoprobe for live-cell single-molecule imaging Cécile Leduc†, Satyabrata Si†, Jérémie Gautier‡, Martinho Soto-Ribeiro#, Bernhard Wehrle- Haller#, Alexis Gautreau‡, Grégory Giannone§, Laurent Cognet† and Brahim Lounis† † Univ Bordeaux, LP2N UMR 5298, Institut d’Optique & CNRS, F- 33405 Talence, France ‡ CNRS, Laboratoire d’Enzymologie et Biochimie Structurales UPR3082, F-91198 Gif-sur-Yvette France # University of Geneva, Department of Cell Physiology and Metabolism, Centre Médical Universitaire, Geneva, Switzerland § Univ Bordeaux, IINS UMR 5297, CNRS, F-33000 Bordeaux, France To whom correspondence should be addressed. E-mail: [email protected] tracking, gold nanoparticles, bio- Keywords: photothermal imaging, single particle functionalization, integrin, microtubule, kinesin… Abstract Single molecule tracking in live cells is the ultimate tool to study subcellular protein dynamics, but it is often limited by the probe size and photostability. Due to these issues, long-term tracking of proteins in confined and crowded environments, such as intracellular spaces, remains challenging. We have developed a novel optical probe consisting of 5-nm gold nanoparticles functionalized with a small fragment of camelid antibodies that recognize widely used GFPs with a very high affinity, which we call GFP-nanobodies. These small gold nanoparticles can be detected and tracked using photothermal imaging for arbitrarily long periods of time. Surface and intracellular GFP-proteins were effectively labeled even in very crowded environments such as adhesion sites and cytoskeletal structures both in vitro and in live cell cultures. These nanobody- coated gold nanoparticles are probes with unparalleled capabilities; small size, perfect photostability, high specificity, and versatility afforded by combination with the vast existing library of GFP-tagged proteins. 1 Over the last decade, it has become essential to work at the individual molecule scale in order to better understand complex cellular processes. Numerous and increasingly sophisticated optical techniques have been developed to explore the dynamics of biomolecules either in vitro or in living systems. Due to its high sensitivity and non-invasiveness, fluorescence microscopy is widely used to detect molecules evolving in biological environments. Single molecule detection requires optimized probes and much effort has been made to improve their properties 1. In general, the ideal probe should be: (i) very specific, (ii) as small as possible (so not to interfere with the dynamics and functions of the target biomolecules), (iii) monovalent, (iv) optically stable for long observation times, (iv) able to deliver intense optical signals for precise localization. As this ideal probe has not yet been designed, long-term single-molecule tracking of proteins in confined and crowded environments remains challenging. Fluorescent molecules have been extensively used for their small size but suffer from poor in physiological environments2. Semi-conducting nanocrystals photostability, especially (quantum dots), however, display superior photostability and can be imaged for minutes in live cells3-5. Nevertheless, due to their relatively large diameter, functionalized quantum dots are not suitable for tracking biomolecules in confined cellular environments such as adhesion sites and synapses 2. An alternative to luminescence relies on the probe absorption properties. Gold nanoparticles of 5 nm diameter are appealing because they present large absorption cross-sections and deliver intense, stable optical signals in photothermal microscopy, even in cellular environments6. To fully benefit from the reduced size of these nanoparticles, the ligand used to bind the particle to the target biomolecule should also be small. Commonly used antibodies have typical sizes of 12 nm and thus significantly increase the overall probe size. Moreover, highly specific primary antibodies are not always available, and when they are, particle coupling must be performed for each protein of interest. To avoid multiple couplings a secondary antibody could be used at the expense of an increase in final probe size. Common antibodies are also divalent and their use raises issues about the probe’s valence state. In this context, the elaboration of general strategies to obtain small, highly specific, monovalent probes that allow for the study of the largest number of proteins is an intense field of research 5, 7-9. 2 In this paper, we designed and characterized a system that most closely approaches the aspects of an ideal probe. We synthesized 5 nm gold nanoparticles functionalized with nanobodies (NB), a small fragment of camelid antibody (2 nm x4 nm) which recognizes widely used GFPs (3 nm x 4 nm) with a very high affinity (Kd~0.23 nM) 10, 11. Our approach allows the production of purified, monovalent nanoparticles. These small, functional gold nanoparticles (NB-Au-NPs) can be detected and tracked for unlimited periods of time in living cells using photothermal heterodyne imaging 12,6 , 13. We demonstrate the versatility and targeting efficiency of our NB-Au-NPs by labeling several types of GFP-tagged proteins, both in in vitro and cellular systems. We also show that NB-Au-NPs can be used to study proteins in confined structures and intracellular compartments by tracking adhesive proteins and microtubule associated proteins in living cells, respectively. NB-Au-NPs were produced in two steps, synthesis and functionalization (Figure 1A, see supporting methods). In the first step, functionalized gold nanoparticles were synthesized using borohydride reduction in the presence of a linker agent, MUA, (11-mercaptoundecanoic acid) and two blocking agents in equimolar amounts: CVVVT-ol peptide (T-ol = threoninol) and alkyl PEG (PEGylated alkanethiol HS-(CH2)11-EG4-OH) (protocol adapted from7, 14). These peptide capped gold nanoparticles are extremely stable, water-soluble and display some chemical properties analogous to proteins, which is responsible for their biocompatibility7, 15. The average diameter of the nanoparticles was estimated to be 4.6 +/- 1 nm (SD) from transmission electron microscopy pictures (Figure 1B-C). Because the molecules of the capping shell are very short the hydrodynamic radius of the capped nanoparticles is less than 1 nm larger16 than the radius obtained from electron microscopy. A photothermal image of isolated particles embedded in a polyvinyl alcohol film is also displayed in Figure 1D. In the second step, the carboxylic function of MUA was activated using EDC/NHS (1-ethyl-3-(3-dimethylaminopropyl) carbodiimide hydrochloride/N-Hydroxysuccinimide) allowing coupling to recombinant nanobodies. We used nanobodies bearing a his-tag purified from bacteria with Ni-NTA columns. We then employed two methods to obtain monovalent probes. In method one, the proportion of MUA was reduced with respect to that of the blocking agents (down to 0.4%, Supporting Figure 1) to achieve at most one MUA per particle when an excess of nanobodies was introduced for the coupling. In method two, a low concentration of nanobodies was introduced into an excess of particles bearing multiple MUA linkers. We favored the second method, which requires a smaller amount of nanobodies, because non-functionalized Au-NPs could be separated from the NB-Au-NPs by using Co-NTA resin able to reversibly bind his-tag nanobodies. This is demonstrated in Figure 3 1E by the shift of nanobody-coated Au-NP band as compared to the control Au-NP in agarose gel electrophoresis. It is also seen that a coating with 20% MUA produces negatively charged particles (line (a) and (f) to (h)), while with 1% and 0.4% MUA, positively charged NB-Au-NPs are obtained and cannot be used for cell electroporation (see below). Of note, NB-Au-NPs bands were relatively broad due to the size dispersion of the nanoparticles (Figure 1C) leading to a relatively large heterogeneity in the number of MUA per particle. NB-Au-NPs were first tested on a well-characterized in vitro assay consisting of purified molecular motors and microtubules in a reconstituted system17,18. GFP-tagged kinesin-1 were incubated with an excess of NB-Au-NPs for 5 min before injection in a flow chamber where taxol-stabilized microtubules were previously fixed (Supporting Figure 2)19. Without ATP in solution, kinesin motors labeled with NB-Au-NPs strongly bound microtubules without any directional motion or unbinding (Figure 2A). We used a high concentration of motors in solution (15 nM, 5 min incubation) to saturate microtubules with motors and then rinsed extensively in the flow chamber to remove unbound motors and probes. A homemade microscopy setup was used to record photothermal and fluorescence images of the sample (see supporting information). Figure 2B shows a photothermal image of microtubules covered by kinesin-GFP motors labeled with NB-Au-NPs and the corresponding fluorescence image. The GFP fluorescence signal and NB-Au-NPs photothermal signal colocalize within a camera pixel size (~250nm at the sample plane), a signature of a specific labeling of the motors by the new probe. Regions with higher photothermal signals correspond to overlapping microtubules (arrow in Figure 2B). More quantitatively, the signal in these regions is, in average, twice the signal of isolated microtubules (Figure 2C). This illustrates the possibility to perform stoichiometric measurements using NB- Au-NP probes detected by photothermal methods12, 20. As a control, we verified that no photothermal signal was observed when microtubules were incubated with non-functionalized Au-NPs or with NB-Au-NPs without kinesin motors (supporting Figure2). To demonstrate that NB-Au-NPs binding specificity is conserved in more complex environments, we tested the particles in fixed cells expressing GFP-tagged proteins in various sub-cellular compartments. We used either COS-7 cells or Mouse Embryonic Fibroblasts (MEFs) platted on fibronectin-coated glass coverslips. The cells were transfected with GFP-tagged proteins enriched in cellular adhesion sites (VASP-GFP), in the actin cytoskeleton (alpha-actinin- 4 GFP), or on microtubules (EB3-GFP) (Figure 3). After fixation, and permeabilization, followed by incubation with a 10 nM solution of NB-Au-NP and extensive rinsing steps, cells were studied by epi-fluorescence and photothermal imaging. For all protein constructs, photothermal images displayed perfect colocalization with the GFP fluorescence images, demonstrating that the probe specificity is maintained in complex biological environments. We verified that non-transfected cells displayed very few non-specific nanoparticle binding (Supporting Figure3). Notably, these results show that NB-Au-NP can access GFP-proteins even in crowded regions of the cells such as adhesion sites or actin rich structures like the lamellipodium. Next, we showed that NB-Au-NPs could label membrane proteins in live cells, allowing single molecule tracking even in highly confined structures. As a model, we focused on mature adhesion sites (focal adhesions, FAs) which are crowded macromolecular platforms where integrins mediate cell-extracellular matrix adhesion21. β3-integrins bearing an extracellular GFP tag were expressed in MEFs and found to be concentrated in FAs (Figure 4A epi-fluorescence). MEFs were incubated with NB-Au-NPs for 10 min and then rinsed extensively. Photothermal images of the same region as in Figure 4A showed that the NB-Au-NPs were also mainly localized in the FAs (delimited regions of the epi-fluorescence image), strong evidence that the probe can enter FAs with minimal steric hindrance. In contrast, his-tagged β3-integrins labeled with streptavidin coated quantum dots via a biotinylated NTA22 could not enter FAs (Figure 4B). Since the space between the membrane and the glass coverslip is close to 30 nm in FAs21, this observation suggests that the size of the quantum dot probe prevents its access to the FA interior. In order to follow the movement of individual GFP-β3-integrins we reduced the labeling density of NB-Au-NPs on MEFs. Using the triangulation method introduced by Lasne et al6, long trajectories (>25 s) of single integrins were recorded at video rate (Figure 5). In Figure 5A, we overlaid two typical trajectories of individual integrins on the fluorescence image of GFP-β3- integrin. The two trajectories illustrate the diffusion properties of integrins found inside and outside FAs. The photothermal signals displayed a constant amplitude and were similar to those shown in Figure 1D obtained from single gold particle with size distribution of Figure 1C, confirming that these trajectories corresponded to individual integrins (Figure 5 B and E inset). We also extracted the mean-square-displacements and instantaneous diffusion coefficients (with a moving window of ~ 1s, see supporting information) of the probes. Inside FAs, probes 5 displayed either immobilization or slow diffusion, whereas outside they displayed free diffusion (Figure 5B-E, supporting Figure 4). These behaviors are similar to those observed using single molecule fluorescence tracking with smaller probes23, which showed that immobilization events observed inside FAs are the consequence of direct integrin binding to the extracellular matrix (fibronectin) and/or to intracellular talins. This confirms that NB-Au-NPs did not alter integrin dynamics after labeling nor did they impede their activation. In addition, combining photothermal imaging and the perfect photostability of our probes, we now are able to investigate the dynamics of single integrins on arbitrarily long time scales. In the specific case of integrins, we can study cycles between immobilizations and free diffusive movements even during the same trajectory. This allows to probe the dynamics of protein interactions, which control membrane receptor diffusion. Finally, we addressed the possibility of targeting and tracking intracellular proteins with NB-Au-NPs in living cells. Negatively charged NB-Au-NPs and 5 µg DNA coding for EB3-GFP, a microtubule end-binding protein, were efficiently internalized in COS-7 cells by electroporation. Figure 6A shows fluorescence and photothermal images of a cell 24h after electroporation. Demonstrating specific intracellular targeting (see supporting Figure 5 for the control), NB-Au-NPs appeared mainly located either along the microtubule lattice or ends, where EB3-GFP proteins are accumulated24 (red arrows in Figure 6A). We further tracked the movement of bound nanoparticles. The trajectory shown in Figure 6B-D represents the movement of a single EB3-GFP protein: it is clearly directional, points towards the cell periphery in the prolongation of a microtubule, and shows an average velocity (~ 0.09 µm/s), similar to the microtubule growth rate (~0.12 µm/s for COS-7 at room temperature, supporting movie 1). This demonstrates that NB-Au-NPs are appropriate probes to target and track individual intracellular proteins in living cells. In this paper, we presented the development of a new functional optical probe, which bears all the qualities to perform non-invasive long-term single particle tracking in different biological environments. It consists of 5 nm gold nanoparticles coated with GFP-nanobodies, the NB-Au-NPs. We showed that their charge and valence can be tuned by varying the composition of the peptide shelves, and that they can be purified using the his-tag present on the nanobodies. 6 These NB-Au-NPs were successfully used in in vitro and live cell experiments, for both surface and intracellular labeling of various GFP-proteins in very crowded environments (adhesion sites, actin networks etc.). They should allow high density labeling of many target proteins in live or fixed cells, which will provide more complete information about the distribution. We believe that NB-Au-NPs are unparalleled probes owing to their small size, perfect photostability, high specificity, and versatility. Possible applications could involve the targeting of nuclear proteins or concern correlative studies between optical and electron microscopy since small gold nanoparticles provide high contrasts for both modalities. Acknowledgement: We would like to thank Marie-Hélène Delville for the use of the TEM microscope, Matthieu Sainlos for the Cobalt resin used to purify the NB-Au-NPs, C. Hoongenrad for the plasmid of EB3-GFP, S. Diez for the plasmid of the kinesin-GFP (rk540-GFP), B. Tessier and O. Rossier for experimental help, and J. Shaver for reading the manuscript. We acknowledge financial support from the Région Aquitaine, Institut universitaire de France, Agence Nationale de la Recherche, and the European Research Council. 7 References 1. 2. 10. 3. 4. 5. 6. Giepmans, B. N.; Adams, S. R.; Ellisman, M. H.; Tsien, R. Y. Science 2006, 312, (5771), 217-‐24. Groc, L.; Lafourcade, M.; Heine, M.; Renner, M.; Racine, V.; Sibarita, J. B.; Lounis, B.; Choquet, D.; Cognet, L. J Neurosci 2007, 27, (46), 12433-‐7. Dahan, M.; Levi, S.; Luccardini, C.; Rostaing, P.; Riveau, B.; Triller, A. Science 2003, 302, (5644), 442-‐ 5. Heine, M.; Groc, L.; Frischknecht, R.; Beique, J. C.; Lounis, B.; Rumbaugh, G.; Huganir, R. L.; Cognet, L.; Choquet, D. Science 2008, 320, (5873), 201-‐5. Pinaud, F.; Clarke, S.; Sittner, A.; Dahan, M. Nat Methods 2010, 7, (4), 275-‐85. Lasne, D.; Blab, G. A.; Berciaud, S.; Heine, M.; Groc, L.; Choquet, D.; Cognet, L.; Lounis, B. Biophys J 2006, 91, (12), 4598-‐604. Duchesne, L.; Gentili, D.; Comes-‐Franchini, M.; Fernig, D. G. Langmuir 2008, 24, (23), 13572-‐80. 7. 8. Medintz, I. L.; Uyeda, H. T.; Goldman, E. R.; Mattoussi, H. Nat Mater 2005, 4, (6), 435-‐46. 9. Howarth, M.; Liu, W.; Puthenveetil, S.; Zheng, Y.; Marshall, L. F.; Schmidt, M. M.; Wittrup, K. D.; Bawendi, M. G.; Ting, A. Y. Nat Methods 2008, 5, (5), 397-‐9. Rothbauer, U.; Zolghadr, K.; Tillib, S.; Nowak, D.; Schermelleh, L.; Gahl, A.; Backmann, N.; Conrath, K.; Muyldermans, S.; Cardoso, M. C.; Leonhardt, H. Nat Methods 2006, 3, (11), 887-‐9. Ries, J.; Kaplan, C.; Platonova, E.; Eghlidi, H.; Ewers, H. Nat Methods 2012. 11. Berciaud, S.; Cognet, L.; Blab, G. A.; Lounis, B. Phys Rev Lett 2004, 93, (25), 257402. 12. 13. Duchesne, L.; Octeau, V.; Bearon, R. N.; Beckett, A.; Prior, I. A.; Lounis, B.; Fernig, D. G. PLoS Biol 2012, 10, (7), e1001361. Zheng, M.; Huang, X. J Am Chem Soc 2004, 126, (38), 12047-‐54. Levy, R.; Thanh, N. T.; Doty, R. C.; Hussain, I.; Nichols, R. J.; Schiffrin, D. J.; Brust, M.; Fernig, D. G. J Am Chem Soc 2004, 126, (32), 10076-‐84. 16. Octeau, V.; Cognet, L.; Duchesne, L.; Lasne, D.; Schaeffer, N.; Fernig, D. G.; Lounis, B. ACS Nano 2009, 3, (2), 345-‐50. Korten, T.; Nitzsche, B.; Gell, C.; Ruhnow, F.; Leduc, C.; Diez, S. Methods Mol Biol 2011, 783, 121-‐ 37. 18. Varga, V.; Leduc, C.; Bormuth, V.; Diez, S.; Howard, J. Cell 2009, 138, (6), 1174-‐83. 19. Helenius, J.; Brouhard, G.; Kalaidzidis, Y.; Diez, S.; Howard, J. Nature 2006, 441, (7089), 115-‐9. Blab, G. A.; Cognet, L.; Berciaud, S.; Alexandre, I.; Husar, D.; Remacle, J.; Lounis, B. Biophys J 2006, 20. 90, (1), L13-‐5. Kanchanawong, P.; Shtengel, G.; Pasapera, A. M.; Ramko, E. B.; Davidson, M. W.; Hess, H. F.; Waterman, C. M. Nature 2010, 468, (7323), 580-‐4. Reichel, A.; Schaible, D.; Al Furoukh, N.; Cohen, M.; Schreiber, G.; Piehler, J. Anal Chem 2007, 79, (22), 8590-‐600. Rossier, O.; Octeau, V.; Sibarita, J. B.; Leduc, C.; Tessier, B.; Nair, D.; Gatterdam, V.; Destaing, O.; Albiges-‐Rizo, C.; Tampe, R.; Cognet, L.; Choquet, D.; Lounis, B.; Giannone, G. Nat Cell Biol 2012, 14, (10), 1057-‐67. Stepanova, T.; Slemmer, J.; Hoogenraad, C. C.; Lansbergen, G.; Dortland, B.; De Zeeuw, C. I.; Grosveld, F.; van Cappellen, G.; Akhmanova, A.; Galjart, N. J Neurosci 2003, 23, (7), 2655-‐64. 14. 15. 24. 17. 21. 22. 23. 8 Figures Figure 1: Synthesis and characterization of nanobodies-coated gold nanoparticles (NB-Au- NPs). (A) Schematics of the two-steps production: synthesis and functionalization. (B) Transmission Electron microscopy (TEM) image of stabilized Au-NPs. (C) Distribution of stabilized Au-NP diameters, extracted from the TEM images. (D) Photothermal image of individual NB-Au-NPs embedded in thin polyvinyl alcohol (PVA) film. (E) Agarose gel of NB- Au-NP samples with varying proportions of MUA in the nanoparticles shelf and of nanobodies used for the functionalization in step 2: (a) 20% MUA, no nanobodies, (b) 1 % MUA, no nanobodies, (c) 0.4 % MUA, no nanobodies (d) 0.4 % MUA, equimolar nanobodies, (e) 1 % MUA, equimolar nanobodies, (f) 20 % MUA, equimolar nanobodies, (g) 20 % MUA, 2 fold excess nanobodies, and (h) 20 % MUA + 10 fold excess nanobodies. 9 Figure 2: NB-Au-NPs target kinesin-GFPs in vitro. (A) Schematics of the assay: GFP-tagged kinesin motors, attached to stabilized microtubules fixed on a glass coverslip, are labeled with NB-Au-NPs (20%MUA in the shelf, 4x excess of Au-NP with respect to the nanobodies). (B) Photothermal image showing the distribution of NB-Au-NPs on the assay. The photothermal signal exactly matches the distribution of the kinesin-GFP epi-fluorescence signals (inset). White arrows point the overlapping of two microtubules. (C) Average photothermal intensity profile along the blue and green boxes in (B) corresponding to one and two microtubules respectively. 10 Figure 3: NB-Au-NPs label all sorts of GFP-tagged proteins on fixed cells with a very high specificity (NB-Au-NPs used: 20% MUA in the shelf, 4x excess of Au-NP with respect to the nanobodies, see text). Top: epi-fluorescence images of a COS-7 cell expressing α-actinin-GFP, of a MEF expressing VASP-GFP and of a COS-7 expressing EB3-GFP. Bottom: corresponding photothermal images. 11 Figure 4: NB-Au-NPs target β3-integrins in focal adhesion sites whereas quantum dots do not. (A) Live MEF expressing β3-integrins tagged with an extracellular GFP and labeled with NB- Au-NPs. An epi-fluorescence image from GFP-β3-integrins (left) and the corresponding photothermal image (middle) are merged (right) showing efficient NB-Au-NP labeling of GFP- β3-integrins concentrated in FAs (delimited by the red-dashed lines). (B) Live MEF co- expressing VASP-GFP and β3-integrins tagged with an extracellular his-tag and labeled with streptavidin coated quantum dots QD-655 via biotinylated TrisNTA. An epi-fluorescence image of VASP-GFP, used as a FAs reporter, allows the delimitation of FAs contours (left). Corresponding TIRF (total internal reflection fluorescence) image of the quantum dots (middle) and merged image (right) are shown. Quantum dot labeling does not reach β3-integrins located in FAs. NB-Au-NPs were prepared with 20% MUA in the shelf and a 4 fold excess of Au-NP with respect to the nanobodies. 12 Figure 5: GFP-β3-Integrins labeled with NB-Au-NPs display a freely diffusive motion outside FAs and cycles of immobilization and slow diffusion motions inside FAs. (A) Two NB-Au-NPs labeled GFP-β3-integrin trajectories are overlaid on FAs localized by GFP-β3-integrin fluorescence. Immobilization sequences and slow movements are displayed in red and free diffusion in blue. (B) x-y trajectory inside a FA. Inset: photothermal signal amplitude vs time. (C) Instantaneous diffusion coefficient vs. time. (D) x-y trajectory outside a FA, and (E) corresponding instantaneous diffusion coefficient vs. time. 13 Figure 6: Single molecule tracking of EB3-GFP inside living cells using internalized NB-Au- NPs. (A) COS-7 cell expressing EB3-GFP 24h after electroporation. Right: zoom on the cell edge using epi-fluorescence and photothermal imaging together with the merged image. Red arrows show microtubule lattice and ends where NB-Au-NPs are colocalized. (B) EB3-GFP fluorescence image in a COS-7 cell overlaid with the trajectory of an internalized NB-Au-NP, color-coded by the elapsed time (t=0 blue, t=10s red). (C) Zoom on the x-y trajectory. (D) Photothermal signal amplitude and walked distance vs. time of the trajectory displayed in (B) highlighting directed movement. The average velocity of the directed track is ~ 0.09 µm/s. NB- Au-NPs were prepared with 20% MUA in the shelf and a 4 fold excess of Au-NP with respect to the nanobodies. 14
1809.10990	1	1809	2018-09-28T12:33:36	Mechanics of tissue competition: Interfaces stabilize coexistence	[ "physics.bio-ph", "q-bio.CB" ]	Mechanical forces influence the dynamics of growing tissues. Computer simulations are employed to study the importance of interfacial effects in tissue competition. It was speculated that mechanical pressure determines the competition, where the determining quantity is the homeostatic pressure - the pressure where division and apoptosis balance; the tissue with the higher homeostatic pressure overwhelms the other. Surprisingly, a weaker tissue can persist in stable coexistence with a stronger tissue, if adhesion between them is small enough. An analytic continuum description can quantitatively describe the underlying mechanism and reproduce the resulting pressures and cell-number fractions. Computer simulations furthermore display a variety of coexisting structures, ranging from spherical inclusions to a bicontinuous state.	physics.bio-ph	physics	Mechanics of tissue competition: Interfaces stabilize coexistence Nirmalendu Ganai1,2, Tobias Buscher1, Gerhard Gompper1, and Jens Elgeti1,∗ 1Theoretical Soft Matter and Biophysics, Institute of Complex Systems, Forschungszentrum Julich and JARA, 52425 Julich, Germany and 2Department of Physics, Nabadwip Vidyasagar College, Nabadwip, Nadia 741302, India (Dated: October 1, 2018) Mechanical forces influence the dynamics of growing tissues. Computer simulations are employed to study the importance of interfacial effects in tissue competition. It was speculated that me- chanical pressure determines the competition, where the determining quantity is the homeostatic pressure - the pressure where division and apoptosis balance; the tissue with the higher homeostatic pressure overwhelms the other. Surprisingly, a weaker tissue can persist in stable coexistence with a stronger tissue, if adhesion between them is small enough. An analytic continuum description can quantitatively describe the underlying mechanism and reproduce the resulting pressures and cell-number fractions. Computer simulations furthermore display a variety of coexisting structures, ranging from spherical inclusions to a bicontinuous state. Mechanical forces influence the growth of cells and tis- sues in several ways [1 -- 3]. This ranges from plants adapt- ing their growth patterns to mechanical loads [4, 5], all the way to tumor growth responding to mechanical forces [6 -- 8]. Cells have been shown to differentiate according to substrate stiffness [9], and divide according to mechan- ical stress and strain [10 -- 16]. Spheroids of many cells, grown in elastic gels [17 -- 19] or shells [20, 21], or even in suspension with osmotic stress [22 -- 25], show strong dependence of growth on the mechanical stress from the embedding medium. Given the evidence of mechanical stress on growth, it seems clear that mechanics should also influence tissue competition, such as the competition between different mutants in the imaginal wing disk of drosophila [26, 27], or clonal expansion in multistep cancerogenesis [28, 29]. Several theoretical studies suggested mechanics as the underlying mechanism for both, competition [1] and size determination [30] in the wing, and tumor growth [8, 31]. Growth is a change of volume and the conjugate force to volume is pressure. It stands to reason that pressure should influence growth. A tissue grown in a finite com- partment exerts a certain pressure onto its surrounding. When reaching a steady state - the homeostatic state - this is the homeostatic pressure PH . Under an external pressure P below this value, the tissue grows; whereas it shrinks if the pressure is above it. This simple ap- proach can be formulated as a linear expansion of the bulk growth rate kb around the homeostatic pressure [31], kb = κ(PH − P ) (1) with the pressure response factor κ. This idea has been developed to understand mechanical tissue comepetition in general, and metastatic inefficiency in particular: It was argued that metastases need to reach a critical size, below which the Laplace pressure from the interfacial ten- sion exeeds the homeostatic pressure difference, and the metastasis disappears [31]. To study the role of pressure on growth, experiments and computer simulations have been developed to explore this effect in cell culture and in silico [22 -- 24, 32 -- 34]. While confirming the general assumption - that mechanical pressure reduces growth - these experiments and simulations have led to another important revelation. Tissues preferentially divide at the surface, even to the extent that they die (on average) in the bulk and sustain a finite size only by surface growth. While consideration of nutrient transport may be nec- cessary for quantitative description of some experiments [35], mechanics alone already suffice. In this work, we study the role of interfacial effects on mechanical tissue competition by numerical simulations, in particular the effect of adhesive interactions between different tissues. We find that similar to free surfaces, cells divide preferentially at the low-adhesive interface. This interfacial growth in turn can stabilize coexistence of two tissues with different homeostatic pressures. ij = G (cid:16) R5 Agent-based modelling of tissue growth has been very successful in the recent years [36, 37]. We follow the approach of Ref. [32] and model growing and dividing cells by two point-like particles, which repel each other with a growth force F G (rij +r0)2 r ij. Once a criti- cal distance is reached, cells divide, and two new parti- cles are inserted, starting the process anew. Apoptosis is modeled by a constant rate of cell removal ka. Vol- ume exclusion is maintained by a relatively soft repul- sive force F V r ij, while adhesion be- tween cells is modeled by a constant attractive force F A ij = −f1 r ij between all cells in range RPP. This model results in pressure-dependent growth, in reason- able agreement with experiments [22 -- 24, 32 -- 34]. For two competing tissues A and B, parameters for each tissue can be set independently. In this work, we only vary the growth strength GA and GB, the self adhesion strengths, f AA := fc. 1 See SI for further details and parameters. and the cross-adhesion strength f AB ij = f0 PP r5 ij − 1 , f BB 1 (cid:17) 1 To our great surprise, very small cross-adhesion strengths fc between cells of different tissues (i.e. fc (cid:28) min(f AA )) result in fundamentally different out- , f BB 1 1 (a) (c) (b) (d) Fig. 1. Snapshots of various structures of tissue coexis- tence. (a) Spherical inclusion. (b) Cylindrical inclusion. (c) Schwarz-P like bicontinuous structure. (d) Flat interface. Other structures observed include perforated lamella, com- binations (e.g. perforated lamellar together with a spheroid), and inverted (e.g. inverse spheroid) structures. ∗ as a function of the distance from the Fig. 2. Growth rate k interface for the competition between two identical tissues ∗ with fc = 0 for different box lengths L z. comes of the tissue competition than predicted previously [31]. Instead of one tissue overwhelming the other or the existence of a critical size threshold explained above, we observe stable coexistence in a variety of different struc- tures depending on initial conditions (see Fig. 1). Even for two identical tissues - just without cross-adhesion - a single A cell in a host of B grows into a stable spheroid occupying about a third of the volume. Similarly, a ran- dom 1:2 mixture of stronger A cells in a host of B can result in a stable 3:1 Schwarz-P bicontiuous structure. In order to understand this puzzling behaviour and the underlying physical mechanisms, we turn to a simpler initial condition of a slab-like tissue arrangement and de- velop an appropriate analytic model. Cells are confined to a finite (periodic) compartment of size Lx × Ly × Lz. All cells in the central half (Lz/4 < z < 3Lz/4) are type 2 B cells, all others type A. Large adhesion between cells of the same tissue and no adhesion between cells of dif- ferent tissues leads to a large surface tension, stabilizing the flat interface. The division profile (see Fig. 2) re- veals that cells divide more in a small region of width a (about one or two cell layers) at the interface. In the bulk of the tissue, the net growth is negative due to an ele- vated pressure. These results motivate a two-rate growth model [22 -- 24, 32, 33] ∂tρ + ∇ · (ρv ) = kbρ + ∆ksΘ(s − a)ρ, (2) where ρ is the cellular density, Θ the Heavyside step func- tion, s the distance to the nearest interface and v the cell-velocity field. The additional growth at the inter- face is modeled as a growth enhancement ∆ks near the interface (less than a away). Division and apoptosis events locally relax stress and thus lead to a liquidification of the tissue on longer timescales [38 -- 40]. Indeed, experiments on tissue rheol- ogy suggest liquid behaviour on long timescales [41 -- 43], while some experiments on drosophila wing discs suggest that not all stress is relaxed by growth [44 -- 46]. Our model tissue clearly behaves as a liquid [38]. With the low velocities and no external forcing, we can thus as- sume a constant pressure across the system. This moti- vates expanding kb as in Eq. (1), and similarly ∆ks (cid:39) ∆k0 s (PH − P ). Under the assumption of constant density and with an integration over the system, the time evolution of the cell number fraction φ = NA/(NA + NB) of type A cells reads s + ∆k1 ∂tφ = kbφ + ∆ksφs, (3) with the fraction φs of A type cells at the surface. Two identical tissues (without cross-adhesion) then develop two interfaces Lz/2 apart. Insertion of the linear expan- sions in Eq. (3) then yields the pressure P = PH + 4a∆k0 s (4a∆k1 s + κLz) , (4) i.e. the additional growth at the interface elevates the pressure above the homeostatic pressure, which in turn causes the negative net growth rate in the bulk. We de- termine the bulk parameters PH, κ from bulk simulations as in Ref. [33], and the surface parameters a∆k0 s , a∆k1 s by fitting Eq. (4) to a tissue with mirror boundary condi- tions in one direction (see SI). As shown in Ref. [33], the homeostatic pressure grows approximately linearly with G, and decreases linearly with f1. κ is essentially inde- pendent off f1, but decreases linearly with G. The surface parameter ∆k0 s is only weakly dependent on G, but grows linearly with f1, while ∆k1 s does not show a clear de- pendence on tissue parameter (see SI). Representatively, we show the pressure dependence on box length Lz for two identical tissues without cross adhesion. With the −3−2−10123Distancefrominterfaces∗−0.010.000.010.020.030.040.05Growthratek∗SpeciesBSpeciesAL∗z=10L∗z=25L∗z=50L∗z=1003 for ∆PH > 0 (see Fig. 4). The weaker tissue suports the higher pressure by decreasing in size, and thus its apop- totic volume, sustained by surface growth. For the sim- ulated tissues, the parameter κ, ∆k0 s only show small variations (see SI). We therefore assume them to be the same for both tissues in order to obtain s and ∆k1 φ = ± + 1 2 (cid:34)(cid:18) κ(P B (cid:19)2 2a∆k0 s H − P A H )Lz 2a∆k0 s H − P A H )Lz κ(P B (cid:18) 1 2 + + 2a∆k1 s κLz (cid:19)2(cid:35) 1 2 . (7) H − P A Note that for ∆PH ≡ (P B H ) → 0, Eq. (7) repro- duces φ = 1/2 as expected. Around ∆PH = 0, φ grows linearly with ∆PH and then slows down (see Fig. 4). For large differences in homeostatic pressure, the model pre- dicts two interfaces less than 2a apart, thus violating its assumptions, and consequently fails to predict the sim- ulation results properly. Equations (6) and (7) are able to reproduce simulation results fairly well without pa- rameter adjustments (see Fig. 4) in a broad parameter regime. Note that this also holds true for negative home- ostatic bulk pressures, where indeed the system pressure is positive, thanks to the surface growth (see Eq. (6)). These results show that indeed the enhanced growth at the interface lies at the heart of the coexistence of tissues observed in our simulations. However, a flat interface is not the only stable structure for two competing tissues. Depending on initial conditions and parameters, a large range of other structures can be found (see Fig. 1). These different structures result in different surface-to-volume ratios (and possibly other interfacial effects), changing the steady-state volume fractions and pressures. We present simulation results for these structures in Fig. 5. Compared to flat interfaces, the number fraction φ of tissues in spherical or cylindrical configuration is smaller, with spheroids being smaller than cylinders. Note that spheroids become unstable with growing homeostatic pressure difference. They then turn into cylinders, which again become unstable and turn into a slab-like structure, which probably becomes unstable as well. Vice versa, cylinders turn into spheroids if the difference in home- ostatic pressure is very negative. The number fraction of the bicontinuous phase is roughly the same as for flat interfaces, but the bicontinuous phase is only stable in a small regime of homeostatic pressure differences. For larger differences in homeostatic pressure it turns into a perforated lamella phase of the weaker tissue inside the stronger tissue. In general, the number fraction φ of all structures changes sigmoidally with homeostatic pressure difference (see Fig. 1). While all of these structures are very stable over time, the question arises how stable they are when the interfa- cial effects become smaller. We study this effect numeri- cally, by observing the structures for two identical tissues Fig. 3. (a) Average pressure measured in competitions be- tween two identical tissue (reference tissue, see SI) with zero ∗ cross-adhesion fc = 0 in terms of the inverse box length L z. Dashed purple line shows the prediction of the two-rate model according to Eq. (4). (b) Solid cyan and red lines show the time evolution of the cell number fractions φA/B for a compe- ∗ tition as in (a), for a box length L z = 100. Dashed black line shows the prediction given by Eq. (5) for both tissues. parameters fixed, the theory reproduces the simulations without further parameter adjustment (see Fig. 3(a)). For identical tissues, the steady-state solution to Eq. (3) is φ = 1/2 by symmetry. For the dynamics we obtain φ(t) = 1 2 + (φ0 − 1 2 )e−κ(P−PH)t, (5) with the initial number fraction φ0. As shown in Fig. 3(b), Eq. (5) reproduces the simulation dynamics. Next, we explore the competition between two different tissues with a planar interface. We balance the pressures on both sides of the interface and get s , s H + = P B H + (2a∆k1A (2a∆k1B P = P A 2a∆k0A s + κALA) 2a∆k0B s + κBLB) (6) where LB and LA(= Lz − LB) are the lengths occupied by tissue A and B. Note that inserting LA,B < Lz in Eq. (6) gives a lower bound for the pressure: The system pressure is always larger than the homeostatic pressure of the stronger tissue, plus a system-size-dependent con- stant. Indeed, this lower bound describes the pressure rather well. The stronger tissue occupies the larger part of the system, and thus LA,B ≈ Lz. Thus the pressure is almost constant for ∆PH < 0, and grows almost linearly 0.000.020.040.060.080.10Inverseboxlength1/(L∗z)0.10.20.3PressureP∗(a)ModelPHSimulation01234Normalizedtimet·τ0.00.20.40.60.81.0NumberfractionφA/BL∗z=100(b)φA,SimulationφB,SimulationφA/B,Model4 H − P A Fig. 4. (a) Cell number fractions φ for various homeostatic pressure differences P B H . Tissue B is fixed (reference tissue) and the homeostatic pressure of tissue A is varied. Symbols are simulation results while the solid lines are pre- dictions by the two-rate model according to Eq. (7), using the parameters of tissue B. Blue corresponds to positive homeo- static pressure of tissue B and yellow to a negative one. (b) Average pressure measured during the simulations shown in (a) together with a plot of Eq. (6), using the parameter of tissue B. Dashed lines are lower bounds from LA,B < Lz. ∗ ∗ ∗ y = 10;L x = L Boxsize L z = 40 Fig. 5. Cell number fractions φ for different homeostatic pres- ∗ H and different structures, as indicated by sure differences ∆P color. Circles correspond to a positive homeostatic pressure of tissues B and squares to a negative one (same parameters as in Fig. 4, except cubic box size L = 10). (b) Average pressure measured in the simulations shown in (a) formed under zero cross-adhesion and continuously in- crease the cross-adhesion fc to the value of self-adhesion (i.e. fc = f AA ). Figure 6 shows that all struc- tures remain almost unchanged up to a cross-adhesion fc approximately two thirds of the self adhesion f1. For higher fc only a mixed, sponge-like state remains. 1 = f BB 1 In summary, the interface between two tissues plays an important role in the competition between them. The enhanced growth at the interface can stabilize coexisting phases even when one tissue has a higher homeostatic pressure. The coexisting phase appears in a variety of different structures, ranging from a spherical inclusion over a flat interface to a bicontinous phase. Interesting future directions are interfacial dynamics, roughness, and shapes, as previously explored for tissues on substrates and without additional interfacial growth [34, 47, 48]. Vice versa, it would be interesting to add interfacial growth to tissues growing on substrates. Finally, our results suggests a tentative explanation for tumor heterogeneity and the abundance of occult tumors: small symptom-free micro-tumors that are fre- quently found in the human body [49]. For the theyroid, it is indeed considered 'normal' to find microscopic le- sions [50]. Our results provide a mechanical explanation ∗ Fig. 6. Variation of cell number fraction φ with time with increasing cross-adhesion fc/f1 = t /240 between two iden- tical tissues. Simulations are started from spherical (blue) and cylindrical inclusions(green) of tissue A in B as well as from flat interfaces (yellow) and a bicontinuous phase (red). Solid lines are marking transition points after which the cor- responding initial structure forms a three dimensional perco- lated cluster. Cubic box size L = 10 how coexsistence of different tissues can be stable by sim- ple mechanical effects. For example, a mutation might downregulate cadherins - an important cellular adhesion protein - as it often happens in tumors [51]. On the one hand, this might reduce survival signaling [52], but the lack of adhesion also favors our mechanism of coexistence, even for weaker tissue growth. 0.00.20.40.60.81.0Numberfractionφa)−0.20.00.2Homeostaticpressuredifference∆P∗H0.00.10.20.30.40.5PressureP∗b)PBH>0,ModelPBH<0,ModelPBH>0,SimulationsPBH<0,Simulations0.00.20.40.60.81.0Numberfractionφ(a)SpheroidCylinderFlatBicontinuous−0.20.00.2Homeostaticpressuredifference∆P∗H0.00.20.40.6PressureP∗(b)0.00.20.40.60.81.0Crossadhesionfc/f10.30.40.5CellnumberfractionφSpheroidCylinderBicontinuousFlat5 ACKNOWLEDGEMENTS and N. Bremond, Lab. Chip 17, 110 (2017). The authors gratefully acknowledge the computing time granted through JARA-HPC on the supercomputer JURECA [53] at Forschungszentrum Jlich [1] B. I. Shraiman, Proc. Natl. Acad. Sci. U. S. A. 102, 3318 (2005). [2] M. A. Wozniak and C. S. Chen, Nat. Rev. Mol. Cell Biol. 10, 34 (2009). [22] F. Montel, M. Delarue, J. Elgeti, L. Malaquin, M. Basan, T. Risler, B. Cabane, D. Vignjevic, J. Prost, G. Cappello, and J. F. Joanny, Phys. Rev. Lett. 107, 188102 (2011). [23] F. Montel, M. Delarue, J. Elgeti, D. Vignjevic, G. Cap- pello, and J. Prost, New J. Phys. 14, 055008 (2012). [24] M. Delarue, F. Montel, O. Caen, J. Elgeti, J.-M. Siaugue, D. Vignjevic, J. Prost, J.-F. Joanny, and G. Cappello, Phys. Rev. Lett. 110, 138103 (2013). [25] A. Taloni, A. A. Alemi, E. Ciusani, J. P. Sethna, S. Zap- peri, and C. A. M. La Porta, PLoS One 9, e94229 (2014). [26] G. Morata and P. Ripoll, Dev. Biol. 42, 211 (1975). [27] B. Diaz and E. Moreno, Exp. Cell. Res. 306, 317 (2005). [28] S. H. Moolgavkar and E. G. Luebeck, Genes. Chromo- somes Cancer 38, 302 (2003). [3] K. D. Irvine and B. I. Shraiman, Development 144, 4238 [29] R. Meza and J. T. Chang, BMC Public Health 15, 789 (2017). (2015). [4] M. Jarvis, S. Briggs, and J. Knox, Plant Cell Environ. 26, 977 (2003). [5] E. Coen, R. Kennaway, and C. Whitewoods, Develop- ment 144, 4203 (2017). [30] L. Hufnagel, A. A. Teleman, H. Rouault, S. M. Cohen, and B. I. Shraiman, Proc. Natl. Acad. Sci. U. S. A. 104, 3835 (2007). [31] M. Basan, T. Risler, J.-F. Joanny, X. Sastre-Garau, and [6] S. Kumar and V. M. Weaver, Cancer Metastasis Rev. 28, J. Prost, HFSP J 3, 265 (2009). 113 (2009). [32] M. Basan, J. Prost, J.-F. Joanny, and J. Elgeti, Phys. [7] D. T. Butcher, T. Alliston, and V. M. Weaver, Nat. Rev. Biol. 8, 026014 (2011). Cancer 9, 108 (2009). [33] N. Podewitz, M. Delarue, and J. Elgeti, Europhys. Lett. [8] A. Taloni, M. Ben Amar, S. Zapperi, and C. A. La Porta, 109, 58005 (2015). Eur. Phys. J. Plus 130, 224 (2015). [34] N. Podewitz, F. Julicher, G. Gompper, and J. Elgeti, [9] A. J. Engler, S. Sen, H. L. Sweeney, and D. E. Discher, New J. Phys. 18, 083020 (2016). Cell 126, 677 (2006). [10] C. M. Nelson, R. P. Jean, J. L. Tan, W. F. Liu, N. J. Sniadecki, A. A. Spector, and C. S. Chen, Proc. Natl. Acad. Sci. U.S.A. 102, 11594 (2005). [11] G. Cheng, J. Tse, R. K. Jain, and L. L. Munn, PLoS One 4, e4632 (2009). [12] J. Fink, N. Carpi, T. Betz, A. Betard, M. Chebah, A. Azioune, M. Bornens, C. Sykes, L. Fetler, D. Cuvelier, and M. Piel, Nat. Cell Biol. 13, 771 (2011). [13] M. Uyttewaal, A. Burian, K. Alim, B. Landrein, D. Borowska-Wykret, A. Dedieu, A. Peaucelle, M. Lu- dynia, J. Traas, A. Boudaoud, D. Kwiatkowska, and O. Hamant, Cell 149, 439 (2012). [14] S. J. Streichan, C. R. Hoerner, T. Schneidt, D. Holzer, and L. Hufnagel, Proc. Natl. Acad. Sci. U.S.A. 111, 5586 (2014). [15] L. LeGoff and T. Lecuit, Cold Spring Harbor Perspect. Biol. 8, a019232 (2015). [16] D. Eder, C. Aegerter, and K. Basler, Mech. Dev. 144, 53 (2017). [17] G. Helmlinger, P. A. Netti, H. C. Lichtenbeld, R. J. Melder, and R. K. Jain, Nat. Biotechnol. 15, 778 (1997). [18] V. D. Gordon, M. T. Valentine, M. L. Gardel, D. Andor- Ardo, S. Dennison, A. A. Bogdanov, D. A. Weitz, and T. S. Deisboeck, Exp. Cell Res. 289, 58 (2003). [19] L. J. Kaufman, C. P. Brangwynne, K. E. Kasza, E. Fil- ippidi, V. D. Gordon, T. S. Deisboeck, and D. A. Weitz, Biophys. J. 89, 635 (2005). [20] K. Alessandri, B. R. Sarangi, V. V. Gurchenkov, B. Sinha, T. R. Kiesling, L. Fetler, F. Rico, S. Scheur- ing, C. Lamaze, A. Simon, S. Geraldo, D. Vignjevic, H. Domejean, L. Rolland, A. Funfak, J. Bibette, N. Bre- mond, and P. Nassoy, Proc. Natl. Acad. Sci. U.S.A. 110, 14843 (2013). [21] H. Domejean, M. d. l. M. Saint Pierre, A. Funfak, N. Atrux-Tallau, K. Alessandri, P. Nassoy, J. Bibette, [35] N. Jagiella, B. Mueller, M. Mueller, I. E. Vignon- Clementel, and D. Drasdo, PLoS Comput. Biol. 12, 1 (2016). [36] P. Van Liedekerke, M. M. Palm, N. Jagiella, and D. Drasdo, Comput. Part. Mech. 2, 401 (2015). [37] Y. Kobayashi, Y. Yasugahira, H. Kitahata, M. Watan- and M. Nagayama, npj Comput. abe, K. Natsuga, Mater. 4, 45 (2018). [38] J. Ranft, M. Basan, J. Elgeti, J.-F. Joanny, J. Prost, and F. Julicher, Proc. Natl. Acad. Sci. U. S. A. 107, 20863 (2010). [39] N. Khalilgharibi, J. Fouchard, P. Recho, G. Charras, and A. Kabla, Curr. Opin. Cell Biol. 42, 113 (2016). [40] D. A. Matoz-Fernandez, E. Agoritsas, J.-L. Barrat, E. Bertin, and K. Martens, Phys. Rev. Lett. 118 (2017). [41] H. Phillips and M. Steinberg, J. Cell Sci. 30, 1 (1978). [42] K. Guevorkian, M.-J. Colbert, M. Durth, S. Dufour, and F. Brochard-Wyart, Phys. Rev. Lett. 104, 218101 (2010). [43] D. Gonzalez-Rodriguez, L. Bonnemay, J. Elgeti, S. Du- four, D. Cuvelier, and F. Brochard-Wyart, Soft Matter 9, 2282 (2013). [44] L. LeGoff, H. Rouault, and T. Lecuit, Development 140, 4051 (2013). [45] Y. Mao, A. L. Tournier, A. Hoppe, L. Kester, B. J. Thompson, and N. Tapon, The EMBO journal 32, 2790 (2013). [46] Y. Pan, I. Heemskerk, C. Ibar, B. I. Shraiman, and K. D. Irvine, Proc. Natl. Acad. Sci. U.S.A. 113, E6974 (2016). and J.-F. [47] J. Ranft, M. Aliee, J. Prost, F. Julicher, Joanny, New J. Phys. 16, 035002 (2014). [48] J. J. Williamson and G. Salbreux, (2017). [49] M. J. Bissell and W. C. Hines, Nat. Med. 17, 320 (2011). [50] H. R. Harach, K. O. Franssila, and V. M. Wasenius, Cancer 56, 531 (1985). [51] R. A. Weinberg, The biology of cancer (Garland Publish- ing, Inc., 2007). [52] Alberts, Bray, Johnson, Lewis, Raff, Roberts, and Wat- son, Molecular Biology of the Cell, 3rd edition (Garland Publishing, Inc., 1994). [53] Julich Supercomputing Centre, Journal of large-scale re- search facilities 4 (2018). 6
1703.08313	1	1703	2017-03-24T08:44:17	Transport efficiency of metachronal waves in 3d cilia arrays immersed in a two-phase flow	[ "physics.bio-ph", "physics.class-ph", "physics.comp-ph" ]	The present work reports the formation and the characterization of antipleptic and symplectic metachronal waves in 3D cilia arrays immersed in a two-fluid environment, with a viscosity ratio of 20. A coupled lattice-Boltzmann-Immersed-Boundary solver is used. The periciliary layer is confined between the epithelial surface and the mucus. Its thickness is chosen such that the tips of the cilia can penetrate the mucus. A purely hydrodynamical feedback of the fluid is taken into account and a coupling parameter $\alpha$ is introduced allowing the tuning of both the direction of the wave propagation, and the strength of the fluid feedback. A comparative study of both antipleptic and symplectic waves, mapping a cilia inter-spacing ranging from 1.67 up to 5 cilia length, is performed by imposing the metachrony. Antipleptic waves are found to systematically outperform sympletic waves. They are shown to be more efficient for transporting and mixing the fluids, while spending less energy than symplectic, random, or synchronized motions.	physics.bio-ph	physics	This draft was prepared using the LaTeX style file belonging to the Journal of Fluid Mechanics 1 TRANSPORT EFFICIENCY OF METACHRONAL WAVES IN 3D CILIA ARRAYS IMMERSED IN A TWO-PHASE FLOW S. Chateau1,2†, J. Favier1, U. D'Ortona1 and S. Poncet1,2 1Aix Marseille Univ, CNRS, Centrale Marseille, M2P2, Marseille, France 2Facult´e de G´enie, Universit´e de Sherbrooke, Sherbrooke, Qu´ebec, Canada (Received xx; revised xx; accepted xx) The present work reports the formation and the characterization of antipleptic and symplectic metachronal waves in 3D cilia arrays immersed in a two-fluid environment, with a viscosity ratio of 20. A coupled lattice-Boltzmann - Immersed-Boundary solver is used. The periciliary layer is confined between the epithelial surface and the mucus. Its thickness is chosen such that the tips of the cilia can penetrate the mucus. A purely hydrodynamical feedback of the fluid is taken into account and a coupling parameter α is introduced allowing the tuning of both the direction of the wave propagation, and the strength of the fluid feedback. A comparative study of both antipleptic and symplectic waves, mapping a cilia inter-spacing ranging from 1.67 up to 5 cilia length, is performed by imposing the metachrony. Antipleptic waves are found to systematically outperform sympletic waves. They are shown to be more efficient for transporting and mixing the fluids, while spending less energy than symplectic, random, or synchronized motions. 1. Introduction Cilia and flagella are contractile hair-like structures, put into motion by biochemical energy, and protruding on the free surface of eukaryotic or prokaryotic cells. While prokaryotes are single-celled organisms, eukaryote cells have membrane-bound organelles and are found in every mammals. Many living organisms, going from the prokaryotic bacteria to mammals, use ciliary and/or flagellar propulsion as a swimming mechanism. Usually, flagella are external appendices used by micro-swimmers such as the algae Chlamydomonas reinhardtii for locomotion purposes, while cilia are generally internal appendices, shorter and more numerous, used for moving materials such as nutrients, dusts, or proteins into living organisms. Ciliary propulsion is a universal phenomenon, and many examples could be cited. For the particular case of the human body, cilia are responsible for the left-right asymmetry of the heart in the early embryonic de- velopment, for the transport of nutrients in the brain, and for the transport of mucus in the mucociliary clearance process which is the background of the present work (see Satir & Christensen (2007) for a review about the structure and function of mammalian cilia). During the breathing process, a large number of foreign particles (bacteria, dusts, pollutants or allergens) can penetrate the organism. The human body has then developed three mechanisms to protect itself from these particles: coughing, alveolar clearance and mucociliary clearance which occurs on the epithelial surface of the respiratory system. † Email address for correspondence: [email protected] 2 S. Chateau, J. Favier, U. D'Ortona and S. Poncet In order to trap the particles, a layer of fluid called the Airways Surface Liquid (ASL) is covering the epithelial surface. Due to differences in the concentration of mucins inside the ASL, it is generally assumed to be the superposition of two different layers: the periciliary layer (PCL) and the mucus. The 7 µm depth PCL is located between the epithelial surface from where the cilia protrude and the mucus phase just above it. Mainly composed of water and of a few mucins of low molecular weight, this phase, often considered as being a Newtonian fluid similar to water, is a kind of lubricant allowing the mucus to slip on it and the cilia to beat without too much viscous resistance. The mucus is a highly non-Newtonian fluid, with a strong visco-elastic behaviour. Additionally, the inner structure of the macromolecules composing it confers to the mucus a clear thixotropic behaviour, which is currently being studied and characterised to under- stand its inner rheological properties which exhibit a large variability (Lafforgue et al. 2016). Mucus is indeed composed of 95% of water, but also contains macromolecules called the mucins (Lai et al. 2009). It serves as a physical barrier against infectious agents and dusts, but also to humidify the air flowing into the respiratory system and to catch the particles. Its height varies between 5 to 100 µm depending on many factors including the position in the respiratory system (Widdicombe & Widdicombe 1995), the pathology for a particular person, and the quality of the air inhaled. Its viscosity can also vary by several orders of magnitude within the same day (Kirkham et al. 2002). Cilia are the other protagonist of the mucociliary process. They are organized as tufts (around 200 to 300 cilia per tuft) at the epithelium surface. Formed by 9 pairs of microtubules placed regularly in a circle and 1 pair of microtubule at the center (which form the so-called axoneme), their purpose is to propel the surrounding fluid layers. Their diameter varies from 0.2 to 0.3 µm and their length from 6 to 7 µm (Sleigh et al. 1988). The cilia motion can be decomposed into two steps: a stroke phase and a recovery phase. The stroke phase is characterized by almost straight cilia orthogonal to the flow in order to maximize the pushing effect; while the cilia are bending during the recovery phase in a more inclined plane to get closer to the epithelial surface in order to minimize the viscous resistance, and therefore to reduce their impact on the flow. During the stroke phase, which takes around 1/3 of the total beating period, the tips of the cilia enter the mucus phase (Widdicombe & Widdicombe 1995). Their beating frequency is estimated to vary between 10 and 20 Hz. Note that the spatial asymmetry is essential for the cilia to generate propulsion in creeping flows, while the temporal asymmetry (recovery phase longer than stroke phase) is not necessary to induce a mucus motion (Khaderi et al. 2010). It has been experimentally observed that cilia usually do not beat randomly (Sleigh 1962), but adapt instead their beatings accordingly to their neighbours, giving birth to the so-called metachronal waves (MCW) observed at the tips of the cilia. MCW occur when adjacent cilia beat with a constant phase lag ∆Φ between each other. For 0 < ∆Φ < π, the MCW move in the opposite direction as the fluid propelled and are called antipleptic MCW. On the contrary, for −π < ∆Φ < 0, the MCW move in the same direction to the flow, and are called symplectic MCW. When the phase lag ∆Φ is null, all cilia beat in a synchronized way. Finally, when the phase lag between neighbouring cilia is ∆Φ = ±π, a standing wave appears. The universality of ciliary propulsion has intrigued scientists for decades, and several studies were conducted in order to understand it. One actual objective is to be able to mimic this process in order to, for instance, create cilia-based actuators for mixing, use them as flow-regulator in microscopic biosensors, or as micropumps for drug-delivery systems (Li et al. 2009; Chen et al. 2013). Moreover, diseases such as asthma or Chronic Obstructive Pulmonary Disease (COPD) are related to the mucociliary clearance process TRANSPORT EFFICIENCY OF METACHRONAL WAVES 3 (Gardiner 2005). The objectives are to understand the underlying mechanism that allows hundreds of cilia to act as a whole for the transport of mucus and how it affects the flow generated. The results could bring a deeper understanding in such pulmonary diseases. Numerically, Gueron and co-authors (Gueron et al. 1997; Gueron & Levit-Gurevich 1999) showed in the 90's that two neighbouring cilia beating randomly will quickly syn- chronize within a few beating cycle due to hydrodynamic interactions and form antiplep- tic MCW. This has been recently confirmed by Elgeti & Gompper (2013), who observed the formation of MCW in a 3D single-phase environment. A certain degree of freedom in the beating pattern is required as shown in the theoretical work of Niedermayer et al. (2008). They modelled the beating pattern of cilia by circular trajectories. By allowing some flexibility in the radii, they managed to introduce the coupling leading to MCW for- mation. Among the different models used for the study of ciliary propulsion, the envelope model (Taylor 1951; Reynolds 1965; Tuck 1968; Brennen & Winet 1977; Blake 1971a,b) assumed that cilia are so densely packed that it is possible to consider their tips as an oscillating surface. Nevertheless, such configuration has only been observed in nature for symplectic metachrony. Moreover, this technique is limited to small amplitude oscillations and impose no slip and impermeability conditions at the oscillating surface. In the sublayer (or stokeslets) model (Keller & Brennen 1968; Lighthill 1976; Phan-Thien et al. 1987; Blake 1972; Gueron et al. 1997; Gueron & Levit-Gurevich 1999; Niedermayer et al. 2008; Gauger et al. 2009; Smith et al. 2007; Ding et al. 2014) the cilia are modelled by a distribution of stokeslets, which impose a force on the surrounding fluid. A proper mirror image of the stokeslets is required to impose the no-slip condition on the surface from where the cilia protrude. However, the presence of a wall is known to alter the nature of the far-field of the stokeslets, and it is thought that this can have important consequences on the hydromechanics of the cilia near the wall (Blake & Chwang 1974). Moreover, stokeslets can only be used for fluids with constant viscosities. Results using this method tend to show that symplectic metachrony would be more efficient for the mucus transport than synchronously beating cilia, and that antipleptic metachrony would induce the lower flow rate. The opposite was nevertheless obtained by Gauger et al. (2009) who found that antipleptic metachrony was more efficient than symplectic metachrony for a particular cilia spacing. For a distance between two cilia of 1.5L, L being the length of the cilia, they obtained an increase in the pumping performance of 40% relative to a single cilium. However, while the beating pattern used in (Gauger et al. 2009) was realistic, they considered a slow stroke phase and a quick recovery phase which is the opposite of what is observed in nature. Some authors (Gueron et al. 1997; Gueron & Levit-Gurevich 1999; Kim & Netz 2006) showed that the energy spent by a cilium would decrease in the presence of metachronal motion. Others tried to model the internal axoneme of a cilium. Among them, Mitran (2007) used an overlapping fixed-moving grid formulation, coupling the finite volume and the finite element methods, to study the emergence of MCW. His model was very detailed and has lot of assets (two-layer flow, viscoelastic mucus) but required many experimental parameters and the impact of the MCW on the flow has not been considered so far. Niedermayer et al. (2008) observed MCW formation in 1D cilia arrays. Those waves were stable only if the wavelength was 4 times higher than the cilia spacing. For the interested reader, a clear review of those computational modelings of the internal axoneme can be found in Fauci & Dillon (2006). A different approach is to let the cilia adapt their motion in order to find the most energetically efficient ciliary beating pattern. In that context, Eloy & Lauga (2012) and Lauga & Eloy (2013) computed the shape and energy-optimal kinematics of cilia from an energetic point of view. They found that the optimal kinematics strongly depend on the cilium bending rigidity, and closely resemble the two-stroke ciliary beating pattern observed in natural 4 S. Chateau, J. Favier, U. D'Ortona and S. Poncet cilia. Similarly, Osterman & Vilfan (2011) computed the ciliary beating pattern with optimal pumping efficiency of isolated cilia and arrays. Recently, new methods have been introduced such as the immersed boundary method (IB) used by Dauptain et al. (2008) to model the swimming of the pleurobrachia. Lukens et al. (2010) used an IB method to study the mixing produced by a carpet of cilia in the context of the mucociliary clearance process. A coupled immersed-boundary lattice Boltzmann method (LBM) was also used by Sedaghat et al. (2016) to study several parameters in a 2D configuration using an Oldroyd-B model for the mucus rheology. They found that the transport of mucus was maximized when considering the mucus as a Newtonian fluid. While many studies regarding the emergence of MCW were conducted, only few addressed 3D configurations of the mucociliary clearance process. Among them, Elgeti & Gompper (2013) managed to observe symplectic and laeoplectic (perpendicular to the power-stroke direction) MCW formations. Ding et al. (2014) did not study the emergence of MCW but performed a comparative study of antipleptic metachrony versus symplectic metachrony in terms of transport efficiency and mixing. Their results showed that both antipleptic and symplectic MCW enhanced the fluid transport and the mixing, with the antipleptic waves being the most efficient ones. However, the two aforementioned studies only considered a single fluid layer. In this work, by using the solver developed by Li et al. (2016) and already validated in similar configurations, the focus is placed on the emergence of MCW in a 3D two-phase flow configuration with a viscosity ratio of 20 and a large number of cilia. In particular, it is shown that a simple hydrodynamical feedback based on mechanical concepts can trigger the emergence of both symplectic and antipleptic MCW, while usually only a single layer of fluid is considered and only antipleptic MCW are seen to emerge. From a numerical point of view, the main advantage of the present method is its ease of implementation. Additionally, the local character of the collisions in the LBM allows an easy and straightforward parallelization of the code, making the simulation of a large number of cilia possible. The numerical method also possesses the following advantages: (i) viscosity ratios up to O (102) can be achieved (Porter et al. 2012), and (ii) the mucus- PCL interface emerges intrinsically from the model. To the knowledge of the authors, this solver is the only one that combines all these capabilities. The main contribution of this work is the thorough analysis of the advantages of the antipleptic and symplectic MCW over synchronized beatings by computing appropriate transport and efficiency ratios, mapping an inter-cilia spacing going from 1.67 to 5 cilia length. Finally, and for the first time, both the PCL and the mucus layer have been taken into account in the present study. The inherent advantages of the MCW for flow transport are studied by (i) considering the efficiency of the waves to transport the flow, (ii) comparing the flow generated and the volume transported by the different kinds of metachrony, (iii) comparing the energetic cost of the MCW, and (iv) analyzing the capacity of the waves to transport particles from an energetic point of view. The following analysis shows that antipleptic MCW are always the most efficient to transport mucus. The remainder of this manuscript is organized as follows. In Section 2, details about the numerical method are given, and the different quantities used are introduced. In Section 3, the results are presented starting from the emergence of MCW by considering the fluid feedback onto the cilia; and then a parametric study is done to quantify the impact of both the MCW and the cilia spacing on the transport of mucus. A summary of the results and the future perspectives conclude this manuscript in Section 4. TRANSPORT EFFICIENCY OF METACHRONAL WAVES 5 (a) (b) Figure 1. (a) Schematic view of the computational domain. The present case corresponds to an antipleptic MCW. The domain is filled with PCL (in blue) and mucus (in red). (b) Beating pattern of a cilium with the parametric equation used. Steps 1 to 6 correspond to the recovery phase, and steps 7 to 9 to the stroke phase. 2. Numerical method From the numerical point of view, modeling the complete problem under realistic conditions remains a huge challenge and several assumptions need to be made to simplify the problem, while keeping the essential ingredients of the physical mechanisms in the model. In particular, it is assumed that there is no mass transfer at the wall, that both fluids are Newtonian and that the cilia are equally spaced filament-like structures. More assumptions are detailed in this section, together with the numerical setup and geometry. 2.1. Geometrical modeling and beating pattern The computational domain is a box with a regular mesh composed of Nx × Ny × Nz points. The cilia are equally spaced along the bottom (x-y) wall, such that their base points are located at z = 0 (see figure 1(a) for a schematic view of the geometry). Their motion is imposed to be in the x-direction only. The spacing between two adjacent cilia is denoted a in the x and y-directions. The wavelength λ of the MCW is set such as λ = Nx when the metachrony is forced. For the case of synchronously beating cilia, Nx is chosen to be 8a. The length of the cilia L is set to 15 lattice units (lu) except when stated differently; and 200 snapshots in time per beating cycle are uniformly distributed to model their motion. The ratio h/H between the PCL thickness and the height of the domain is fixed to 0.27 for all simulations. The equations of motion for the cilia are inspired from Chatelin (2013) and reproduce the beating pattern by resolving a 1D transport equation along a parametric curve. Let P (ζ, t) be the position of the curve at time t and at a normalized distance ζ from the base point of a cilium. With appropriate boundary conditions, a realistic beating pattern can be obtained using the following transport equation: ∂P ′ ∂t + ν(t) ∂P ′ ∂ζ = 0 BC:(cid:26) P (0, t) = (0, 0, 0) P ′(0, t) = (2 cos(2πt/T ), 0, cos(2πt/T )) (2.1) with ν(t) = [1 + 8 cos2(π(t + 0.25T )/T )]/T being the viscosity of the surrounding fluid, T the beating period, and P ′ = ∂ζP . The resulting angular amplitude between the beginning and the end of a stroke phase is θ = 2π 3 , which agrees well with experimental data (Sleigh et al. 1988). Note that a 3D beating pattern would allow to achieve more realistic simulations while being more computationally expensive. Hence, the choice has been made to use this 2D beating pattern. It captures the essential ingredients of the beating, and as cilia have a diameter smaller than a lattice unit, the difference of induced flow between two cilia overlapping in 2D or slipping onto each other in 3D is very small. Figure 1(b) gives a view of the beating pattern obtained by resolving Eq. (2.1). Note that 6 S. Chateau, J. Favier, U. D'Ortona and S. Poncet with the present model both phases take the same amount of time while moving in the same (x,y) plane. This choice has been made in order to only study spatially asymmetric motions, which are the only mechanisms effective at low Reynolds numbers. The temporal asymmetry is indeed a mechanism that can enhance the flow only when inertial forces are no longer negligible (Khaderi et al. 2010). Nevertheless, when the feedback of the fluid is taken into account (as in §3.1), a non symmetrical motion will develop, with a stroke phase slower than the recovery phase. More details regarding this temporal asymmetry are given in §2.3. The PCL is set such that it fills the region going from the bottom of the domain (z = 0) up to an altitude of h. In all the simulations, the value of h = 0.9L has been used in order to allow the tips of the cilia to emerge into the mucus layer during their stroke phase as observed in real epithelium configurations. Both PCL and mucus are considered to be Newtonian fluids. The kinematic viscosity of the mucus is νm = 10−3 m2/s, and the viscosity ratio rν = νm/νP CL between the mucus and PCL is set to 20. It has indeed been recently shown (Chatelin & Poncet 2016) that mucus transport was maximized for viscosity ratios ranging from 10 to 20 with a stiff transition between the two fluid layers. The beating period of the cilia is equal to Nit × dt (with dt = 1 using the classical LBM normalization), Nit being the number of iterations for a cilium to perform a complete beating cycle. An oscillatory Reynolds number Reosc, based on the velocity of the cilia's tips Ucil = 2θL Tosc can now be defined: Reosc = UcilL νmucus = 4π 3 L2 Tosc νmucus = ωL2 νmucus (2.2) with ω being the angular frequency of the cilia beating. With realistic physical quantities corresponding to the ciliated epithelium surface, the value of Re osc is of the order of 10−5 (using L ≈ 10−5 m, νmucus ≈ 10−3 m2/s and Ucil ≈ 10−3 m/s). To avoid running simulations at such a low Reynolds number which would require a very high number of iterations using a lattice Boltzmann scheme, a higher Reynolds number was chosen: Re = 20. Indeed, achieving simulations at Re < 1 using a LB scheme, while still describing well the cilia (L = 15 lu at least), requires a lot of CPU resources. Thus, inertial effects are introduced in the model but it has been carefully checked that the results remain the same for creeping flows (Re = O(10−2)) by comparing them to the results obtained with LBM formulation designed for Stokes flows (Guo & Shu 2013; Zou et al. 1995). The differences are found to be less than 7% on the transported velocity for all phase lags ∆Φ 6= 0. The corresponding beating patterns of the cilia shown in the following are similar also in terms of vorticity generation and the same qualitative coordination concerning the antipleptic/symplectic behaviour is observed. However, for the particular case of synchronously beating cilia (i.e. ∆Φ = 0), the inertial effects will play a non-negligible role in the flow dynamics. Despite being weak, they will indeed cancel the reversal of the flow that should occur when the cilia are in the recovery phase. More details regarding the inertial effects will be given in §3.2.7. Finally, it is worth noticing that the lattice has been chosen such that the numerical diffuse diameter of the cilia due to IB method corresponds to real cilia diameter (≈ 0.3µm). Hence, a realistic drag is taken into account in the present study. 2.2. Algorithm The numerical model is described in Li et al. (2016), and validated on several config- urations involving flexible and moving boundaries in multiphase flows, with a 2nd order accuracy. Briefly, the idea is to add a forcing term F σ to the discrete i = F SC σ + F IB σ TRANSPORT EFFICIENCY OF METACHRONAL WAVES 7 LB equation for each σ fluid component. F SC takes into account the fluid-fluid cohesion forces (Shan & Chen 1994), and F IB σ IB-related force for ensuring the no-slip condition at the fluid-solid interface. is an interparticle potential force that is an σ The fluid part is first solved on a Cartesian grid with LBM using the Bhatnagar -- Gross -- Krook (BGK) operator and a D3Q19 scheme. The collision and streaming steps proper to the LBM method are first performed. The model of Porter et al. (2012) is used to model the two-phase flow as it allows to minimize the magnitude of spurious currents near the fluid-fluid interface. More importantly, it also allows to consider higher density or viscosity ratios. Then, values for the fluid velocity are interpolated at the Lagrangian points. It allows to compute an IB force to be spread onto the neighbouring Eulerian fluid nodes in order to ensure the no-slip condition along the cilia. The macroscopic fluid velocity is then updated. The motion of each cilium is decomposed into a finite number of steps (snapshots) during a period. If necessary, an interpolation can be done in order to have the velocity values along the cilia in between two steps. Note that the geometric shape of the beating is fixed in all simulations and is not impacted by the feedback law introduced in §2.3, which only affects the duration of the recovery and stroke phases. Since the model of Porter et al. (2012) uses a Shan-Chen (SC) repulsive force (Shan & Chen 1993, 1994), surface tension effects emerge intrinsically at the PCL- mucus interface. Hence, a sharp interface between the mucus and the PCL can be maintained at any time. However, small diffusion effects might occur on the lattices bordering the interface. Additionally, the cilia which enter the PCL-mucus interface may punctually induce a small mixing. This is however corrected by the SC force which "unmixes" the two fluids. Periodic boundary conditions are used in the x and y directions, while no-slip and free-slip boundary conditions are used at the bottom and top walls respectively. The size of the computational domain ranges from 50 lu to 400 lu depending on the configuration considered, except for the size along the z direction which is always set to 50 lu. Taking advantage of the local character of the LBM algorithm, the code is parallelized using MPI libraries (Message Passing Interface), by splitting the full computational domain into 9 sub-domains of size (Nx/3, Ny/3, Nz). More details on the numerical model can be found in Li et al. (2016). 2.3. Feedback of the fluids onto the cilia The basic idea is to modulate the beating motion of the cilia as a function of the fluids motion. To do so, it is assumed that all cilia follow the same beating pattern, meanwhile a feedback of the fluids, which consists in accelerating or slowing down the motion of the cilia, is introduced. Each cilium is discretized with Ns = 20 Lagrangian points. Let s be the subscript corresponding to the sth Lagrangian point, starting from the base tip at s = 1, and V s i the velocity on the sth Lagrangian point of the ith cilium. For each cilium, we define the average velocity over all Lagrangian points V i, which is linked to the number of steps (snapshots) this cilium will skip during one iteration of the fluid solver. The fluid feedback onto the cilia thus consists in modifying the norm of the velocity vector kV ik, while its direction remains unchanged. The feedback is computed in three steps. First, the IB forces corresponding to mucus and PCL are projected onto the corresponding velocity vectors for each Lagrangian point. Then, an estimate of the feedback is computed based on the torques of the forces for each Lagrangian point. Finally, the beating pattern of the cilia is adjusted at the beginning of the next time step. The different forces and geometrical variables used are illustrated in figure 2, and the three steps explained below. 8 S. Chateau, J. Favier, U. D'Ortona and S. Poncet (i) The interpolated IB forces applied by the ith cilium onto the fluids - respectively m for the force imposed on the mucus phase and F i P CL for the force imposed on the F i PCL phase-, are projected on V s i : F i,proj ∗ = ∗ · V s F i i kV s i k2 V s i , (2.3) where ∗ stands for "m" or "PCL" depending on the position of the Lagrangian node. In order to take into account the difference of viscosity between the two layers, the forces F i,proj . The total projected force of the fluids onto the sth segments of the ith cilium F i P CL are weighted by a term under the form: τm+τP CL f luids→cilia writes: and F i,proj τ∗ m F i f luids→cilia = − (τmF i m + τP CLF i (τm + τP CL)kV s P CL) · V s i i k2 V s i (2.4) (ii) For each Lagrangian point s, the norm of the torque of F i f luids→cilia with respect to the base point O of the ith cilium is computed by: kMi O(F i f luids→cilia)k = kF i f luids→ciliakLp (2.5) with Lp = kX p⊗V s i k being the lever arm. The total fluid feedback onto the ith cilium i k considered is then computed by summing the computed quantities over all Lagrangian points: kV s T i = Nsegments Xs=2 kF i f luids→ciliakLp The velocity of each cilium is finally modified as follows: kV ik = kV0k + αT i (2.6) (2.7) where V 0 is the initial speed of the cilia, and α a coupling parameter allowing the tuning of the feedback strength. (iii) Then, at the beginning of the next iteration, the beating pattern of the ith cilium is adjusted: N t i = mod (N t−1 i + kV ik, N total i ) (2.8) i where N total N t−1 number of snapshots) at the current iteration. the previous snapshot, and N t is the total number of snapshots defining the beating pattern of the cilia, i the new position of the ith cilium (in terms of i 2.4. Injection of passive tracers In the context of real epithelial systems in human body, mucus acts as a barrier against particles and pollutants. To gain insight on how particles are dispersed and advected in the mucus and PCL layers, passive tracers are injected into the domain on a (y,z) plane, when the flow has reached an established regime (see figure 3 for a view of a domain filled with tracers). Their displacements are then computed and averaged over several cilia beating cycles. Their equations of motion are solved by a second-order Runge-Kutta (RK2) scheme, using the interpolated fluid velocity at each time step and the same procedure as in the IB method. By taking a closer look at figure 3, one can observe that particles initially seeded into the PCL are greatly mixed while staying in the PCL. On the contrary, particles initially TRANSPORT EFFICIENCY OF METACHRONAL WAVES 9 Figure 2. Schematic view of a cilium with the corresponding forces exerted on the fluids. Figure 3. Domain filled with passive tracers. The present case corresponds to an antipleptic MCW with ∆Φ = π/4. An array of 8 cilia with a cilia spacing a/L = 1.67 is considered on a computational domain of size (Nx = 201, Ny = 26, Nz = 50). The mucus phase (in red) is located above the PCL phase (in blue), and the ratio of viscosity is set such as rν = 10. The colorbar represents the dimensionless velocity magnitude of the fluids on the periodic boundary over the x-direction. seeded into the mucus layer stay in the mucus layer and are not mixed. This is mainly due to the surface tension effects present at the mucus-PCL interface which prevent a too strong mixing at the interface. This shows how wetting particles that are deposited to the air-mucus interface might enter the mucus layer but never reach the PCL one. 3. Results & Discussion 3.1. Emergence of metachronal waves Using the feedback force introduced in §2.3, randomly beating cilia are observed to synchronize with their immediate neighbours giving birth to symplectic or antipleptic 10 S. Chateau, J. Favier, U. D'Ortona and S. Poncet Figure 4. Symplectic MCW emerging from an initially random state of the cilia. 24 cilia arranged in a 8 × 3 rectangle are considered on a computational domain of size (Nx = 361, Ny = 136, Nz = 50) with a cilia spacing a/L = 3. The mucus phase is in red and the PCL phase in blue. The color bar indicates the phase of a particular cilium within one beating period, which is represented by a circle at its base. Top: 3D view of the system. Bottom: 2D view of the same system in a (x,y) plane to highlight the 3D modulation in the z-direction. metachrony. The time for the synchronization to occur depends on the set of parameters used. Both local and global synchronizations can be observed. The empirical parameter α plays a role in the emergence of the waves. When α is set to a too low value (α < 0.5), the cilia will beat randomly, while if set to a too high value (α > 7 for the case a/L = 1.67 for example), the cilia will fully synchronize with each other without any phase lag. Additionally, higher absolute values of α usually decrease the times needed for metachronal waves to emerge. Figure 4 shows a symplectic MCW emerging from an initially random state (every cilium were initially beating at a random step of their beat cycle). In the present case, 3 rows of 8 cilia on a computational domain of size (Nx = 361, Ny = 136, Nz = 50) are considered. The PCL is set such as h = 0.9L, and the ratio of viscosity between the PCL and the mucus phase is 15. The feedback coefficient α used is α = −3.5. The spacing between two neighbouring cilia is a/L = 3 in the x and y directions with L = 15 lu. Hence cilia never collide and overlapping of kernels from neighbouring cilia never occurs. One can clearly see the symplectic MCW that emerged from the initially random state of the cilia, with a wavelength λ = 180 lu, and a phase lag ∆Φ ≈ −π/2, confirming that hydrodynamic interactions only suffice to account for the emergence of MCW. Figure 5 shows a similar configuration (h = 0.60L, rν = 15, α = −3.5) where 1024 cilia arranged in a 32 × 32 square are considered on a computational domain of size (Nx = 161, Ny = 161, Nz = 32). One can clearly see the formation of an antipleptic MCW with a wavelength λ = 80 lu, and a phase lag ∆Φ ≈ π/8. Notice that in the simulation presented in figure 5 the cilia spacing has been reduced (a/L = 0.23 and L = 22 lu), to take into account a larger number of cilia. Thus, neighbouring cilia may overlap, which can cause spurious numerical effects, due to the 2D beating pattern of the cilia, when two Lagrangian points with opposite velocities occupy the same mesh point. However, it has been verified on some simulations, that these effects do not play any role in the emergence of the MCW by setting to zero the IB forces contribution of these Lagrangian points, therefore canceling the spurious effects that may occur at TRANSPORT EFFICIENCY OF METACHRONAL WAVES 11 the overlapping points of neighbouring cilia. In other words, the fluid velocity at the Eulerian node surrounding these particular Lagrangian points was not modified by the IB method, proving that it is not the source of the cilia synchronization. As shown in figures 4 and 5, weakly 3D effects may be observed but do not play a key role in the physics of emerging waves. In all the simulations presented, the synchronization along the x-direction is quite strong. In some simulations, particular cilia may lose synchronization over time, but quickly readjust their beating accordingly to the others. It also appears that once an equilibrium is reached, the synchronization along the y-direction is stable too. A remarkable fact is that for particular configurations (as the one displayed in figure 5), the interface between the mucus and the PCL also forms a wave traveling in the same direction as the MCW. If the mucus viscosity is increased, the torques felt by the cilia become stronger. Hence, the cilia are either more accelerated (α > 0) or decelerated (α < 0), and the mucus will adapt its displacement accordingly. However, when the stationary regime is reached, the clearance velocity should not be strongly modified as shown by Chatelin & Poncet (2016). According to these authors, the mucus velocity is decreased by only 50% when the viscosity ratio is increased by a factor of 100000. Preliminary results (not shown) seem to indicate that, assuming a least effort behaviour of the cilia (meaning α < 0), antipleptic waves are obtained for small cilia spacings (a/L 6 1.5) while symplectic waves are seen to emerge for higher cilia spacings (a/L > 1.5). Since in nature, antipleptic waves are often observed for densely packed cilia, it implies that natural cilia adopt a least effort behaviour. It also suggests the existence of a critical value for the cilia density, ρc = ρ(αc) where αc is the corresponding critical value of the coupling parameter, from which the kind of waves emerging can be controlled. On the contrary, with the present model, assuming that the cilia beat faster when encountering a resistance (meaning α > 0), symplectic waves are seen to emerge for small cilia spacings (a/L 6 1.5) while antipleptic waves emerge for higher cilia spacings (a/L > 1.5), which is not observed in nature. Movies of the MCW are provided as online supplementary materials. As already explained in §2.1, when the feedback is taken into account, a non sym- metrical motion develops, with a stroke phase slower than the recovery phase for both antipleptic and symplectic MCW. For the antipleptic case, kV ik is smaller during the stroke phase, and since α is negative and the torques T i are always positive, it means that the values of T i computed during the stroke phase are larger than the values computed during the recovery phase. It results in a weaker velocity for the cilia which cover less snapshots during the stroke phase. It is the opposite for the symplectic case where α is positive: T i takes larger values during the recovery phase as kV ik. In other terms, the feedback depends on the clustering character of the motion: when cilia are clustered, the torque exerted by the fluids onto the cilia is weaker, and when they are far from each other (during the stroke phase for antipleptic motion and during the recovery phase for symplectic motion), the torques are stronger. This makes sense since the cilia encounter more viscous resistance when they are not clustered, as it will be discussed further down in §3.2.4. The resulting motion of the cilia then differs from what is observed in nature, but it indicates that cilia experience stronger stresses during the stroke phase of antipleptic motion and during the recovery phase of symplectic motion. It supposes that the beating kinematics of the real cilia are not dictated only by hydrodynamical interactions and suggests that other biological parameters or functions (such as sensing) may play a role. This is in agreement with Guo et al. (2014) who compared the performance of pumping-specialized cilia and swimming-specialized cilia as a function of the metachronal coordination, and found that the later almost always outperforms the pumping-specialized cilia. As it will be further detailed in §3.2.5, their 12 S. Chateau, J. Favier, U. D'Ortona and S. Poncet Figure 5. Antipleptic MCW emerging from an initially random state of the cilia. 1024 cilia arranged in a 32 × 32 square are considered on a computational domain of size (Nx = 161, Ny = 161, Nz = 32) with a cilia spacing a/L = 0.23. The mucus phase is in red and the PCL phase in blue. The color bar indicates the phase of a particular cilium within one beating period, which is represented by a circle at its base. Top: 3D view of the system. Bottom: 2D view of the same system in a (x,y) plane to highlight the 3D modulation in the z-direction. results are also in accordance with the outcome of the present paper: antipleptic waves are the most efficient ones to transport fluids. Finally, it is worth noticing that the degree of asymmetry is much higher for antipleptic MCW compared to symplectic MCW. 3.2. Quantitative study of the metachronal waves This section presents the results obtained, once the flow is well-established, for the three configurations studied: (a) synchronized case (all cilia beat together with no phase lag); (b) symplectic MCW where two neighbouring cilia beat with a negative phase lag, i.e −π < ∆Φ < 0; and (c) antipleptic MCW where two neighbouring cilia beat with a positive phase lag, i.e 0 < ∆Φ < π. Note that in all following results, contrary to §3.1, the metachrony is imposed in order to study specific phase lags ∆Φ. Hence, the size of the domains must be changed accordingly, since different phase lags ∆Φ imply different wavelengths, and hence more or less cilia. However, the quantities presented here have been averaged over the domain, and there are no effects in changing the size of the box. In order to study only spatially asymmetric motions, the recovery and stroke phases now take the same amount of time in all following results. Also note that the standard deviations are not displayed in any of the figures that will be presented as they are extremely small (less than 0.01%) for all considered quantities. Hence they do not give additional information. Schematic views of synchronized beating and metachronal motions are displayed in figure 6. Another configuration, corresponding to randomly beating cilia is also represented in figure 6. One can observe that the synchronized motion of cilia creates vorticity only in the periciliary zone, whereas metachronal motions also induce vorticity into the mucus layer. It is worth noticing that for symplectic motion TRANSPORT EFFICIENCY OF METACHRONAL WAVES 13 (a) ∆Φ = 0 (b) ∆Φ = −π/4 (c) ∆Φ = π/4 (d) Random Figure 6. Different kinds of collective coordination for the beating cilia. 32 cilia arranged in a 8 × 4 square are considered on a computational domain of size (Nx = 241, Ny = 121, Nz = 50) in each case, with a cilia spacing a/L=2. (a) Synchronized motion; (b) Symplectic metachronal motion; (c) Antipleptic metachronal motion; (d) No synchronization (random state of cilia). The figures show contours of the magnitude of the dimensionless vorticity k~ωk using a logarithmic scale. The black lines show the frontier between the PCL at the bottom and the mucus layer above. (case (b)), vortex trails emerge from the cilia tips performing their stroke phase. The same phenomenon is observed in the antipleptic case during the recovery motion of cilia. The presence of coherent vortices is clearly visible in both cases. 3.2.1. Transport and mixing zone 0 The displacement field is calculated such as d(x)=R T u(x(t), t)dt, where x is the position vector, and u the fluid velocity. Its component over the x-direction as a function of the axial position is averaged over several beating cycles and displayed in figure 7. The transport and mixing zones defined in Ding et al. (2014) are clearly visible. It shows that the antipleptic MCW perform better than the other kind of coordination to transport the particles. Note that in figure 7b, the two curves with ∆Φ = ±π/4 have a similar shape, whereas in figure 7a symplectic MCW induce a negative transport (i.e. a counter-flow) in the mixing zone for the same phase lag. For a given value of phase lag, the transport depends on the density of cilia. One can also notice the importance of the phase lag 14 S. Chateau, J. Favier, U. D'Ortona and S. Poncet (a) (b) Figure 7. Normalized average displacement in the x-direction as a function of z/L for (a) ∆Φ = ±π/9 and ∆Φ = ±π/4 with a/L = 0.8; (b) ∆Φ = ±π/4 and a/L = 1.67. As in Ding et al. (2014), the mixing zone extends from 0 to 1L, and the transport zone from 1L to 1.4L. In both cases (a) and (b), the mucus-PCL interface is located at z/L = 0.9 and is indicated by a dashed horizontal line. by looking at figure 7a: for a phase lag ∆Φ = −π/9, the symplectic MCW induce no counter-flow in the mixing zone while the opposite happens for ∆Φ = −π/4. Finally, it is worth noticing that, as the cilia spacing increases, the influence of the kind of metachrony decreases. Eventually, for very large cilia spacings, one expects that the 3 curves displayed in figure 7b would merge into a single one corresponding to the displacement generated by an isolated cilium. Additionally, even with a larger cilia spacing, the particles are better transported in the case (b). As in Ding et al. (2014), the displacement reaches a plateau for an axial position of 1.4L in both cases (a) and (b). The beating pattern used in Ding et al. (2014) is different, but similar trends are observed. Nevertheless, a major difference must be pointed out: for ∆Φ = −π/4 and a/L = 1.67 (figure 7b), the present results show that the displacement induced by the symplectic wave has the same shape as the synchronized and antipleptic cases, i.e it induces transport in the mucus phase even if it is smaller. On the contrary, Ding et al. (2014) obtained no transport at all for the symplectic wave in the mucus phase, but instead a peak of transport under the cilia tips in the PCL at z = 0.7L due to the presence of a vortex-like structure both below and above the cilia tips for this particular value of the phase lag ∆Φ. Since the main difference between both studies is the beating pattern of the cilia (same phase lag and cilia spacing), it highlights the sensitivity of the system to the beating pattern. The displacement in the z and y-directions have also been analyzed. The results show that the y-component of the displacement could be neglected and that the z-component of the displacement was almost null above the cilia tips, but not zero under. As it will be discussed later, the z-component of the displacement reaches its maximal value in regions where the shear rate is maximal too. It agrees with the existence of a mixing zone under the cilia tips. By doing an analogy with the strain-rate tensor classically used in solid mechanics, the gradient of the displacement field ∇d can be computed by considering each Eulerian node of the Cartesian grid as passive tracers. Since the displacement in the transverse direction is very small, the y-component of the displacement field can be neglected. Henceforth, the gradient is computed over every (x,z) plane, and an average is then done over the y-direction. Following the methodology described in Ding et al. (2014), ∇d is decomposed into an antisymmetric component R = (∇d − (∇d)T )/2 corresponding to rotation, and a symmetric component S = (∇d + (∇d)T )/2 corresponding to shear TRANSPORT EFFICIENCY OF METACHRONAL WAVES 15 deformation. The two eigenvalues of S are of the form ±γ and indicate the rates of stretching (+γ) and compression (−γ). The unit eigenvector eγ corresponding to the positive eigenvalue indicates the direction of stretching, and the other eigenvalue the direction of compression. Because of the incompressibility condition, both eigenvectors are orthogonal, and so plotting only one of them is sufficient to have the complete set of information. In figure 8, the stretching rate and its direction are presented in a (x,z) plane for the three cases studied. The stretching rate is maximal near the upper part, and at little distance above the cilia tips in all three cases. For the synchronized motion (case (a)) there is almost no stretching away from the cilia during the recovery phase whereas for cases (b) and (c), corresponding respectively to symplectic and antipleptic motion, a weak stretching can always be observed (see the right side of cases (b) and (c) in figure 8). A complex shape of the stretching rate is observed at the cilia tips during the stroke phase in antipleptic motion (see cilium 3 for instance); during the stroke phase in symplectic motion (see cilia 6 and 7) and during the stroke phase in synchronized motion (results not shown). From their orientations, one can expect an enhancement in the mixing in this region. As in Ding et al. (2014), the stretching direction is a nonlinear function of space. Except for zones where the shear rate is maximal, the γ~eγ field is then oriented at 45◦ compared to the x-direction, and almost uniform. This is reminiscent of a linear shear profile dx = cz where c is a constant. The nonuniform aspect of the stretching orientation indicates the presence of "folding" in the displacement field d which also plays a role in the mixing as explained by Kelley & Ouellette (2011). Ding et al. (2014) obtained that the stretching rate was maximal for the antipleptic case for ∆Φ = π/4 which is exactly what is obtained here, even if the beating patterns used are different. One can then expect that this value of phase lag ∆Φ is the most efficient in mixing fluids by enhancing the stretching near the upper part of the cilia. In order to assess the reliability and robustness of the solver, a comparison with experimental data is also performed. The average clearance velocity was computed on the plane (x,y,3.2L) for all simulations. The highest clearance velocity is reached for the case a/L = 1.67 with ∆Φ = π/4 and equals 33.47 µm/s. It is in good agreement with the experimental results of Matsui et al. (1998) who observed a clearance velocity of 39.8 ± 4.2 µm/s for the mucus. Matsui et al. (1998) also observed that the PCL flows in the same direction as the mucus, which is the case in the present study. Finally, the PCL-mucus interface remains approximately flat in most of the simulations presented, as it is the case in the micrographs performed by Sanderson & Sleigh (1981) on rabbit tracheal epithelium. 3.2.2. Directional pushing efficiency Inspired by the works of Khaderi et al. (2011), Gauger et al. (2009), and Kim & Netz (2006), a positive flux Qp and a negative flux Qn are defined as follows: the x-component of the velocity field is considered over the (Nx,y,z) plane and, at each time step of a full beating cycle, the negative and positive velocity values are separated. The instantaneous negative and positive fluxes are then computed over a sufficiently large number of cycles. The difference (Qp − Qn) gives the net instantaneous flux; and the directional pushing efficiency ǫP N is defined as : ǫP N = (Qp − Qn)/(Qp + Qn). Note that here, contrary to the previous works of the aforementioned authors, the domain is not restricted to the transport area but covers the whole flow region instead. Such a choice is justified by the fact that it has been experimentally observed, using confocal microscopy and fluorescent markers, that PCL and mucus are in reality transported at approximately the same rate (Matsui et al. 1998). As a matter of fact, the PCL transport seems to depend on the presence of a mucus layer above it, and ciliary mixing is thought to be responsible for 16 S. Chateau, J. Favier, U. D'Ortona and S. Poncet (a) (b) (c) Figure 8. Stretching rate and direction for: (a) Synchronized motion (∆Φ = 0), (b) Symplectic MCW (∆Φ = −π/4), (c) Antipleptic MCW (∆Φ = π/4). In each case, the cilia spacing is a/L = 1.67, and the size of the computational domain is (Nx = 201,Ny = 26,Nz = 50). the diffusion of momentum from mucus to PCL. Taking into account the transport zone and the mixing zone in the computation of ǫP N then allows one to collect information about how transport and mixing work together during a beating cycle. The variation of ǫP N over one beating cycle for five different phase lags ∆Φ, all for antipleptic MCW with a cilia spacing such as a/L = 2, is illustrated in figure 9. Let us remind that, although the quantities considered here are the results of the motion of all cilia, the calculations have been made through the beating cycle of a single cilium of reference. Therefore, the fact that the corresponding curves have a huge decrease around t = π is related to the choice of this particular cilium. Indeed, the plane used to compute the positive and negative fluxes is located close to a particular cilium. This cilium is in the recovery phase at t = π, inducing a small value of ǫP N at this particular time. Choosing another cilium or another plane would shift horizontally the drop observed around t = π. To interpret the evolution of ǫP N in figure 9, it is necessary to examine the topology of the flow during this beating cycle, shown in figure 10. As cilium 1 begins its recovery phase (case (a) of figure 10), there is no reversal flow at the periodic boundary frontier. Hence, ǫP N = 1, as it is visible in figure 9 at the point (a). Then, when cilium 1 carries on its recovery phase (case (b) of figure 10), no reversal flow can be observed since cilium 1 is still at the beginning of its recovery phase and cilium 8 is finishing its stroke phase. TRANSPORT EFFICIENCY OF METACHRONAL WAVES 17 Figure 9. Directional pushing efficiency ǫP N over one beating cycle for three different phase lags ∆Φ. Results obtained for an antipleptic MCW and a/L = 2. At t = π/2 (case (c) of figure 10), both cilia 1 and 8 are in their early recovery phase and the directional pushing efficiency begins to decrease (see points (b) and (c) on figure 9). At t = 3π/4 (case (d) of figure 10), cilium 1 is reaching the maximal speed of its recovery phase. The negative flow generated is important, and the efficiency quickly drops: see point (d) on figure 9. Between t = 3π/4 and t = π, the directional pushing efficiency weakly increases. This is due to the positive velocity near the base of cilium 1 which is doing a "whip-like" motion: while its free end is still going to the left, its base starts moving on the right. Case (e) of figure 10 corresponds to the end of the recovery phase of the first cilium and at the same moment, cilium 8 is at its maximal negative velocity: henceforth the directional pushing efficiency is still decreasing. At t = 5π/4 (case (f) of figure 10), cilium 8 has finished its recovery phase and cilium 1 is in the middle of its stroke phase. It counteracts the reverse flow created earlier and ǫP N finally increases. At t = 3π/2 and t = 7π/4 (cases (g) and (h) of figure 10), both cilia 1 and 8 are in their stroke phases, and the flow generated is purely in the positive x-direction: see points (g) and (h) on figure 9. It is now interesting to compare the average directional pushing efficiency < ǫP N > of the different kinds of synchronization over a full beating cycle (see figures 11 (a) and (b)). Before going further into details, let us recall the fact that the current efficiency < ǫP N > is a criterium qualifying the non-isotropy of the transport. Thus, it only gives an insight of the capacity of the cilia to transport the flow in an unidirectional way, but does not give any quantitative information on the volume of fluid flow, which is effectively displaced. Figures 11 (a) and (b) show an interesting phenomenon: systems with larger cilia spacing have a better capacity to transport the flow in the same direction. Another intriguing fact is that, for the smallest cilia spacing (a/L = 1.67, figure 11), antipleptic MCW seem to have a better ability to create an unidirectional flow compared to synchronized or symplectic motion, with a peak for the average efficiency around π/2. It agrees particularly well with the results of Gauger et al. (2009) who reported that antipleptic MCW were more efficient than the synchronized motion of cilia, itself being more efficient that the symplectic case. It also partially agrees with the results of Ding et al. (2014) who obtained peaks of efficiency for phase lags around ±π/2, with a stronger one for the antipleptic case. In the present study, the minimal efficiency for 18 S. Chateau, J. Favier, U. D'Ortona and S. Poncet (a) t = 0 (b) t = π 4 (c) t = π 2 (d) t = 3π 4 (e) t = π (f) t = 5π 4 (g) t = 3π 2 (h) t = 7π 4 1 2 3 4 5 6 7 8 1 2 3 4 5 6 7 8 1 2 3 4 5 6 7 8 1 2 3 4 5 6 7 8 1 2 3 4 5 6 7 8 1 2 3 4 5 6 7 8 1 2 3 4 5 6 7 8 1 2 3 4 5 6 7 8 Figure 10. Snapshots of the fluid velocity taken at 8 different instants during the beating cycle of the cilium located at the left of the images. The cilia are beating in an antipleptic motion with a phase lag ∆Φ = π/4, and a spacing of a/L = 2. (a) t = 0: the recovery phase of the first cilium begins; (b-e) recovery phase; (f-h) stroke phase. Once (h) is completed, a new cycle begins: (h) → (a). The figures show contours of the norm of the normalized fluid velocity k~Vf k such that: ~Vf = with V ref = λ/T the reference speed of the present system. ~V dim f V ref f TRANSPORT EFFICIENCY OF METACHRONAL WAVES 19 (a) (b) Figure 11. Mean directional pushing efficiency < ǫP N > over a beating cycle as a function of the phase lag ∆Φ for different cilia spacing a/L. metachronal motion is reached for a symplectic MCW with a phase lag of approximately −π/2. When the cilia spacing is increased to a/L = 2 (figure 11), a peak of efficiency appears for the antipleptic MCW for ∆Φ = 2π/3; and the lower values are reached in the symplectic case for ∆Φ ≈ −π/2 and ∆Φ ≈ −π/3. These trends will be found again in the results shown in §3.2.3. Then, if the cilia spacing is increased above a limit value (a/L = 3.33, figure 11), the directional pushing efficiency becomes equal to 1 for all cases (antipleptic, symplectic, and synchronized). It agrees well with the results obtained by Khaderi et al. (2011) who concluded that "the amount of flow enhancement depends on the inter-cilia spacing [but] the efficiency is not significantly influenced". Indeed, as the cilia are set away from each other, the influence of their neighbours become negligible. Nevertheless, the authors reported a very low efficiency for synchronized cilia, whereas the present results always show a positive flow in the mucus phase for synchronized motions. As mentioned in §2.1, this is a direct consequence of the inertial effects (Re > 1) in the present simulations. It is recalled that they only affect the synchronized case. It has carefully been checked that all other conclusions drawn for antipleptic and symplectic metachrony remain the same for Re < 1 and Re = 20. For a detailed study of the inertial effects at Re = 20, see §3.2.7. The present results are different from the results of Gueron et al. (1997), as it appears that even when cilia spacing is higher than 2 cilium lengths, the influence of neighbouring cilia cannot be neglected (see the case a/L = 2.53 for example). It is important to remember that this efficiency does not take into account the actual net flow volume transported. 3.2.3. Compared hydrodynamic efficiency After having investigated the capacity to transport the flow in the desired direction, it is now interesting to look at the actual flow volume displaced to see if a better efficiency to transport the flow directionally corresponds to a better flow transport. To do so, the global volumetric flow rate Qv over a unit volume of size (1 × 1 × Nz) is introduced: Qv = Nz U ∗dx2 L2 (3.1) 20 S. Chateau, J. Favier, U. D'Ortona and S. Poncet Figure 12. Normalized volumetric flow rate Qv as a function of time over a beating cycle for different phase lags ∆Φ and different cilia spacings a/L. The recovery phase occurs for t ∈ [0, π] and the stroke phase for t ∈ [π, 2π]. where dx = 1 using the classical LBM normalization, and U ∗ = U av/U ref , with U ref =(λ/Ncil)/T being the reference velocity and U av= 1 fluid velocity inside the whole domain. ninj nk Pi,j,k Uijk the average On figure 12, the dimensionless volumetric flow rate Qv is plotted over one beating cycle for arrays of cilia with different phase lags (∆Φ = 0, ∆Φ = −π/4, and ∆Φ = π/4) and different cilia spacings. For cilia spacings equal to a/L = 1.67 and a/L = 2, the antipleptic MCW is the most efficient for transporting fluids. However, note that as the cilia spacing is increased, the ability of the antipleptic wave to transport fluids exhibits a huge decrease compared to the symplectic wave. For a larger cilia spacing (a/L = 3.33), the fluid transport is not impacted anymore by metachrony. Finally, no flow reversal occurs for the synchronized cases: this is the consequence of working at Re = 20. The total volume of fluid displaced during a beating cycle for the different phase lags is compared on figures 13 (a) and (b). For a small cilia spacing (a/L = 1.67), the efficiency of the antipleptic metachrony is obvious. It agrees well with the results of Khaderi et al. (2011) who observed a larger net flow produced by antipleptic metachrony for this value of cilia spacing. Symplectic waves appear to be less or at best equally efficient than antipleptic motion, except for ∆Φ = −7π/8 for a/L = 1.67 where there is a peak in the total displaced volume of flow. On the contrary, negative peaks are found for a/L = 1.67 and a/L = 2 for ∆Φ = −π/3. There are two neighbouring maxima at ∆Φ = π/4 and ∆Φ = π/2 for a/L = 1.67 and a/L = 2 respectively, indicating that specific phase lags are more able to generate a strong flow. At this point, it is worth remembering that in the present model, the recovery and stroke phases occur in the same plane which is not the case in real configurations. Accordingly to Downton & Stark (2009), who studied both 2D and 3D beating patterns, it is expected that in presence of 3D beating patterns, both the directional pushing efficiency ǫP N and total displaced volume of flow would increase. It is also expected that a small flow might occur in the y-direction. 3.2.4. Energetic cost of the metachronal waves The previous sections have been dedicated to the study of the directional transport and the volume displacements of fluids. An energetic perspective is now introduced, to characterize completely the potential benefits of metachrony. TRANSPORT EFFICIENCY OF METACHRONAL WAVES 21 (a) (b) Figure 13. Total dimensionless displaced flow volume generated by an array of cilia over a beating cycle for different phase lags and cilia spacings. +: a/L = 1.67; (cid:13): a/L = 2; ∗: a/L = 2.53; (cid:3): a/L = 3.33. (a) (b) Figure 14. Power spent by the system for different phase lags ∆Φ and different cilia spacings. +: a/L = 1.67; (cid:13): a/L = 2; (cid:3): a/L = 3.33 The average power spent by the cilia during a beating cycle is given by: Pcil = Ps,i V s Ncilia i · (F i m + F i P CL) (3.2) using the forces illustrated in figure 2. The power spent is averaged over several beating cycles. To have a dimensionless power P ∗, the power P ∞ spent by an isolated cilium (a/L = 10) is computed such that P ∗=Pcil/P ∞. Before going any further, it is important to remember here that the only parameter that differs between each value of phase lag ∆Φ for a given cilia spacing a/L, is the size of the domain Nx over the x-direction, hence the number of cilia acting in one wavelength. Moreover, between different cilia spacings a/L, the space a between two cilia is modified in both the x and y-directions, and the sizes Nx and Ny of the domain must be changed accordingly. As a consequence, if the cilia spacing is increased, the density of cilia in both the x and y-directions is decreased. In all the simulations presented in §3.2, all the other parameters (ratio of viscosity rν , beating frequency f , length of the cilia L, etc.) are fixed. Figures 14 (a) and (b) show the average dimensionless power as a function of ∆Φ. 22 S. Chateau, J. Favier, U. D'Ortona and S. Poncet For small values of phase lags (∆Φ 6 π/2), the average power spent decreases to a minimal value for ∆Φ ≈ π/4. For this particular value, the system spends less power than the synchronized case: the antipleptic MCW allows the system composed of all cilia to encounter a smaller resistance from the surrounding fluids. Nevertheless, it is obvious, from a fluid mechanics point of view, that when cilia beat together in a synchronized way, the viscous resistance felt by each cilium is reduced. In the case of metachrony, the behaviour is dual. Indeed, cilia in the stroke phase of the antipleptic motion encounter a stronger viscous resistance from the flow due to their added respective remoteness. Therefore, compared to the synchronized motion, they will transfer more energy to the flow; and the energy transferred will be entirely used to propel the fluids in the desired direction. The opposite is true during the recovery stroke of the antipleptic motion: the clustered behaviour of the cilia allows them to experience a lower viscous resistance from the flow, and therefore to limit the amount of power transferred during this phase. But, globally, the system with antipleptic MCW and ∆Φ < π/2 encounter less viscous resistance than the synchronized motion of cilia, as it can be seen on figures 14 (a) and (b). It results in a better efficiency of the cilia with antipleptic metachronal motion to transfer their momentum to the flow, while in the meantime requiring less power. For the symplectic motion, a similar phenomenon happens: cilia in their stroke phase are clustered and therefore unable to fully exert their pushing action, while cilia in their recovery phase are away from each other and generate a stronger reversal flow compared to the antipleptic motion. It results in a lower capacity of the symplectic MCW to transport mucus, while still allowing the system to spend less power than synchronously beating cilia for ∆Φ < π/2. An energetic ratio will be introduced in §3.2.5 to quantify, for such value of ∆Φ, the capacity of the cilia with antipleptic or symplectic metachrony to transfer their momentum to the flow. This is not true for large phase lags (∆Φ > π/2): the system with antipleptic or symplectic metachrony spends more power than the synchronous case. One can suppose that it is a direct consequence of the reduced number of cilia in a wave length. Then, to explain the better efficiency of the antipleptic MCW over the symplectic ones for large phase lags, an investigation of the flow topology is necessary. Figure 15 shows an antipleptic MCW (on the left) and a symplectic MCW (on the right) at the same time for a phase lag ∆Φ = ±2π/3. For such value of ∆Φ, the antipleptic wave is more efficient for transporting the mucus (see figures 13 (a) and (b)), although the average power spent by both system is relatively similar (see figures 14 (a) and (b)). One can easily see the main difference between both cases: in the symplectic case, the cilium in stroke phase encounters the reversal flow generated by the other cilium in recovery phase. When these two secondary flows meet, vortices are generated and the global transport of fluid is less efficient, as a fraction of the energy transferred to the flow is used to cancel this reversal flow. On the contrary, in the antipleptic case, the cilium in stroke phase does not immediately feel the influence of the reversal flow created by the other cilium in recovery phase, which is behind him. Henceforth, this cilium is able to fully exert its pushing effect on the mucus phase. Moreover, for the antipleptic case, there is a suction effect due to the combined motion of the cilia in the stroke and recovery phases, maximizing the propulsion of mucus. One can then expect a weak blowing effect between the cilia in the stroke and recovery phases of symplectic motion. When the cilia spacing is large (see the case a/L = 3.33 of figures 14 (a) and (b) for example), the power spent by the system is the same for all phase lags: the cilia are too far from each other to be impacted by the phase difference of their immediate neighbours. One can observe that for ∆Φ = 0 (i.e. for synchronized beating), the average power spent by a cilium P ∗ is equal to 1 in figure 14 (a) for the cilia spacings a/L = 1.67 TRANSPORT EFFICIENCY OF METACHRONAL WAVES 23 (a) (b) Figure 15. Comparison of the flow generated by an antipleptic and a symplectic wave for a the cilia spacing a/L = 1.67. (a): Antipleptic MCW with ∆Φ = 2π/3. (b): Symplectic MCW with ∆Φ = −2π/3. The plane is colored with the magnitude of the dimensionless fluid velocity. and a/L = 3.33, meaning that synchronous systems spend the same amount of power as an isolated cilium. It shows the beneficial cost of antipleptic metachrony for small phase lags. 3.2.5. Displacement ratio Inspired by the work of Kim & Netz (2006), a displacement ratio, which can be seen as the transport efficiency of the waves, is introduced to quantify the capacity of a given system to transport particles, with respect to a given amount of power. In that context, η1 is defined by the mean displacement over the x-direction during one beating cycle, divided by the mean power that a cilium had to spend during this beating cycle. Since the main purpose of mucociliary clearance is to transport mucus, and since experimental data (Winters & Yeates 1997) report that the total thickness in the vertical direction of the mucus layer is in the range [1.4L; 10L], values for the displacement were taken on an arbitrary plane z/L = 3.2 near the extremity of the domain. To obtain a value for the displacement, the instantaneous average fluid velocity over the x-direction is computed, and the resulting value is then multiplied by the period of a full beating cycle, giving the mean displacement < dx > over one period on the (x,y,3.2L) plane. By dividing this mean displacement with appropriate quantities, a dimensionless expression of the displacement ratio is obtained: η1 = < dx > P ∗ Ncil λ (3.3) In the synchronized case, i.e. ∆Φ = 0, λ is infinite and thus the size of the domain over the x direction was used and divided by the number of cilia. In figures 16 (a) and (b), one can see, from an energetic point of view, the superiority of the antipleptic wave to transport mucus for small cilia spacings, which confirms the previous findings. More than that, for the smallest cilia spacing (a/L = 1.67), a clear peak of efficiency can be seen for antipleptic MCW with ∆Φ = π/4. If the cilia spacing is increased (a/L = 2), similar results are found. Nevertheless, for a/L = 2.53, a different behaviour occurs: the displacement ratio is found to be the worst for ∆Φ = ±π/2. Then, for cilia far away from each other, the displacement ratio η1 remains constant for all phase lags (the cilia do not feel the influence of the others anymore). 24 S. Chateau, J. Favier, U. D'Ortona and S. Poncet (a) (b) Figure 16. Displacement ratio η1 as a function of the phase lag ∆Φ for different cilia spacings a/L. Similar results regarding the efficiency of the antipleptic MCW are found in Osterman & Vilfan (2011), where the optimal beating pattern is investigated. In their study, they observed that antipleptic MCW were often the most efficient. They also found, as it is the case in the present study, that increasing the density of cilia results in an increase of the efficiency up to a critical point where clustering becomes counter-productive. 3.2.6. Mixing To investigate how the mixing can be enhanced by metachrony, the average stretching rate over the transport and mixing areas during a beating period is computed. Figure 17 shows the results obtained for all cilia spacings. Clearly, the antipleptic MCW is the most efficient for stretching fluids, hence to mix fluids. Clear peaks are visible for antipleptic waves with ∆Φ = π/4 and a/L = 1.67; with ∆Φ ≈ π/3 and a/L = 2; and with ∆Φ ≈ π/3 and a/L = 2.53. On the contrary, symplectic MCW are almost always less efficient for mixing than antipleptic MCW, except for ∆Φ = −7π/8 and a/L = 1.67 where there is an enhancement in the mixing. This peak is also present for the opposite phase lag ∆Φ = 7π/8 where it reaches approximately the same value. This is perfectly coherent with all previous results, and partially with the ones of Ding et al. (2014) who observed, for an unique layer of fluid with an uniform viscosity and a cilia spacing a/L = 1.67, the existence of two peaks in the shear rate: a weak one for symplectic waves for ∆Φ = −π/2, and a stronger one for antipleptic waves for ∆Φ = π/2. The main difference here is that in the present study, the symplectic MCW with ∆Φ ≈ −π/2 seem to have the worst capacity for mixing, while the antipleptic MCW reach their full mixing capacity for ∆Φ = π/4. The combined study of figures 14 and 17 gives a good insight of the mixing efficiency of the system. 3.2.7. Quantification of inertial effects In this part, the influence of the Reynolds number is considered. The present objective is to compare the qualitative behaviour of the MCW between Re < 1 and Re > 1. To reach Reynolds numbers of the order of 10−2, the size of the cilia has been reduced to half the size of real cilia. The geometry is then divided by a factor 2 in the three spatial directions. The main consequence of such choice is that the computed quantities (fluxes, displaced volume of fluid, etc.) displayed here are approximately 8 times smaller compared to the results previously shown (as in figure 13). However, the qualitative TRANSPORT EFFICIENCY OF METACHRONAL WAVES 25 (a) (b) Figure 17. Average stretching rate in the transport and mixing areas as a function of the phase lag ∆Φ for different cilia spacings. (a) +: a/L = 1.67; (cid:13): a/L = 2; ∗: a/L = 2.53; (cid:3): a/L = 3.33 (b) 3D view of the corresponding plot. Figure 18. Total displaced volume of fluid for Re = 0.02, Re = 1 and Re = 20 as a function of the phase lag ∆Φ. The results of these simulations are obtained with L = 7 lu and a/L = 1.67. behaviour of the MCW remains the same, as well as the drag exerted by the cilia. Thus, a comparison of the MCW behaviour between Re < 1 and Re > 1 is possible. On Figure 18, one can see the total displaced volume of fluid for Re = 0.02, Re = 1 and Re = 20, and a cilia spacing a/L = 1.67. A transition between Re < 1 and Re > 1 for the synchronous cases is expected, and can be seen on figure 18. As mentioned in §3.2.3, for Re 6 1, the synchronized cases are less efficient than the symplectic cases. Except for ∆Φ = 0, no other notable quantitative differences are present between the cases Re < 1 and Re = 20. The general behaviour of both the antipleptic and symplectic MCW remains similar for all the other phase lags ∆Φ. Especially, the presence of the peak around ∆Φ = π/4 is present for all the Reynolds numbers tested (Re = 0.047, 2, 5, 10 not shown). The similarity of the metachronal cases (i.e. ∆Φ 6= 0) for this range of Reynolds numbers is certainly due to the fact that there are always cilia in the stroke phase. Hence, a transport is observed during all the beat cycle, and the inertial effects thus remain minor. This is no longer the case for fully synchronously beating cilia: despite being weak, inertial effects cancel the reversal of the flow. 26 S. Chateau, J. Favier, U. D'Ortona and S. Poncet 4. Conclusions By using a lattice Boltzmann solver coupled to an Immersed Boundary method, and considering a purely hydrodynamical feedback from the fluid, symplectic and antipleptic metachronal waves can emerge in a 3D two-phase flow configuration with a viscosity ratio of 20. It is known that metachronal waves may emerge due to hydrodynamic interactions (Elgeti & Gompper 2013). However, for the first time, both types of metachrony are observed to emerge in a two-layer fluid with different viscosities, using a simple feedback law. This feedback depends on a coupling parameter α that can be used to tune the strength and direction of the waves. It is observed that cilia experience weaker torques when being in a clustered configuration, i.e. during the recovery phase of antipleptic motion, as well as during the stroke phase of symplectic motion. The resulting beating pattern is a slow stroke phase and a fast recovery phase for both cases, suggesting that even if a simple hydrodynamic interaction is enough to let emerge metachronal waves, it is not sufficient to reproduce all features of the beating patterns observed in real ciliated surfaces, and other biological issues may play a role in the transport mechanism. The study of Hussong et al. (2011) has shown that metachronal motion can switch from symplectic to antipleptic with increasing inertial effects. Here, the influence of cilia density is highlighted: by assuming a least effort behaviour for the cilia, the present model shows that the metachrony can switch from antipleptic to symplectic by lowering the cilia density. A more detailed study of the quality of synchronization along the x and y- directions is the next step toward a better understanding of the emergence of MCW. A thorough comparative study of the antipleptic and symplectic MCW has been performed and the results show that the antipleptic MCW are the most efficient ones for transporting mucus. This is in accordance with most recent studies addressing this point (Khaderi et al. 2011; Osterman & Vilfan 2011; Ding et al. 2014); and now confirmed in a two-layer environment. In the range of cilia spacing studied, and especially for small phase lags (∆Φ < π/2) the antipleptic MCW have a better ability to (i) transport the flow in the same direction compared to symplectic MCW with the same wavelength, (ii) generate higher flow rate, (iii) advect particles at a given power input, and (iv) generate a higher stretching and hence, are more able to mix fluids. On the contrary, symplectic MCW do not appear to have a great impact on the flow and are often less efficient than antipleptic MCW. This is not in agreement with Elgeti & Gompper (2013) who reported that symplectic waves are almost nearly as efficient as antipleptic waves, using an optimized-efficiency model. This difference may be due to several factors, including the two-layer flow character of the present work, the ratios of viscosity considered, or even the cilia beat shape used. However, these results are in accordance with Khaderi et al. (2011) who explained this better transport efficiency as being due to vortices obstructing the flow for symplectic motions. Among the quantities introduced earlier, some are more decisive for the transport of particles. The most important ones are the net volume of mucus displaced, the transport efficiency η1, and the mixing capacity of the system. It has been shown all along this work that antipleptic MCW with ∆Φ = π/4 generate the larger flow while in the meantime showing the higher value of the transport efficiency (the power spent P ∗ for ∆Φ = π/4 being minimal). A preliminary study of the mixing also strongly indicates that they might also be the more suitable to mix fluids. The best transport efficiency of the antipleptic MCW is certainly due to the clusterized aspect of the cilia in recovery phase which minimizes their impact on the flow, while cilia in stroke phase are able to fully exert their pushing action. While being more able to advect particles than any other form of coordinated motions, antipleptic MCW with ∆Φ < π/2 also require less power. Note TRANSPORT EFFICIENCY OF METACHRONAL WAVES 27 that the results presented all along concern a viscosity ratio rν of 20. Chatelin & Poncet (2016) has shown that mucus transport was maximized for viscosity ratios ranging from 10 to 20. With the present model, and within this interval of viscosity ratios, the same trends are always observed (results not shown), and the best transport capacities are always obtained for antipleptic MCW with ∆Φ = π/4. The transport properties of the waves also remain similar and do not vary significantly. However, it is worth noticing that the best transport capacity was obtained for rν = 20. Hence, the optimum is probably outside the range of viscosity ratios studied. It also strongly indicates that increasing the mucus viscosity results in an increase of the transport properties. As in previous studies (Norton et al. 2011; Chatelin & Poncet 2016), an optimum viscosity for the transport of particles can certainly be found. A more detailed study of how different mucus viscosity affect the transport properties of the MCW is the next step toward a better understanding of the mucociliary clearance. While in the present study only one prescribed beating pattern was investigated, the results can certainly be generalised to other beating patterns. An interesting path of investigation would be to study how this beat shape is influenced by different mucus viscosities, mucus rheologies, or thicknesses of the PCL. Note that a Reynolds number of 20 was used for computational reasons. While this does not affect the behaviour of both the antipleptic and symplectic MCW, weak inertial effects are seen to change the dynamics of the flow in the particular case of synchronously beating cilia. Finally, the mucus was considered as being a Newtonian fluid in the present work, while it is in reality highly non-Newtonian. It is expected that the clearance velocity of both the antipleptic and symplectic MCW would be increased by considering the mucus as being a visco- elastic fluid. One could also expect the symplectic waves to become more efficient than the antipleptic MCW for highly viscous mucus. Indeed, the clusterized aspect of the cilia during the stroke phase of symplectic motion could help them overcome more easily the viscous resistance of the mucus (Knight-Jones 1954). Nevertheless, note that studies (Norton et al. 2011; Chatelin & Poncet 2016) all tend to show, however, that the mucus can not be too "solid" nor too "liquid". The numerical results obtained here constitute a useful basis of investigation to progress in the understanding of respiratory diseases linked to cilia beating disorders, such as COPD or severe asthma (Chanez 2005). It is also of interest for industrial purposes, such as the design of cilia-based actuators for mixing (Chen et al. 2013) or flow-regulator in microscopic biosensors. Future works include the implementation of a realistic rheological model (Lafforgue et al. 2016) for the mucus in order to take into account its highly non-Newtonian behaviour, and the study of the mass transfer occurring at the epithelial surface. A long term objective of this work is to build a numerical environment to predict the transport and mixing of drugs inside both the mucus and PCL layers. With this tool, the effect of drugs could be tested virtually before proceeding to clinical trials. For instance, drugs acting on microscopic parameters such as beating frequency, viscosity of the mucus, density of cilia could be tested and their effects on macroscopic quantities such as mucus flow rate and mixing could be explored, in order to progress in the understanding and treatment of respiratory diseases. The authors would like to thank Dr. Zhe Li and master intern Jean Mercat for the development of the numerical code. S. Chateau and S. Poncet acknowledge also the Natural Sciences and Engineering Research Council of Canada for its financial support through a Discovery Grant (RGPIN-2015-06512). This work was granted access to the HPC resources of Compute Canada and Aix-Marseille University (project Equip@Meso ANR-10-EQPX-29-01). 28 S. Chateau, J. Favier, U. D'Ortona and S. Poncet REFERENCES Blake, J. R. 1971a Infinite model for ciliary propulsion. J. Fluid Mech. 49 (2), 209 -- 222. Blake, J. R. 1971b A spherical envelope approach to ciliary propulsion. J. Fluid Mech. 46 (1), 199 -- 208. Blake, J. R. 1972 A model for the micro-structure in ciliated organisms. J. Fluid Mech. 55 (1), 1 -- 23. Blake, J. R. & Chwang, A. T. 1974 Fundamental singularities of viscous flow. Part II. J. Eng. Math. 8 (1), 113 -- 124. Brennen, C. & Winet, H. 1977 Fluid mechanics of propulsion by cilia and flagella. Annu. Rev. Fluid Mech. 9, 339 -- 398. Chanez, P. 2005 Severe asthma is an epithelial disease. Eur. Respir. J. 25 (6), 945 -- 946. Chatelin, R. 2013 M´ethodes num´eriques pour l'´ecoulement de Stokes 3D : fluides `a viscosit´e variable en g´eom´etrie complexe mobile; Application aux fluides biologiques. PhD thesis, Institut de Math´ematiques de Toulouse. Chatelin, R. & Poncet, P. 2016 A parametric study of mucociliary transport by numerical simulations of 3D non-homogeneous mucus. J. Biomech. 49 (9), 1772 -- 1780. Chen, C. Y., Chen, C. Y., Lin, C. Y. & Hu, Y. T. 2013 Magnetically actuated artificial cilia for optimum mixing performance in microfluidics. Lab Chip 13 (14), 2834 -- 2839. Dauptain, A., Favier, J. & Bottaro, A. 2008 Hydrodynamics of ciliary propulsion. J. Fluid. Struct. 24 (8), 1156 -- 1165. Ding, Y., Nawroth, J. C., McFall-Ngai, M. J. & Kanso, E. 2014 Mixing and transport by ciliary carpets: a numerical study. J. Fluid Mech. 743, 124 -- 140. Downton, M. T. & Stark, H. 2009 Beating kinematics of magnetically actuated cilia. Europhys. Lett. 85 (44002). Elgeti, J. & Gompper, G. 2013 Emergence of metachronal waves in cilia arrays. PNAS 110 (12), 4470 -- 4475. Eloy, C. & Lauga, E. 2012 Kinematics of the most efficient cilium. Phys. Rev. Lett. 109 (038101). Fauci, L. J. & Dillon, R. 2006 Biofluidmechanics of reproduction. Annu. Rev. Fluid Mech. 38, 371 -- 394. Gardiner, M. B. 2005 The importance of being cilia. HHMI Bulletin 64, 32 -- 36. Gauger, E. M., Downton, M. T. & Stark, H. 2009 Fluid transport at low Reynolds number with magnetically actuated artificial cilia. Eur. Phys. J. E 28 (2), 231 -- 242. Gueron, S. & Levit-Gurevich, K. 1999 Energetic considerations of ciliary beating and the advantage of metachronal coordination. PNAS 96 (22), 12240 -- 12245. Gueron, S., Levit-Gurevich, K., Liron, N. & Blum, J. J. 1997 Cilia internal mechanism and metachronal coordination as the result of hydrodynamical coupling. PNAS 94 (12), 6001 -- 6006. Guo, H., Nawroth, J., Ding, Y. & Kanso, E. 2014 Cilia beating patterns are not hydrodynamically optimal. Phys. Fluids 26 (091901). Guo, Z. & Shu, C. 2013 Lattice Boltzmann Method and Its Applications in Engineering. World Scientific Publishing Company, Singapore. Hussong, J., Breugem, W.P. & Westerweel, J. 2011 A continuum model for flow induced by metachronal coordination between beating cilia. J. Fluid Mech. 684, 137 -- 162. Keller, S. R. & Brennen, C. 1968 A traction-layer model for ciliary propulsion. Proceedings of the Symposium on Swimming and Flying in Nature, held at the California Institute of Technology, Pasadena, 253 -- 271, Plenum Press, New York. Kelley, D. H. & Ouellette, N. T. 2011 Separating stretching from folding in fluid mixing. Nat. Phys. 7, 477 -- 480. Khaderi, S. N., Baltussen, M. G. H. M., Anderson, P. D., den Toonder, J. M. J. & Onck, P. R. 2010 Breaking of symmetry in microfluidic propulsion driven by artificial cilia. Phys. Rev. E 82 (027302). Khaderi, S. N., Den-Toonder, J. M. J. & Onck, P. R. 2011 Microfluidic propulsion by the metachronal beating of magnetic artificial cilia: a numerical analysis. J. Fluid Mech. 688, 44 -- 65. TRANSPORT EFFICIENCY OF METACHRONAL WAVES 29 Kim, Y. W. & Netz, R. R. 2006 Pumping fluids with periodically beating grafted elastic filaments. Phys. Rev. Lett. 96 (15), 158101. Kirkham, S., Sheehan, J. K., Knight, D., Richardson, P. S. & Thornton, D. J. 2002 Heterogeneity of airways mucus: variations in the amounts and glycoforms of the major oligomeric mucins MUC5AC and MUC5B. Biochem. J. 361 (3), 537 -- 546. Knight-Jones, E.W. 1954 Relations between metachronism and the direction of ciliary beat in Metazoa. J. Cell Sci. s3 -- 95, 503 -- 521. Lafforgue, O., Poncet, S., Seyssiecq-Guarente, I. & Favier, J. 2016 Rheological characterization of macromolecular colloidal gels as simulant of bronchial mucus. 32nd International Conference of the Polymer Processing Society (PPS-32), Lyon, France. Lai, S. K., Wang, Y. Y., Wirtz, D. & Hanes, J. 2009 Micro- and macrorheology of mucus. Adv. Drug Deliver Rev. 61 (2), 86 -- 100. Lauga, E. & Eloy, C. 2013 Shape of optimal active flagella. J. Fluid Mech. 730 (R1), 1 -- 11. Li, H., Tan, J. & Zhang, M. 2009 Dynamics modeling and analysis of a swimming microrobot for controlled drug delivery. IEEE T-ASE 6 (2), 220 -- 227. Li, Z., Favier, J., D'Ortona, U. & Poncet, S. 2016 An improved explicit immersed boundary method to couple with lattice Boltzmann model for single- and multi-component fluid flows. J. Comput. Phys. 304, 424 -- 440. Lighthill, J. 1976 Flagellar hydrodynamics. SIAM Rev. 18, 161 -- 230. Lukens, S., Yang, X. & Fauci, L. 2010 Using lagrangian coherent structures to analyze fluid mixing by cilia. Chaos 20 (1), 017511. Matsui, H., Randell, S. H., Peretti, S. W., Davis, W. C. & Boucher, R. C. 1998 Coordinated clearance of periciliary liquid and mucus from airway surfaces. J. Clin. Invest. 102 (6), 1125 -- 1131. Mitran, S.M. 2007 Metachronal wave formation in a model of pulmonary cilia. Comput. Struct. 85 (11-14), 763 -- 774. Niedermayer, T., Eckhardt, B. & Lenz, P. 2008 Synchronization, phase locking, and metachronal wave formation in ciliary chains. Chaos 18 (3), 037128. Norton, M. M., Robinson, R. J. & Weinstein, S. J. 2011 Model of ciliary clearance and the role of mucus rheology. Phys. Rev. E 83 (011921). Osterman, N. & Vilfan, A. 2011 Finding the ciliary beating pattern with optimal efficiency. PNAS 108 (38), 15727 -- 15732. Phan-Thien, N., Tran-Cong, T. & Ramia, M. 1987 A boundary-element analysis of flagellar propulsion. J. Fluid Mech. 184, 533 -- 549. Porter, M. L., Coon, E. T., Kang, Q., Moulton, J. D. & Carey, J. W. 2012 lattice Boltzmann model for fluids with large Multicomponent interparticle-potential viscosity ratios. Phys. Rev. E 86 (036701). Reynolds, A. J. 1965 The swimming of minute organisms. J. Fluid Mech. 23 (2), 241 -- 260. Sanderson, M.J. & Sleigh, M.A. 1981 Ciliary activity of cultured rabbit tracheal epithelium: Beat pattern and metachrony. J. Cell. Sci. 47, 331 -- 341. Satir, P. & Christensen, S. 2007 Overview of structure and function of mammalian cilia. Annu. Rev. Physiol. 69, 377 -- 400. Sedaghat, M. H., Shahmardan, M. M., Norouzi, M., Jayathilake, P. G. & Nazari, immersed boundary lattice M. 2016 Numerical simulation of muco-ciliary clearance: Boltzmann method. Comput. Fluids 131, 91 -- 101. Shan, X. & Chen, H. 1993 Lattice boltzmann model for simulating flows with multiple phases and components. Phys. Rev. E 47 (3), 1815 -- 1819. Shan, X. & Chen, H. 1994 Simulation of nonideal gases and liquid-gas phase transitions by the lattice Boltzmann equation. Phys. Rev. E 49 (4), 2941 -- 2948. Sleigh, M. A. 1962 The biology of Cilia and Flagella. Pergamon Press, Oxford. Sleigh, M. A., Blake, J. R. & Liron, N. 1988 The propulsion of mucus by cilia. Am. Rev. Respir. Dis. 137 (3), 726 -- 741. Smith, D. J., Gaffney, E. A. & Blake, J. R. 2007 Discrete cilia modelling with singularity distributions: Application to the embryonic node and the airway surface liquid. B. Math. Biol. 69 (5), 1477 -- 1510. Taylor, G. 1951 Analysis of the swimming of microscopic organisms. Proc. R. Soc. A 209 (1099), 447 -- 461. 30 S. Chateau, J. Favier, U. D'Ortona and S. Poncet Tuck, E. O. 1968 A note on a swimming problem. J. Fluid Mech. 31 (2), 305 -- 308. Widdicombe, J. H. & Widdicombe, J. G. 1995 Regulation of human airway surface liquid. Resp. Physiol. 99 (1), 3 -- 12. Winters, S. L. & Yeates, D. B. 1997 Roles of hydration, sodium, and chloride in regulation of canine mucociliary transport system. J. Appl. Physiol. 83 (4), 1360 -- 1369. Zou, Q., Hou, S., Chen, S. & Doolen, G. D. 1995 A improved incompressible lattice Boltzmann model for time-independent flows. J. Stat. Phys. 81 (1), 35 -- 48.
1704.08394	1	1704	2017-04-27T01:00:53	Changes in lipid membranes may trigger amyloid toxicity in Alzheimer's disease	[ "physics.bio-ph" ]	Amyloid beta peptides (A\b{eta}), implicated in Alzheimers disease (AD), interact with the cellular membrane and induce amyloid toxicity. The composition of cellular membranes changes in aging and AD. We designed multi component lipid models to mimic healthy and diseased states of the neuronal membrane. Using atomic force microscopy (AFM), Kelvin probe force microscopy (KPFM) and black lipid membrane (BLM) techniques, we demonstrated that these model membranes differ in their nanoscale structure and physical properties, and interact differently with A\b{eta}. Based on our data, we propose a new hypothesis that changes in lipid membrane due to aging and AD may trigger amyloid toxicity through electrostatic mechanisms, similar to the accepted mechanism of antimicrobial peptide action. Understanding the role of the membrane changes as a key activating amyloid toxicity may aid in the development of a new avenue for the prevention and treatment of AD.	physics.bio-ph	physics	Changes in lipid membranes may trigger amyloid toxicity in Alzheimer's disease. Elizabeth Drolle1,2, Alexander Negoda3, Keely Hammond4, Evgeny Pavlov3,5, and Zoya Leonenko1,2,4* 1Department of Biology, University of Waterloo, 2Waterloo Institute of Nanotechnology, University of Waterloo, 3Department of Physiology and Biophysics, Dalhousie University, Canada, 4Department of Physics and Astronomy, University of Waterloo, 5Department of Basic Sciences, New York University College of Dentistry, USA. * corresponding author: [email protected] Abstract Amyloid-beta peptides (Aβ), implicated in Alzheimer's disease (AD), interact with the cellular membrane and induce amyloid toxicity. The composition of cellular membranes changes in aging and AD. We designed multi-component lipid models to mimic healthy and diseased states of the neuronal membrane. Using atomic force microscopy (AFM), Kelvin probe force microscopy (KPFM) and black lipid membrane (BLM) techniques, we demonstrated that these model membranes differ in their nanoscale structure and physical properties, and interact differently with Aβ. Based on our data, we propose a new hypothesis that changes in lipid membrane due to aging and AD may trigger amyloid toxicity through electrostatic mechanisms, similar to the accepted mechanism of antimicrobial peptide action. Understanding the role of the membrane changes as a key activating amyloid toxicity may aid in the development of a new avenue for the prevention and treatment of AD. Introduction Alzheimer's disease (AD) is a progressive neurodegenerative disease which leads to severe impairment of memory and cognitive function and is characterized by the formation of amyloid-beta (Aβ) protein aggregates on neurons and cerebral blood vessels1, 2. While all amyloid aggregates, such as oligomers, fibrils, and plaques serve as cellular hallmarks of AD, small soluble oligomers have recently been shown to be more toxic to cells than larger fibrils3. There is currently no cure or prevention for AD; prospective strategies to prevent amyloid toxicity include inhibiting the formation of 1 toxic oligomers, as well as preventing amyloid-damaging effect to the cellular membrane. In this work we propose and test a new hypothesis that changes in lipid membrane structure and properties may trigger amyloid toxicity. It is known that Aβ aggregation occurs on the surfaces of neuronal cells, leading to amyloid plaque formation in the brain tissues of individuals diagnosed with AD1, 2. The cellular membrane is therefore recognized as a target for amyloid attack. Aβ-membrane interactions may occur through specific membrane receptors4 as well as non-specifically with the lipid membranes themselves. Many studies have reported the effect of the membrane in general, and of lipid rafts on amyloid binding and toxicity3, 5-13. Despite these efforts, the molecular mechanism of amyloid toxicity remains unclear, which delays the development of a treatment for AD. Previous studies on the brain membrane lipid composition of AD patients have reveal changes in lipid composition that occur during disease progression. These include lowering the content of several types of phospholipids found in the inner leaflet of the membrane14 and a decrease in sphingomyelin (SM) content due to increased sphingomyelinase activity15. Perhaps surprisingly, the role of these changes has not been investigated in relation to amyloid toxicity. One type of neuronal lipids – gangliosides - is of special interest, with some contradicting results as to what occurs to their levels as a result of AD. Reductions in the amount of gangliosides present in the membrane have been observed in several regions of AD brains compared to that of control brains16-18 while other studies have suggested ganglioside plays a role in the ganglioside monosialotetrahexosylganglioside (GM1) results in an increase of Aβ aggregation in vitro 19-21. However, changes in membrane lipid composition may occur before the onset of AD symptoms and its corresponding cellular pathology. Recently, researchers demonstrated the predictive power of such changes in lipid composition in blood plasma as an early indicator of AD22. Changes in the composition of lipids found in blood plasma may be related to the changes in the lipid composition of neuronal membranes and /or membrane damage. Therefore it is of great interest to study the changes of structure and composition in neuronal membranes and their relevance to the amyloid-induced membrane damage, as these membrane changes may serve as an important switch to activate amyloid toxicity. formation increase an of plaques and in 2 Biological cellular membranes are very complex and therefore model monolayers and bilayers are widely used to mimic the cellular membrane23, 24. While lipid models are very valuable for studying the mechanism of Aβ toxicity, earlier studies on model membranes cannot be easily related to in vivo animal and cellular studies, due to the fact that often, very simple models, composed of one or few lipid types, are used5, 6, 8, 12, 13, 25- 30. Investigation of more complex model membranes will bridge our understanding of model systems and in vivo systems. In recent work, Sasahara et al. investigated the behaviour of Aβ in association with a lipid model containing five lipid constituents31, and Bennett et al examined 29 neurolipidomic datasets and found evidence to support the idea of phospholipid metabolism having an important determinant in the conversion of AD32. Here, our goal is to not only increase membrane complexity to better mimic neuronal membrane, but most importantly to mimic healthy and AD states of neuronal cell membranes and elucidate the role of membrane changes in amyloid toxicity. Currently, there are no lipid models mimicking healthy and AD neuronal membranes available in the literature, despite analysis of brain tissues showing changes in lipid composition with aging and AD19, 33, 34. Based on previous reports31, 35, 36, we developed a membrane model that incorporates DPPC, POPC, sphingomyelin, cholesterol, and ganglioside GM1. These lipids are found in the outer leaflet of neuronal cell membranes6, 19, 33, 37. We hypothesize that changes in lipid composition of the AD brain affect the physical and structural properties of the neuronal cell membrane, compared to a healthy membrane, and that these changes in membrane fluidity, permeability, and lipid domain (raft) distribution, affect amyloid binding and make the neuronal membrane more susceptible to Aβ induced injury. In order to test this hypothesis, we designed multicomponent lipid models that mimic healthy and diseased states of neuronal cell membrane, with the goal of elucidating structural differences between these models as well as the differences in Aβ binding and the damage that Aβ produces to the healthy versus AD model membranes. We used AFM and KPFM imaging to reveal the surface morphology and electrical surface potential of monolayers associated with these models. We used AFM imaging in liquid 3 to visualize the binding of Aβ to the membrane as well as the BLM technique to monitor the membrane damage induced by Aβ upon binding. Materials and Methods Lipids. 1,2-dipalmitoyl-sn-glycero-3-phosphocholine (DPPC), 1-palmitoyl-2-oleoyl-sn- glycero-3-phosphocholine (POPC), sphingomyelin (SM), cholesterol (Chol), and ganglioside monosialotetrahexosylganglioside (GM1) were purchased from Avanti Polar Lipids (Alabaster, AL) in powder form. Complex mixtures of these five constituents were made for analysis, and are outlined in Table 2. All other chemicals used were of reagent grade. Supported lipid monolayers. Preparation via Langmuir-Blodgett monolayer technique. Phospholipid monolayers were deposited on freshly cleaved mica (Ashville-Schoonmaker Mica Co., Newport News, VA) by the method of Langmuir-Blodgett (LB) deposition using a KSV-Nima LB microtrough (Biolin Scientific, Stockholm, Sweden). For sample preparation, solutions of lipid dissolved in chloroform at a concentration of 1 mg/mL (lipid/chloroform) were spread at the surface of the subphase and deposited on the mica substrates at a pressure of 35 mN/m with a dipper arm speed of 2 mm/min. The mica slide was allowed to air-dry for 10 minutes before being placed in a dessicator for a 24-hour period, prior to further analysis. AFM / KPFM Imaging. AFM and KPFM imaging of monolayers supported on mica was performed using SmartSPM 1000 (AIST-NT) in air at room temperature and normal humidity using a MicroMasch gold coated cantilever (HQ:NSC14/Cr-Au) with a resonance frequency of 160 kHz and a spring constant of 5.0 N/m. KPFM imaging was performed in amplitude modulation mode (AM-KPFM) to achieve higher resolution and high sensitivity required for our biological samples than typical KPFM is capable of achieving. In this mode AM-KPFM imaging was done simultaneously with AFM imaging, AFM and AM-KPFM images of the sample correspond to the sample location38. 4 Data Processing and Analysis. Data collected was processed using SPIP and AIST-NT image processing software. The AFM topography images were plane corrected by means of global leveling and global bow removal and filtered using noise reduction caused while scanning with high resolution Z-scale (picometers). KPFM images were not processed with any filters, to ensure the proper potential measurements; the raw data was used for average differences in electrical surface potential. Data was collected on 2 μm by 2 μm and 5 μm by 5 μm high-resolution images of monolayer samples. At least 10 images were analyzed for each sample. At least 3 samples were prepared for each membrane type. All quantitative results are presented as mean ± standard error of the mean (SEM), with significance determined using ANOVA tests. Any results determined to be significant are reported with a 95% confidence level. Aβ binding to the membrane. Preparation of Lipid Bilayers. Hydrated phospholipid bilayers were deposited on freshly cleaved mica (Ashville-Schoonmaker Mica Co., Newport News, VA) via vesicle fusion, as described in previous publications12, 39. Bilayers were covered with nanopure water and imaged in liquid using AFM. Aβ incubation on lipid membrane. Aβ1-42 (rPeptide, Bogart GA) was pretreated to ensure monomeric form according to Fezoui procedure40. Additional Aβ1-42 was made in the lab of Dr. Paul Fraser (University of Toronto, Toronto ON) and was also studied to confirm and compare the findings obtained for the Aβ form rPeptide. Aβ was suspended in HEPES buffer (20 mM HEPES, 100 mM NaCl, pH 7.4) at a concentration of 40 mM (Aβ/buffer). 100 μL of the Aβ solution was added to pre-formed membranes and incubated for increasing time periods; at the end of the time period for each membrane, excess Aβ was gently rinsed away in order to stop the fibrilization process, with complete hydration maintained at all times. The membrane with Aβ deposits was kept in Nanopure water. At least two repeats of the incubations were completed for amyloid from each source, totaling to 4 trials for each time point. 5 AFM imaging of lipid membrane in liquid. AFM imaging of hydrated membrane and Aβ incubated membrane samples on a mica substrate was performed using Magnetic- Alternating-Current (MAC) mode on an Agilent AFM/SPM 5500 using Keysight Type II MAC mode rectangular cantilevers (force constant of 2.8 N/m and a resonance frequency in water of 30 kHz). Membrane imaging was conducted at ambient room temperature in liquid cell in Nanopure water, with hydration of the membrane maintained at all times throughout imaging. Data processing and analysis. Data collected was processed using SPIP data processing software. The topography images were corrected via global levelling and global bow removal. Data was collected on 2 μm by 2 μm and 5 μm by 5 μm high-resolution images of membrane samples. All quantitative results are presented as mean ± SEM, with significant difference determined using ANOVA tests. Any differences determined to be significant are reported with a 95% confidence level. At least 3 samples were analyzed, each utilized a fresh batch of Aβ and lipid membranes, with at least 10 images taken for each sample from multiple locations on the membrane. Black Lipid Membrane (BLM) Studies. Preparation of BLM samples. Planar lipid bilayers were formed from a 15 mg/mL lipid solution in n-decane (Aldrich). The suspended bilayer was formed across the 200 µm aperture of a Delrin cup (Warner Instruments, Hamden, CT) by direct application of lipids41, Both cis (voltage command side) and trans (virtual ground) compartments of the cup cuvette contained 150 mM NaCl, 2 mM CaCl2, 10 mM Tris-HCl pH 7.4. 5 µM Aβ was added to the cis compartment of the cuvette. All measurements were performed at room temperature. Data recording and analysis. Currents across lipid bilayers were recorded with a Planar Lipid Bilayer Workstation (Warner Instruments, Hamden, CT). The cis compartment was connected to the head stage input and the trans compartment was held at virtual ground via a pair of matched Ag/AgCl electrodes. Signals from voltage-clamped BLM were high-pass-filtered at 2.1 kHz using an eight-pole Bessel filter LPF-8 (Warner 6 Instruments), digitized (Data Translation digitizer) and recorded on PC using in‐house analog‐to‐digital converter acquisition software developed by Elena Pavlova. For the statistical analysis data were averaged from at least three independent experiments and analyzed using Origin software. Experiments were performed in three separate trials for each sample. Each recorded trace was analyzed to get obtain the mean value of conductance. Results are expressed as mean ± SEM. Results Lipid Composition of Model Systems Mimicking Healthy and AD Membrane States. Though there have been numerous studies on the interaction of lipid monolayers and membranes with Aβ6, 8, 12, 13, 25-30, 42, many studies are carried out using simple models, consisting of one to three lipid types, which do not provide a good model for neuronal membrane. Based on previous studies on the composition of the outer leaflet of the neuronal membrane6, 19, 33, 37, we designed three different multicomponent lipid models consisting of DPPC, POPC, sphingomyelin (SM), cholesterol (Chol), and ganglioside monosialotetrahexosylganglioside (GM1), which mimic healthy and AD neuronal membranes. The structure and properties of these lipids are shown in Table 1. Table 1: Structures and Properties of Lipids Studied. This table outlines information about the five constituents of the models studied: DPPC, POPC, SM, Chol, and GM1. Phospholipid phase at ambient room temperature is indicated as samples were studied under these conditions. Dipole moment value is included due to its relevance in the KPFM study portion of this work. Lipid structures were adapted from Avanti Polar Lipids. Structure and Name Abbreviation Phase at 25⁰C Phase Transition Temperature (⁰C) Dipole Moment (D) DPPC Gel 41 +0.8243 POPC Fluid ‐2 +0.47344 7 SM Gel ~ 37 +0.3045 Chol ‐ ‐ +0.4046 GM1 Fluid ~ 20 ‐0.17143 Three models consist of varying lipid ratios of the same composition: DPPC-POPC-SM- Chol-GM1, shown in Table 2, and mimic healthy model membrane (HM), diseased model 1 accounting for decrease in GM1 content (D1) and diseased model 2 accounting for the decrease in both GM1 and SM content (D2). Such changes in membrane composition as a result of AD were observed in vivo19, 33. Table 2: Complex lipid models mimicking healthy and AD neuronal membranes. Lipid mixtures are all comprised of the same components but differ in their ratios (by weight) based on documented changes in membrane composition as a result of AD. Lipids Ratio (by weight, %) 37 : 37 : 10 : 10 : 6 Ratio (by molarity, %) 35.3 : 34.1 : 9.8 : 18.1 : 2.7 Model Name Healthy Model 39 : 39 : 10 : 10 : 2 DPPC – POPC – SM – Chol – GM1 ‐mimics a "healthy" neuron Diseased 1 DPPC – POPC – SM – Chol – GM1 (D1) Model ‐mimics a neuron beginning to enter the "diseased" state with a decrease in the GM1 content (Diseased 1) DPPC – POPC – SM – Chol – GM1 ‐mimics an increasingly "diseased" neuron with a decrease in both GM1 and SM content (Diseased 2) 36.5 : 35.2 : 9.7 : 17.8 : 0.09 39.4 : 38.0 : 3.9 : 17.8 : 0.09 Diseased 2 (D2) Model 42 : 42 : 4 : 10 : 2 The healthy model (Table 2) has five constituents commonly found in the outer leaflet of a general healthy neuronal cell membrane. Mass ratios were utilized to allow for easier comparison to our earlier studies12, 39 and relative ratios were decided upon based on extrapolation from studies of lipid content in neuronal cells47. The D1 model was chosen 8 to correlate with the reductions in gangliosides observed in membranes in several regions of AD brains compared to that of control brains, as well as with decreases in GM1 content as AD progresses19, 29. The D2 model has a decrease in both GM1 and SM compared to our model of a healthy neuronal membrane. This decreased amount of SM was chosen based on reported decrease in SM content due to increased sphingomyelinase activity in association with AD15. Study of Monolayer Morphology and Electrical Surface Potential. Using these combinations of lipids, we prepared supported monolayers of the three models and used AFM and KPFM in order to study the morphology and electrical surface potential distribution of each model monolayer. AFM allows for topographical imaging of a sample with nanoscale resolution, making it ideal for investigating the changes in monolayer morphology. KPFM is a variation of an AFM and has been shown useful for mapping electrical surface potential of a lipid monolayer at the nanoscale38. Figure 1 depicts the changes in topography (AFM), and electrical surface potential (KPFM), observed between the three models. The healthy model (Figure 1A) shows the network of interconnected nanodomains spread across the monolayer. The topographical nanodomains have the average differences in height (Δh) of 0.986 ± 0.024 nm between higher and lower domains. The lateral dimensions of the higher domains ranged from 22.5 x 40.1nm up to over 200 x 52 nm. The lower domains were mostly small, less than 20 nm across, with few larger areas up to 150nm across. KPFM images of the monolayer corresponding to this model show some minor fluctuations in electrical surface potential (V) (Figure 1D), though no discernible patterns in ΔV are observed. The average roughness (variation in ΔV between higher and lower domains) of the sample is 24.64 ± 1.10 mV (as seen in Table 3). 9 Figure 1: Comparison of monolayer topography and electrical surface potential for three models. AFM topography (in gold) and KPFM electrical surface potential (in blue) images are shown for healthy (A/D), diseased: D1 (B/E), and D2 (C/F) model systems in monolayer form. Cross sections of topography features are below each topography image. Scale bar denotes 500 nm. In the AFM topography of the D1 model (Figure 1B), irregularly shaped interconnected higher and lower domains are observed. The higher domains are larger in area than in the healthy model, spanning up to 525 nm. The lower domains range from 35 to 200 nm across. The higher domains appear only slightly higher than those in the healthy model, with average Δh values of 1.051 ± 0.016 nm as compared to the healthy model (0.986 nm). The KPFM image of the D1 model (Figure 1E) shows organized, nanoscale electrostatic domains with the differences in electrical surface potential ΔV of 70.47 ± 5.41 mV, the highest average ΔV observed across the three samples. Topographic (AFM) and electrostatic (KPFM) nanodomains correlate with each other for the D1 model. Lower domains that are more disordered in nature are present in both the D1 and healthy model. However in the D1 model, these lower domains are larger in area than the healthy model; in the D1 model lower domains reached sizes of 200 nm across, whereas domains 10 in the healthy model spanned up to 20 nm across. According to Connelly et al., Aβ is able to directly interact with a DOPC lipid bilayer and insert itself to form ion-conductive pores with an average outer diameter of 7.8 - 8.3 nm48. We expect that Aβ would most easily form pores in less ordered areas of the membrane corresponding with the regions in our monolayer models. The D2 model (Figure 1C) shows a disruption of the larger, more ordered domains observed in the D1 model; it contains irregularly shaped features but smaller in area and more plentiful in number. They are quite narrow, with an average width of about 25 nm, and do not exceed 180 nm in length. The size of the lower domains are decreased when compared to the D1 model, with widths ranging from 27 to 40 nm compared with the widths of up to 200 nm seen in the D1 model. We also observe a drastic decrease in the average difference in height between higher and lower domains for the D2 model when compared to the other samples studied. The nanodomains observed in the D2 model had an average Δh of 0.500 ± 0.03 nm, which is significantly smaller than both the healthy and D1 models. This is likely due to the reduced SM content of this model, causing reduced lipid tail ordering. As seen from KPFM images (Fig 1F), the D2 model electrostatic domains present in the KPFM images did not correlate to AFM topography domains for this model (Figure 1C). An average ΔV of 11.63 ± 0.59 was measured for this model the smallest average ΔV observed between all the models. Results of the monolayer study are summarized in Table 3, which compares the average Δh and ΔV values for each of the three models studied. Table 3: Summary of Statistical Analysis of Mixed Lipid Monolayer Samples. Average numbers for Δh (difference in height) and ΔV (difference in electrical surface potential), determined from AFM and KPFM images of monolayers corresponding to healthy, D1 and D2 models. Diseased 1 Model Diseased 2 Model Healthy Model 1.051 ± 0.016 nm 0.500 ± 0.03 nm 70.47 ± 5.41 mV 11.63 ± 0.59 mV Topographical Domains Δh Features in Electrical Surface Potential ΔV 24.64 ± 1.10 mV 0.986 ± 0.02 nm 11 These results demonstrate that differences in topographical nanoheterogeneity and electrical surface potential are clearly seen in all three models (healthy, D1 and D2), which influence the interaction of each membrane with the charged Aβ peptide, discussed in the next section. Model Membranes and their Interactions with Aβ. BLM Study: Amyloid Effect on Membrane Permeability. We used Black Lipid Membrane (BLM) techniques to study the effects of Aβ on lipid bilayer conductance. This method allows for the measurement of ion currents across the membrane and membrane permeability to ions49. We compared currents measured across lipid bilayers of three different compositions corresponding to our model membranes shown in Table 2. These membranes were studied both in the absence and the presence of Aβ (1-42) peptide in order to investigate changes in membrane permeability caused by Aβ binding. The conductance was observed in the control/"healthy" model without the addition of the Aβ (Figure 2A). The addition of 5 µM Aβ to the compartment of the cuvette containing the model led to an increase in noise amplitude with the following increase in conductance level. Although we recorded the increase in current, this increase was not statistically significant for the healthy model (Figure 2A Left panel, p=0.053, n=4). For the D1 membrane model, we found that the addition of Aβ caused a significant increase in the conductance of the membrane, which further increased with time. This effect is apparent from Figure 2B (p=5x10-6, n=11), which illustrates an increase in conductance after 15 minutes of Aβ incubation compared to the conductance after 5 minutes of Aβ incubation. Finally, for the D2 model, we also observed that the addition of Aβ led to a significant increase (p=0.003, n=8) in the conductance of the membrane, which developed over the period of time. However, this increase was less than the increase in the conductance observed in the D1 model (Figure 2C). Overall, we established that the current across all of the tested membranes progressively increased with time reflecting membrane disintegration upon interaction with Aβ. Results of the quantitative comparison of the Aβ-induced conductance of tested model membranes are summarized in Figure 3. 12 Figure 2: Ion currents observed across the membrane: healthy model (A), D1 model (B), and D2 model (C). Lipid membranes were suspended between symmetric aqueous solutions of 150 mM NaCl, 2 mM CaCl2, 10 mM Tris pH 7.4. The left panel for each section shows average current at the voltage amplitude of 50 mV under control conditions (no additions), in 5 min after induction of conductance by the Aβ, in 15 min after induction of conductance by the Aβ. Representative currents at 50 mV are shown in the right panel. * indicates significantly lower current. 13 Figure 3: Comparison of the currents induced by 5 µM of Aβ on different model membranes at voltage amplitude of 50 mV. * indicates significantly lower current. Of the three models, we found that in the presence of Aβ the D1 and D2 models both had higher conductance measurements than the healthy model. The highest current amplitude was observed in the D1 model membranes (in Fig 3, for 5 min (left): p=0.03, n=11; for 15 min (right): p=0.3, n=3). Finally, we studied the interactions of Aβ with the membrane models using AFM, in order to determine how the nanoscale heterogeneity in topography and electrical surface potential observed in the monolayer study affects peptide-membrane interaction. Figure 4 illustrates schematically the presence of lipid domains (differing in height and electrical surface potential) in the complex multi-lipid membrane. AFM Study: Amyloid Incubation on Model Membrane Systems. We formed supported membranes (bilayers) for each of the three lipid models and incubated solutions of Aβ in HEPES buffer in its monomeric form atop the membrane for 1, 4, 6, and 24 hours in buffer and imaged them with AFM in nanopure water, in order to maintain membrane hydration. We looked at four main factors: how Aβ binds to the membrane; the amount of Aβ binding and accumulating over time; the morphology of the Aβ aggregates on the membrane; and the presence of Aβ -induced membrane damage. 14 The differences in Aβ binding to the membrane and accumulation over time are shown in Figure 5 and results are summarized in Table 4. Figure 4: Schematic of Aβ interacting with a model membrane (not to scale). Arrangement of lipids present model bilayer system and phase separation leads to membrane nonhomogeneity, i.e., the presence of nanodomains, both topographical (Δh) and electrostatic (ΔV). We followed the changes in the surface roughness with time which reflects Aβ binding to the membrane and accumulation of amyloid deposits on the membrane. In the healthy model, we observed the surface roughness increasing over time as the larger clusters are formed. This suggests that Aβ accumulation progressively increases with increasing incubation time. After 1 hour of incubation (Figure 5A) we see a uniform layer of aggregates randomly spread across the surface of the membrane, with a roughness of 0.267 ± 0.05 nm. The accumulation of Aβ aggregates on the membrane increase with time, with roughness increasing as incubation time increases: after 4 hours of incubation, average surface roughness was 0.536 ± 0.11 nm; after 6 hours of incubation (Figure 5D), roughness was 0.541 ± 0.074 nm; and after 24 hours of incubation, Aβ accumulation gave a surface roughness of 1.596 ± 0.19 nm. This is a progressive increase in the size of the Aβ clusters on the membrane. 15 Figure 5: Comparison of Aβ Incubation on Three Different Model Membranes for 1 and 6 hours. AFM images in liquid illustrate the difference in Aβ accumulation between 1 and 6 hours of incubation time on a healthy model membrane (A and D respectively), a diseased 1 membrane (B and E) and a diseased 2 membrane (C and F). Image sizes are shown via the ruler scale bar in the left bottom hand corner of each image (the scale bar for A, D, E, and F corresponds to 1 µm; the scale bar for B and C corresponds to 500 nm). Table 4: Changes of Surface Roughness with Time due to Aβ Accumulation on the Membrane. Aβ accumulation is quantified via surface roughness analysis for the HM, D1 and D2 models during 1h, 4h, 6h, and 24 h of incubation. Healthy Model (nm) 0.267 ± 0.05 0.536 ± 0.11 0.541 ± 0.074 1.59 ± 0.19 1 Hour 4 Hour 6 Hour 24 Hour Surface Roughness Diseased 1 Model (nm) 0.487 ± 0.027 0.42 ± 0.03 0.57 ± 0.11 0.44 ± 0.19 Diseased 2 Model (nm) 0.38 ± 0.033 1.366 ± 0.12 0.366 ± 0.045 0.546 ± 0.026 In the D1 model, we saw a change in the accumulation pattern over time, in comparison to the healthy model. After 1 hour of incubation (Figure 5B), we observed a higher surface roughness in the D1 model, 0.487 ± 0.027 nm, than in the healthy model. This 16 indicates more Aβ accumulation than in the healthy model. Although there is no discernible difference in the size and shape of Aβ species on the surface, in both models we saw small, spherical and irregularly shaped oligomers and aggregates. However, as time progressed, the roughness fluctuated: after 4 hours of incubation, we saw a decrease in surface roughness to 0.42 ± 0.03 nm; after 6 hours (Figure 5E), the surface roughness increased slightly to 0.57 ± 0.11 nm; and after 24 hours of incubation time, the average roughness decreased again to its lowest of all the time periods, to 0.44 ± 0.19 nm. Finally, in the D2 model, we saw an initial large accumulation of Aβ species, indicated by an initial large increase in roughness, followed by a dramatic decrease. After 1 hour of incubation (Figure 5C), the surface roughness was 0.38 ± 0.03 nm, which almost quadrupled to 1.36 ± 0.12 nm after 4 hours. This indicates an initial large accumulation of Aβ on the surface of the membrane, without much membrane damage. However, after 6 hours (Figure 5F), we saw a large decrease in roughness of the model, with an average roughness of 0.366 ± 0.045 nm. At this point, Aβ clusters likely penetrated into the membrane, leading to a dramatic decrease in surface roughness, similar to what has been observed in fluid membranes12, 50. After 24 hours, the roughness increased slightly, to 0.546 ± 0.026 nm, suggesting a continuation of accumulation atop the Aβ-disrupted membrane. Discussion In this investigation, summarized in Table 5, we designed and studied three complex lipid models mimicking a healthy neuronal membrane, and two diseased states of the membrane (D1 and D2), mimicking changes in lipid compositions occurring in AD neurons. AFM and KPFM studies of monolayers show that the D1 and D2 models have different nanoscale surface morphologies (topographical domains and electrostatic domains) from each other and as well as compared with the healthy model. 17 Table 5: Summary of Results. AFM/KPFM study on topographical and electrical surface potential features of the models in monolayer form; Black Lipid Membrane analysis on the permeability of each model and the effect of Aβ on this conductance; and AFM liquid imaging of Aβ accumulation over time on each model membrane. Monolayer Morphology – AFM / KPFM Analysis Topographical Δh Diseased 1 Model Diseased 2 Model Lower than Healthy Healthy Model 1.051 ± 0.016 nm 0.986 ± 0.02 nm 0.500 ± 0.03 nm Model 70.47 ± 5.41 mV Higher than Healthy Model 11.63 ± 0.59 mV Lower than Healthy Model 24.64 ± 1.10 mV Electrostatic Surface Potential ΔV Black Lipid Membrane (BLM) Analysis Pore Forming Activity (PFA) No significant increase in PFA with addition of amyloid Highest PFA of three systems studied Significant increase in PFA with addition of amyloid Aβ Binding to Model Membranes Aβ Accumulation Increases with Over Time measured via Roughness Measurements time Fluctuate between increases and decreases over time Indicative of amyloid penetrating into membrane Significant increase in PFA with addition of amyloid Initial increase followed by large decrease Indicative of an initial accumulation event preceding membrane disruption/penetration Lipid domains originate from lipid separation commonly observed in multi-component lipid systems. Different lipids exist in different phases at ambient room temperature, such as liquid crystalline (Lc), liquid disordered (Ld), or with the presence of Chol, cholesterol- induced liquid-ordered phase domains, (Lo). These higher domains are likely to be rich in DPPC, SM, and Chol molecules, (Table 1), as well as GM1 molecules, known to associate with saturated phospholipids, SM, and Chol in lipid rafts. The lower domains are likely areas of high POPC concentration, as POPC is found in Ld phase at room temperature. The changes in lipid composition in HM, D1 and D2 models, including GM1 and cholesterol, result in changes in membrane morphology, i.e. domain organization, size and ordering. These domains also differ in electrical surface potential. 18 We previously showed that similar nanoscale topographical and electrostatic domains are formed in a simple DOPC-Chol model and their presence causes preferential amyloid binding39. We showed that such changes in domain morphology and electrical surface potential in HM, D1 and D2 models influence their interaction with Aβ, and amyloid- induced damage. Our BLM study shows that Aβ binds to the membrane and induces an increase in membrane conductance (ion permeability), which is a result of pore formation induced by amyloid. The significantly higher pore forming activity of the two diseased membrane models compared to the healthy model (Figure 2 and Table 5) supports the idea that the differences in membrane composition can have a strong effect on the interaction of the Aβ with the membrane and the extent to which the Aβ can cause damage and alteration of normal cell function by changing membrane permeability. This is consistent with different domain distribution and morphologies observed in each model with the monolayer studies. Our AFM images illustrate Aβ binding to the membrane and indicate less penetration of Aβ into the healthy model membrane in good correlation with BLM results. According to the proposed mechanisms of Aβ interaction with the membrane 50, depending on the membrane state, Aβ may either adsorb onto the surface of the lipid membrane (as seen in the healthy model) or partially penetrate into the membrane, causing membrane disruption and pore formation48, 50. The multiple types of membrane interaction that Aβ is capable of can be attributed to its complex charge distribution; this distribution allows for Aβ to bind to surfaces of varying charges and hydrophobicity51. Both the D1 and D2 models show higher penetration of Aβ into the membrane, inducing membrane damage, which correlates with the much higher membrane permeability recorded by BLM in D1 and D2 models. The differences in Aβ accumulation as well as differences in membrane disruption further show the significant effect of membrane lipid composition as well as nanoheterogeneity on its interaction with Aβ. 19 The importance of the effect of the composition of the membrane itself on amyloid - membrane interactions is of even greater interest as it has been shown recently that Aβ peptides share many similarities with antimicrobial peptides (AMP), which specifically recognize and kill bacterial cells through selective membrane-mediated recognition mechanism, causing disintegration of the bacterial membrane without affecting the host cell52. AMP and Aβ share common characteristics, including the ability to form fibrils in solution, capabilities of membrane interaction, ability to form ion channels and defects in the membrane52. In vitro studies have shown that Aβ has antimicrobial activity against eight common and clinically relevant microorganisms with a potency equivalent to or greater than AMP, which suggests the potential of Aβ is an unrecognized AMP of the innate immune system53. In fact, an exciting recent study showed the ability of Aβ to mediate the entrapment of unattached microbes in the brain, further suggesting that Aβ has protective roles in innate immunity 54acting as an AMP in the brain. AMPs are known to be very specific in recognizing the structure of bacterial membranes through electrostatic interactions. Similar to AMPs, Aβ is negatively charged (-3)51 and thus may share this electrostatic mechanism. This allows Aβ to recognize changes in membrane structure and integrity through electrostatic interactions and the presence of electrostatic nanodomains in neuronal model membranes. This alteration in membrane composition and structure being a factor in the onset of AD also may help to explain why some people maintain normal cognitive abilities despite the presence of Aβ plaques55. Conclusions In summary, we designed model lipid membranes which mimic the neuronal cell membrane in healthy and diseased states. We demonstrated that healthy and diseased model membranes differ in their nanoscale structure, which significantly influence the interaction of these membranes with Aβ (1-42). The diseased membrane models are more susceptible to interaction with Aβ and its damaging effects than a healthy membrane model. Based on our data and Aβ – AMP similarities reported in literature we propose a new hypothesis for the mechanism of amyloid toxicity in which the neuronal membrane changes play a crucial role: i.e. when neuronal cellular membrane changes in composition 20 and properties due to aging or AD it becomes recognized by Aβ as foreign or "bacteria- like" membrane through electrostatic interactions and therefore undergo amyloid attack and disintegration. Therefore, sustaining neuronal cellular membrane in a healthy state may reduce the damaging effects of Aβ and serve as a preventative strategy against Alzheimer`s disease. Conflicts of interest The authors declare no competing financial interest. Acknowledgements The authors acknowledge contribution of Prof Paul Fraser and Ling Wu from University of Toronto, who provided Amyloid beta peptide for part of this study and critically read the manuscript. This work was supported by the Natural Science and Engineering Council of Canada (NSERC) grant [ZL], Heart and Stroke Foundation of Canada grant and American Heart Association grant [EP], an NSERC Canada Graduate Scholarship and WIN Fellowship [ED], and an NSERC Undergraduate Student Research Award [KH]. Technical support from AIST-NT and Keysight is greatly appreciated. References L. Puglielli, R. E. Tanzi and D. M. Kovacs, Nat Neurosci, 2003, 6, 345‐351. P. J. Whitehouse, D. L. Price, R. G. Struble, A. W. Clark, J. T. Coyle and M. R. Delon, Science, 1982, 215, 1237‐1239. J. V. Rushworth and N. M. Hooper, Int J Alzheimers Dis, 2010, 2011, 603052. J. W. Um, H. B. Nygaard, J. K. Heiss, M. A. Kostylev, M. Stagi, A. Vortmeyer, T. Wisniewski, E. C. Gunther and S. M. Strittmatter, Nat Neurosci, 2012, 15, 1227‐1235. G. P. Eckert, W. G. Wood and W. E. Muller, Curr Protein Pept Sci, 2010, 11, 319‐325. M. Vestergaard, T. Hamada and M. Takagi, Biotechnol Bioeng, 2008, 99, 753‐ 763. T. L. Williams, B. R. Johnson, B. Urbanc, A. T. Jenkins, S. D. Connell and L. C. Serpell, Biochem J, 2011, 439, 67‐77. R. Friedman, R. Pellarin and A. Caflisch, J Mol Biol, 2009, 387, 407‐415. K. A. Burke, E. A. Yates and J. Legleiter, Front Neurol, 2013, 4, 17. F. Tofoleanu and N. V. Buchete, Prion, 2012, 6, 339‐345. E. Terzi, G. Holzemann and J. Seelig, Biochemistry, 1997, 36, 14845‐14852. 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 21 F. Hane, E. Drolle, R. Gaikwad, E. Faught and Z. Leonenko, J Alzheimers Dis, 12. 2011, 26, 485‐494. K. Matsuzaki, Biochim Biophys Acta, 2007, 1768, 1935‐1942. 13. J. W. Pettegrew, K. Panchalingam, R. L. Hamilton and R. J. McClure, Neurochem 14. Res, 2001, 26, 771‐782. X. He, Y. Huang, B. Li, C. X. Gong and E. H. Schuchman, Neurobiol Aging, 2010, 15. 31, 398‐408. L. Svennerholm and C. G. Gottfries, Journal of Neurochemistry, 1994, 62, 16. 1039‐1047. I. Kracun, H. Rosner, V. Drnovsek, M. Heffer‐Lauc, C. Cosovic and G. Lauc, Int J 17. Dev Biol, 1991, 35, 289‐295. K. Mlinac and S. K. Bognar, Translational Neuroscience, 2010, 1, 300‐307. 18. T. Ariga, M. P. McDonald and R. K. Yu, J Lipid Res, 2008, 49, 1157‐1175. 19. K. Yanagisawa and Y. Ihara, Neurobiol Aging, 1998, 19, S65‐67. 20. K. Yanagisawa, A. Odaka, N. Suzuki and Y. Ihara, Nat Med, 1995, 1, 1062‐1066. 21. 22. M. Mapstone, A. K. Cheema, M. S. Fiandaca, X. Zhong, T. R. Mhyre, L. H. MacArthur, W. J. Hall, S. G. Fisher, D. R. Peterson, J. M. Haley, M. D. Nazar, S. A. Rich, D. J. Berlau, C. B. Peltz, M. T. Tan, C. H. Kawas and H. J. Federoff, Nat Med, 2014, 20, 415‐418. O. G. Mouritsen and K. Jorgensen, Pharmaceutical Research, 1998, 15, 1507‐ 23. 1519. 24. H. van De Waterbeemd, D. A. Smith, K. Beaumont and D. K. Walker, J Med Chem, 2001, 44, 1313‐1333. 25. M. Vestergaard, M. Morita, T. Hamada and M. Takagi, Nagoya, 2010. E. Maltseva and G. Brezesinski, Chemphyschem, 2004, 5, 1185‐1190. 26. L. Qiu, A. Lewis, J. Como, M. W. Vaughn, J. Huang, P. Somerharju, J. Virtanen 27. and K. H. Cheng, Biophys J, 2009, 96, 4299‐4307. 28. E. Y. Chi, S. L. Frey and K. Y. Lee, Biochemistry, 2007, 46, 1913‐1924. 29. J. Fantini and N. Yahi, Expert Rev Mol Med, 2010, 12, e27. A. Choucair, M. Chakrapani, B. Chakravarthy, J. Katsaras and L. J. Johnston, 30. Biochim Biophys Acta, 2007, 1768, 146‐154. 31. K. Sasahara, K. Morigaki and K. Shinya, Phys Chem Chem Phys, 2013, 15, 8929‐ 8939. S. A. Bennett, N. Valenzuela, H. Xu, B. Franko, S. Fai and D. Figeys, Front 32. Physiol, 2013, 4, 168. 33. M. Soderberg, C. Edlund, K. Kristensson and G. Dallner, Lipids, 1991, 26, 421‐ 425. 34. S. C. Cunnane, J. A. Schneider, C. Tangney, J. Tremblay‐Mercier, M. Fortier, D. A. Bennett and M. C. Morris, J Alzheimers Dis, 2012, 29, 691‐697. D. Lingwood and K. Simons, Science, 2010, 327, 46‐50. 35. K. Yanagisawa, J Neurochem, 2011, 116, 806‐812. 36. A. Prinetti, V. Chigorno, S. Prioni, N. Loberto, N. Marano, G. Tettamanti and S. 37. Sonnino, J Biol Chem, 2001, 276, 21136‐21145. 38. B. Moores, F. Hane, L. Eng and Z. Leonenko, Ultramicroscopy, 2010, 110, 708‐ 711. 39. E. Drolle, R. M. Gaikwad and Z. Leonenko, Biophys J, 2012, 103, L27‐29. 22 Y. Fezoui, D. M. Hartley, J. D. Harper, R. Khurana, D. M. Walsh, M. M. Condron, 40. D. J. Selkoe, P. T. Lansbury, Jr., A. L. Fink and D. B. Teplow, Amyloid, 2000, 7, 166‐178. E. Pavlov, C. Grimbly, C. T. Diao and R. J. French, FEBS letters, 2005, 579, 41. 5187‐5192. Y. H. An and R. J. Friedman, J Biomed Mater Res, 1998, 43, 338‐348. 42. H. Beitinger, V. Vogel, D. Mobius and H. Rahmann, Biochimica et biophysica 43. acta, 1989, 984, 293‐300. H. Brockman, Chem Phys Lipids, 1994, 73, 57‐79. 44. E. Prenner, G. Honsek, D. Honig, D. Mobius and K. Lohner, Chem Phys Lipids, 45. 2007, 145, 106‐118. D. Paiva, G. Brezesinski, C. Pereira Mdo and S. Rocha, Langmuir, 2013, 29, 46. 1920‐1925. 47. R. O. Calderon, B. Attema and G. H. DeVries, J Neurochem, 1995, 64, 424‐429. L. Connelly, H. Jang, F. T. Arce, R. Capone, S. A. Kotler, S. Ramachandran, B. L. 48. Kagan, R. Nussinov and R. Lal, J Phys Chem B, 2012, 116, 1728‐1735. 49. M. Winterhalter, Current Opinion in Colloid & Interface Science, 2000, 5, 250‐ 255. 50. E. Drolle, F. Hane, B. Lee and Z. Leonenko, Drug Metab Rev, 2014, 46, 207‐223. B. Moores, E. Drolle, S. J. Attwood, J. Simons and Z. Leonenko, PLoS One, 2011, 51. 6, e25954. 52. H. Jang, F. T. Arce, M. Mustata, S. Ramachandran, R. Capone, R. Nussinov and R. Lal, Biophys J, 2011, 100, 1775‐1783. 53. S. J. Soscia, J. E. Kirby, K. J. Washicosky, S. M. Tucker, M. Ingelsson, B. Hyman, M. A. Burton, L. E. Goldstein, S. Duong, R. E. Tanzi and R. D. Moir, PLoS One, 2010, 5, e9505. D. K. Kumar, S. H. Choi, K. J. Washicosky, W. A. Eimer, S. Tucker, J. Ghofrani, A. 54. Lefkowitz, G. McColl, L. E. Goldstein, R. E. Tanzi and R. D. Moir, Sci Transl Med, 2016, 8, 340ra372. 55. J. A. Elman, H. Oh, C. M. Madison, S. L. Baker, J. W. Vogel, S. M. Marks, S. Crowley, J. P. O'Neil and W. J. Jagust, Nat Neurosci, 2014, 17, 1316‐1318. 23
1712.01981	2	1712	2017-12-14T23:12:28	Multi-modal transport and dispersion of organelles in narrow tubular cells	[ "physics.bio-ph" ]	Intracellular components explore the cytoplasm via active motor-driven transport in conjunction with passive diffusion. We model the motion of organelles in narrow tubular cells using analytical techniques and numerical simulations to study the efficiency of different transport modes in achieving various cellular objectives. Our model describes length and time scales over which each transport mode dominates organelle motion, along with various metrics to quantify exploration of intracellular space. For organelles that search for a specific target, we obtain the average capture time for given transport parameters and show that diffusion and active motion contribute comparably to target capture in the biologically relevant regime. Because many organelles have been found to tether to microtubules when not engaged in active motion, we study the interplay between immobilization due to tethering and increased probability of active transport. We derive parameter-dependent conditions under which tethering enhances long range transport and improves the target capture time. These results shed light on the optimization of intracellular transport machinery and provide experimentally testable predictions for the effects of transport regulation mechanisms such as tethering.	physics.bio-ph	physics	Multi-modal transport and dispersion of organelles in narrow tubular cells Department of Physics, University of California San Diego, La Jolla, California Saurabh Mogre and Elena F. Koslover∗ (Dated: May 17, 2018) Intracellular components explore the cytoplasm via active motor-driven transport in conjunction with passive diffusion. We model the motion of organelles in narrow tubular cells using analyt- ical techniques and numerical simulations to study the efficiency of different transport modes in achieving various cellular objectives. Our model describes length and time scales over which each transport mode dominates organelle motion, along with various metrics to quantify exploration of intracellular space. For organelles that search for a specific target, we obtain the average capture time for given transport parameters and show that diffusion and active motion contribute com- parably to target capture in the biologically relevant regime. Because many organelles have been found to tether to microtubules when not engaged in active motion, we study the interplay between immobilization due to tethering and increased probability of active transport. We derive parameter- dependent conditions under which tethering enhances long range transport and improves the target capture time. These results shed light on the optimization of intracellular transport machinery and provide experimentally testable predictions for the effects of transport regulation mechanisms such as tethering. I. INTRODUCTION Transport of cargo within the intracellular environ- ment is a highly essential and tightly regulated process. Most eukaryotic cells have an active transport machin- ery consisting of molecular motors moving on a network of cytoskeletal polymers such as microtubules or actin filaments. Organelles can couple directly to motor pro- teins via specialized adaptors [1], or hitch-hike on other motile organelles [2]. This mode of transport results in motion that is processive over variable length scales up to many microns. Many organelles execute bidirectional motion, switching direction between processive runs by either engaging alternate motor types or transferring to a cytoskeletal track with different orientation [3–7]. In addition to this motor-driven processive transport, effectively diffusive motion of organelles can arise due to thermal noise, active fluctuations of cytoskeletal net- works [8], or hydrodynamic entrainment in flow set up by moving motors and cargo [9]. Evidence has shown that the short time-scale movement of organelles appears ef- fectively diffusive even when the underlying cytoplasmic medium is primarily elastic [8, 10, 11]. For brevity, we will refer to this stochastic motion of organelles as passive diffusion, while acknowledging that the fluctuations un- derlying the motion can have a number of actively driven origins. The interplay between active and passive transport modes gives rise to length-scale dependent effects. While active transport is efficient at transporting cargo over relatively long cellular distances, diffusion can more ef- fectively spread organelles over smaller length scales. Prior modeling work demonstrated the role of multi- modal transport in intracellular signaling and reduction ∗ [email protected] of noise, particularly in the case where organelles can- not carry out their biological functions in the actively walking state [12, 13]. Other theoretical studies have demonstrated that limited processivity can enhance tar- get search even in the absence of a passive or diffusing state, highlighting the importance of bidirectional mo- tion [14]. These results suggest that the transport ma- chinery in the cell may be optimized to allow substantial contributions from both processive and diffusive trans- port. Endosomes, peroxisomes, lipid droplets, mitochon- dria and mRNA are some example intracellular species known to employ multi-modal transport to move around within the cell [15–18]. The organization of the cytoskeletal network has a po- tentially important role to play in the distribution of in- tracellular particles. While a number of past models for intracellular transport employed a continuum approxi- mation for cytoskeletal density [12, 19, 20], it is becom- ing clear that the specific arrangement of distinct cy- toskeletal tracks has a substantial impact on cargo trans- port [21]. Obstructions due to intersecting microtubules may cause particles to pause or switch tracks and change the direction of movement [6]. Localized traps arising from heterogeneous filament polarity have been found to hinder transport in cell-scale computational models [21]. In tubular cell projections such as neuronal axons and fungal hyphae tips, the arrangement of cytoskeletal fil- aments is highly simplified, with microtubules aligned along the tubular axis and in many cases uniformly po- larized towards the distal tip [22, 23]. These projections range in length from tens to many hundreds of microns, and require cargo to be efficiently transported from the cell body to the distal tips and back again. In addi- tion to being particularly amenable to theoretical models of transport phenomena, these cell types are of funda- mental biological importance. Defects in axonal trans- port in neurons have been implicated in a number of human pathologies, ranging from multiple sclerosis to Alzheimer's to prion diseases [24] . Due to their sim- plified morphology and long length, these tubular cells form an ideal system for investigating the length-scale dependent effects of multimodal transport. The discrete nature of cytoskeletal tracks within tubu- lar cell projections limits active transport to narrow ax- ially oriented bundles of microtubules [23]. It has been proposed in several cellular systems that transport effi- ciency is increased by directly tethering organelles to the microtubules in order to prevent them from losing access to the tracks [7, 25, 26]. Tethering can occur by special- ized adaptor proteins binding the organelle to cytoskele- tal tracks, as in the case for axonal mitochondria that become preferentially anchored in cellular regions with high metabolic needs [27–29]. Alternately, the binding of multiple motor proteins to individual vesicles results in a tethering effect that is believed to contribute to ob- served motor cooperativity [7, 30]. Because tethering is expected to hinder short-range dispersion while en- hancing the ability of organelles to engage in long-range processive walks, it can potentially serve as a regulatory mechanism for length-dependent transport. A variety of cellular processes rely on efficient trans- port to achieve distinct objectives necessary for biologi- cal function. One such objective is the establishment of a uniform distribution of particles throughout the cell, as is observed for peroxisomes, mitochondria and lipid droplets [32–34]. Establishing this distribution, starting from the point of genesis of particular organelles, requires rapid transport and broad dispersion across long cellu- lar length-scales. Another objective is the delivery of organelles to specific subcellular regions. Examples in- clude the motion of synaptic vesicles from the cell body to the pre-synaptic terminal of neuronal axons [35], and the transport of vesicles containing newly synthesized membrane-bound proteins from the Golgi apparatus to the cell boundary [36]. The role of different transport modes in this process depends on the length scale of separation between the site of organelle synthesis and their eventual target. A third cellular objective is the rapid encounter between an intracellular target and any one of a uniformly distributed population of organelles. For instance, peroxisomes serve to neutralize oxidative metabolic byproducts, and the health of a cell is depen- dent on rapid removal of these toxic species as soon as they appear [37]. Similarly, early endosomes rely on con- tact with any of a population of lysosomes that aid in releasing the endosomal contents into the cytoplasm [38]. The efficiency of such target encounter depends both on the nature of transport processes for the organelles and on their density within the cell. In this article, we present a simplified model for trans- port in a tube through a combination of processive walks and diffusion. We analyze the relative contributions of the two transport modes, as well as the possibility of tethering to cytoskeletal tracks, in achieving the differ- ent transport objectives of the cell. Sec. II establishes our halting creeper model and its behavior in terms of 2 the rate with which particles explore a one-dimensional environment. In Sec. III, we use the developed model to study the effects of bidirectional transport on distributing particles uniformly within a domain. In Sec. IV, we ex- plore the contribution of different transport modes to the delivery of individual particles, as well as target clearance by a dispersed particle population. Sec. V introduces an expanded model that accounts for particle tethering, de- lineating the effects of this mechanism on organelle dis- persion and target capture times. II. HALTING CREEPER MODEL We define a simplified stochastic model for intracellular particles undergoing multi-modal transport, focusing on motion along a single dimension. Each "halting creeper" particle exists in either a passive diffusive state charac- terized by diffusion coefficient D or an actively moving state with constant speed v in either the positive or neg- ative direction (Fig. 1). Switching between the states is a Markovian process with constant starting rate γ for transitioning from the passive to the active state and constant stopping rate λ for transitioning from active to passive. Selection of the direction of motion is random at each initiation of an active run, and we assume com- plete symmetry between forward and backward motion. We note that this model is a more general form of pre- viously defined creeping particle models [14], which have been analyzed in the limit γ → ∞, D → 0. Further- more, a three-dimensional version of our halting creeper model has previously been explored in the context of mean squared displacement and local concentration fluc- tuations [12]. By contrast, in this work we focus explicitly on the efficiency with which such two-state transport dis- tributes organelles throughout a cell and delivers them to intracellular targets. Two important quantities which describe the behavior of a halting creeper particle are the active run-length (ℓ = v/λ) and the equilibrium fraction of particles in the active state (f = γ λ+γ ). For much of the subsequent discussion, we non-dimensionalize all length units by the run-length ℓ and all time units by the run-time 1/λ. We denote the remaining dimensionless parameters as D = ℓ2λ , γ = γ D λ , and dimensionless time as t = λt. FIG. 1. Schematic for the transition between particle states. 'D' denotes diffusive particles, '+' denotes particles moving processively in the rightward direction, whereas '-' denotes particles moving processively in the leftward direction. The arrows are labeled with transition rates between states. The Markovian nature of the transitions between ac- tive and passive states allows the calculation of a spatio- temporal propagator function G(x, t) for the halting creeper, which gives the distribution of positions at time t given the particle started at the origin at time 0. This propagator is obtained by convolution in the space and time domain of the individual propagators for passive and active transport. After a Fourier transform in space (x → k, G → eG) and a Laplace transform in time (t → s, eG → beG), the multi-modal propagator for particles ini- tially in an equilibrium distribution between passive and active states is given by 3 (λ + s)(γ + λ)(γ + λ + s) +(cid:0)Dγ(λ + s) + v2λ(cid:1) k2 (γ + λ) [s(λ + s)(γ + λ + s) + (D(λ + s)2 + v2(γ + s)) k2 + v2D k4] , (1) beG(k, s) = as derived in Appendix A. This propagator serves as the basis for our subsequent calculations on the efficiency of particle spreading and target site search. A. Particle spreading: mean squared displacement The mean squared displacement (MSD) is a commonly used measure of spreading speed for diffusing particles. For the halting creeper model, it can be calculated di- rectly from the propagator as (cid:10)x(t)2(cid:11) = L−1"− ∂k2(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)k=0# ∂2bG = 2(1 − f ) Dt + 2fht + (e−t − 1)i , where the Laplace inversion L−1 is carried out analyti- cally via the residue theorem. This expression for the MSD is composed of a linear superposition of fraction 1 − f of diffusing particles and fraction f of particles undergoing active walks that are persistent over a dimensionless time-scale of 1. The latter component corresponds to an MSD that scales ballisti- cally as f t2 for t ≪ 1 and diffusively as 2f t for t ≫ 1. In the case of small diffusivity, there is an additional tran- sition time when the ballistic motion begins to dominate over the passive diffusion. This occurs at t∗ = 2 D γ . When t∗ ≪ 1, the MSD transitions from diffusive to ballistic and back to diffusive scaling (Fig. 2a). The long-time behavior of the particle is defined by an effective diffu- sion coefficient Deff = (1 − f ) D + f, (2) which in the limit of t∗ ≪ 1 is dominated by the term corresponding to bidirectional active walks ( Deff → f ). The relative importance of processive versus diffusive transport over a length-scale x can be characterized by the P´eclet number [12]: Pe(x) = vx/D, which is a dimen- sionless quantity often used to compare the contributions from advection and diffusion for particles in a flowing fluid [47]. A large P´eclet number Pe ≫ 1 corresponds to transport that is dominated by the processive motion. In the case where active motion remains processive only up to distances comparable to the run length ℓ, the relevant P´eclet number for long-range transport is P e(ℓ) = 1/ D. Our dimensionless diffusion constant thus describes the relative contribution of diffusion above processive mo- tion over a length scale comparable to the average run- length. For the remainder of the discussion, we focus on the regime where the transition time t∗ = 2 D γ < 1, so that a distinct regime of processive motion appears between the regimes dominated by passive diffusion and effectively diffusive bidirectional walks. This is the case for the organelle transport examples listed in Table I. We note in passing that the presence of a discernible proces- sive motion is key to identifying active runs in experi- mental particle-tracking data[32, 42, 48], so that systems not in this regime are unlikely to be selected for studies of active transport. B. Particle spreading: range An alternate metric for the efficiency of particle spread- ing is the overall range – or the average size of the domain that has been explored by a halting creeper particle after an interval of time. For a one-dimensional model, the range of each particle is given by its maximum position minus its minimum position over the course of its trajec- tory. As will be discussed further in Section IV, the range is directly related to rate at which a dispersed population of particles first encounters a target. Our model permits calculation of the range over time for a halting creeper using the renewal equation method [14, 49]. Namely, we define the distribution of first passage times to a target at position x > 0 (for a particle starting in a diffusive state at the origin) as FD(t; x) = FD+(t; x) + FDD(t; x), where FD+ gives the distribution of first passage times for the fraction of par- ticles that arrive at the target while walking in the posi- tive direction and FDD gives the distribution of times for particles arriving in the passive diffusive state. Similarly, we consider the components of the propagator function defined in Appendix A, where GDW (x; t) gives the spa- 4 Transport System Peroxisomes in fungal hyphae [32] Lysosomes in kidney cells [6, 39] Mouse neuron transport vesicles in vitro [40] Mitochondria in Drosophila axons [41] Dense core vesicles in Aplysia neurons [42, 43] PrPC vesicles in mouse axons [40, 44] Rate of Rate of switching Velocity of switching to active transport to passive transport active transport Diffusivity Density of the population Approx. size of cellular region γ (s−1) λ (s−1) v (µm/s) D (µm2/s) ρ (µm−1) L (µm) 0.015a 0.17 0.33a 0.17 0.22a 0.36 0.29 0.15 2.7 0.15 2.2b 0.15 1.9 0.52 0.8 0.35 0.014 0.071 0.03 – 0.36c 0.002c 0.85 − 1.5 – 0.14 1.3 1.7 0.4 50 20 – 1000 100 100 a Estimated from equilibrium fraction in active state. b Estimated from single particle trajectory. c Estimated from MSD plot TABLE I. Estimated values of transport parameters for some biological systems. Run length can be obtained as ℓ = v/λ. Parameters can be converted to dimensionless units according to: D = Dλ/v2, γ = γ/λ, ρ = ρv/λ. GDD(x; t) =Z t 0 dt′(cid:2)FDD(t′; x)GDD(0+; t − t′) + FD+(t′; x)G+D(0+; t − t′)(cid:3) , tial distribution at time t of particles that began in a diffusive state at the origin at time 0, and are found in the actively walking state at time t. The other com- ponents GDD, GW D, GW W are defined analogously, with additional expressions for G+D, G+W giving the propaga- tor for particles that are initially walking in the positive direction, and end up in either the diffusive or the ac- tively walking state. One of the renewal equations for this system is then given by, ǫ→0+ where G+D(0+; t′) = lim G+D(ǫ; t′). This expression describes a convolution between the probability that the particle first hits the target x at time t′ and then returns to position x within the remaining time t− t′. Analogous renewal equations are derived for GDW , GW D, GW W . Af- ter a Laplace transform in time, this convolution struc- ture can be expressed as a product, which yields a system of equations, This system can be solved to calculate the first passage bGDW (x)(cid:21) . (3) bGDW (0+) bG+W (0+)(cid:21)(cid:20) bFDD (cid:20) bGDD(0+) bG+D(0+) time bFD for particles that started in the diffusive state. bGW W (x)(cid:21) . (4) bGDW (0+) bG+W (0+)(cid:21)(cid:20) bFW D (cid:20) bGDD(0+) bG+D(0+) bFD+ (cid:21) =(cid:20) bGDD(x) bFW + (cid:21) =(cid:20) bGW D(x) An analogous system yields the first passage time for particles that began in the active state: The range of the halting creeper particles over time (Z(t)) can be related to the Laplace-transformed first (5) Z(t) = L−1(cid:20) 1 passage time bF (s; x) according to [14], −∞ bF (s; x)dx(cid:21) , sZ ∞ where bF is a linear combination of bFD and bFW , weighted bGDD(0+), bG+D(0+), bGDW (0+), bG+W (0+). of the renewal equation being expressed as beGDD(k = 0), by the equilibrium probability that the particle starts in an active or a passive state. To calculate the range func- tion, we analytically perform the Fourier inversion of the propagators The spatial integral over x results in the right hand side etc. While short-time and long-time limits of the range can be obtained directly from the large s and small s lim- its of the renewal equations, the relevant time-scales for biological processes can span across many orders of mag- nitude, thus making it desireable to calculate the particle spreading efficiency over all time scales. To this end, we invert the Laplace transform numerically using Talbot's algorithm [50]. As shown in Fig. 2, the range exhibits similar tran- sitions in scaling as the MSD. However the transitions between the different regimes are shifted to longer times. In the short time limit, the average range of particles with an equilibrium initial distribution between active and passive states is given (in dimensionless units) by Z(t) → 4(1 − f )s D πpt + f t. (6) 5 FIG. 2. Contribution of passive and active motion to spreading of particles at different length and time scales. (a) Mean squared displacement for halting creepers. Black dash-dotted lines show scaling regimes. Vertical dashed lines indicate transition times between the regimes. Horizontal dashed lines indicate transition length scales. (b) Dimensionless range versus time for a halting creeper particle, with scaling regimes, transition times, and transition length scales indicated. Dotted curve shows the root mean squared displacement for comparison. (c) Range versus time for typical parameter values for intracellular organelles, showing the increase in long-range transport with increasing run length, above a length scale indicated by the black dashed line. Thus, the ballistic motion dominates over the diffusive motion above a critical transition time time required for a single particle to transition between an active and a passive state and back again. t∗ range = 16 D πγ2 . (7) In the case where particles spend very little time in active motion (γ ≪ 1), this time-scale is substantially longer than the transition time t∗ for the MSD. The correspond- ing length scale for the transition from primarily diffusive to primarily ballistic motion is x∗ range = 16 D πγ(1 + γ) . (8) At longer times, there is a subsequent transition from the ballistic scaling of the range to the effectively diffusive long-time scaling, Z(t) → 4s Deff π pt, which occurs at a secondary transition time t∗∗ corresponding length scale x∗∗ range given by, t ≫ t∗∗ range, (9) range and t∗∗ range = x∗∗ range = 16 Deff πf 2 , 16 π 1 + D γ! . (10) This result highlights the fundamental insufficiency of the MSD in describing the efficiency with which the par- ticles explore their domain. Specifically, for a very small equilibrium walking fraction f , the time required for the active walks to contribute substantially to the average range can be well above the time-scale 1/γ for an indi- vidual particle to start walking. Similarly, in this regime the range will only exhibit diffusive scaling at time-scales long enough for individual particles to execute multiple starting and stopping transitions. Examples of parti- cle motion where the pause time substantially exceeds the processive run time include organelles (such as per- oxisomes) whose active transport is mediated by hitch- hiking on other organelles [2], and particles whose motion is driven by hydrodynamic entrainment due to cytoplas- mic flow associated with nearby passing particles [9]. In such cases, the MSD does not accurately represent the rate at which these particles explore their domain. We note that in the case where D < 1, which corre- sponds to most biologically relevant examples, increas- ing the run length (e.g.: by decreasing the stopping rate λ) raises the particle range for all length scales above x > 16D (Fig. 2), corresponding to the length at which πv the Peclet number Pe(x) becomes substantial. The impli- cation is that longer processive runs improve the ability of particles to explore their domain at all length scales where active walks move faster than diffusion. In the case of small fraction of time spent walking, this transition time is again shifted substantially above what would be expected from the MSD behavior, where the corresponding transition occurs at t∗∗ = 2. In the limit D/γ ≪ 1, the transition time for the range can also be expressed as t∗∗ π (1+1/γ), comparable to the cycle range = 16 III. PARTICLE DISPERSION THROUGH BIDIRECTIONAL TRANSPORT Having established the speed of particle spreading via multi-modal bidirectional transport, we now turn to con- 6 B length units are non- FIG. 3. Dispersion of particles towards a uniform distribution via bidirectional transport. All dimensionalized by domain length L and all time units by L/v. (a) Particle distribution density for different run lengths, at dimensionless time 0.3. (b) Entropy vs time for different run lengths. The horizontal dash-dot line denotes the threshold entropy for the system to be considered well-mixed. The vertical green dash-dot line is at dimensionless time 0.3 (c) Time to reach a well-mixed state as a function of the run length (ℓ) and rate of transition to an active state (γ). Points A and B are drawn at corresponding transport parameters for lysosomes in monkey kidney cells, and PrPC vesicles in mouse axons, respectively (Table I). sider explicitly the efficiency with which such transport can achieve a particular cellular goal. Certain metabolic and regulatory needs of the cell require a well-dispersed distribution of organelles throughout the cell interior. For instance, mitochondria are found throughout neu- ronal axons, providing a locally available energy source through glucose metabolism [28]. In fungal hyphae, peroxisome organelles are maintained at near uniform distribution[32], allowing for rapid neutralization of toxic metabolic byproducts [37]. Establishing a well-mixed distribution relies not only on the ability of particles to move rapidly through the cell, but also on the ability of a transport mechanism to disperse and flatten regions of highly concentrated parti- cles. We focus specifically on the rate with which a bolus of particles is spread over a cellular region. Such a pro- cess becomes necessary, for instance, in the case of rapid organelle production in response to an external signal, where the organelles must then be spread through long cellular projections such as axons or hyphae. We use the halting creeper model to explore how dif- ferent transport parameters affect the efficiency of such dispersion. Because we are interested in the initial es- tablishment of an equilibrium spatial distribution, we consider particles that originate at x = 0 in the pas- sive state, whose distribution is given by GD(x, t) = GDD(x, t) + GDW (x, t). This function can be evalu- ated by numerical Fourier-Laplace inversion of the trans- formed distribution, as described in Appendix A. The time evolution of the distribution is plotted in Fig. 3a. Note that long walk lengths result in little dispersion of particles, with the distribution splitting into two narrow, processively moving peaks. Short walk lengths lead to an effectively diffusive motion, with the particle distri- bution assuming the form of a slowly spreading Gaus- sian. An intermediate walk length combines both the rapid spreading of the distribution with the flattening of localized peaks to enable more efficient dispersion. The limits for large and small walk lengths suggest that there exists an optimal run length ℓ for which the particles are most efficiently mixed. A number of different metrics have been developed for quantifying the rate of mixing driven by stochastic pro- cesses [51–53], including several that track the approach of a bolus of particles towards uniform spread [54, 55]. A commonality of these measures is their dependence on a particular length scale of interest [53] over which par- ticles are to be mixed. For our one-dimensional system, we introduce a length L corresponding to the size of the domain on which uniform distribution is desired. This length represents the approximate extent of the tubular cell region across which particles are being dispersed. It can range over many orders of magnitude, with mam- malian axons reaching up to a meter in length. Example values for some cellular systems are listed in Table I. We calculate the spatial distribution of halting creeper particles originating in the center of a domain of length L with reflecting boundary conditions, implemented us- ing the standard image method [56]. The mixing of the particles is quantified via the Shannon entropy of the dis- tribution [55, 57], defined as S = − NXi=1 pi log(pi) log(N ) , (11) where the domain is broken up into N bins, and pi is the probability of a particle being located in bin i. Optimal mixing is achieved when the organelles are uniformly dis- tributed, in which case pi = 1 N and S = 1. Conversely, 7 FIG. 4. Target capture times for a single particle. (a) Cumulative encounter probability for different initial distances to the target (x). The dotted and dashed lines denote the encounter probability to a target at distance x = 0.01 by diffusive particles with diffusivity D, and to a target at distance x = 10 by diffusive particles with diffusivity Deff, respectively. The dash-dot line denotes the average time required for a particle in the active state to cover a distance of x = 1. All length units are non-dimensionalized by ℓ and all time units by ℓ/v. (b) Time to reach 90% capture probability for different run lengths, assuming rapid starting rate γ = 1s−1 and distances appropriate for intracellular organelle transport. The dotted and dashed lines denote t90% for diffusive particles with diffusion coefficient D and Deff , respectively. a distribution with all particles in a single section is the least mixed state, with S = 0. The entropy has an inher- ent dependence on the number of bins used for discretiz- ing the probability distribution, and we employ N = 5000 throughout our calculations. The time evolution of the entropy is dependent on the dimensionless run-length ℓ/L (Fig. 3b), with long runs corresponding to an initially slow rise in entropy as the bolus of particles evolves into two coherent spatial peaks until sufficient reversals are achieved to disperse the par- ticles throughout the domain. Short run lengths limit the rate of entropy increase over long times, because the par- ticle distribution spreads slowly as an effectively Gaus- sian peak. We consider the system to be well-mixed when the entropy crosses a threshold value St = 0.9 and de- fine the time taken to reach this state as the mixing time tmix. This mixing time depends in a non-monotonic fash- ion on both the starting rate (γ) and run length (ℓ) of processive walks (Fig. 3c). High values of γ, correspond- ing to particles that spend most of their time in the active state, give rise to an optimum run length to achieve the most rapid mixing. This effect arises from the need to reverse active walking direction in order to efficiently dis- perse particles within the domain. However, each such reversal necessitates a waiting time of 1/γ during which the particles are in a passive state and spreading very slowly. Consequently, at low values of γ mixing is most efficiently achieved by particles that carry out very long walks. The results shown in Fig. 3 assume a small value of passive diffusivity ( D Lv = 0.01). Increasing this diffu- sivity would lead to a monotonic rise in the entropy, as diffusion enhances the particle mixing. IV. TARGET SEARCH BY MULTIMODAL TRANSPORT A. Search by a single particle In addition to achieving uniform dispersion of parti- cles, another goal of intracellular transport is to deliver organelles to specific cellular regions. This transport ob- jective arises, for instance, when synaptic vesicles must reach the presynaptic bouton of a neuron [35]. Using our one-dimensional halting creeper model, we consider the first passage time of a single particle towards a stationary target located at distance x. For simplicity, we consider the case where x is much smaller than the overall extent of the domain, so that the distance to the target x and the processive run length ℓ are the only relevant length scales in the problem. As in the case of our dispersion calcu- lations (Sec. III), we consider particles that are initially in the passive state, as applicable to the distribution of newly synthesized organelles. The distribution of first passage times can be obtained from the renewal equation Eq. 3, by carrying out analytic Fourier inversion followed by numerical Laplace inversion of the propagators (see Appendix A). The cumulative distribution of encounter times to the target is plotted in Fig. 4a, showing the transition from a passively diffu- sive process at small distances (x < x∗ range) to an effec- tively diffusive process (with diffusivity Deff) at distances much longer than the run length (x > 1). Intermediate distances show a sharp increase in the cumulative prob- ability of target encounter at time t = x, corresponding to the arrival of the first processively walking particles. 8 FIG. 5. Average time for target cap- ture by a population of uniformly distributed particles, with two dif- ferent densities (left: ρ = 10 and right: ρ = 0.3). All units are non- dimensionalized by run length ℓ and run time ℓ/v. Marked points show estimated parameters for two cellular systems (A: peroxisome transport in fungal hyphae [2], and B: vesicle trans- port in Aplysia axons [42]; see Ta- ble I). Black lines mark the transi- tion between diffusion-dominated and transport-dominated motion on the length scale of inter-particle distance (Eq. 16). B. Search by a population of particles A closely related objective of intracellular transport is the capture of a target by any one of many moving par- ticles. In this case we assume particles that are initially uniformly distributed with some density ρ, and consider the mean first passage time (MFPT) for the first of them to hit the target. Some biological examples include the clearance of toxic cytoplasmic metabolites by any one of a uniformly scattered field of peroxisome organelles [37], the influx of peroxisomes to plug septal holes in damaged fungal hyphae [59], or the arrival of lysosomes to fuse with a phagosome and digest its engulfed contents [38]. For simplicity, we assume the target is itself immobile and must wait for the particles to come to it via some combi- nation of active and passive transport. In this situation, the relevant length scale is defined by the typical ini- tial spacing between the particles (1/ρ). In the limit of a uniform distribution over a very long domain, the MFPT can be related directly to the range of the moving parti- cles [14]. Specifically, the mean first passage time is given by MFPT =Z ∞ 0 e−ρZ(t)dt (13) where Z(t) is the average range of particles over time t. The behavior of the MFPT is dictated by the dimen- sionless length scale for the separation between particles [1/ ρ = 1/(ρℓ)]. When this length scale is short enough that active walks remain processive (1/ ρ ≪ x∗∗ range), we can approximate the particle range as a linear combina- tion of a diffusive and a ballistic process (Eq. 6). The integral for the MFPT can then be approximated ana- lytically as 1 f ρ −s 4 D γ2 ρf MFPT ≈ range ρ exp(cid:20) x∗ 4 range ρ (cid:21) erfc(cid:20) x∗ 4 (cid:21) , (14) A Because the average first passage time of a random walk in a semi-infinite domain diverges [56], we focus on the time required for particles to hit the target with suffi- ciently high probability. Analogously to our calculations of particle dispersion in Section III, we define the hitting time t90% as the time by which there is a 90% chance that the particle has hit the target. We note that t90% is well-defined even on a semi-infinite domain due to the recurrent nature of random walks in one dimension, en- suring a finite hitting time for all particles [56]. For short distances, the time for probable encounter of the parti- cle scales as expected for purely diffusive motion with diffusivity D (Fig. 4b), 90 (x; D) = tdiff (12) x2 4 D(cid:2)erf−1(0.1)(cid:3)2 where erf−1 is the inverse of the error function. At long times, a similar scaling is observed with effective diffu- sivity Deff. As is the case when the transport objective is to achieve a uniform distribution of particles, increasing the length of processive runs does not necessarily result in more effi- cient transport. This is true despite the fact that, unlike previous models of multimodal transport [12, 13, 58], we consider our particles capable of accessing their target in both the passive and active states. A run length that is much longer than the distance to the target can hin- der particle delivery, because particles have a 50% chance of initiating their motion in the wrong direction. They then require a long time to stop, turn, and return towards the target. At the same time, very short processive runs decrease the overall rate of spread for the particle distri- bution and thus slow down the target encounter. These two effects give rise to an optimum in the efficiency of tar- get delivery, with minimal values of target hit time t90% occuring at intermediate run lengths ℓ (Fig. 4b). This effect is a direct analogue to the optimum walk-length for achieving uniform distribution. The existence of this optimum walk length has also previously been noted for creeper models without any paused or passive state [14]. where erfc is the complementary error function. The lim- its for high and low particle concentration are given by MFPT → π(1 + γ)2 8 D ρ2 , MFPT → 1 f ρ , ρ ≪ 1 range ≪ ρ ≪ x∗ 1 , x∗ range 1 x∗∗ range , (15) where the high density limit corresponds to diffusive scal- ing of MFPT with the distance between particles while the low density limit corresponds to ballistic scaling. Set- ting these two limits equal to each other indicates that a transition in the encounter times occurs at a critical length scale 1 ρcrit = x∗ range 2 , which can be equivalently expressed as f Pe(cid:18) 1 ρ(cid:19) = 8 π(1 + γ)2 . (16) (17) This transition corresponds to a particle density where processive walks begin to dominate the ability to rapidly encounter targets, which occurs when the P´eclet num- ber for the distance between particles, multiplied by the fraction of time spent walking, is of order unity. A calculation of the mean first passage time accurate at all length scales can be carried out by numerical inversion of the Laplace-transformed range function (Sec. II B), and the results are plotted in Fig. 5 for two values of particle density. The black line indicates the transition between behavior dominated by diffusive versus by pro- cessive particle motion (Eq. 16). Below this line, active transport dominates the motion of the particles and the time to reach the target is insensitive to the passive dif- fusivity. Above this line, passive diffusion dominates and the target search is insensitive to the fraction of time that the particles spend in processive motion. The parame- ters relevant to two example biological systems (peroxi- some transport in fungal hyphae and vesicle transport in Aplysia neurons) are marked with dots. The two example cases fall near the transition region, where both passive diffusion and active processive walks contribute to the ability of these organelles to reach any target position within the cell. While previous modeling studies have indicated that both transport mechanisms are important to the maintenance of a uniform distribu- tion of peroxisomes in hyphae [32], we demonstrate here that the particle density falls in an intermediate regime such that diffusion and active walks both contribute to efficient target search by the population of peroxisomes. V. TRANSPORT IN A TUBE AND THE BENEFITS OF TETHERING Active transport in a cell occurs via motor proteins attached to microtubule tracks. Even very narrow cel- lular projections are typically substantially wider than 9 the diameter of a single microtubule. Consequently, or- ganelles must navigate transversely through the cytoplas- mic environment in order to encounter a microtubule and engage in active processive motion. A mechanism to keep organelles located close to the microtubule can improve transport efficiency by reducing this search time. In many cases, organelles are believed to be tethered to the microtubule tracks, preventing them from dissociat- ing and diffusing even when they pause after a processive walk. This tethering can be accomplished by additional inactive motors attached to the organelle [25, 31] or by specific molecular adaptors linking the organelle directly to the microtubule [27, 60]. It has been speculated that tethering can enhance transport by forcing the organelle to remain in proximity to the microtubule tracks, thereby effectively increasing the rate at which processive walks are initiated [7]. At the same time, tethering can severely limit the intracel- lular space that can be explored by an organelle in the passive state, either by reducing the axial diffusivity in the case where inactive motors slide diffusively along mi- crotubules [25, 26], or by halting it entirely in the case of organelle docking [27]. The benefits of tethering thus de- pend on the relative balance between active and passive transport, as well as the radial size of the domain around the microtubule, which determines the delay associated with encountering the track. The former aspect is de- pendent on the length scale over which transport must be achieved, as discussed in the previous sections. We extend our halting creeper model to a three- dimensional cylindrical domain of radius R, wherein ac- tive runs can be initiated only within a radius of size a < R, corresponding to a small region surrounding a central track. While cellular projections such as hy- phae and axons generally have multiple microtubule bun- dles [32, 61], this model serves as an approximation where the size of the cylindrical domain sets the cross-sectional density of the microtubule bundles. In addition to bidi- rectional walking and passive diffusion states, the parti- cles in this extended model can also enter a tethered state with rate kb while within the encounter radius a. For simplicity, we assume particles in the tethered state are entirely immobilized. The model could be extended in a straight-forward manner to limited but non-zero diffu- sivity while in the tethered state. Exit from the tethered state occurs at rate ku, with the particle unbinding to a uniform radial distribution within the capture radius a. A dimensionless binding strength for tethering is defined by Keq = kb/ku. We note that this model assumes that tethering does not in any way hinder the initiation of an active run, so that particles transition to the active state with the same rate regardless of whether they are bound or freely diffusing within the capture radius. While it is possible for tethering to either speed up or slow down the associ- ation of an organelle with a motor or a carrier particle, depending on the length, flexibility, and configuration of the tether, we neglect this effect here. Our model for 10 FIG. 6. Effects of tethering on transport. a) Schematic for the model of a halting creeper in a cylinder, with tethering. The smaller cylinder denotes the region within which particles can tether to the microtubule or initiate active transport. Rates of transition between states are labeled with corresponding arrows. b) Range vs time for weak (Keq = 0.1) and strong (Keq = 500) tethering. Dashed black lines show analytical approximations in the limits of no tethering and infinitely strong tethering, accurate for short to intermediate times (Eq. 21). Horizontal dash-dotted line indicates the transition length-scale Lcrit where tethering becomes advantageous. c) Average time for target capture by a population as a function of the starting rate γ and binding strength Keq. The solid line indicates the transition from diffusive to active transport as the dominant transport mode at different values of tethering strength. The dashed line shows the transition where strong tethering becomes advantageous for target encounter. tranport in a cylindrical tube around a microtubule track is summarized schematically in Fig. 6a. In the limit of rapid transverse diffusivity or small domain size (D/R2 ≫ γ, kb), diffusive particles remain equilibrated throughout the cross section of the domain, and the effective rates of starting a walk or binding be- come α2γ and α2kb, respectively, where α = a/R. In this limit, the delays associated with transverse diffusive transport are elimitated, and the equilibrium fraction of particles in each state can be easily calculated. For par- ticles starting at equilibrium, the long-time diffusivity is then given by Deff = Dfdiff + fwalk, γα2 fwalk = γα2 +(cid:16) α2Keq+γ/ku+1 Keq+γ/ku+1 (cid:17) , (cid:16) γ/ku+1 Keq+γ/ku+1(cid:17) γα2 +(cid:16) α2Keq+γ/ku+1 Keq+γ/ku+1 (cid:17) fdiff = (18) where fwalk and fdiff are the fraction of particles in the active and diffusive state, respectively. We again non- dimensionalize all length units by the run-length ℓ and all time units by the run-time ℓ/v, for consistency with previous calculations. In the more general case where the delay due to trans- verse diffusion is included, it can be shown (see Ap- pendix B) that for a particle which begins uniformly dis- tributed in the diffusive state within radius a, the mean waiting time to enter a walking state is identical to the fast-diffusion limit and is given by, 1 γ" α2Keq + 1 + γ/ku α2(Keq + 1 + γ/ku)# . (cid:10)tw(cid:11) = (19) This average time ranges from 1/(α2γ) in the limit of low binding strength to 1/γ in the limit of strong binding, and is independent of the diffusivity D. In the case of very slow diffusion, those particles that escape the binding radius a take a long time to return, but such escape be- fore initiating a walk becomes concomitantly less likely, with these two effects canceling each other out in the calculation of the average time to start walking. Because particles are assumed to distribute uniformly across ra- dius a when leaving the tethered state, this equivalence of the average time to initiate a subsequent walk means that the long-time behavior of particles matches the fast- diffusivity limit, regardless of the actual value of D. By contrast, we note that the standard deviation in the time required to start a walk, for a particle that starts diffusive and uniformly distributed within a, is depen- dent on the diffusivity (see Appendix B). Slow diffusion and strong binding can greatly increase the variance in the time required for a particle to start a walk, lead- ing to large variability in the amount of time individual particles remain in a passive or tethered state over a par- ticular time interval of observation. This extreme case may contribute to the identification of apparently im- mobile populations of particles observed in some in vivo organelle tracking studies [33]. The effectiveness of tethering in improving transport over a long time can be inferred from the derivative of the effective diffusivity Deff with respect to the binding strength Keq. A positive derivative signifies that long- range transport is accelerated by tethering, whereas a negative value indicates that tethering hinders transport. Tethering is advantageous in the long-time limit when the following criterion is satisfied: (1 − α2)(cid:18) γ γ + 1(cid:19) Pe(ℓ) > 1. (20) This expression summarizes the idea that tethering is helpful for long-range transport in situations where the domain is wide (α ≪ 1), where the rate of walking is substantial compared to the pausing rate (γ ≫ 1), and where active runs move the particles faster than diffusion over the longest processive length scale (Pe(ℓ) ≫ 1). Below the long-time diffusive limit, the extent to which tethering aids transport depends on the length scale of interest. In particular, at times much shorter than the cycle time to initiate and stop an active walk, the dimen- sionless particle range can be approximated by Z(t) ≈ 4fdiffs Dt π + fwalkt, (21) in a manner analogous to Eq. 6. This expression can be inverted to calculate the time at which a particular range is reached. Comparing the low Keq and high Keq limits indicates that the ability of particles to tether to the track allows them to explore more rapidly over length scales above Lcrit = x∗ range (1 + γ)2 (1 − α2)2 . (22) For large domains (α ≪ 1) and low propensity for active walking (γ ≪ 1), the tethering is helpful over all length scales where processive active motion is the dominant form of transport, as defined by the critical length x∗ (Eq. 8). range We use kinetic Monte Carlo methods to simulate the spreading of particles within our cylindrical model. The simulations are accelerated with the use of analyti- cally calculated Green's functions to propagate the par- ticles within homogeneous cylindrical domains (see Ap- pendix C), allowing for efficient sampling of particle be- havior over a broad set of parameters. The average axial range for a population of particles can be obtained as a function of time from the simula- tions. Fig. 6b shows the time evolution of the range for weak and strong tethering. The transport parameters used are relevant for peroxisomes in fungal hyphae (Ta- ble I), with the domain width assumed to be R = 1µm and a central region of width a = 0.1µm. For consis- tency with previous calculations, results are reported in dimensionless units, using the run length (ℓ ≈ 7µm) and processive walking time (1/λ ≈ 3 s) as the length and time units. The critical length scale for this system is Lcrit ≈ 0.12, below which the average range for strongly tethered particles is lower than the weakly tethered ones. 11 For length scales above Lcrit, strongly tethered particles explore over a greater range. The full extent of a hy- phal growth tip ( L ≈ 8) is several times longer than this critical length scale, highlighting the potential benefit of tethering for distributing peroxisome particles over the entire growth tip. Having established the length scales over which teth- ering is advantageous, we now calculate explicitly the effect of tethering on the average search time by a pop- ulation of particles with dimensionless density ρ. The capture time is defined as the first passage time to an ar- bitrary cross-section of the cylinder, by a population of particles equilibrated between states and uniformly dis- tributed along the axis of the cylinder. A surface plot of the average capture times versus binding strength Keq and walking rate γ is shown in Fig. 6c. The effect of tethering on the average time to target capture varies depending on γ. For particles with a very small proba- bility of engaging in active runs, tethering hinders target search by limiting mobility in the passive state. For par- ticles with a high propensity for active motion, tethering can aid their ability to encounter targets by increasing the amount of time spent in the region where active runs can be initiated. We approximate the parameter regime where this transition occurs by analytically calculating the integral for the MFPT (Eq. 13), using the short time approximation of the particle range (Eq. 21). Comparing the low Keq and high Keq limits yields a transition at a critical particle density ρtether =(cid:18) 2 1 − α2 (cid:19)2 Lcrit(cid:19)(cid:18) 1 + α2γ . (23) For small values of α, γ, this transition is equivalent in form to the critical length scale where processive walks first begin to play an important role, as calculated in Eq. 16. For parameters relevant to the motion of per- oxisomes in fungal hyphae, we compare the critical par- ticle density (ρtether ≈ 17) with the observed density of peroxisomes (ρ ≈ 10). Because the observed density is comparable to the critical density, we expect that teth- ering would have only a weak benefit on the ability of the peroxisomes to patrol the cytoplasm and encounter targets within the cell. For a given finite binding strength Keq, the MFPT to the target will be dominated by either diffusive or pro- cessive motion, depending on the fraction of particles in each state. The transition to the regime where encounter times are sensitive to the initiation of active walks occurs when the spacing between particles hits a critical length scale where such walks between to dominate. This length can be obtained analogously to the expression for x∗ range (Eq. 8) by replacing the starting rate γ with an effective starting rate based on the average time to initiate a walk: In the case of rapid binding/unbinding γeff = 1/ htwi. (ku ≫ γ), this rate is approximated as γeff ≈ γ(cid:20) α2(Keq + 1) α2Keq + 1 (cid:21) (24) 12 Finally, we investigate an extension of the one- dimensional model to a cylindrical domain, where active transport can only occur in a narrow region along the axis and where particles can enter a halted tethered state that both enhances the effective rate of initiating an ac- tive run and limits their ability to explore the domain while in the passive state. The advantages of tethering to microtubule tracks have been a topic of much specula- tion in the literature on intracellular transport [7, 25, 26]. We delineate the parameter regime in which tethering is expected to aid the long-time dispersion of particles (Eq. 20) and identify a critical length scale Lcrit (Eq. 22) below which tethering hinders the ability of the particles to explore their domain. For several example intracellu- lar transport systems (Table I), this critical length is on the order of a few hundred nanometers, confirming the advantages of tethering for transport over many micron length scales. The results derived in this work highlight the com- plementary role of diffusion and processive transport in fulfilling cellular goals for delivering and distributing cy- toplasmic organelles. The derived expressions can be employed for analyzing data on measured transport pa- rameters to determine the length scales and transport objectives where active motor-driven motion is expected to dominate, where bidirectional transport with limited processivity is advantageous, and where tethering to cy- toskeletal tracks can aid overall organelle dispersion. VII. ACKNOWLEDGEMENTS We thank A. Agrawal and C. Niman for helpful com- ments on the manuscript and S. Reck-Peterson for fruit- ful discussions. The critical particle density is then given by ρ(cyl) crit = 2 ∗(cyl) range x = πγeff 8 D , (25) ∗(cyl) range is the length scale for transition between where x diffusive and processive motion in the model of a halting creeper within a cylindrical domain. This transition is shown with a solid black line in Fig. 6c. VI. SUMMARY Specifically, this model We have employed a simplified "halting creeper" model, consisting of stochastic interchange between pas- sive diffusion and active processive walks, to investigate the efficiency of transport within an extended cylindri- cal domain. is applicable to the transport of organelles within long narrow cellu- lar processes such as neural axons and fungal hyphae. We explore the space of relevant parameters, including the rates of transition between passive and active states and the relative speed of diffusion versus active trans- port, as characterized by the P´eclet number over differ- ent length scales. Our results highlight the importance of the relevant length scale in determining the contri- butions of the different transport modes and we iden- tify simple expressions for the time [t∗ πv2γ 2 ] the length [x∗ πvγ(λ+γ) ] at which processive motion dominates particle spreading. We emphasize the use of the average range as a metric for the ability of particles to explore their domain via multimodal transport, demon- strating passive diffusion can play an important role over longer length scales than expected based on the classic analysis of the mean squared displacement. range = 16Dλ2 range = 16Dλ2 We focus specifically on the contributions of active and passive transport to several key objectives relevant to the cell. First, we consider the establishment of a uniform distribution from a bolus of particles, demon- strating that efficient dispersion is achieved at interme- diate run-lengths that can be substantially smaller than the domain size. This result indicates the importance of bidirectional active transport with frequent reversals in the movement of particles that must be spread broadly throughout a large domain, as is the case with metabolic organelles such as peroxisomes and mitochondria. Sec- ond, we quantify the rate at which a single particle first encounters a stationary target, showing again an advan- tage to intermediate run lengths that minimize the time wasted pursuing a long processive walk in the wrong di- rection. Third, we consider the rate of encounter to a target by the first of a population of halting creeper parti- cles, identifying the parameter regime where active trans- port or diffusion dominate the motion, and showing that several examples of biological interest fall in the interme- diate regime where both modes of transport contribute substantially to target encounter. Appendix A: Propagator for a one-dimensional halting creeper We calculate the position distribution of a particle switching between diffusive transport with diffusivity D and processive motion with speed v. Switching between states is a Poisson process with rate γ for entering an active state and rate λ for leaving an active state (see Fig. 1). The overall spatial distribution can be obtained by a convolution of propagators for individual states, summed over all possible state transitions. Starting at an intial position x = 0, the spatial distri- bution of a diffusive particle at a time t is RD(x, t) = 1 √4πDt e− x2 4Dt (A1) We define the joint distribution that the particle first switches to an active state at time t while at position x by, HD(x, t) = γe−γtRD(x, t). (A2) The position distribution of particles starting at x = 0 in the active state, moving with a velocity v at time t is given as R±(x, t) = δ(x ∓ vt). (A3) The corresponding joint distribution for the time and location of switching from the active to the passive state is H±(x, t) = λe−λtδ(x ∓ vt) We define a "step" in the particle's trajectory as a switch from the passive to the active state and back to the passive state again. If the particle starts in the passive state at zero time, the position and time distribution at the end of one such step can be expressed as M (x, t) =Z ∞ −∞ dx′Z t −∞ dt′ HD(x′, t′) ×(cid:20) H+(x − x′, t − t′) + H−(x − x′, t − t′) 2 (A4) (cid:21) , where the first term denotes a particle reaching x′ at time t′ via diffusion and the second term denotes the particle covering a distance x − x′ in the remaining time t − t′ by walking, integrated over all values of x′ and t′. To get the spatial propagator of a halting creeper par- ticle that both starts and ends in a passive state, we sum over all possible paths between the active and passive states, convolved with the probability that the particle does not leave the passive state in the final time interval (given by HD(x, t)/γ). The resulting expression for the propagator can then be expressed as: GDD(x, t) = HD γ γ (cid:19) + M ∗x,t(cid:18) HD + M ∗x,t M ∗x,t(cid:18) HD γ (cid:19) + ..., (A5) 13 where ∗x,t denotes convolution with respect to x and t. The first term in the summation corresponds to a par- ticle that never left the passive state, the second term to a particle that performs a single active step before re- turning to the passive state, the third term includes two active steps, and so forth. the convolutions in x and t to products of functions in k and s respectively. Applying these transforms to the expression for GDD(x, t) yields a geometric series that sums to A Fourier transform in space (x → k, M → fM ) and a Laplace transform in time (t → s, M → cM ) transform beGDD(k, s) = (cid:18) beHD (k,s) γ (cid:19) 1 −cfM (k, s) (s + λ)2 + k2v2 = (s + γ + Dk2) ((s + λ)2 + k2v2) − γλ(s + λ) (A6) . The distributions for the other quantities appearing in Eqs. 3, 4 of the main text can be derived similarly. The transformed distributions are: beGDW = beGW W = beGW D = beG+D = beG+W = γ(s + λ) (s + γ + Dk2) ((s + λ)2 + k2v2) − γλ(s + λ) (s + γ + Dk2)(s + λ) (s + γ + Dk2) ((s + λ)2 + k2v2) − γλ(s + λ) λ(s + λ) (s + γ + Dk2) ((s + λ)2 + k2v2) − γλ(s + λ) λ(s + λ + ikv) (s + γ + Dk2) ((s + λ)2 + k2v2) − γλ(s + λ) (s + γ + Dk2)(s + λ + ikv) (s + γ + Dk2) ((s + λ)2 + k2v2) − γλ(s + λ) . (A7) A linear combination of these distributions, weighted by the equilibrium fraction of particles in each state is used to derive the overall propagator in Eq. 1: γ γ + λ(cid:18)beGW D +beGW W(cid:19) + beG = λ γ + λ(cid:18)beGDD +beGDW(cid:19) . (A8) The expressions obtained can be transformed back to real space and real time by a combination of analyti- cal and numerical methods. To calculate the Laplace- transformed expressions in Eq. 3, 4, and 5 we in- 1 vert the Fourier transform analytically as bG(x, s) = 2πR ∞ −∞ eikxbeG(k, s) dk. The Laplace transform of the range and first passage time distribution can then be in- verted numerically using Talbot's algorithm [50]. An alternate approach for efficiently calculating the propagator in real-space and real-time is to first invert the Laplace transform analytically using the residue the- orem, followed by a numerical integral over k to invert the Fourier transform. We use this approach to calculate the particle distribution at a high spatial resolution in order to find the entropy over time (Fig. 3). Appendix B: Analytical model for multimodal transport in a cylinder, with tethering In this section we develop the full analytical model for axial transport in a cylinder of radius R = 1 for particles capable of passive diffusion with diffusivity D, of initi- ating active processive walks with a rate γ while within a region of radius α of the central axis, and of entering a stationary tethered state with binding rate kb while in the same region. The rate constant for unbinding from a tethered state is ku and for transitioning between an active walk and passive diffusion is λ (see Fig. 6(a) for illustration of the model). For ease of the derivation, all length units in this section as well as Appendix C are non-dimensionalized by the cylinder radius R and all time units are nondimensionalized by R/v where v is the processive velocity of actively walking particles. We give our final results in fully dimensional units to facilitate comparison with other sections of the manuscript. Our model is developed in an analogous manner to the approach previously used for modeling facilitated diffu- sion by DNA-binding proteins that occurs via a combina- tion of 3D diffusion and 1D sliding along a filament [62]. We describe the particle motion by a system of individ- ual states with Markovian transitions between them. The rates of transition between the states are time-varying, depending specifically on the time interval since the par- ticle first entered the state. These states ( Fig. 7) consist of: a tethered state (h), a walking state (w), a state (n) wherein the particle started at radius α − ǫ and has re- mained within a radius α, a state (nu) where the particle started uniformly distributed within radius α and has re- mained within that inner region, a state (f ) where the particle started at radius α + ǫ and has remained outside the inner region at a radius greater than α and a state fu where the particle started uniformly distributed in the outer region and has remained in the outer region. When computing statistics for the overall motion of the particle, we take the limit ǫ → 0. The axial propagation of a parti- cle in states n, nu, f, fu is given by the propagator func- tion for diffusive motion RD(x, t) (Eq. A1). The axial propagation in state w is given by 1 2 [R+(x, t) + R−(x, t)] (Eq. A3). We construct a transition matrix of propagators H, where Ha,b(x, t) is the joint probability density for the time and position of a particle initially at the origin in state a making its first transition out of that state, into state b. A Fourier transform in space x → k and a Laplace transform in time t → s is carried out to yield the transformed propagator of this propagator matrix are derived from the Laplace- eH(k, s). The components transformed solutions for first passage times to an inner or outer absorbing boundary for a particle diffusing in a cylindrical domain [63]. These components are given by, 14 , Dσ2 kb Dσ2 γ , s + γ + ku , , where σD = I0((α − ǫ)σb) I0(ασb) γ 2 ασb γ Dσ2 kb Dσ2 I0(ασD)K1(σD) + K0(ασD)I1(σD) s + γ + ku λ(s + λ) (s + λ)2 + k2 , beH n,w = beH n,h = ku b (cid:18)1 − beH n,f(cid:19) b (cid:18)1 − beH n,f(cid:19) 2α I1(σD)K1(ασD) − I1(ασD)K1(σD) (1 − α2)σD I0(ασD)K1(σD) + K0(ασD)I1(σD) I0((α + ǫ)σD)K1(σD) + K0((α + ǫ)σD)I1(σD) I1(ασb) I0(ασb) beH n,f = b (cid:18)1 − beH nu,f(cid:19) , b (cid:18)1 − beH nu,f(cid:19) , beH h,nu = beH fu,n = beH f,n = beH nu,f = beH nu,w = beH nu,h = beH h,w = beH w,nu = p(s + γ + kb + Dk2)/D and Iν , Kν are the modi- kind, respectively. All other components of beH not listed of propagators beF. Each component beF a corresponds to D(cid:18)1 − beH f,n(cid:19) b (cid:18)1 − beH n,f(cid:19) the Fourier-Laplace transformed spatial distribution of particles that first reached state a at time 0 and have moved a displacement x at time t, without having left that state. These components are given by in Eq. B1 correspond to transitions not allowed in the model and are equal to 0. To calculate the overall distribution of particles, we additionally define a vector p(s + Dk2)/D, fied Bessel functions of order ν of the first and second beF f = beF n = D(cid:18)1 − beH fu,n(cid:19) , b (cid:18)1 − beH nu,f(cid:19) , beF h = s + λ beF fu = beF nu = beF w = (s + λ)2 + k2 , 1 Dσ2 1 Dσ2 1 1 Dσ2 1 Dσ2 (B1) σb = s + γ + ku . (B2) The overall propagator for a particle moving through this system of states can be found by a convolution over all possible transition paths, analogous to the discrete path sampling technique used for calculating kinetics on potential energy surfaces [64]. Specifically, the spatial density of a particle that started at the origin in state i at time 0 and is in state j at time t is given by Gi,j(x, t) = =δi,jFi + ∞Xn=0 Xk1,k2,...kn Hi,k1 ∗ . . . ∗ Hkn−1,kn ∗ Hkn,j ∗ Fj (B3) The average time to start walking for a particle initially uniformly distributed within the inner radius α can be evaluated as 15 (B7) htwi = "(cid:18)I − cfH∗(cid:19)−1#nu,. ·cfF∗(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)k=0,s=0 where the subscript (nu, .) indicates the corresponding row of the inverse matrix. The resulting expression is given in Eq. 19. We similarly calculate the mean squared time to initi- ate a walk, using ∂ ∂s "(cid:18)I − cfH∗(cid:19)−1#nu,. w(cid:11) = −2 (cid:10)t2 w(cid:11) − htwi2. While the full closed-form expression σ2 =(cid:10)t2 is too cumbersome to include here, in the limit of rapid unbinding from the tethered state (ku ≫ γ, ku ≫ D/a2), the variance in the walking time is ·cfF∗(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)k=0,s=0 The variance in the time to start walking is given by . (B8) σ2 = ku→∞ lim 4D(1 + α2Keq)2 − α2γ(3 − 4α2 + α4 + 4 log α)(1 + Keq) (B9) 4α4Dγ2(1 + Keq)2 Fig. 8 shows the Fano factor, a measure of the variability in a stochastic process defined as the standard deviation in the time to start walking, relative to the average time. Large variability in how long it takes a passively diffus- ing particle to start walking is seen in the case of slow diffusion and strong binding. Appendix C: Simulation details We simulate moving particles within a cylindrical do- main of unit radius and unbounded length. The axial po- sition of each particle is tracked to determine the range and the mean squared displacement. We also track the radial position to determine the probability of state tran- sitions for the particles. Each particle is assigned to a walking, diffusive or teth- ered state at initialization. The fraction of particles in each state is determined by the equilibrium distribution, fwalk = fdiff = fbound = γα2 kb+γ+ku (cid:17) γα2 + λ(cid:16) α2kb+γ+ku λ(cid:16) γ+ku kb+γ+ku(cid:17) kb+γ+ku (cid:17) γα2 + λ(cid:16) α2kb+γ+ku λ(cid:16) α2kb kb+γ+ku(cid:17) kb+γ+ku (cid:17) γα2 + λ(cid:16) α2kb+γ+ku (C1) FIG. 7. Schematic state diagram illustrating the particle states used to develop the analytical model for multi-modal transport in a cylinder. Allowed transitions are labeled with arrows and the rates for the constant-rate transition processes (to and from tethered or actively walking state) are indicated. The transitions between diffusive states occur with a time- varying rate that can be derived by evaluating the matrix components in Eq. B1 at k = 0. where n is the number of intermediate states over which the particle transitions and kl is the identity of the l-th intermediate state. Replacing the convolutions with multiplication of the Fourier-Laplace transformed propagators we find the overall spatial distribution for a particle that started in a linear combination of initial states described by the vector P. beG(k, s; P) = lim ǫ→0 where I is the identity matrix. P ·(cid:18)I − beH(cid:19)−1 (B4) ·beF The Laplace-transformed mean squared displacement can be found directly from the propagator by taking derivatives with respect to k. Its long time limit is found by expanding to lowest order in s and taking the coeffi- cient of the 1/s2 term: MSD =(cid:20)− lim s→0 lim t→0 ∂k2beG(k, s)(cid:12)(cid:12)(cid:12)(cid:12)k=0(cid:19)(cid:21) t = 2Defft, s2(cid:18) ∂2 (B5) where the effective long time diffusivity Deff is given in Eq. 18. The average time for a particle with initial distribution P among the different states to first initiate a walk can be found as the time integral of the probability that no walk has yet occurred: htw(P)i =Z ∞ 0 dtZ ∞ 0 dx G∗(x, t; P) (B6) where G∗ is obtained from Eq.B4 with alternate transi- tion matrices cfH∗,cfF∗ defined by removing the rows and columns of beH,beF corresponding to the walking state (w). 16 sage time distribution for diffusive particles between two cylindrical boundaries. The cumulative encounter prob- ability to an absorbing inner boundary of radius α with a reflective outer boundary of unit radius is given by, Φ(t) = 1 − π2 2 ∞Xn=1(cid:20) J 2 0 (βnα)βnα 0 (βnα) − J 2 J 2 1 (βn)(cid:21) × [J0(βnr)Y1(βn) − Y0(βnr)J1(βn)] × [Y1(βnα)J1(βn) − J1(βnα)Y1(βn)] e−β2 nDt (C2) FIG. 8. Fano factor σ/ htwi quantifying the variability in the time required for a particle to first begin a processive walk. The particle is assumed to start uniformly distributed within the inner radius α. Results shown are for parameters α = 0.1, ku = 100, γ = 10−2 Unbound particles in the diffusive state start uniformly distributed radially throughout the cross section. We divide the cylindrical domain in two concentric sections (Fig. 6a). The inner domain of radius α de- notes the region within which particles can transition from the diffusive state to the walking or tethered state. Particles execute explicit Brownian dynamics with a time-step ∆t when their radial position is smaller than 3α/2. This includes the inner domain along with a buffer region of radius α/2. The time step is chosen to be smaller than all relevant time-scales in the model: ∆t ≪ min(1/kb, 1/γ, α2/2D). Note that this choice of time-step prevents multiple events occuring within a sin- gle step. Particles outside the capture domain can spend a long time diffusing before reaching the region of interest. To accelerate the simulation, we make use of the first pas- where Jν and Yν are the bessel functions of the first and second kind respectively, with order ν [63]. The βn are eigenvalues of the equation J1(βn)Y0(βnα) − J0(βnα)Y1(βn) = 0. The time required to reach the inner domain starting from an initial radial position r is drawn from the above distribution and the particles are propa- gated along the axis according to the diffusive propagator RD (Eq. A1) over this time interval. The simulation is run using a hybrid Brownian Dy- namics – kinetic Monte Carlo algorithm where the prob- ability of a state transition depends on the position of the particle. Particles in the diffusive state within the inner domain (r < α) can transition to the tethered or walking states at a combined rate kb + γ. A transition is attempted at every diffusion time-step based on the relative probabilities for tethering and walking. Transi- tions leading away from the tethered state occur with a rate ku to the diffusive state, and with a rate γ to the walking state. Particles in the walking state transition to the diffusive state a rate λ. Each time particles re-enter the diffusive state, they are uniformly distributed in the radial dimension within the inner region (of radius α), ensuring symmetry between the binding and unbinding position. A schematic of these transitions is shown in Fig. 6(a). For each transition out of a tethered or walk- ing state, the waiting time is drawn from an exponential distribution with the mean equal to the corresponding transition rate. The particles are propagated in space according to the distribution for the given state over the duration of the waiting time. The simulation continues until all particles have covered a predetermined time in- terval. [1] M.-m. Fu and E. L. Holzbaur, Trends Cell Biol 24, 564 [8] C. P. Brangwynne, G. H. Koenderink, F. C. MacKintosh, (2014). and D. A. Weitz, Trends Cell Biol 19, 423 (2009). [2] J. Salogiannis, M. J. Egan, and S. L. Reck-Peterson, J [9] M. Mussel, K. Zeevy, H. Diamant, and U. Nevo, Biophys Cell Biol , 201512020 (2016). J 106, 2710 (2014). [3] C. Kural et al., P Natl Acad Sci 104, 5378 (2007). [4] J. L. Ross, M. Y. Ali, and D. M. Warshaw, Curr Opin Cell Biol 20, 41 (2008). [5] H. V. Mudrakola, K. Zhang, and B. Cui, Structure 17, [10] K. Jaqaman et al., Cell 146, 593 (2011). [11] V. Ananthanarayanan et al., Cell 153, 1526 (2013). [12] A. Godec and R. Metzler, Phys Rev E 92, 010701 (2015). [13] O. B´enichou, C. Loverdo, M. Moreau, and R. Voituriez, Rev Mod Phys 83, 81 (2011). 1433 (2009). [6] S. B´alint, I. V. Vilanova, ´A. S. ´Alvarez, and [14] D. Campos, E. Abad, V. M´endez, S. Yuste, and K. Lin- M. Lakadamyali, P Natl Acad Sci 110, 3375 (2013). [7] W. O. Hancock, Nat Rev Mol Cell Bio 15, 615 (2014). denberg, Phys Rev E 91, 052115 (2015). [15] Y. Tanaka et al., Cell 93, 1147 (1998). 17 [16] C. Kural et al., Science 308, 1469 (2005). [17] M. Schuster, R. Lipowsky, M.-A. Assmann, P. Lenz, and G. Steinberg, P Natl Acad Sci 108, 3618 (2011). [18] P. Targett-Adams et al., J Biol Chem 278, 15998 (2003). [19] P. C. Bressloff and J. M. Newby, Rev Mod Phys 85, 135 (2013). [41] A. D. Pilling, D. Horiuchi, C. M. Lively, and W. M. Sax- ton, Mol Biol Cell 17, 2057 (2006). [42] W. W. Ahmed and T. A. Saif, Sci. Rep. 4 (2014). [43] E. V. Romanova, S. P. Oxley, S. S. Rubakhin, P. W. Bohn, and J. V. Sweedler, Biomaterials 27, 1665 (2006). [44] S. E. Encalada, L. Szpankowski, C.-h. Xia, and L. S. [20] J. Gou, L. Edelstein-Keshet, and J. Allard, Mol Biol Cell Goldstein, Cell 144, 551 (2011). 25, 2408 (2014). [45] G. Steinberg and J. Perez-Martin, Trends in cell biology [21] D. Ando, N. Korabel, K. C. Huang, and A. Gopinathan, 18, 61 (2008). Biophys J 109, 1574 (2015). [22] L. S. Goldstein and Z. Yang, Annu Rev Neurosci 23, 39 (2000). [23] M. J. Egan, M. A. McClintock, and S. L. Reck-Peterson, Curr Opin Microbiol 15, 637 (2012). [24] X.-A. Liu, V. Rizzo, and S. Puthanveettil, J Transl Neu- rosci 3, 355 (2012). [46] R. T. Ambron, X.-P. Zhang, J. D. Gunstream, M. Povelones, and E. T. Walters, Journal of Neuroscience 16, 7469 (1996). [47] L. G. Leal, Advanced transport phenomena: fluid me- chanics and convective transport processes, Cambridge University Press, 2007. [48] K. Chen, B. Wang, J. Guan, and S. Granick, Acs Nano [25] J. R. Cooper and L. Wordeman, Curr Opin Cell Biol 21, 7, 8634 (2013). 68 (2009). [49] W. Feller, On the integral equation of renewal theory, in [26] T. L. Culver-Hanlon, S. A. Lex, A. D. Stephens, N. J. Selected Papers I, pages 567–591, Springer, 2015. Quintyne, and S. J. King, Nat Cell Biol 8, 264 (2006). [27] J.-S. Kang et al., Cell 132, 137 (2008). [28] G. Pekkurnaz, J. C. Trinidad, X. Wang, D. Kong, and T. L. Schwarz, Cell 158, 54 (2014). [29] R. L. Frederick and J. M. Shaw, Traffic 8, 1668 (2007). [30] M. J. Muller, S. Klumpp, and R. Lipowsky, P Natl Acad [50] A. Talbot, Ima J Appl Math 23, 97 (1979). [51] P. Danckwerts, Appl. Sci. Res., Section A 3, 279 (1952). [52] Z. Stone and H. Stone, Phys Fluids 17, 063103 (2005). [53] J.-L. Thiffeault, Nonlinearity 25, R1 (2012). [54] P. Ashwin, M. Nicol, and N. Kirkby, Physica A 310, 347 (2002). Sci 105, 4609 (2008). [55] M. Camesasca, M. Kaufman, and I. Manas-Zloczower, [31] J. L. Ross, H. Shuman, E. L. Holzbaur, and Y. E. Gold- Macromol Theor Simul 15, 595 (2006). man, Biophys J 94, 3115 (2008). [56] S. Redner, A guide to first-passage processes, Cambridge [32] C. Lin et al., Nat. Commun. 7 (2016). [33] D. T. Chang, A. S. Honick, and I. J. Reynolds, J Neurosci 26, 7035 (2006). [34] A. M. Valm et al., Nature 546, 162 (2017). [35] N. Hirokawa and R. Takemura, Nat Rev Neurosci 6, 201 (2005). University Press, 2001. [57] K. Ogawa and S. Ito, J Chem Eng Jpn 8, 148 (1975). [58] C. Loverdo, O. B´enichou, M. Moreau, and R. Voituriez, Nat Phys 4, 134 (2008). [59] G. Jedd and N.-H. Chua, Nat Cell Biol 2, 226 (2000). [60] S. R. Chada and P. J. Hollenbeck, Curr Biol 14, 1272 [36] M. A. De Matteis and A. Luini, Nat Rev Mol Cell Bio 9, (2004). 273 (2008). [37] I. Singh, Mol Cell Biochem 167, 1 (1997). [38] N. A. Bright, M. J. Gratian, and J. P. Luzio, Curr Biol 15, 360 (2005). [61] S. J. Peter and M. R. Mofrad, Biophys J 102, 749 (2012). [62] E. F. Koslover, M. A. D. de la Rosa, and A. J. Spakowitz, Biophys J 101, 856 (2011). [63] M. N. Ozisik, Heat transfer: a basic approach, McGraw- [39] D. Bandyopadhyay, A. Cyphersmith, J. A. Zapata, Y. J. Hill New York, 1985. Kim, and C. K. Payne, PloS one 9, e86847 (2014). [64] D. J. Wales, Mol Phys 100, 3285 (2002). [40] A. G. Hendricks et al., Curr Biol 20, 697 (2010).
1808.09383	1	1808	2018-08-28T16:13:56	Stabilization of overlapping biofilaments by passive crosslinkers	[ "physics.bio-ph" ]	The formation, maintenance and reorganization of the cytoskeletal filament network is essential for a number of cellular processes. While the crucial role played by active forces generated by motor proteins has been studied extensively, only recently the importance of passive forces exerted by non-enzymatic crosslinkers has been realized. The interplay between active and passive proteins manifests itself, e.g., during cell division, where the spindle structure formed by overlapping microtubules is subject to both active sliding forces generated by crosslinking motor proteins and passive forces exerted by passive crosslinkers, such as Ase1 and PRC1. We propose a minimal model to describe the stability behaviour of a pair of anti-parallel overlapping microtubules resulting from the competition between active motors and passive crosslinkers. We obtain the stability diagram which characterizes the formation of stable overlap of the MT pair, identify the controlling biological parameters which determine their stability, and study the impact of mutual interactions between motors and passive crosslinkers on the stability of these overlapping filaments.	physics.bio-ph	physics	Stabilization of overlapping biofilaments by passive crosslinkers Sougata Guha1,2, Subhadip Ghosh2 Ignacio Pagonabarraga3 and Sudipto Muhuri1 1Department of Physics, Savitribai Phule Pune University, Ganeshkhind, Pune 411007, India 2Indian Institute of Science Education and Research Mohali, Knowledge City, Sector 81, SAS Nagar - 140306, Punjab, India. 3CECAM, Centre Europ´een de Calcul Atomique et Mol´eculaire, ´Ecole Polytechnique F´ed´erale de Lasuanne, Batochime, Avenue Forel 2, 1015 Lausanne, Switzerland The formation, maintenance and reorganization of the cytoskeletal filament network is essential for a number of cellular processes. While the crucial role played by active forces generated by motor proteins has been studied extensively, only recently the importance of passive forces exerted by non-enzymatic crosslinkers has been realized. The interplay between active and passive proteins manifests itself, e.g., during cell division, where the spindle structure formed by overlapping micro- tubules is subject to both active sliding forces generated by crosslinking motor proteins and passive forces exerted by passive crosslinkers, such as Ase1 and PRC1. We propose a minimal model to describe the stability behaviour of a pair of anti-parallel overlapping microtubules resulting from the competition between active motors and passive crosslinkers. We obtain the stability diagram which characterizes the formation of stable overlap of the MT pair, identify the controlling biological parameters which determine their stability, and study the impact of mutual interactions between motors and passive crosslinkers on the stability of these overlapping filaments. The structural integrity of an eukaryotic cell is provided by the cytoskeleton filament network comprised of micro- tubules (MTs) and actin filaments [1]. The ability of the cell to reorganize itself to develop spatio-temporal structures out of these cytoskeletal filaments is vital for the process of cell division, morphogenesis and movement [1, 2]. This ability hinges crucially on the activity of the motor proteins which attach and move on these cellular filaments and generate active stresses that cause reorganization of the cytoskeletal filaments [2 -- 10]. A paradigmatic example in this context is the formation of a spindle during mitotic cell division. The spindle structure comprises of overlapping anti-parallel MTs originating from centrosome poles within the cell. These overlapping arrays of anti-parallel MTs are subject to active sliding forces exerted by bipolar motor proteins such as Eg5 and kinesin-8 [6] which attach simultaneously to the two overlapping filaments and slide them with respect to each other. These molecular motors use the stored chemical energy from ATP hydrolysis to generate the sliding forces, driving the cytoskeleton out of equilibrium. While in-vitro and in-vivo studies of mitotic spindle have established the centrality of the role played by the molecular motors in the organization of the cytoskeleton network [7 -- 12], relatively less attention has been paid to the role of passive filament crosslinkers such as Ase1, PRC1 and MAP65 [13 -- 17]. While it has been known earlier that these passive crosslinkers provide structural integrity of the filament network and increases the effective friction between the sliding filaments, recent experiments have also shown that these entities are capable of generating entropic forces by harnessing the thermal energy from the cellular environment and are also involved in determining the stability and reorganization of the cytoskeletal filament network along with the molecular motors [13, 14, 18, 19]. Two mechanisms that lead to entropic forces between filaments have been identified in out of equilibrium cytoskeletal filament networks. On one hand, in a suspension of cytoskeletal filaments and small non-adsorbing polymers, such as Polyethylene Glycol (PEG), the volume occupied by the filaments is unavailable for the depletant PEG polymers. Decreasing the excluded volume due to the filaments which is achieved by filament bundling and maximizing the overlap of the filaments, is entropically favorable. The gain in free energy for two overlapping filaments is proportional to their overlap length, l, and the associated entropic force acting along the long axis of the two filaments is constant, independent of the l, set by the osmotic pressure of the depletant polymers and associated geometric factors [19, 20]. On the other hand, overlapping microtubules in a mitotic spindle confine freely diffusing, passive crosslinkers, such as Ase1 and MAP65 [14, 19]. As a result, the confined proteins exert an effective pressure which tends to increase l, with a strength that increases linearly in the number of crosslinkers and inversely proportional to l [19]. These studies have highlighted the relevance of entropic forces in biological networks [14], which are relevant also to understand the self-assembly of biofilament bundles [20], and shown to be potentially relevant in spindle geometries [21, 22]. 2 + kb uk 3D kb o ku − C P l − B + U FIG. 1: (color online) Schematic representation of the system comprising of two overlapping anti-parallel microtubules with passive crosslinkers (P), motors which is bound to only one filament (B), crosslinked motors (C) which are bound to both the filaments, and the unbound motors (U). The motors transform from bound state to crosslinked state with rate kb and from crosslinked state to bound state with rate ku. ko is the binding rate of free motors to the filament. u is the unbinding rate of the motor from the MT filament and k3D b In this letter we identify an alternative mechanism for the generation of sliding forces by passive crosslinking proteins. In contrast to previous mechanisms [13, 14, 19], we take into account the exchange of passive protein crosslinkers with the environment in which the filaments are suspended. As a result, if the evolution of the overlap length, l, is much slower than the exchange rate of the passive crosslinkers with the environment, the passive crosslinkers decrease the effective free energy of the complex by an amount that increases linearly with l. As we will show, passive crosslinkers induce an entropic force that increases with their density, φ, independent of l, unlike the entropic forces exerted by confined passive crosslinkers [9, 13, 14]. We first describe the model and obtain the phase diagram associated with the linearly stable behaviour of a pair of overlapping anti-parallel MTs. We then identify the relevant biological parameters which control the system stability. We also consider the stability of overlapping MTs for the situation where the passive crosslinking proteins in the overlap interact with the active motor crosslinkers. I. MODEL We consider a pair of anti-parallel microtubules with an overlap length l as depicted in Fig.1. In the overlap region, kinesin motors can crosslink anti-parallel MTs and slide them away from each other, generating an active outward force which tends to decrease the overlap length l. The motor proteins (un)binding kinetics in the overlap region is expressed in terms of rate equations for two populations of motors, e.g; nc which is the average number of motors crosslinked to the MT filaments, and nb which is the average number of motors when they are attached to only one of the two overlapping filament. The passive crosslinkers have a linear density, φ, in the overlapping region of the two filaments, set by the Langmuir kinetic process of attachment and detachment onto the filaments. First we describe the case corresponding to the situation where the passive crosslinkers do not interact with the motor proteins. Subsequently we look at the situation where these passive crosslinkers also interact with motor proteins. A. Non-interacting motor -passive crosslinkers - MTs: The binding free energy of the passive crosslinkers to the microtubule, −ǫo , induces a net inward force on the pair of anti-parallel MT, Fp ≡ − ∂E ∂l , associated with the total free energy reduction, E = −ǫoφl, due to the passive crosslinkers. We consider the relative motion of a pair of anti-parallel filaments. While in general under in-vivo 3 conditions the (−) end of the filament would be coupled to the cell cortex, we focus our attention only on the overlap dynamics resulting from the interplay of the motor proteins and the forces exerted by the passive crosslinkers in the overlap region. In the absence of the crosslinkers, the overlapping filaments will slide with respect to each other with a velocity 2vo, where vo is the motor velocity of the crosslinking motors in the absence of any external load. The load force exerted by the passive crosslinkers on the molecular motors will reduce their average sliding velocity. Assuming a linear force-velocity relation for the motor velocity with equal load sharing between the crosslinked motors, the overlap length l evolves as, where, fs stands for the stall force for a single motor and Γ is the friction coefficient associated to the motion of the filaments against their embedding medium [22]. dl dt = −2vo(cid:18)1 − Fp ncfs(cid:19) + 1 Γ Fp (1) Due to the passive force experienced by the crosslinked motors, the unbinding rate of the motors from the crosslinked state to single filament bound state grows exponentially with the load force felt by the individual motors [23]. The corresponding dynamics of the average number crosslinked motors nc can then be expressed as, dnc dt = kbnb − ko unc exp(cid:20) bFp nckBT(cid:21) (2) where, kb and ko u are the rates with which the uncrosslinked motors are converted to crosslinked state and vice-versa in the absence of external load force, and b is the length scale characterizing the activation process leading to unbinding from the crosslinked state of the motor. The number of bound motors in overlapped region which are not crosslinked, nb, evolves due to interconversion between bound and crosslinked motor states, (un)binding process of free unbound motor on to the filament, and the incoming flux of bound motors 2J from the two boundaries of the overlap region. When the overlap length is much smaller than the MT length, J = k3d + Fp b ρ3d ko u dnb dt = ko ncfs Γ i [7, 23]. Thus the dynamic equation for nb can be expressed as, h voFp nckBT(cid:21) − (ko unc exp(cid:20) bFp Γ (cid:21) (cid:20) voFp u + kb)nb + k3D 2k3d b ρ3d ko u b ρ3dl ncfs Fp + (3) + where, ρ3d is the density of suspended motors in the bath, and k3d b in the overlap region. is the binding rate of the free motors to the filament The behaviour of a pair of anti-parallel filaments that is subject to sliding forces due to motors and passive crosslinkers is determined by nc, nb and l , and correspondingly, Eqs.(1)-(3) describe the dynamics of the composite composed of filaments, motors and passive crosslinkers. We rescale time in units of 1/ko u, overlap length in terms of the average run length of single motor, lp = vo/ko u and forces in terms of kBT /b. We also introduce a length scale associated with the characteristic forces experienced by the passive crosslinkers, le = ǫob/kBT . We define dimensionless variables τ = tko kB T and φo = leφ. Then Eqs. (1)-(3) can be re-expressed in terms of these dimensionless variables as, u, fs = bfs/kBT , l = fsl/lp, Γ = 2voΓb dl dτ dnc dτ dnb dτ = Γ 2φo nc + 2 fs(cid:18) φo − 1(cid:19) nc(cid:19) = γnb − nc exp(cid:18) φo = nc exp(cid:18) φo nc(cid:19) + ∆n l + − (1 + γ)nb 2φo nc + 4 fsφo Γ ! (4) (5) (6) We identify the relevant control parameters associated with motor-filament interaction. The linear density of the passive proteins, φo, is one of the biologically relevant control parameters. We also identify ∆n = ρ3dk3d b vo , which fsko uko u measures the strength of motor flux reaching the overlap region, and γ = kb ko u motor attachment/detachment at vanishing load force, as relevant biological control parameters. , which quantifies the asymmetry in 4 B. Interacting motor -passive crosslinkers - MTs: Until now, in formulating the evolution equations governing the dynamics of overlapping filaments, we have not taken into account the mutual interaction between the passive crosslinkers with the motors. However in general the passive crosslinker proteins will interact with motor proteins [15 -- 17, 21]. For instance, in experiments with reconstituted system comprising of MT and crosslinking protein PRC1, it has been observed that these crosslinking proteins had an effective attractive interaction with kinesin motors in the overlap region [16, 17]. We quantify the interaction strength between passive proteins and motor crosslinkers with a coupling parameter, a, in terms of which we propose a simple interaction free energy E = −φl(ǫo + anc + anb) and a corresponding force, F ∗ p = φ(ǫo + anc + anb). Although it has been argued on theoretical grounds that excluded volume interactions between diffusing passive proteins and motors can lead to phase segregation in the overlap region [21], for the sake of simplicity, we consider homogeneous admixture of motors and passive crosslinkers in the overlap region and focus on the impact of the simplest coupling between the passive crosslinkers and the molecular motors. We quantify this coupling through the dimensionless coupling constant αo ≡ ab/lekBT . Accordingly, the relevant dimensionless force becomes F ∗ p = φo + αoφonc + αoφonb. While positive values of αo correspond to a repulsive interaction between the passive crosslinkers and the motors, negative values of αo corresponds to an attractive interaction between the crosslinking motors and the passive proteins. The corresponding scaled equations of motion describing the dynamics of the system read as dl dτ dnc dτ dnb dτ + nc = 2φo nc Γ nb nc − 1(cid:19) + 2 fs(cid:18) φo + 2αoφo 1 + = γnb − nc exp(cid:18) φo = −(1 + γ)nb + nc exp(cid:18) φo + ∆n l + Γ ! + 2∆nαoφo 1 + 2φo nc nb nc + 2 fs Γ 4 fsφo + Γ nc fs(nc + nb) ! + αoφo + αoφo nb nc(cid:19) + αoφo + αoφo (nc + nb)! . nb nc(cid:19) (7) (8) (9) Equations (7)-(9) describe the dynamical behavior of the system when the motors interact with passive proteins. In order to analyze the stability behavior for this system ( with and without mutual interaction between crosslinking motors and passive proteins), we first obtain steady state solutions for nc, nb and l and perform a linear stability analysis about the fixed point to obtain the stability diagram for obtaining a stable overlap of the anti-parallel MTs. For the case of motors interacting with the passive crosslinkers, the fixed points can be obtained numerically and the corresponding stability curve separating regions of stable overlap with the unstable region is also obtained numerically. If the passive proteins interact preferentially with the crosslinked motors alone then the corresponding sliding force acting on the overlapping filaments simplifies to F ∗ p = ǫoφ + aφnc. Eqs.(7)-(9) simplify accordingly, and analytical 5 ✶ ✶ ✶ ✵✿✼✺ ✵✿✼✺ ✁♥ ✁♥ ✵✿✺ ✵✿✺ ✵✿✺ ✵✿✷✺ ✵✿✷✺ ✵ ✵ ✵ ✵ ✵ ✵ ✝✞✟ ✉■ s ✉■■ ✝ ✝ ✟ ⑦❧❢ ❁ ⑦❧❢ ❁ ⑦❧❢ ❃ ⑦❧❢ ❃ st❛❜❧❡✭s✮ st❛❜❧❡✭s✮ ✵✿✷✺ ✵✿✷✺ ✵✿✺ ✵✿✺ ✵✿✺ ✵✿✼✺ ✵✿✼✺ ✶ ✶ ✶ ✣♦ ✣♦ FIG. 2: Phase diagram of the system in φo − ∆n plane, when there is no interaction between passive cross-linkers and the molecular motors (αo = 0). Solid (red) line is the linear stability boundary. Dashed (blue) line corresponds to lf = 0.The dotted (green) line is corresponds to nf c = 0. No physical fixed point of nc exists on the right of green dashed curve. s denotes the region with stable overlap with nf c > 0 and lf > 0. uI and uII are the subregions within the linearly unstable region where the fluctuations about the fixed points grows with and without oscillations respectively. Here fs = 1.9, γ = 1, Γ = 1. expression for the the fixed points and the corresponding linear stability boundaries can be obtained. II. RESULTS We analyze the stability of a complex composed by overlapping anti-parallel MT that traps kinesin motors and passive crosslinkers. We analyze the impact that the different types of interactions induced by the passive cross-linker have on the stability of the MT-motor complex as a function of the relevant biological control parameters. Eqs.(7)-(9) highlights the relevance of ∆n and concentration of passive crosslinkers φo, and we focus on their impact on the mechanical properties of these active complexes. A. Non-interacting motor -passive protein-MT system: The passive crosslinker entropy is the only contribution to the stability of the complex in this regime. The steady state state solutions, which correspond to the fixed points of Eqs.(4)-(6), read nf c = φo/g nf b = φo exp(g)/γg lf = φo exp(g)/γg∆n − 2g − 4 fsφo/Γ (10) where g = fs(cid:16)1 − φo that l > 0 implies ∆n < ∆c Γ (cid:17). Since nf c > 0, all steady state morphologies must satisfy φo < Γ. Analogously, the restriction n, which implicitly depends on the force Fp. A linear stability analysis around these steady ✂ ✄ ❝ ☎ ✆ ✂ ✄ ❝ ☎ ✆ 6 ✵✿✶ ✵✿✶ ✵✿✶ ✁♥ ✁♥ ✁♥ ✵✿✵✺ ✵✿✵✺ ✵✿✵✺ ✵ ✵ ✵ ✵ ✵ ✵ ✵✿✶ ✵✿✶ ✵✿✶ ✁♥ ✁♥ ✁♥ ✵✿✵✺ ✵✿✵✺ ✵✿✵✺ ✭❛✮ ✉ ✉ ✉ ✭❝✮ s s s ✵✿✺ ✵✿✺ ✵✿✺ ✣♦ ✣♦ ✣♦ ✉ ✉ ✉ ✵ ✵ ✵ ✵ ✵ ✵ s s s ✵✿✺ ✵✿✺ ✵✿✺ ✣♦ ✣♦ ✣♦ ✵✿✶ ✵✿✶ ✵✿✶ ✁♥ ✁♥ ✁♥ ✵✿✵✺ ✵✿✵✺ ✵✿✵✺ ✭❜✮ ✵ ✵ ✵ ✵ ✵ ✵ s s s ✉ ✉ ✉ ✵✿✺ ✵✿✺ ✵✿✺ ✣♦ ✣♦ ✣♦ ✵✿✶ ✵✿✶ ✵✿✶ ✁♥ ✁♥ ✁♥ ✵✿✵✺ ✵✿✵✺ ✵✿✵✺ ✄☎✄✄✝ ❞✂ ✄ ✄ ✵ ✵ ✵ ✵ ✵ ✵ ✵✿✺ ✵✿✺ ✵✿✺ ✣♦ ✣♦ ✣♦ ✶ ✶ ✶ ✶ ✶ ✶ ✶ ✶ ✶ ✶ ✶ ✶ ✄☎✆ s s s FIG. 3: Effect of variation of γ on stability: (a) γ = 3, (b) γ = 5, (c) γ = 10 and (d) γ = 20. u denotes the unstable region, while s indicates stable region. Here αo = 0 (No interaction between passive cross-linkers and motors). Other parameter values are, fs = 1.9, Γ = 1. ✵✿✶ ✵✿✶ ✁♥ ✁♥ ✵✿✵✺ ✵✿✵✺ ✭❛✮ ✉ ✉ s s ✵ ✵ ✵ ✵ ✵✿✶ ✵✿✶ ✵✿✶ ✁♥ ✁♥ ✁♥ ✵✿✵✺ ✵✿✵✺ ✵✿✵✺ ✵ ✵ ✵ ✵ ✵ ✵ ✵✿✺ ✵✿✺ ✣♦ ✣♦ ✭❝✮ ✉ ✉ ✉ s s s ✵✿✺ ✵✿✺ ✵✿✺ ✣♦ ✣♦ ✣♦ ✵✿✶ ✵✿✶ ✁♥ ✁♥ ✵✿✵✺ ✵✿✵✺ ✭❜✮ ✉ ✉ s s ✵ ✵ ✵ ✵ ✵✿✺ ✵✿✺ ✣♦ ✣♦ ✵✿✶ ✵✿✶ ✵✿✶ ✁♥ ✁♥ ✁♥ ✭❞✮ ✵✿✵✺ ✵✿✵✺ ✵✿✵✺ ✉ ✉ ✉ s s s ✵ ✵ ✵ ✵ ✵ ✵ ✵✿✺ ✵✿✺ ✵✿✺ ✣♦ ✣♦ ✣♦ ✶ ✶ ✶ ✶ ✶ ✶ ✶ ✶ ✶ ✶ FIG. 4: Effect of interaction strength αo on stability: (a) αo = −0.2 and (b) αo = −0.02 corresponds to the attractive interaction between crosslinked motors and passive crosslinkers, while (c) αo = 0.02 and (d) αo = 0.2 corresponds to repulsive interaction between them. u denotes the unstable region, while s indicates stable region. Here fs = 1.9, Γ = 1 and γ = 1. solutions leads to the eigenvalue equation which is of the form, λ3 + a1λ2 + a2λ + a3 = 0 When one of the roots is real and negative, the other two are complex conjugate, the stability boundary corresponds to the curve a3 = a2a1 as the real part of the complex conjugate pair changes sign [28]. Accordingly, the phase 7 0.1 0.1 0.1 (a) (a) (a) 0.1 0.1 0.1 (b) (b) (b) ∆n ∆n ∆n 0.05 0.05 0.05 uI uI s s s ∆n ∆n ∆n 0.05 0.05 0.05 suI suI s uI I uI I uI I uI I uI I uI I 0 0 0 0 0 0 0.5 0.5 0.5 φo φo φo 1 1 1 0 0 0 0 0 0 0.5 0.5 0.5 φo φo φo 1 1 1 FIG. 5: Comparison of the phase diagrams for the cases (a) when the passive cross-linkers (a) interact only with crosslinked molecular motors and (b) when the passive crosslinkers interact with both bound and crosslinked motors. u denotes the unstable region, while s denotes region of stable physical overlap. Here fs = 1.9, Γ = 1 γ = 1 and αo = 0.2. ✵✿✶ ✵✿✶ ✵✿✶ ✁♥ ✁♥ ✁♥ ✵✿✵✺ ✵✿✵✺ ✵✿✵✺ ✵ ✵ ✵ ✵ ✵ ✵ ✭❛✮ ✭❛✮ ✭❛✮ s s s ✉ ✉ ✉ ✵✿✺ ✵✿✺ ✵✿✺ ✣♦ ✣♦ ✣♦ ✵✿✶ ✵✿✶ ✵✿✶ ✁♥ ✁♥ ✁♥ ✵✿✵✺ ✵✿✵✺ ✵✿✵✺ ✭❝✮ ✭❝✮ ✭❝✮ ✵ ✵ ✵ ✵ ✵ ✵ s✉ s✉ s✉ ✵✿✺ ✵✿✺ ✵✿✺ ✣♦ ✣♦ ✣♦ ✶ ✶ ✶ ✶ ✶ ✶ ✵✿✶ ✵✿✶ ✵✿✶ ✁♥ ✁♥ ✁♥ ✵✿✵✺ ✵✿✵✺ ✵✿✵✺ ✵ ✵ ✵ ✵ ✵ ✵ ✭❜✮ ✭❜✮ ✭❜✮ s s s ✉ ✉ ✉ ✵✿✶ ✵✿✶ ✵✿✶ ✁♥ ✁♥ ✁♥ ✵✿✵✺ ✵✿✵✺ ✵✿✵✺ ✭❞✮ ✭❞✮ ✭❞✮ ✵ ✵ ✵ ✵ ✵ ✵ ✵✿✺ ✵✿✺ ✵✿✺ ✣♦ ✣♦ ✣♦ s s s ✵✿✺ ✵✿✺ ✵✿✺ ✣♦ ✣♦ ✣♦ ✶ ✶ ✶ ✶ ✶ ✶ FIG. 6: Effect of variation of γ on stability when passive crosslinkers interact with both crosslinked and bound motors: (a) γ = 3, (b) γ = 5, (c) γ = 10 and (d) γ = 20 . Here fs = 1.9, Γ = 1 and αo = 0.2. u denotes the unstable region, while s indicates stable region. boundary separating stable from unstable morphologies of anti-parallel overlapping MTs, reads ∆n = φoG 2g2γ (cid:18)1 + 1 γ − G(cid:19) (11) where G = (g − 1) exp(g) is a function of φo ≡ leφ. On the linear stability boundary the real part of the relevant complex eigenvalue vanishes. Hence the pure imaginary eigenvalues lead to a finite frequency fω, which characterizes the intrinsic oscillations of the MT complex at the threshold of linear stability. The expression for fω reads as, fω = 1 2πs 2g2∆nγ φo − G As shown in Fig.2, linear stability analysis has allowed us to span the whole phase diagram of the anti-parallel MT complex as a function of concentration of passive crosslinkers, φo, and ∆n, a measure of motor flux in the overlap 8 region. Overlapping anti-parallel MTs requires a positive overlap lf > 0, which restricts the physically meaningful region of phase diagram, and leads to an upper bound for the strength of the motor flux ∆c as function of passive crosslinker density. γg(2gΓ+4 fsφo) n = φoeg Γ In Fig. 2, the unshaded region corresponds to physically meaningful overlapping MTs for which both (lf > 0) and nc > 0. This region is bounded by the dashed blue curve, above which (lf < 0), and dash-dotted green line, beyond o = Γ. The critical concentration which nc < 0. A positive number of crosslinked motors requires φo < φc o depends only on the scaled friction coefficient Γ and does not depend on the (un)binding motor rates.The red φc curve separates the linearly stable regime of the anti-parallel MT arrangement with the unstable regime. The linearly unstable region can be further classified into two sub-regions based on how the small deviations about the fixed points grow with time. While in one region the instability grows with a characteristic frequency, in the other the instabilities grow monotonously in time. o, where φc Fig.3 depicts the effect of γ on the stability of the overlapping MTs. Increasing γ diminishes the region of stable physical overlap. It is worthwhile to point out that although the region of stable overlap diminishes, beyond a threshold value of γ, even for arbitrary small concentration of passive proteins, φo, there always exists a range of ∆n for which the overlap configuration of the MTs can be stabilized, as illustrated in the insets of Fig.3c and Fig.3d. B. Interacting motor -passive protein-MT system : From Eqs.(7-9) we can identify the two distinct sets of fixed points that quantify the steady states when there is mutual interaction only between the passive crosslinkers and crosslinked motor proteins, which read as, nf c = 1/ω± =(cid:26)−η ±qη2 − 4αoφ2 o fs Γ(cid:27) /2αoφo fs nf b = exp(φoω± + αoφo)/γω± lf = exp(φoω± + αoφo)/γ∆nω± − 2φoω± − 2αoφo where η = Γ(αoφo − g) − 4 fsφo/Γ − 4 fsαoφo/ω± Γ For attractive interactions between passive crosslinkers and motors ( αo < 0), The solution of fixed point cor- responding to ω+ cannot be sustained since ω+ is always negative, leading to nf c < 0. For repulsive interactions (αo > 0), the solution of fixed point corresponding to ω+ leads to an unstable fixed point. [29] Thus only the solution corresponding to ω− leads to physically relevant steady states. The corresponding expression for stability boundary is, ∆n = −φo + αoφo fs Ω(cid:16)1 − 1 Ω−γ(cid:17) (cid:16) 1 Γ Ω−γ 2γhω2 − 2(cid:17)i (12) where, Ω = (φoω− − 1) exp(φoω− + αoφo). In this regime also intrinsic, spontaneous, oscillations develop at the boundary of stability of magnitude fω = 1 2πs2φoω2 −∆nγ − 4αoφo∆n fsγ Γ − Ω 9 (13) Fig.4 displays the phase diagram in φo − ∆n plane for different interaction strengths, αo, when passive proteins only interact with crosslinked motors. Both for attractive, αo < 0, and repulsive, αo > 0, interactions , increasing the interaction, αo, reduces the region of stable physical overlap of MTs. Significantly, when αo < 0, there is no upper bound of passive crosslinkers, φo (Fig.4a and Fig.4b), beyond which a stable MT physical overlap can be obtained. This is in contrast to the situation for which there is no interaction (αo = 0) as illustrated in Fig.1 or when there is attractive interaction between the crosslinked motors and crosslinking passive proteins as shown in Fig. 4c and Fig.4d, for which there is an upper bound of φo, beyond which physical overlap of MTs cannot be obtained. Fig.5a and Fig.5b shows the comparison of the phase diagrams for the case when passive proteins interact only with crosslinked motors with the case for which the passive crosslinkers interact with both bound and crosslinked motors. For the latter case, there is substantial reduction of the region in the phase diagram for which physical stable overlapping MTs can be obtained. Finally, Fig.6 shows that increasing γ decreases the region where overlapping MTs are stable when passive proteins interact with both the bound and the crosslinked motors. C. Experimental relevance: We can estimate the typical overlap length of MTs and motor numbers using experimental data known for biological systems. For single motor velocity, vo = 1 µms−1,[24, 25] and the bare unbinding rate ko u = 1 s−1 [24, 25], the typical processivity length lp = 1 µm. Using the value of thermal energy, kBT= 4.2 pN-nm [27], and length scale associated with unbinding process of motors, b = 1.3 nm [26], an estimate for the additional reference length scale, le = 2.6 nm is obtained. Experimental measurements for the stall force for kinesin, fs = 6 pN [24, 25], in turn implies fs ∼ 1.9. For an experimental estimate of the binding rate, kb = 5 s−1[24, 25], we get γ ∼ 5. Indeed under different physiological conditions, γ may vary one order of magnitude [27]. Furthermore, we consider that passive crosslinkers bind with a characteristic energy, ǫo = 2kBT. Quantitative estimates of the friction coefficient of passive crosslinkers at low concentrations [13] implies that the scaled friction coefficient Γ ∼ 1. Assuming a linear density of passive crosslinker, φ = 0.35 nm−1, and choosing the bath motor concentration such that ρ3Dk3D b = 1.7 µm−1 s−1 (corresponding to φo = 0.9, ∆n = 0.9, the point is marked with '∗' in Fig.2) we obtain an estimate of the stable overlap length ∼ 0.45 µm with bound and crosslinked motor numbers being nb ∼ 7, and nc ∼ 6, respectively. These estimates fall well within the scale of cellular processes and thus suggests that the interplay between the active forces due to motors and the passive crosslinkers is a biologically relevant mechanism in determining the stability of overlapping MT filaments. III. SUMMARY AND DISCUSSION In summary we have shown that the presence of passive crosslinkers qualitatively impacts the mechanical properties of arrangements of overlapping biofilaments that are subject to active sliding forces due to molecular motors. These forces arise due to the interaction of passive crosslinkers with overlapping filaments in the presence of a passive crosslinker bath. This is a distinct mechanism from other passive force generating mechanism such as entropic forces due to confinement or forces due to depletant molecules. We have shown that for a pair of anti-parallel MTs, the interplay of sliding forces due to such passive proteins crosslinkers with active sliding forces due to molecular motors can stabilize the overlapping filament pair. Further we have identified that this passive mechanism is necessary to ensure the emergence of stable overlaps. The effect of attractive mutual interaction between passive crosslinkers and active motors leads to an enhanced range of passive crosslinker density for which stable overlapping MT pair can be obtained while a repulsive interaction between the motors and passive crosslinkers tend to diminish the domain of stable overlaps. Acknowledgment IP acknowledges MINECO and DURSI for financial support under projects FIS2015-67837- P and 2017SGR844, respectively. SM acknowledges SERB project EMR/2017/001335 for financial support. 10 [1] B. Alberts at. al , Molecular Biology of the cell ( Garland Science, New York, 2007, 6th ed) [2] J. Howard, Mechanics of motor proteins and the cytoskeleton (Sinauer Associates, Sunderland, 2001) [3] F. Nedelec, T. Surrey, A.C. Maggs, and S. Leibler, Nature 389, 305 (1997) [4] Y. Hatwalne, S. Ramaswamy, M. Rao, and R. A. Simha, Phys. Rev. Lett. 92, 118101 (2004) [5] S. Sankararaman, G. I. Menon, and P. B. Sunil Kumar, Phys. Rev. E 70, 031905 (2004) [6] N. P. Ferenz, R. Paul, C. Fagerstrom, A. Mogilner, and P. Wadsworth, Curr. Biol. 19 1833 (2009) [7] S. Muhuri, I. Pagonabarraga, and J. Casademunt, EPL 98, 68005(2012) [8] P. Malgaretti and S. Muhuri, EPL 115, 28001 (2016) [9] S. Ghosh, V. N. S. Pradeep, S. Muhuri, I. Pagonabarraga, and D. Chaudhuri, Soft Matter 13, 7129 (2017) [10] S. W. Grill, K. Kruse, and F. Julicher, Phys. Rev. Lett. 94, 108104 (2005) [11] X. Su, H. Arellano-Santoyo, D. Portman, J. Gailard, M. Vantard, M. Thery, and D. Pellman, Nat. Cell. Biol. 15, 948 (2013) [12] Y. Fukuda, A. Luchniak, E. R. Murphy, and M. L. Gupta, Curr. Biol. 24, 1826 (2014) [13] Z. Lansky, M. Braun, A. Ludecke, M. Schlierf, P. R. ten Wolde, M. E. Janson, and S. Diez, Cell 160, 1159 (2015) [14] M. Braun, Z. Lansky, F. Hilitski, Z. Dogic, S. Diez, Bioessays 38, 474 (2016) [15] R. Subramanian, E. M. Wilson-Kubalek, C. P. Arthur, M. J. Bick, E. A. Campbell, S. A. Darst, R. A. Milligan, and T. M. Kapoor, Cell 142, 433 (2010) [16] P. Bieling, I. A. Telley, and T. Surrey, Cell 142, 420 (2010) [17] H. S. Kuan and M. D. Betterton, Biophys. J. 110 2034 ( 2016) [18] M. Braun, Z. Lansky, F. Hilitski, Z. Dogic, and S. Diez, Bioessays 39, 474 (2016) [19] Z. Lansky, M. Braub, A. Ludecke, M. Schlierf, P.R. ten Wolde, M.E. Janson, S. Diez, Cell 160, 1159 (2015) (and comment by D. J. Odde, Cell 160 1041 (2015)). [20] F. Hilitski, A.R. Ward, L. Cajamarca, M.F. Hagan, G.M. Grason, Z. Dogic, Phys. Rev. Lett. 114, 138102 (2015) [21] D. Johann, D. Goswami, K. Kruse, Phys. Rev. E 93, 062415 (2016) [22] D. Johann, D. Goswami, and K. Kruse, Phys. Rev. Lett. 115, 118103 (2015) [23] O.Campas, J. Casademunt, and I. Pagonabarraga, EPL 81, 48003 (2008) [24] S. Klumpp and R. Lipowsky, Proc. Natl. Acad. Sci. 102, 284 (2005) [25] M. J. I. Muller, S. Klumpp and R. Lipowsky, Proc. Natl. Acad. Sci. 105, 4609 (2008) [26] Schnitzer M. J. et al., Nat. Cell Biol., 2, 718 (2000) [27] A. Chaudhuri and D. Chaudhuri, Soft Matter 12, 2157 (2016) [28] The other possible sets of roots for the eigenvalue equation are (i) all roots are real and negative, (ii) one positive real root and a pair of complex conjugate roots and (iii) all roots are real and at least one is positive. While (i) always corresponds to the stable fixed points, both (ii) and (iii) corresponds to unstable fixed points. [29] When αo > 0, the set of fixed points corresponding to ω+ has always at least one real positive root for the eigenvalue equation for the fluctuation spectrum which corresponds to linearly unstable fixed point.
1207.3188	1	1207	2012-07-13T09:55:43	How simple can a model of an empty viral capsid be? Charge distributions in viral capsids	[ "physics.bio-ph", "q-bio.MN" ]	We investigate and quantify salient features of the charge distributions on viral capsids. Our analysis combines the experimentally determined capsid geometry with simple models for ionization of amino acids, thus yielding the detailed description of spatial distribution for positive and negative charge across the capsid wall. The obtained data is processed in order to extract the mean radii of distributions, surface charge densities and dipole moment densities. The results are evaluated and examined in light of previously proposed models of capsid charge distributions, which are shown to have to some extent limited value when applied to real viruses.	physics.bio-ph	physics	How simple can a model of an empty viral capsid be? Charge distributions in viral capsids Anze Losdorfer Bozic∗, Antonio Siber†, and Rudolf Podgornik∗‡1 1∗Department of Theoretical Physics, Jozef Stefan Institute, SI-1000 Ljubljana, Slovenia †Institute of Physics, Bijenicka cesta 46, P.O. Box 304, 10001 Zagreb, Croatia ‡Department of Physics, Faculty of Mathematics and Physics, 2 1 0 2 l u J 3 1 ] h p - o i b . s c i s y h p [ 1 v 8 8 1 3 . 7 0 2 1 : v i X r a University of Ljubljana, SI-1000 Ljubljana, Slovenia (Dated: July 24, 2018) We investigate and quantify salient features of the charge distributions on viral capsids. Our analysis combines the experimentally determined capsid geometry with simple models for ionization of amino acids, thus yielding the detailed description of spatial distribution for positive and negative charge across the capsid wall. The obtained data is processed in order to extract the mean radii of distributions, surface charge densities and dipole moment densities. The results are evaluated and examined in light of previously proposed models of capsid charge distributions, which are shown to have to some extent limited value when applied to real viruses. I. INTRODUCTION Starting from the early structural studies of tobacco mosaic virus gels, Bernal and Fankuchen [1] already in- voked electrostatic interactions that are "probably due to the ionic atmospheres surrounding [viruses]" to ex- plain their behavior in ionic solutions. Virus architec- ture, cell attachment, penetration, progeny assembly and egress should be dependent on long-range colloidal in- teractions between and within viruses and various other structural components of the cell [2]. Though the impor- tance of electrostatic interactions in the context of viruses is well recognized (see the review by Siber et al. [3] and references therein) and electrostatic models on various levels of sophistication abound [4 -- 12], preciously little systematic effort [13, 14] has been directed towards de- tailed quantification of the charge distributions on and within viral capsids. Models of electrostatic interactions in the context of viruses as well as virus-like nanoparticles [15, 16] only make sense if they are derived from detailed observed charge distributions on the epitopal and hypo- topal surfaces[43] of the capsid, as well as charge buried inside the capsomeres. Therefore, to evaluate previous modeling attempts, to propose better models, and to find out whether there is a prototypical charge distribution of a virus capsid, we embark on a detailed study of charge distribution on empty viral capsids. Our focus will not reside upon the distribution of charged amino acids along the 1D primary sequences of capsomeres [13] but exclusively on the 3D geometry of the charge distribution on the capsid. While the details of the large-scale nature of the electronic structure of pro- teins that would allow the assessment of partial charge distribution buried inside the protein core are presently unavailable [17, 18], the charges of the amino acids re- siding on the surface of the capsomers in contact with the aqueous solvent at physiological pH are known and readily available [19]. We will use the charge distribu- tion on the epitopal and hypotopal capsid surfaces of a large number of viruses in order to analyze and model its statistical signature among the various virus types. In order to describe any charge distribution one first needs to identify the spatial region in which such a distri- bution resides and then quantify its geometry via a set of lowest multipolar moments [20]. With this goal in mind we will examine a number of available X-ray scattering and cryo-electron microscopy structural data on capsids of various viruses in order to extract a small set of pa- rameters that would characterize simple models of charge distribution pertaining to these capsids. This minimal set of parameters includes the average size and thickness of the capsid, the surface charge density, and surface dipole density magnitude of the charge distribution. The structure of the paper is as follows: We first ex- plain how we construct two simple capsid models from the experimental data and obtain the parameters per- taining to them. We then briefly analyze the geometrical properties of the two models before proceeding to the monopolar and dipolar charge distributions on the cap- sids. We focus on different surface charge distributions pertaining to both models, and the effect of charge on the disordered protein N-tails. Lastly, we consider the surface dipole density in capsids, and conclude with the discussion of our results. II. FROM STRUCTURES TO MODEL(S) We focus on two simple models most widely used: a single, infinitely thin charged shell of radius RM and sur- face charge density σ as shown in Fig. 1a [4, 9], and two thin shells of inner and outer radius Rin and Rout (giving a capsid thickness of δM = Rout − Rin), carrying surface charges of σin and σout (Fig. 1b) [4, 5]. We will refer to the two models as the single-shell and double-shell model, respectively. Besides the monopole (total) charge distri- bution, we also consider the dipole distribution on such model capsids. The analysis is done solely for empty viral capsids not encapsidating any genetic material. In our analysis we use experimental data deposited 2 and 5. We analyze approximately 130 viruses from dif- ferent families and compare their corresponding model charge and mass distribution parameters. We classify the different viruses by their genome (single-stranded (ss) DNA and ssRNA on one hand and double-stranded (ds) DNA and dsRNA on the other) [26] and Caspar-Klug triangulation number T [27, 28]. These are the most conspicuous properties that classify the an- alyzed viruses; there are others, for example the sec- ondary/tertiary structure of capsid proteins (i.e. pres- ence of α-helices, β-barrels, . . . ). However, we expect such additional properties play a smaller role in the task at hand [29], and their inclusion would yield no addi- tional insight in our analysis. We consider separately the bacteriophages (which come with either DNA or RNA genome), as well as the T = p3 capsids[44] of RNA viruses (which are abundant in our sample), since they might dif- fer in their properties [29]. III. SINGLE- AND DOUBLE-SHELL MODELS We begin our analysis by constructing single-shell and double-shell models from the mass distributions in differ- ent viral capsids. The single, infinitely thin shell model is characterized by one parameter only, the mean capsid radius RM (Fig. 1a). The latter is extracted from the radial mass distribution in the capsid ρ(r) = ∆m 4πr2∆r , (1) where the angular coordinates have already been pro- jected out. This can be done for either the distribution of capsid atoms, centers-of-mass of amino acids, or centers- of-mass of proteins. The differences between these are within a couple of angstroms for most capsids, so we concern ourselves mainly with the distribution of capsid protein atoms. The double-shell model on the other hand is charac- terized by two radii, the inner (hypotopal) and the outer (epitopal) radius Rin and Rout (Fig. 1b). Their differ- ence is the capsid thickness δM = Rout − Rin. These parameters are again obtained from the radial density distribution, with the thickness defined as the full-width- half-maximum (FWHM) of the distribution, and the in- ner and outer radius defined as the inner and outer half- maximum of the distribution. The bin size of the distri- bution influences the result to some extent, but the effect is still lower than the usual experimental precision. Also, since the exact half-maxima are never achieved due to the discreteness of the distribution, the condition they have to satisfy is to lie within 5% around the half-maximum. To illustrate how this analysis is done we consider the example of cucumber mosaic virus (CMV, PDB ID 1f15). Figure 2 shows the radial mass distribution in the cap- sid, where we can see that the root parts of the protein N-tails, prominent in this example, are protruding into the capsid interior as defined by the hypotopal radius of Figure 1: Schematic representation of the single-shell and double-shell models treated in the paper. Left: single-shell model with mean radius RM and surface charge distribution σ. Right: double-shell model with the inner shell of radius Rin and outer shell of radius Rout. The surface charge dis- tributions pertaining to the two shells are denoted by σin and σout. in the VIPERdb database [21]. This allows us to con- struct three-dimensional structures of viral capsids, from which we obtain the various mass and charge distribu- tions within the capsid. We consider not only the distri- bution of atoms inside a capsid, but the distribution of amino acids (their positions taken as centers-of-mass of their constituent atoms) and complete protein chains as well. Some capsid data do not contain the positions of all atoms but only the positions of alpha carbons -- in such cases we equate their positions with the positions of the amino acids to which they belong. Due to the methods of detection there are also no hydrogen atoms included in the experimental data. We have tested the effect of the lack of hydrogen atoms on our analysis by adding the hydrogen atoms via the MolProbity web server [22] to several different capsid entries. As expected, their effect on the mass distributions can be neglected, and we did so throughout our analysis. To obtain the charge distributions of the capsids we extract the positions of charged amino acids from the experimental data by using Tcl scripting language in VMD [23]. At physiological pH of 7.4 we consider the fol- lowing amino acids as charged [24]: aspartic acid (ASP) and glutamic acid (GLU) carrying a charge of −1.0 e0, lysine (LYS) and arginine (ARG) carrying a charge of +1.0 e0, and histidine (HIS) carrying a fractional charge of +0.1 e0 (where e0 is the elementary charge). The available experimental data cannot capture the usually disordered N-tails of proteins, which in certain cases do carry a significant charge [11]. To estimate to what extent this affects our analysis we also com- pare the capsid protein sequences of viruses deposited in VIPERdb with the full sequences obtained from the UniProt database of protein sequences [25]. In the following sections we extract and analyze the parameters of these simple models from the experimen- tal data which look like the examples shown in Figs. 2 3 To a good approximation, the mean capsid radius of the single-shell model increases with the square root of the capsid T -number, which means that one can idealize the capsid as consisting of uniformly distributed copies of a disk-shaped (or prism-shaped) elementary protein with a fixed area. A minimal model of this type for equilib- rium capsid structure with explicit interaction between capsomeres on a spherical shell has received much atten- tion recently [7, 30]. An additional point of interest is also the ratio of the capsid thickness and the mean capsid radius δM /RM , as this can influence the validity of mechanical mod- els of viruses, for instance continuum elasticity models of thin elastic shells [26, 31]. Analysis of this ratio is shown in Fig. 4. For the average virus analyzed this ra- tio lies around 0.2, but is (expectedly) no longer small for smaller, T = 1 viruses, where the idealization of a thin protein shell is misleading. These characteristics of the capsid architecture turn out to be insensitive to taking the mass distribution in- stead of the position distribution, which barely affects the calculated mean radius or the thickness of the cap- sid. A more detailed analysis of the conserved geometri- cal properties of viruses and their elastic properties will be published elsewhere (A. Losdorfer Bozic, A. Siber, and R. Podgornik, in preparation). Figure 3: Outer capsid radius compared to inner cap- sid radius of the double-shell model. The thickness of the capsid emerges naturally from this linear dependence: the dashed line shows a thickness of 4.5 nm (i.e. Rout = Rin + 4.5 nm), and the dot-dashed line shows a thickness of 1.5 nm. Approximately two-thirds of the analyzed capsids have a thickness between 2-4 nm. Symbols encode some dif- ferent virus types: single-stranded genome (circles), double- stranded genome (squares), bacteriophages (diamonds), and T = p3 ssRNA viruses (triangles). Same symbols are used throughout the paper in other similar figures. Figure 2: Cross-section of (experimentally determined) capsid mass distribution in the example of the cucumber mosaic virus (ssRNA) capsid (strain FNY), constructed from RCSB Protein Databank entry 1f15. The drawing was constructed with a procedure described in Ref. [3] with W = 1.34 nm and t = 0.85, where all amino acids were as- signed strength ("q/e0") 1. Protrusions can be seen on the capsid interior which are the roots of protein N-tails; com- parison with the full protein sequence shows that they are not complete. The inset shows the radial mass distribution (Eq. 1) across the capsid, normalized so that total area of the histogram equals 1; marked are the mean capsid radius RM (single-shell model) and the inner and outer radii Rin and Rout (double-shell model). the distribution Rin. Any significant outer protrusions such as spikes are located in the exterior of the capsid as defined by the epitopal radius Rout. These details are not included in the simpler single-shell model, characterized only by the mean capsid radius RM . In Fig. 3 we next plot the inner and outer radius of the double-shell model for the entire dataset of analyzed viruses[45]. Capsid thickness naturally follows from the apparent linearity of their relation and is generally well defined. For more than 75% of viruses in our sample the thickness is confined to a narow range, δM ∼ 1.5-4.5 nm. 0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 9 10 11 12 13 14 15 16ρ(r)r [nm]RRRinoutM(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)1020301510203015RinnmRoutnm4 a very different distribution is shown in Fig. 6 for the case of simian virus 40 (PDB ID 1sva). Here, it is difficult to separate the charge distribution into a positively charged hypotopal surface and a negatively charged epitopal sur- face, and there is a good deal of charge variation within the capsid wall. Figure 4: Ratio of capsid thickness and capsid mean ra- dius δM /RM as a function of the triangulation number. The capsid thickness becomes more and more comparable to the capsid radius as the triangulation number gets lower. IV. CHARGE DISTRIBUTIONS As specified in Section II, there are five amino acids in proteins that carry charge at physiological pH. However, there is some uncertainty as to whether these ionizable amino acids are charged or not when buried inside a pro- tein. Either the dissociation cost for charges buried in the protein interior is too high and the buried charges are therefore virtually absent [19], or the converse is true and the majority of ionizable amino acids buried inside the protein are ionized [32]. Yet another possibility is that the local environment of a buried ionizable amino acid is changed, so that its charge is modified [33, 34]. With this in mind we consider two limiting cases: in the first one, we take all the ionizable amino acids as charged, no matter where they are located. In the sec- ond case we consider as charged only the ionizable amino acids lying on the periphery of the capsids as defined by their inner and outer radii. This is admittedly a simpli- fied picture, but it enables us to cover the extreme cases. Only a complete ab initio quantum chemical calculation of the electronic properties of capsid proteins in contact with aqueous solvent and neighboring proteins could re- solve the issue of the correct charging model for the amino acids [17, 18]. A sample radial charge distribution is again shown for the CMV in Fig. 5. All the ionizable amino acids are taken as charged, regardless of their position in the cap- sid. In this case, we observe that the charges on the hypotopal and epitopal surfaces are mostly positive and mostly negative, respectively; there are also some charges buried in the capsid wall. These are the only distinguish- ing features of an otherwise very complicated charge dis- tribution. The distribution of charges in the capsid can vary significantly from virus to virus, and there appears to be no simple way of classifying them. One example of Figure 5: Cross-section of charge distribution in the ex- ample of cucumber mosaic virus (PDB ID 1f15). The 3D representation is constructed as described in Ref. [3] with W = 1.34 nm and t = 0.85. The histogram plot shows corresponding radial charge distribution across the capsid. Note that the 3D representation separately represents neg- ative (blue) and positive (red) charge densities, while the histogram shows the total charge density distribution, calcu- lated by weighing both charge distributions. As the negative and positive charge distributions overlap, in order to clearly show both of them, the positive and negative distributions are infinitesimally shifted with respect to each other, so that on the right (left) half of the 3D representation the positive (negative) distribution is infinitesimally closer to the viewer. Marked are the capsid mean mass radius RM and the inner and outer radii Rin and Rout of the single- and double-shell models, respectively. (cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)1234713250.100.500.200.300.150.70T∆MRM-0.2-0.1 0 0.1 0.2 0.3 9 10 11 12 13 14 15 16RRRinoutMρ (r) [e /nm ]Q03r [nm]given in terms of a surface charge density σ: σ = Q 4πR2 M ; 5 (3) this is the surface charge density of the single-shell model with all the ionizable amino acids being charged. Here we could equally well use RQ (Eq. 2) instead of RM but we stick with the latter for consistency. The surface charge densities of the double-shell model σin and σout are similarly defined as σin/out = Qin/out The charge on the inner shell is Qin =P 4πR2 in/out . i qi ; ri(qi) < Rin, and the charge on the outer shell is calculated in an analogous fashion. In order to compare the total charge of the two models, we also define (4) σIO = Qin + Qout 4πR2 M . (5) This can be considered as the surface charge density of the single-shell model with only peripheral amino acids (i.e. not buried inside the capsid as defined by the double- shell model) taken as charged. The dependence of the total charges of the single-shell model in both limits (σ and σIO) on the capsid T -number is shown in Fig. 7. We can see that for the majority of viruses the total charge becomes more positive when we exclude the buried charges. The values of capsid surface charge densities mostly lie within the range from −0.4 to +0.4 e0/nm2. Invoking the previously obtained RM this implies net charge values in the range Q (cid:46) 4500 e0. Empty viral capsids are obviously quite charged and their interactions either between themselves or with other structural components of the cell must be to a large ex- tent modulated by electrostatics. In Fig. 8 we then compare the inner and outer surface charge densities of the double-shell model. An emerging feature, which can be also discerned from the histogram in Fig. 9, is that the outer charges of viruses are close to zero or slightly negative; on the contrary, there are quite some viruses that carry a significant positive inner charge, even though a lot of them still carry an inner charge close to zero. The viruses carrying a positive inner charge in this case are mostly viruses with single-stranded genome (with the exceptions of T = 1 and T = p3 capsids) as well as bacteriophages with single-stranded genome. B. Effect of Missing (Disordered) N-tails The basic (positively charged) N-tails of capsid pro- teins are largely unresolved in X-ray scattering experi- ments and Belyi and Muthukumar [11] have shown that due to their positive charges they can strongly interact with the oppositely charged RNA genome. This inter- action is also a major factor in constraining the length Figure 6: Cross-section of charge distribution in the ex- ample of simian virus 40 (PDB ID 1sva). The figure is con- structed in the same manner as Fig. 5. However, the radial charge distribution in this example cannot be easily cat- egorized, and, most notably, does not have a pronounced positively charged inner part of the capsid and negatively charged outer part of the capsid. Marked are again the cap- sid mean mass radius RM and the inner and outer radii Rin and Rout of the single- and double-shell models, respec- tively. A. Total Charge sum of all charged amino acids in the capsid, Q =P The total charge of the capsid Q is calculated as the i qi, within the two limiting models described above. We also introduce the mean radius of the distribution of absolute P charge (mean charge radius) RQ, i qiri Q RQ = (2) , where qi are the charges of amino acids located at radii ri. The charge mean radius of most of the viruses differs from the mass mean radius by up to a few percent. The total charge of the single-shell model is usually -0.15-0.1-0.05 0 0.05 0.1 0.15 0.2 18 19 20 21 22 23 24 25 26ρ (r) [e /nm ]Q03r [nm]RRRinoutM6 the new values for the total surface charge density σ0 in the single-shell model and new values for the inner sur- face charge density σ0 in in the double-shell model. From the latter we can also obtain the corrected total surface charge density of Eq. 5, σ0 IO; all the surface charge den- sities are again normalized with RM . The distributions of the new surface charge densities of the single-shell model as a function of the triangula- tion number are shown in Fig. 10. In general, a trend toward more positive charge is observed by corrections up to ∆Q ∼ 6000 e0. The same is true also for the double-shell model where the total surface charge density decomposes into the hypotopal in epitopal contributions (Figs. 8 and 9). The rationalization for this rescalings of the capsid charge once one adds explicit charges on the disordered N-tails could be envisioned as stemming Figure 7: Distributions of the total capsid surface charge density depending on the triangulation number, taking into account either all the charged amino acids (σ; Eq. 3) or only the charged amino acids lying outside the mean capsid thickness (σIO; Eq. 5). In the latter case the capsids tend to carry slightly more positive charge; the relevant range of surface charge densities is in both cases well described by the interval [−0.4, 0.4] e0/nm2. of viral genome, implying a linear relation between the number of positive charges on the tails and the length of the encapsidated RNA [11, 35, 36]. The effect of missing disordered tails in the experimen- tal structure data can be most easily estimated from the changes in the total capsid charge brought about by the positively charged N-tails. The missing charge is calcu- lated from the full primary sequences of capsid proteins. Since nothing can be said about their position (other than that they are most likely disordered and located on the hypotopal side of the capsid), we take all the missing charges to be located in the interior of the capsid, that is on the inside of RM or Rin within the single- and double- shell models, respectively. This is an assumption which should hold true for most of the analyzed viruses, but cannot be easily verified. By adding the charge contributed by the N-tails we get an estimate of the charge correction ∆Q and from there Figure 8: Top panel: comparison of the surface charge densities on the inner and outer shells of capsids (Eq. 4). The majority of the viruses tend to have at least slightly negatively charged outer shell. There is more diversity con- cerning the charge on the inner shell, which is in our sample centered around zero net charge, with viruses having ei- ther negatively or positively charged interior. Bottom panel: same as above, with added disordered N-tails of the pro- teins. There is a noticeable shift of the inner shell charge (to which the missing sequences were attributed) towards more positive values. (cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)123471325(cid:45)1.5(cid:45)1.0(cid:45)0.50.0TΣe0nm2(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)123471325(cid:45)0.4(cid:45)0.20.00.20.4TΣIOe0nm2(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:45)0.20.00.20.40.60.8(cid:45)0.6(cid:45)0.5(cid:45)0.4(cid:45)0.3(cid:45)0.2(cid:45)0.10.00.1Σine0nm2Σoute0nm2(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:45)0.50.00.51.0(cid:45)0.6(cid:45)0.4(cid:45)0.20.00.2Σ'ine0nm2Σoute0nm2as P(r0) =X i qi(ri − r0). 7 (6) where qi are again the charges of amino acids located at radii ri within the capsid shell. The dipole distribution is not invariant with respect to geometric description and has to be calculated with respect to some particular refer- ence point r0 [20]. We choose for the origin the radius of the centre of absolute charge RQ. Apart from the abso- lute magnitude of the dipolar moment we again consider the surface dipole density, normalized with the capsid mean radius. The surface dipole density is completely analogous to the surface charge density introduced be- fore. Since the dipolar moment and its local surface den- sity are vectors, we can decompose them into a radial and a tangential component -- across and along the cap- sid wall -- and compare their respective magnitudes. We calculate the dipolar moment for the basic asym- the conglomeration of a T - metric unit of a capsid: Figure 10: Surface charge density of the capsids with added disordered N-tails of the proteins plotted against the triangulation number, for both limiting cases consid- ered (compare with Fig. 8). The total charge moves towards more positive values in both cases; this trend is least pro- nounced in bacteriophages and T = p3 ssRNA viruses. Figure 9: A histogram showing the distribution of inner and outer surface charge densities of the double-shell model in the sample of viruses used in the analysis. The upper part shows the outer surface charge density (blue), and the lower part shows the inner surface charge density without (red) and with added charge of the N-tails (magenta). primarily from the strong N-tail genome electrostatic in- teractions [11]. Since the genome is negatively charged, the hypotopal N-tails effectively act locally to completely screen this charge, conferring much needed stability to the virus. The most pronounced and consistent changes can be observed in the case of viruses with single-stranded genome, with a clear separation of the total charge be- tween T = 3 single-stranded viruses and the rest, and a slightly less pronounced separation in the T = 1 viruses as well. The charges of the bacteriophages remain mostly unchanged after the explicit addition of N-tail charges, as do the charges of T = p3 ssRNA viruses. The latter case is somewhat surprising, as the majority of single- stranded viruses undergo an increase of charge. The ef- fect on double-stranded viruses is not so systematic. From these results we conclude that the surface charge of the capsid is quite large, being comparable to the equivalent surface charge of a DNA molecule. In ab- solute terms the number of effective charges can go into tens of thousands, which is an impressive charge even af- ter all the screening and condensation effects are taken into account, making viral capsids quintessential charged nano-objects [3]. The electrostatic interactions stemming from this huge capsid charge are therefore important and cannot be neglected. C. Dipole Distribution Lastly, we analyze the first higher order multipolar mo- ment of the capsid charge distribution, the dipole mo- ment. The electric dipole of the capsid shell is defined (cid:45)0.50.00.51.0101020302030400Σinoute0nm2NΣinΣout(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)123471325(cid:45)1.0(cid:45)0.50.00.51.01.5TΣ'e0nm2(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:236)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)(cid:242)123471325(cid:45)0.50.00.51.0TΣ'IOe0nm2number of proteins which, upon applying 60 rotation ma- trices of the icosahedral group, compose the entire cap- sid. This is done to simplify the analysis and enable us to make a good comparison of the results; in principle it would be possible to calculate the dipolar moment of each capsid protein, but we believe this would not serve any additional purpose in our analysis. It would even make sense to calculate the dipolar moments of either dimers, trimers, pentamers, hexamers, or whatever the basic structural units of each capsid is [37], so as to see if the dipolar moment plays a role in their interaction. However, these units differ from virus to virus, and would be difficult to address within our approach. In any case, we find that the magnitudes of the dipolar moments in capsid proteins are small, and these effects are thus likely to be small as well. The majority of viruses have small surface dipole den- sities, below 0.02 e0/nm. For comparison, one could note that the surface dipole density of a completely oriented layer of water molecules at close packing would be 0.55 e0/nm. The obvious conclusion then is that if there is any ordered water on the periphery of the capsid, its ef- fect will overwhelm the intrinsic dipolar moments of the capsid proteins. Note however that the surface water or- dering in "hydration layers" would be highly contingent on the local protein charge distribution [38]. One should nevertheless remark here that the dipolar moment cal- culated above does not take into account the complete electronic structure of the proteins with implied partial charges within the protein cores that may eventually con- tribute to the total dipolar moments of the capsid pro- teins. Regardless, compared to monopolar, the dipolar surface charge density seems to be much less important. V. DISCUSSION & CONCLUSIONS We have performed a detailed statistical analysis of mass and charge distributions in approximately 130 empty viral capsids, and extracted the relevant parame- ters needed to construct simple single- and double-shell models of them. The complete list of analyzed viruses, their (triangulation) T -numbers, and genome types, as well as a compilation of the results presented in the pa- per, is available from the authors upon request. The analysis of the charge distribution in capsids was based on several assumptions that do not have a uni- versal validity, but are at present necessary to take as given. In structuring our models we ignored the depen- dence of the dissociation constants of amino acids on the detailed molecular environment as it would, even though possible in a case-by-case analysis [39], make our gen- eral approach completely untransparent. Nevertheless, these features should be investigated in the context of an improved model that would consider fully dissociated charge of amino acids on solvent accessible surface of a protein as well as the rearrangement of charges inside the protein due to quantum electron charge transfer [17]. 8 However, the calculation of the latter is at present not feasible for such a large number of amino acids, and we thus focused only on the dissociated charge. Some viruses are also reported stable under in vitro conditions at non- physiological pH [40]. Apart from the fact that this at- tests to the importance of electrostatic interactions in self-assembly of viruses it also has implications for their charges. Therefore, some approximation for calculating the dis- sociation charges of amino acids in a protein has to be made, and it can be done in several ways [11, 21, 41]. We chose a straightforward and simple method for extracting the charges from 3D experimental data at a single value of solution pH that enabled us to perform a consistent and general analysis. It is only one possibility though, and different approaches can yield quantitatively different re- sults especially if the solution pH variation is considered in full. Within the limitations described above, we were able to quantify the radial capsid charge distribution, its cor- responding surface charge densities, dipole moments, and some of their geometric properties. This is clearly an im- portant information to be had when using simple models of viral capsids. The monopolar surface charge density of the capsids was found to be quite large when com- pared with other charged biomolecules, being in the range [−0.4, 0.4] e0/nm2. We have also shown that for the over- all charge of the virus capsids the disordered N-tails con- tribute significantly to the net charge, often changing its sign. Consequently, this also results in strongly positively charged interiors of ssRNA viruses, for which it has been suggested that the interior charge is correlated with the genome length [11, 12, 42]. While the dipolar charge contribution turned out to be on the other hand overall much smaller, it can never- theless play an important role whenever stabilization of high energy structures hinges on important subdominant contributions. It is in fact this secondary dipolar density that most probably governs the short range interactions between capsomeres [38]. Contrary to some of the capsid geometrical proper- ties, the distribution of capsid charges does not seem to possess any regularity among viruses with similar trian- gulation numbers, genome types, or species, as was also observed by Michen and Graule [14] in the study of their isoelectric points. The choice of the dataset used in such a study can certainly influence the result to some ex- tent [12], and a future increase in the number and va- riety of available experimental data would undoubtedly improve the analysis. Acknowledgments One of us (A.L.B.) thanks A. Ljubetic for introducing him to Tcl scripting language in VMD. A.L.B. acknowledges the support from the Slovene Agency for Research and Development under the young researcher grant. A.S. acknowledges support from the Ministry of Science, Education, and Sports of Republic of Croatia (Grant No. 035-0352828-2837). R.P. acknowl- edges support from the Slovene Agency for Research and Development through research program P1-0055 and re- search project J1-4297. 9 [1] J. Bernal and I. Fankuchen, "X-ray and crystallographic studies of plant virus preparations," J. Gen. Physiol. 25 (1941) 111 -- 165. [q-bio.BM] (2011) . [19] A. V. Finkelstein and O. B. Ptitsyn, Protein Physics: A Course of Lectures. Academic Press, 2002. [2] K. Iwasaki and T. Omura, "Electron tomography of the [20] J. D. Jackson, Classical Electrodynamics. Wiley, New supramolecular structure of virus-infected cells," Curr. Opin. Struct. Biol. 20 (2010) 632 -- 639. [3] A. Siber, A. Losdorfer Bozic, and R. Podgornik, "Energies and pressures in viruses: contribution of nonspecific electrostatic interactions," Phys. Chem. Chem. Phys. 14 (2012) 3746 -- 3765. [4] A. Siber and R. Podgornik, "Role of electrostatic interactions in the assembly of empty spherical viral capsids," Phys. Rev. E 76 (2007) 061906. [5] P. Prinsen, P. van der Schoot, W. M. Gelbart, and C. M. Knobler, "Multishell structures of virus coat proteins," J. Phys. Chem. B 114 (2010) 5522 -- 5533. [6] A. Losdorfer Bozic, A. Siber, and R. Podgornik, "Electrostatic self-energy of a partially formed spherical shell in salt solution: Application to stability of tethered and fluid shells as models for viruses and vesicles," Phys. Rev. E 83 (2011) 041916. [7] C. J. Marzec and L. A. Day, "Pattern formation in icosahedral virus capsids: the papova viruses and nudaurelia capensis β virus," Biophys. J. 65 (1993) 2559 -- 2577. [8] R. Zandi, P. van der Schoot, D. Reguera, W. Kegel, and H. Reiss, "Classical nucleation theory of virus capsids," Biophys. J. 90 (2006) 1939 -- 1948. York, 3rd ed., 1999. [21] M. Carrillo-Tripp, C. M. Shepherd, I. A. Borelli, S. Venkataraman, G. Lander, P. Natarajan, J. E. Johnson, C. L. Brooks, III, and V. S. Reddy, "Viperdb2: an enhanced and web api enabled relational database for structural virology," Nucleic Acids Res. 37 (2009) D436 -- D442. [22] V. B. Chen, W. B. Arendall III, J. J. Headd, D. A. Keedy, R. M. Immormino, G. J. Kapral, L. W. Murray, J. S. Richardson, and D. C. Richardson, "Molprobity: all-atom structure validation for macromolecular crystallography," Acta Crystallogr. D66 (2010) 12 -- 21. [23] W. Humphrey, A. Dalke, and K. Schulten, "VMD -- Visual Molecular Dynamics," J. Mol. Graphics 14 (1996) 33 -- 38. [24] M. J. Betts and R. B. Russell, "Amino acid properties and consequences of substitutions," in Bioinformatics for Geneticists, M. R. Barnes and I. C. Gray, eds. Wiley, 2003. [25] The UniProt Consortium, "Ongoing and future developments at the universal protein resource," Nucleic Acids Res. 39 (2011) D214 -- D219. [26] W. H. Roos, R. Bruinsma, and G. J. L. Wuite, "Physical virology," Nat. Phys. 6 (2010) 733 -- 743. [9] W. K. Kegel and P. van der Schoot, "Competing [27] T. S. Baker, N. H. Olson, and S. D. Fuller, "Adding the hydrophobic and screened-coulomb interactions in hepatitis b virus capsid assembly," Biophys. J. 86 (2004) 3905 -- 3913. [10] W. K. Kegel and P. van der Schoot, "Physical regulation of the self-assembly of tobacco mosaic virus coat protein," Biophys. J. 91 (2006) 1501 -- 1512. [11] V. A. Belyi and M. Muthukumar, "Electrostatic origin of genome packing in viruses," Proc. Natl. Acad. Sci. USA 103 (2006) 17174 -- 17178. [12] C. L. Ting, J. Wu, and Z. G. Wang, "Thermodynamic basis for the genome to capsid charge relationship in electrostatically-driven viral encapsidation," Proc. Natl. Acad. Sci. USA 108 (2011) 16986 -- 16991. [13] S. Karlin and V. Brendel, "Charge configurations in viral proteins," Proc. Natl. Acad. Sci. USA 85 (1988) 9396 -- 9400. [14] B. Michen and T. Graule, "Isoeletric points in viruses," J. Appl. Microbiol. 109 (2010) 388 -- 397. [15] N. Steinmetz and M. Manchester, Viral Nanoparticles -- Tools for Materials Science and Biomedicine. Pan Stanford Publishing, 2011. [16] A. Siber, R. Zandi, and R. Podgornik, "Thermodynamics of nanospheres encapsulated in virus capsids," Phys. Rev. E 81 (2010) 051919. [17] W.-Y. Ching, 2011. UMKC Department of Physics, personal communication. [18] F. Pichierri, "Quantum proteomics," arXiv:1107.5853v1 third dimension to virus life cycles: three-dimensional reconstruction of icosahedral viruses from cryo-electron micrographs," Microbiol. Mol. Biol. Rev. 63 (1999) 862 -- 922. [28] R. V. Mannige and C. L. Brooks III, "Periodic table of virus capsids: implications for natural selection and design," PLoS ONE 5 (2010) e9423. [29] J. A. Speir and J. E. Johnson, "Virus particle structure: Nonenveloped viruses," in Encyclopedia of Virology, 5 vols., B. W. J. Mahy and M. H. V. V. Regenmortel, eds., pp. 380 -- 393. Elsevier, Oxford, 2008. [30] R. Zandi, D. Reguera, R. F. Bruinsma, W. M. Gelbart, and J. Rudnick, "Origin of icosahedral symmetry in viruses," Proc. Natl. Acad. Sci. USA 101 (2004) 15556 -- 15560. [31] A. Siber and R. Podgornik, "Stability of elastic icosadeltahedral shells under uniform external pressure: Application to viruses under osmotic pressure," Phys. Rev. E 79 (2009) 011919. [32] M. R. Gunner, J. Mao, Y. Song, and J. Kim, "Factors influencing the energetics of electron and proton transfers in proteins," Biochim. Biophys. Acta 1757 (2006) 942 -- 968. [33] D. G. Isom, B. R. Cannon, C. A. Castaneda, A. Robinson, and B. Garcia-Moreno E., "High tolerance for ionizable residues in the hydrophobic interior of proteins," Proc. Natl. Acad. Sci. USA 105 (2008) 17784 -- 17788. [34] D. G. Isom, C. A. Castaneda, B. R. Cannon, P. D. Velu, and B. Garcia-Moreno E., "Charges in the hydrophobic interior of proteins," Proc. Natl. Acad. Sci. USA 107 (2010) 16096 -- 16100. [35] T. Hu, R. Zhang, and B. I. Shklovskii, "Electrostatic theory of viral self-assembly," Physica A 387 (2008) 3059 -- 3064. [36] A. Siber and R. Podgornik, "Nonspecific interactions in spontaneous assembly of empty versus functional single-stranded RNA viruses," Phys. Rev. E 78 (2008) 051915. [37] G. A. Petsko and D. Ringe, Protein Structure and Function. New Science Press, 2004. [38] V. A. Parsegian and T. Zemb, "Hydration forces: Observations, explanations, expectations, questions," Curr. Opin. Colloid Interface Sci. 16 (2011) 618 -- 624. [39] J. Langlet, F. Gaboriaud, C. Gantzer, and J. F. L. Duval, "Impact of chemical and structural anisotropy on the electrophoretic mobility of spherical soft multilayer particles: the case of bacteriophage MS2," Biophys. J. 94 (2008) 3293 -- 3312. 10 [40] P. Pfeiffer, M. Herzog, and L. Hirth, "RNA viruses: stabilization of brome mosaic virus," Phil. Trans. R. Soc. Lond. B. 276 (1976) 99 -- 107. [41] C. E. Felder, J. Prilusky, I. Silman, and J. L. Sussman, "A server and database for dipole moments of proteins," Nucleic Acids Res. 35 (2007) W512 -- W521. [42] P. Ni, Z. Wang, X. Ma, N. C. Das, P. Sokol, W. Chiu, B. Dragnea, M. Hagan, and C. C. Kao, "An examination of the electrostatic interactions between the N-terminal tail of the brome mosaic virus coat protein and encapsidated RNAs," J. Mol. Biol. (2012) . in press. [43] Outer and inner surfaces, respectively. [44] The T = pseudo 3 icosahedral capsids do not obey the Caspar-Klug principle of quasi-equivalence because the basic unit is composed of three different (but morphologically similar) proteins. [45] In all such plots the following legend is used for different virus types: single-stranded genome (circles), double-stranded genome (squares), bacteriophages (diamonds; both ss and ds genome), and T = p3 ssRNA viruses (triangles).
1107.5816	1	1107	2011-07-28T20:06:56	Modeling the topology of protein interaction networks	[ "physics.bio-ph", "cond-mat.stat-mech", "q-bio.MN" ]	A major issue in biology is the understanding of the interactions between proteins. These interactions can be described by a network, where the proteins are modeled by nodes and the interactions by edges. The origin of these protein networks is not well understood yet. Here we present a two-step model, which generates clusters with the same topological properties as networks for protein-protein interactions, namely, the same degree distribution, cluster size distribution, clustering coefficient and shortest path length. The biological and model networks are not scale free but exhibit small world features. The model allows the fitting of different biological systems by tuning a single parameter.	physics.bio-ph	physics	Modeling the topology of protein interaction networks Christian M. Schneider,1, ∗ Lucilla de Arcangelis,1, 2 and Hans J. Herrmann1, 3 1Computational Physics, IfB, ETH Zurich, Schafmattstrasse 6, CH-8093 Zurich, Switzerland 2Department of Information Engineering and CNISM, Second University of Naples, I-81031 Aversa (CE), Italy 3Departamento de F´ısica, Universidade Federal do Cear´a, 60451-970 Fortaleza, Cear´a, Brazil (Dated: September 28, 2018) A major issue in biology is the understanding of the interactions between proteins. These interac- tions can be described by a network, where the proteins are modeled by nodes and the interactions by edges. The origin of these protein networks is not well understood yet. Here we present a two-step model, which generates clusters with the same topological properties as networks for protein-protein interactions, namely, the same degree distribution, cluster size distribution, clustering coefficient and shortest path length. The biological and model networks are not scale free but exhibit small world features. The model allows the fitting of different biological systems by tuning a single parameter. PACS numbers: 64.60.aq, 89.75.Fb, 87.15.km, 87.23.Kg I. INTRODUCTION The study of complex networks in biology promises new fruitful insights about the functionality of genes and proteins [1 -- 5]. Since the interactions between proteins determine their functionality, the properties and the ori- gin of the interaction networks have attracted much at- tention [6 -- 8]. They consist of protein complexes, which are connected in a large, constantly evolving, cluster [9]. The analysis of hundreds of protein complexes has es- tablished that some of the relevant structural features are the contact area, the shape of the interfaces, the complementarity of surface shapes, and the interaction- mediating forces. Although not all interactions have been discovered yet, numerous studies have been performed and many data sets are available [10 -- 16]. One important outcome of these studies is that most protein networks show a wide range of variability in the number of nodes and edges and the average connectivity degree (Table I). They appear not to be scale-free, namely, the distribu- tion of connectivity degrees is not a power law although it stretches over a significant number of orders of magni- tude. Moreover, they do not consist of one single cluster but in addition to a large component many small clusters of interactions are also detected. These results suggest that the specific features of bio- logical networks express different underlying mechanisms than do other networks, like social interaction networks or the internet [17, 18]. it has been specu- lated that gene duplication is the dominant evolution- ary force in shaping biological networks [10, 19]. Con- versely, non-biological networks are typically driven by additive growth processes [18] such as, for instance, pref- erential attachment [20], but many other mechanisms like rewiring [21], aging [22], or fitness [23] have been inves- tigated. However, none of these models can reproduce In fact, ∗ [email protected] N 9050 8212 M (cid:104)k(cid:105) α Organism 3582 12045 6.7 1.50 Nocardia farcinica (NF) Bradyrhizobium japonicum (BJ) 4883 19261 7.9 1.00 6.7 0.75 Aeromonas hydrophila (AH) 2708 3373 Citrobacter koseri (CK) 4.9 0.50 3204 13091 8.2 0.75 Escherichia coli (EC) Pseudomonas aeruginosa (PA) 3794 14252 7.5 0.75 4.9 0.75 3373 Serratia proteamaculans (SP) 2512 Vibrio cholerae (VC) 6.9 1.00 4771 54607 22.9 1.75 Saccharomyces Cerevisae (SC) Homo Sapiens (HS) 11102 136930 24.7 1.75 8187 8612 TABLE I. List of organisms from STRING 8.2 data set [26] investigated here. Columns report the number of nodes N , the number of edges M , the average degree (cid:104)k(cid:105), and the value of the model parameter α used here. Edges between pairs of proteins represent an 80% reliability of protein interaction. NF belongs to Acatinobacteria, BJ to Alphaproteobacteria, and all other bacteria belong to the Gammaproteobacteria class. the full topology of protein networks like, for instance, the emergence of isolated clusters found in real biological networks (Fig. 1). Here we propose a different model, which reproduces many topology properties of protein interaction net- works. We do not consider the details of the biochemical mechanisms at the basis of each interaction, nor classify proteins in classes as in other approaches [3, 9, 24]. Con- versely, we follow a simple probabilistic approach. II. THE MODEL The procedure starts with a fully connected network of N sites and M = N (N − 1)/2 edges. The number of nodes is equal to the number of nodes of the biological network considered, N = Nbio. The evolution is per- formed according to the following steps: (i) Choose at random a node i. (ii) Choose at random an edge eij and remove it with a 1 1 0 2 l u J 8 2 ] h p - o i b . s c i s y h p [ 1 v 6 1 8 5 . 7 0 1 1 : v i X r a 2 FIG. 1. (Color online) The topology of (left) the AH network and (right) the model network with α = 0.75. The largest clusters are drawn in the center and the smaller clusters on the border. The largest clusters are drawn in the center and the smaller clusters on the border. The color code and the size (from small to large) represent the degree of each site on a logarithmic scale: blue k < 3, green 3 ≤ k < 5, cyan 5 ≤ k < 10, yellow 10 ≤ k < 21, red 21 ≤ k < 43 and purple k ≥ 43[31]. probability pi,j related to the degree kj of the neighbor j of node i: (cid:40) k−α 0 j kj > 1 otherwise (1) pj Ni pi,j = and with Ni the normalization Ni =(cid:80)ki with pj = l=1 pl. α > 0 is the only free parameter of the model and controls the rel- ative robustness of edges belonging to highly connected nodes with respect to edges of sites with low k. This rule implies that "the poor get poorer. The case α = 0 im- plies that all sites have the same probability to lose edges and the process reduces to a random depletion. (iii) Repeat this procedure for another node i until the number of edges M in the network equals the number of nodes N . (iv) Choose at random two nodes i and j. Add an edge between these nodes with probability pi,j = [Nc(i, j)]2/(kikj), (2) where Nc(i, j) is the number of neighbors that nodes i and j have in common. This step supposes that, if two given nodes are able to interact with the same nodes, they have a high probability to interact with each other. (v) Repeat this procedure for another random pair of nodes i and j until the number of edges M in the net- work equals the number of edges of the modeled biologi- cal network Mbio. These rules are based on the assumption that the evolu- tion is controlled by two basic mechanisms: (i) preferen- tial depletion: the lower the node degree, the lower the probability to maintain interactions [25]; (ii) similarity: the more common neighbors two nodes share, the higher is the probability to have an interaction. The first mechanism is important for the emergence of isolated clusters and a maximal degree, while the second one is necessary to generate networks with a high clus- tering coefficient and assortativity. It is interesting to notice that the implementation of the depletion mech- anism alone generates scale free networks and does not FIG. 2. (Color online) The degree distribution p(k) for AH (circles), BJ (triangles), CK (stars) and HS (squares) and their corresponding model networks (lines) with α obtained from Table I. Star, triangle, and square data sets are shifted vertically by factors of 0.5, 2, and 5, respectively, for better visibility. reproduce the topology of protein-protein interacion net- works [25]. III. RESULTS The biological networks are obtained from the STRING 8.2 data set [26], where a combined score of 80% is used to decide whether two proteins interact. We tested our algorithm on the ten different biological net- works listed in Table I. For each organism we determine a value of the parameter α which provides a good fit (Ta- ble I) for the degree distribution. All results for model networks are averages over 100 independent runs for bac- teria and 10 runs for the other two networks. In Fig. 1 we show an example for a biological network and the corre- sponding model network, with the same number of nodes and edges and α = 0.75. Both networks have one large 100101102k10-410-310-210-1100p(k)3 FIG. 3. (Color online) Frequency f (S) of finding a cluster with a given number of nodes SN for CK (circles), EC (trian- gles), and VC (stars) and with a given number of edges SM (inset) for AH (circles), BJ (triangles), and VC (stars), and their corresponding model networks (lines). Top and bottom data sets are shifted vertically by one decade, upward and downward. FIG. 4. (Color online) Visualization of small-world properties of biological networks. Average shortest path length lk of sites of degree k versus k for AH (circles), EC (triangles), and VC (stars) and clustering coefficient Ck(inset) of sites of degree k versus k for AH (circles), BJ (triangles), and EC (stars), and their corresponding numerical networks (lines). Top and bottom data sets for lk are shifted vertically by factors of 2 and 0.5. cluster with dangling ends, shown in the center of both graphs. Moreover, both networks have a large number of small clusters, placed on the border of each network. For both networks highly connected nodes are placed in the largest cluster, whereas small clusters are made of low-degree nodes. Since the topology is not a quanti- fied differentiation property to decide whether two net- works are similar, we calculate some fundamental prop- erties characterizing the connectivity and the structure of the two networks. The model has by construction the same numbers of nodes N and of edges M as the bio- logical one and therefore the average degrees per node (cid:104)k(cid:105) are exactly the same. To provide more information on the connectivity level of the two networks, we mea- sure first the degree distribution. In Fig. 2 we show the degree distribution of different biological networks and their numerical counterparts. The biological networks are not scale-free and the numerical data reproduce the data very well by tuning the parameter α. We observe that the value of the exponent α controls the maximum degree and the exponential cutoff of the distribution. For α = 0 the exponential cutoff is at k = 1 and therefore the degree distribution a pure exponential. By increas- ing α, the range of the initial regime increases and the exponential cutoff moves toward larger k values. To tune the parameter, we compare the tail of the degree distri- bution for different α values and choose the one which fits best. In the procedure the smallest allowed degree is k = 1; the model then generates one large network and many small clusters, as in biological systems. We char- acterize this complex structure by evaluating the cluster size distribution. The cluster size is defined in terms of both the number of nodes, SN , and the number of edges, SM , belonging to the cluster. Figure 3 shows the cluster size distributions for different biological and numerical networks. Both distributions exhibit a regime consistent with a power law with an exponent (cid:39) −4.4, for the size Model 7398 N M N M Smax 8390 Biological Cmax lmax Smax Smax System Smax Cmax lmax 1996 0.5 0.65 5.08 1785 7986 3551 18043 0.61 4.85 0.5 2807 16453 2471 0.5 0.59 5.81 6609 2354 12250 0.71 4.92 0.5 2677 12620 2569 11257 0.49 4.33 0.5 1683 9435 2778 13263 0.68 5.09 0.5 2613 13024 0.5 2484 0.54 5.33 1778 5911 1842 0.5 0.60 4.74 1717 7726 3351 53012 0.81 3.89 4711 54570 0.4 10890 136799 0.4 7864 133576 0.66 4.15 AH BJ CK 2032 EC NF PA SP VC SC HS 5.9 6.2 8.1 6.2 6.8 7.3 6.4 5.6 3.7 3.9 7468 8028 TABLE II. Properties of the largest connected cluster for the biological networks and their model counterparts: the num- ber of nodes Smax M , the average clustering coefficient Cmax, and the shortest path length lmax. The error bars are 1%, 2%, 4% and 2%, respectively. N , the number of edges Smax in terms of sites, and an exponent (cid:39) −2.7 for the size in terms of edges. The faster decay found for the first distri- bution suggests that the structure is highly clustered, as will be confirmed later. Furthermore, in most cases the size of the largest connected cluster is comparable (Table II). Interestingly, numerical data for f (SM ) also repro- duce the fluctuations at small sizes observed in biological data. These are not the effect of statistical noise, but measure the relative weight of the population of clusters with few edges, whose patterns can be simply identified. The level of connectivity in the system is measured by the average clustering coefficient of nodes of degree k and the average shortest path between nodes of degree k (Fig. 4). Both quantities vary smoothly with k for biological and numerical data. Both the model and biological networks are highly clustered. Moreover, biological data show that the average shortest path length slowly increases with k for low connectivity degrees and then reaches a fairly stable value for a wide range of k, in agreement with 100101102103SM100102104f(SM)100101SN100102f(SN)10010110210-1100Ck100101102k10-1100101lk4 the topological properties exhibit similar behavior. It is also possible to infer the α value by analysis of only 75% of the entire protein data set. From the statistical point of view our model seems to be a good candidate for modeling the topology of protein interaction networks. However, the ingredients we im- plement are not well established for protein interaction networks, although they are present in other biological systems. Stem cells are an example of depletion. When a stem cell specializes and becomes a particular cell (a red blood cell, a muscle cell, or even a neuron) it loses the ability to interact with cells from other types [28]. Moreover, the similarity concept can be interpreted as the establishment of interacting protein families [29, 30]. IV. DISCUSSION In conclusion, we present a statistical model, which reproduces surprisingly well many topological properties of protein interaction networks. The model is based on a twofold mechanism for evolution, namely, preferential depletion and similarity. By fitting a single parameter, we are able to generate networks that reproduce protein interaction networks for different bacteria as well as Sac- charomyces cerevisae and Homo sapiens. We wish to stress that not only do the largest clusters exhibit the same connectivity properties but also the small-cluster distributions show very good agreement between biolog- ical and model data. The clustering coefficient and the average path length suggest that highly connected nodes are placed in the largest cluster and preferentially con- nected to nodes with high degree. The systematic anal- ysis of the network structure for a number of biological systems indicates that protein interaction networks are not scale-free but rather exhibit small-world properties. Further research should be performed to better under- stand the origin of this dual mechanism in protein inter- action networks. We acknowledge financial support from the ETH Com- petence Center "Coping with Crises in Complex Socio- Economic Systems (CCSS) through ETH Research Grant No. CH1-01-08-2 and FUNCAP. FIG. 5. (Color online) Average nearest neighbor degree knn(k) of nodes of degree k versus k for AH (circles), PA (triangles) and VC (stars) and their corresponding model net- works (lines). Top and bottom data sets are shifted vertically by a factor of 2 and 0.5. N numerical data. This result suggests that the model net- work reproduces not only the distribution of connectivity degrees, but also the relative position in the network of nodes with the same k value. Moreover, the high value of the clustering coefficient and the small shortest path length suggest that biological and model networks have small world properties [27]. Finally the average cluster- ing coefficient Cmax and the average shortest path length lmax evaluated for the largest cluster show a very weak dependence on the cluster size Smax and exhibit (Table II) a good agreement between biological and model data. A further confirmation that our model captures the struc- ture of the network at both a global and local level is given by the evaluation of the average degree of the neigh- bors of a site of degree k, knn(k) (Fig. 5). This quantity increases with the node degree as k0.67±0.02 for biological networks, and k0.61±0.01 for numerical data. This scaling behavior suggests that highly connected nodes tend to be connected with each other. Finally we notice that, for each system, topological prop- erties are very stable with respect to changes in the fitting parameter and the calculation of the similarity. Even if the fitting value of α is changed by ±0.25 or the similar- ity rule is modified [ e.g., using pnew i,j = Nc(i, j)/(ki +kj)], [1] L.H. Hartwell et al., Nature 402, C42 (1999). [2] H. Jeong et al., Nature 407, 651 (2000). [3] P. Uetz et al., Nature 403, 623 (2000). [4] A.C. Gavin et al., Nature 415, 141 (2002). [5] A.L. Barab´asi and Z.N. Oltvai, Nature Rev. Genet. 5, 101 (2004). [6] T. Ideker, T. Galitski and L. Hood, Annu. Rev. Genomics Hum. Genet. 2, 343 (2001). [7] L. Hood, Mech. Ageing Dev. 124, 9 (2003). [8] Editorial, Nat. Cell Biol. 8, 1179 (2006). [9] B. Schwikowski, P. Uetz and S. Fields, Nat. Biotechnol- ogy, 18, 1257 (2000). [10] R. Friedman and A. Hughes, Genome Res. 11, 373 (2002). [11] P.T. Spellman et al., Mol. Biol. Cell 9, 3273 (1998). [12] J. DeRisi, V. Iyer and P.Brown, Science 278, 680 (1997). [13] Z. Gu et al., Mol. Biol. Evol. 19, 256 (2002). [14] http://interactome.dfci.harvard.edu/. [15] A.H.Y. Tang et al., Science 303, 808 (2004). [16] L. Hakes, D.L. Robertson and S.G. Oliver, BMC Ge- nomics 6, 131 (2005). [17] R. Albert, H. Jeong and A.L. Barab´asi, Nature (London) 401, 130 (1999). 100101102k100101102knn(k)[18] R. Albert and A.L. Barab´asi, Rev. Mod. Phys. 74, 47 [24] T. Ito et al., Proc. Natl. Acad. Sci. U.S.A. 98, 4569 (2002). (2001). [19] A. Bhan, D. Galas, and D.G. Dewey, Bioinformatics 18, [25] C.M. Schneider, L. de Arcangelis and H.J. Herrmann, 1486 (2002). Europhys. Lett. 5, 16005 (2011). [20] A.L. Barab´asi and R. Albert, Science 286, 509 (1999). [21] P.L. Krapivsky and S. Redner, Phys. Rev. E 63, 066123 [26] http://string-db.org [27] D.J. Watts and S.H. Strogatz, Nature (London)393, 440 (2001). (1998). [22] L.A.N. Amaral et al., Proc. Natl. Acad. Sci. U.S.A. 97, 11 149 (2000). [28] F.H. Gage, Science 287, 1433 (2000). [29] J.Park, M. Lappe and S.A. Teichmann J. Mol. Biol. 307, [23] G. Bianconi and A.L. Barab´asi, Europhys. Lett 54, 436 929 (2001). (2001). [30] S.A. Teichmann, J. Mol. Biol. 324, 399 (2002). [31] V. Batagelj and A. Mrvar, Connections 21(2), 47 (1998). 5
1506.04835	2	1506	2015-06-17T21:38:29	Application of the reciprocity theorem to EEG inversion and optimization of EEG-driven transcranial current stimulation (tCS, including tDCS, tACS, tRNS)	[ "physics.bio-ph", "physics.med-ph" ]	Multichannel transcranial current stimulation (tCS) systems offer the possibility of EEG-guided optimized, non-invasive brain stimulation. In this brief technical note I explain how it is possible to use tCS electric field realistic brain model to create a forward "lead-field" matrix and, from that, an EEG inverter for cortical mapping. Starting from EEG I show how to generate 2D cortical surface dipole fields that could produce the observed EEG electrode voltages. The main tool is the reciprocity theorem derived by Helmholtz. The application of reciprocity for the generation of a forward mapping matrix (lead field matrix as is sometimes known) is well known [Rush and Driscoll, 1969], but here we will use it in combination with the realistic head models of [Miranda et al 2013] to provide cortical mapping solutions compatible with realistic head model tCS optimization. I also provide a generalization of the reciprocity theorem [Helmholtz 1853] to the case of multiple electrode contact points and dipole sources, and discuss its uses in non-invasive brain stimulation based on EEG. This, as far as I know, is a novel result. Applications are discussed.	physics.bio-ph	physics	Neuroelectrics Barcelona SL - TN0008 1 Application of the reciprocity theorem to EEG inversion and optimization of EEG-driven tCS (tDCS, tACS and tRNS) Author: Giulio Ruffini ([email protected]) Date: June 10 2015 Summary Multichannel transcranial current stimulation systems offer the possi- bility of EEG-guided optimized brain stimulation. In this brief technical note I explain how it is possible to use transcranial current stimulation tCS (which includes tDCS, tACS and tRNS among others1) electric field real- istic brain models to create a forward "lead-field" matrix and, from that, an EEG inverter for cortical mapping. Starting from EEG we show how to generate 2D cortical surface dipole fields that could generate the observed EEG electrode voltages. The main tool is the reciprocity theorem derived by Helmholtz. The application of reciprocity for the generation of a forward mapping matrix (lead field matrix as is sometimes known) is well known [Rush19692], but here we will use it in combination with the realistic head models of [Miranda20133] to provide cortical mapping solutions compatible with realistic head model tCS optimization. I also provide a generalization of the reciprocity theorem [Helmholtz18534] to the case of multiple electrode contact points and EEG dipole sources, and discuss its uses in non-invasive brain stimulation based on EEG. This can be used to guide for optimization of transcranial current stimulation with multiple channels based on EEG. This, as far as I know, is a novel result. 1 Ruffini G, Wendling F, Merlet I, Molaee-Ardekani B, Mekonnen A, Salvador R, Soria- Frisch A, Grau C, Dunne S, Miranda PC, Transcranial current brain stimulation (tCS): models and technologies, IEEE Trans Neural Syst Rehabil Eng. 2013 May;21(3):333-45. 2Rush and Driscoll, 1969, IEEE Trans. Biom. Eng. 3Miranda et al., NeuroImage 70 (2013) 4858 4H. Helmholtz, "Uber einige Gesetz der Vertheilung elektrischer Strdme in korperlichen Leitern, mit Anwendung auf die thierisch- elektrischen Versuche," Ann. Phys. Chem., ser. 3, vol. 29, pp. 211-233 and 353-377, 1853. 5 1 0 2 n u J 7 1 ] h p - o i b . s c i s y h p [ 2 v 5 3 8 4 0 . 6 0 5 1 : v i X r a Neuroelectrics Barcelona SL - TN0008 Contents 1 Introduction 2 Reciprocity 2.1 Inversion with EEG data as constraints . . . . . . . . . . . . 2.2 Inversion with standard spatial regularisation . . . . . . . . . 3 Application of reciprocity for EEG-guided multichannel cur- rent brain stimulation (MtCS) 2 3 4 6 7 9 3.1 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11 3.2 Spatial resolution of tCS . . . . . . . . . . . . . . . . . . . . . 11 Neuroelectrics Barcelona SL - TN0008 3 1 Introduction The motivation of this note is to develop ways to derive targeted mul- tichannel transcranial current stimulation (tCS) protocols from EEG data. Since both EEG and tCS are mostly cortically relevant, we would like to de- velop ways to compare the generated electric fields of tCS with EEG dipole cortical field sources. There are now many groups developing realistic, FEM-based models of current propagation in the human head, solving Poisson's equation. In the Starstim software5 we have now integrated such realistic head modeling of the generated electric fields by non-invasive multichannel brain stimulation, based on the work of [Miranda2013]. How can we use such current propagation (electric field) models in to infer an EEG dipole forward model? They are certainly related by the reciprocity theorem and by the (not-unrelated) fact that are both governed by Poisson's equation. Once we derive a realistic forward model we can develop inverse modeling approaches, to go from scalp to source space. In this note I concentrate first on the problem of cortical mapping, i.e., finding a dipole distribution on the cortical surface that will generate the EEG pattern. Although the method described can be used to create volume dipole vector field forward operators, given the fact that in our own current multichannel systems the number of electrodes is limited to 32, and the issues with volume inverse modeling (a very ill-posed problem) it is reasonable to work with more constrained models: we fix the locations of the dipoles to the cortical surface and their orientation normal to it, reflecting our current understanding that EEG sources are mostly cortical and normally aligned to the surface (pyramidal neuron populations). Surface mesh points in our models on the cortical surface are of the order of 35,000. The transformation from voltage space to dipole space is linear. The main task is to find the linear transformation (the "lead field" matrix). What follows is based on the reciprocity theorem. Once we derive a cortically mapper, tCS montages can be optimized to target the desired EEG features in cortical space. Finally, I provide also an extension of this theorem to include the case of multiple contact points and current sources. This can be used to guide for optimization of transcranial current stimulation with multiple channels based on EEG. 5 http://www.neuroelectrics.com Neuroelectrics Barcelona SL - TN0008 4 2 Reciprocity The reciprocity theorem states that there is a relationship between the following variables: • V (1) ab and (cid:126)J (1): V (1) is the observed potential difference between two points in the scalp due to a lone dipole current source in the brain, (cid:126)J (1) (in a volume δV ) ab • I (2) ab and (cid:126)J (2): I (2) is an imposed, stimulation, current between these two points in the scalp and the resulting current density in the brain, (cid:126)J (2) ab The reciprocity relation is σV (1) ab I (2) ab = − (cid:126)J (1) · (cid:126)J (2)δV or V (1) ab I (2) ab = − (cid:126)J (1) · (cid:126)E(2)δV Dropping the reference and other indices, we can just write: Va Ia = − (cid:126)J(x) · (cid:126)E(x)δV (2.1) (see the figure below). This is nicely explained by [Plonsey19636] paper and later book. With regards to units, if we input uV and we work with 1 mA current as a reference, the output will be in nA m. The problem we wish to address is: what is the voltage Vab at a point b w.r.t. point a due to a lone dipole (cid:126)J (1)? The answer is: For more dipoles the superposition principle applies, (2) δV V (1) I (2) ab ab = − (cid:126)J (1) · (cid:126)E(2)δV ab = −(cid:88) (cid:126)Jn (1) · (cid:126)En I (2) ab V (1) where n refers to voxel location ID. n 6Plonsey R., (1963) Reciprocity applied to volume conductors and the EEG. IEEE Trans. Biomed. Electron. 10:912 Neuroelectrics Barcelona SL - TN0008 5 Now, in general, the volume corresponding to each dipole will not be the same. A more refined equation is ab = −(cid:88) V (1) (1) (cid:126)En (2) · (cid:126)Jn I (2) ab δVn. n Now, we have computed the solutions to all combinations of electrodes in our EEG cap with respect to Cz. Using this expression we can thus compute the voltage of any channel location s w.r.t Cz. We can now define the algorithm for the forward model. 1. Put dipole where you want (a 3D vector). 2. Dot product with the electric field (cid:126)E0 n produced by a the reciprocal montage at a and b using the realistic head model with a unitary current (I=1). (2) 3. Multiply by volume of dipole. You now have the potential at site a w.r.t b (Vab = Va − Vb). ab = −(cid:88) V (1) n (2) · (cid:126)Jn (cid:126)E0 n (1) δVn. In a triangular mesh, each mesh triangle can be associated to its defining three mesh points. So if we define the area of a mesh point to be the area of all the triangles it touches, and then this area over all mesh points, each triangle will show up three times. Thus, to account for dipole area/volume, we can properly assign an area to each mesh m point by the formula (cid:88) Am = 1 3 triangles t touched by meshpoint m ATm t Do this for every single component of the dipole sources you want your model to represent (e.g., all those in the cortex) ... Algorithmically, we have as many basis functions as electrodes minus 1, defining currents from Cz to other points. We need to loop around them, and then loop over dipole space (3N) to create the matrix. The output will be a 26 x 3N matrix, the famous "lead" or forward-mapping K matrix. An option for use here is to focus on the cortical surface and only on normally oriented dipoles. This will make the problem much more amenable. Interesting, only the normal E fields generated by stimulation are relevant according to first order models of neuron electric field interaction. Neuroelectrics Barcelona SL - TN0008 6 2.1 Inversion with EEG data as constraints Once we have the forward mapping K matrix, we need to smooth the prob- lem for inversion. An approach I've used in the past is to minimise the "cur- vature" of the solution subject to the data constraints. See [Ruffini20027]. Here we do the same, with a different implementation of the Laplacian operator which can be used on a mesh. A simpler approach we can employ here is to use the (over the surface) distance matrix across cortical surface points. The matrix to use for regu- larisation would be The equation RJ = 0 says the "the value of J at a point is equal to the average of points nearby weighted by distance". We can start with q = 2, R(x, x(cid:48)) = δ(x, x(cid:48)) − d(x, x(cid:48))−q x(cid:54)=x(cid:48) d(x, x(cid:48))−q R(x, x(cid:48)) = δ(x, x(cid:48)) − d(x, x(cid:48))−2 x(cid:54)=x(cid:48) d(x, x(cid:48))−2 (cid:80) (cid:80) More generally, we can write R(x, x(cid:48)) = δ(x, x(cid:48)) − (cid:80) f (d(x, x(cid:48))) x(cid:54)=x(cid:48) f (d(x, x(cid:48))) E.g., we can use f (d(x, x(cid:48))) = exp[−d(x, x(cid:48))/d0] We can follow the work in [Ruffini2002]. We aim to minimise "curvature" here defined by R subject to the constraints that the EEG data impose (V = KJ), which are added using a family of Lagrange multipliers: χ = 1 2 J T RT RJ + λT · (V − KJ) and the solution to this problem is given by J =(cid:2)RT R + KT K − KT (KKT )−1KRT R(cid:3)−1 KT V as explained in the aforementioned technical note on arxiv. 7 "Spherical Harmonics Interpolation, Computation of Laplacians and Gauge Theory" at http://arxiv.org/abs/physics/0206007 Neuroelectrics Barcelona SL - TN0008 7 2.2 Inversion with standard spatial regularisation A related approach is to start from a functional in which both data and curvature constraints appear treated equally. Mathematically this means λ is input by hand and not solved for. The solution is J = [KT K + αRT R]−1KT V where the parameter α is input by hand. Another option is Tikhonov regularisation (using I instead of R), J = [KT K + αI]−1KT V where the parameter α is input by hand. Implicit referencing to Cz is used. Figure 1 -- Top: Reciprocity relation symbols and actors Neuroelectrics Barcelona SL - TN0008 8 Figure 2 -- Sample tomographic solutions with 26 and 38 electrodes. On the top, source dipole patch and reconstruction. On the bottom, reconstructed EEG from a sample subject (20 electrodes). Red colors represent outgoing sources, blue ones ingoing sources. Neuroelectrics Barcelona SL - TN0008 9 3 Application of reciprocity for EEG-guided multichannel current brain stimulation (MtCS) Here we discuss how we can derive scalp potentials from EEG to design optimal stimulation protocols. Using our tomographic approach, the EEG data can be inverted to source space to define a target map, and then optimize as discussed [Ruffini20148]. That is, from EEG we derive a cortical (or volume) dipole field map, and then we can specify that the generated electric fields by MtCS align or anti-align with them with specified strengths and so on. The advantage of this approach is that it allows for flexible specification of targets and of weight maps. Spontaneous EEG or ERP data can be pre-filtered to bands of interest or treated in other ways (ICA, etc). Here we discuss a simpler approach (although a less flexible one as well). Consider the theorem for two dipole sources measured again at a single scalp point a: Va1 Ia + Va2 Ia = − (cid:126)J(x1) · (cid:126)E(x2)δV + − (cid:126)J(x2) · (cid:126)E(x2)δV (3.1) but this is just the resulting potential, of course, Va Ia. Hence, it is clear that for arbitrary sources, we can generalize the reciprocity theorem a bit, (cid:90) Va Ia = − dx (cid:126)J(x) · (cid:126)E(x) (3.2) We can see where this is going ... but let us get there faster. Using the reciprocity relations we can use the lead matrix to relate a dipole source distribution in the brain to scalp potentials, Va = Kax J(x) with implicit summation over repeated indices (vector indices on densities dropped for simplicity, also summed). The reference point for currents and potentials b has been dropped for notational simplicity. Here we recall that K = − Eax, is the lead field matrix associated to unitary currents, mapping interior to scalp space (thanks to reciprocity). Hence (summing over x implicit), Va = −Eax J(x) 8Ruffini et al., NeuroImage 89 (2014) 216225 Neuroelectrics Barcelona SL - TN0008 10 Let's multiply (and implicitly sum over repeated indices using the Einstein convention) both sides by a vector of injected currents Ia (meaning Iab): Ia Va = −Ia Eax J(x) Now, if we inject currents Ia, the generated electric fields are E(x) = Eax Ia Hence, Ia Va = − E(x) · J(x) or, with the explicit summations written out (cid:88) a (cid:90) (cid:90) Ia Va = − E(x) · J(x)dx (3.3) Finally, we can imagine a future current stimulation system which is capable of controlling scalp current density. The corresponding equation would be (cid:90) Va dA · Ja = − E(x) · J(x)dx (3.4) A V This rather elegant expression says that if you want generated electric fields and EEG sources to be correlated, currents and potentials have to be anti-correlated. This gives a simple way to determine optimal stimulation currents given scalp potential. Make currents and potentials to be as "par- allel" as possible, given the constraints. E.g., maximize IaVa subject to some constraints on maximal current, current sum, total injected current, etc. We note that this will not optimize things like the electric field magnitude at the sources. Rather the specific component of the electric field parallel to the sources. A limitation of this model-free optimization is that we will not really know what size electric fields we are generating, cannot add a weight map to work with weighted correlation, etc. But still, give some constraints on currents it provides a recipe to optimize currents to EEG sources. This can be especially useful in close-loop applications. Neuroelectrics Barcelona SL - TN0008 11 3.1 Applications The reciprocity theorem above can be used for various applications, includ- ing • Online optimization of MtCS from EEG (model-blind optimization based on EEG data) • Closed-loop applications from EEG: monitor EEG, create stimulation waveform (e.g., of the form I=c V or other optimizations as discussed above) to amplify or reduce EEG • Experiments involving "Playing back EEG using MtCS currents (I ? V), which may actually make some sense based on the above equation. • Theoretical analysis purposes (e.g., EEG/tCS electrode density needed, see next section) 3.2 Spatial resolution of tCS We can also use this equation to study the issue of how many electrodes (density) we need for stimulation. If we assume that the sources lie in a 2D surface (cortical mapping), for example, it is easy to argue that the map from sources to EEG is 1-1, and that, vice-versa, the map from scalp currents to cortical space E-fields is 1-1. Thus, in principle, if we could measure the full 2D surface with EEG (continuum EEG) we could invert the problem for sources, or if we could stimulate over all the entire scalp surface (continuum stimulation), we could reproduce any cortical target. Then, the needed number of EEG or stimulation electrodes would be transferred to the question of what the spatial scale of cortical sources or target is. In other words, the density of EEG or tCS electrodes needed for full inversion or target in this simplified model depends on the spatial scale (autocorrelation length) of the dipole sources on the cortex. We are ignoring here the issue of the finite size of electrodes, which act as a spatial filter. That is, we assume infinitely small electrodes.
1202.4619	1	1202	2012-02-21T12:22:00	Polarization Sensitive Optical Coherence Tomography for Blood Glucose Monitoring in Human Subjects	[ "physics.bio-ph", "physics.ins-det", "physics.med-ph", "physics.optics" ]	A device based on Polarization sensitive optical coherence tomography is developed to monitor blood glucose levels in human subjects. The device was initially tested with tissue phantom. The measurements with human subjects for various glucose concentration levels are found to be linearly dependent on the degree of circular polarization obtainable from the PS-OCT.	physics.bio-ph	physics	Polarization Sensitive Optical Coherence Tomography for Blood Glucose Monitoring in Human Sub jects Jitendra Solanki and Om Prakash Choudhary Department of Applied Physics, Shri G S Institute of Technology & Science, Indore - 452 003 India. P. Sen Laser Bhawan, School of Physics, Devi Ahilya University, Khandwa Road, Indore - 452 007 India. J. T. Andrews∗ Department of Applied Physics, Shri G S Institute of Technology & Science, Indore - 452 003 India. (Dated: Received: date / Accepted:date) Abstract A device based on Polarization sensitive optical coherence tomography is developed to monitor blood glucose levels in human sub jects. The device was initially tested with tissue phantom. The measurements with human sub jects for various glucose concentration levels are found to be linearly dependent on the degree of circular polarization obtainable from the PS-OCT. Keywords: Circularly polarized light, chiral glucose, scattering coefficient, tissue phantom, and human sub jects. ∗ [email protected] 1 I. INTRODUCTION The optically active nature of glucose inspires us to use polarization state of light as key for blood glucose monitoring. In an optically active medium the velocities of right circularly polarized (RCP) and left circularly polarized (LCP) lights differ. Consequently, such medium has the ability to rotate the plane of polarization of the resultant linearly polarized (LP) light. Larger glucose concentration is expected to yield larger polarization rotation. Despite the wealth of different properties that can be probed with polarized light, the turbid nature of biological samples makes them nontrivial. For example multiple light scattering in turbid media can hinder measurement and data interpretation. Over the years, many attempts were made to overcome these complexities involved in the measurement of polarization properties of scattered light from biological tissues. Several techniques are proposed and adopted to interpret the data obtained from turbid media. According to Ghosh et al. [1], the study of polarization properties of biological tissue using light scattering carries a wealth of morphological and functional information and has potential biomedical importance. They discussed comprehensive turbid polarimetry plat- form consisting of forward Monte Carlo modeling and inverse polarization decomposition analysis. Hee et al. [2], used optical coherence domain reflectometer for characterizing phase retardation between orthogonal linear polarization modes at each reflection point in a birefringent sample. They used the reflectometer to characterize attenuation and bire- fringence of tunic media in calf coronary artery. Kohl et al. [3], showed that presence of glucose dissolved in an aqueous solution increases refractive index of the solution and also influences scattering properties of any particles suspended within it. They demonstrated the effect of the glucose on light transport in highly scattering tissue simulating phantoms using diffusion theory. Their results pave way to use this effect for noninvasive glucose monitoring in diabetic patients. The next step in this direction was to correlate scattering coefficient with blood glucose concentration. In 1997 Bruulsema et al. [4], examined reduced scattering coefficient of tissue in response to steep changes in blood glucose levels of diabetic volunteers. They observed correlation between steep changes in blood glucose concentration and tissue reduced scattering coefficients in 30 out of 41 sub jects. Wang et al. [5] reported existence of linear relationship between angles of rotation in Muller matrix elements and concentration of glucose. 2 In a pilot study of specificity of noninvasive blood glucose monitoring using Optical Co- herence Tomography (OCT) technique, Larin et al. [6] report that several osmolytes may change refractive index mismatch between interstitial fluid and scattering centers in tissue. They found that the effect of glucose is approximately two to three orders higher and con- cluded that OCT technique may provide blood glucose monitoring with sufficient specificity under normal physiological conditions. They also used phase-sensitive optical low-coherence reflectometry (PS-OLCR) for measurement of analyte concentration [7] and demonstrated a high degree of sensitivity and accuracy of phase measurements of analyte concentrations in their experiment. Cameron and Li [8] examined the polarization sensitivity of glucose concentration in physiological range for highly scattering biological media and concluded that OCT based glucose monitoring is approaching standard invasive and minimally inva- sive techniques. Lee et al. [9] measured changes of degree of circular polarization (DOCP) in intralipid suspensions and human cervical tissue using polarization sensitive OCT (PS-OCT) system and proposed that this technique may provide a unique diagnostic tool. Motivated by the studies reported above, we have used PS-OCT technique for analyzing blood glucose concentration in tissue phantom and noninvasive blood glucose in human sub jects. We report the measurement of change in degree of polarization as a function of glucose concentration in tissue phantom as well as in human sub jects. The experimental observations have been explained using Mie scattering theory by incorporating the refractive index dependence on the state of polarization in the optically active medium. II. EXPERIMENT The optical coherence tomography setup developed by us is custom designed for the measurements with tissue phantom, blood samples as well as for non-invasive measurements with human sub jects. A schematic of the Michelson interferometer type PS-OCT is shown in Figure 1. A low coherence broad band light source viz; Hamamatsu Superluminescent diode (SLD) having a center wavelength of 835 nm and bandwidth 50 nm is used. The sample arm of the device has the option to accommodate samples of tissue phantom or blood or any one finger of the sub ject for non-invasive measurements. Low coherence beam of light emitted from SLD was passed through the linear polarizer and a quarter wave plate at 45◦ to the polarizer axis as shown in the figure to provide circularly polarized (CP) light. The 3 Figure 1. Schematic of the Polarization Sensitive Optical Coherence Tomography setup: SLD - Superluminescent diode, P-Polarizer, W-Quarter Wave plate, BS-Non-polarizing Beam Splitter, S-Sample Arm, R-Reference arm with scanning assembly, PBS-Polarizing Beam Splitter, D1,D2- Photo Diodes, DAQ-Data acquisition system, ED-Envelope detector, PC-Laptop and other elec- tronic processing units. CP beam was delivered to the a 50-50 non-polarizing beam splitter (BS). The reference arm was scanned repetitively at a constant speed of (3 mm/s). The back reflected light from the reference arm and the backscattered light from the sample arm form an interferogram, which is delivered to the polarizing beam splitter (PBS). The orthogonal components of the interference signal are detected by two photo diodes D1 and D2 . These orthogonal components of the OCT signal were filtered, amplified and delivered to a laptop for further processing and analysis. A. Signal Processing The sample arm of the device consists of a reference glass plate. As shown in Figure 1, the sample under study makes direct contact with the top surface of the glass plate. The OCT signal obtained from samples such as tissue phantom or human sub ject is expected 4 to give two distinct peaks, corresponding to the air/glass and glass/tissue interfaces. The signal obtained for air/glass interface is a pure Gaussian having signatures of the light source while the signal obtained for glass/tissue interface have the signatures of the sample under study. Since the optically active medium rotates the LCP and RCP at different magnitudes, finite degree of circular polarization occurs for different values of glucose concentrations. Accordingly, the PS-OCT setup employs a polarizing beam splitter (PBS) to disentangle the orthogonal components of the light polarization. The amplitudes Ex [= (σ+ + σ− )/2] and Ey [= (σ+ − σ− )/2] of the backscattered signal from the sample are measured using the photo diodes D1 and D2 with σ+(−) being the electric filed component of the RCP (LCP). B. Tissue Phantom Intralipid is widely used in many optical experiments to find the scattering properties of biological tissues and can be used as a good scatterer like RBCs in human blood. Intralipid is an emulsion of soybean oil, egg phospholipids and glycerin. The ma jor advantages of intralipid are its well known optical properties and the similarity of its microparticles to lipid cell membranes and organelles that constitute the source of scattering in biological tissue. We used intralipid as a tissue phantom that provides the backscattered component of an incident light. Average size of scatterers in intralipid measured using laboratory confocal microscope is found to be 3.5 µm with refractive index of 1.42. The measurement with tissue phantom was taken under the hypoglycemic, normal, and hyperglycemic glucose levels of humans as (< 80 mg/dl), (80-110 mg/dl), and (>110 mg/dl), respectively. Following measurement modality is adopted during the experiment for use with tissue phantom. First, a fixed volume (1 ml) of intralipid (20% w/V) was diluted to 100 ml using distilled water and the diluted solution of 0.01% of tissue phantom was prepared. In the second step 200 mg of glucose was dissolved in 100 ml distilled water to make glucose solution. In the third step 100µl of glucose solution was added to 1 ml of tissue phantom using U-TEK Chromatography syringe with least count of 5µl. This gives us 20mg/dl glucose concentration. Two minutes of settlement time was given between tissue phantom and every glucose concentrations. The glucose concentration was increased upto 200mg/dl. Only 10µl of whole solution was placed on 140µm thick coverslip in the sample arm and five measurements of backscattered light were taken for each concentration. 5 The results obtained from the intralipid were verified using Mie scattering theory. We have further extended the analysis with voluntary human sub jects. C. Blood samples from human sub jects Our device has the flexibility of monitoring blood glucose invasively as well as non- invasively. In the present case, the thumb finger of the sub ject is used in the sample arm instead of tissue phantom in the previous case. Before placing the finger the distal pulp region of thumb was cleaned by mild spirit and glycerin was applied for better optical contact between thumb skin of human sub ject and coverslip. Measurements were performed simultaneously using PS-OCT setup and with standard glucometer or through chemical route (GOD/BOD Method). It is well known that after a meal, a temporary rise in blood sugar occurs. The extent and duration of sugar level depends on the characteristics of the food as well as the sub jects. For a normal sub ject the increase in blood sugar level may not exceed 160-180mg/dl. In case of diabetic sub ject the rise and reduction of glucose level after meal is slower and elevated. The sub jects volunteered in the current experiment are asked to consume 100g glucose with water. After a settling time of 10 minutes, the measurement of blood glucose were carried out using PSOCT system and through glucometer, simultaneously. The measurement were repeated for every 10-15 minute interval. A correlation between the glucose concentration and DOCP is obtained and analyzed. III. THEORETICAL ANALYSIS A theoretical model is developed to understand the experimental observations. Intralipid and human blood can be theoretically categorized as turbid media in which the backscattered light from sample arm contributes to OCT signal. The signal from the sample arm comprises of single and multiple scattered components of light. Strong multiple scattering typical of most biological tissues leads to loss of phase, direction and polarization of incident radiation. It has been observed that for circularly polarized light randomization of polarization requires more scattering events than the randomization of its direction [10–12]. Multiply scattered polarized light from a turbid medium carries information about refractive index variation. 6 The incident light in the present experimental setup is circularly polarized. Tinoco et al. [13] discussed differential scattering of circularly polarized (CP) light and concluded that circular differential scattering retains information concerning the chiral properties of the scattering ob ject even when the sample is partially or completely disordered. The circular intensity differential scattering (CIDS) or degree of circular polarization (DOCP) is defined as [14, 15] DOC P = IL − IR IL + IR = σL − σR σL + σR = Ix Iy (1) Here, σL and σR being the scattering cross section corresponding to left and right circular polarization components and Ii represent the intensity. The RBCs which are considered as the principle scatterers in blood lie in the Mie scattering domain. Accordingly the scattering cross section σL and σR are given by [14] σi = 2πa2 Pn (−1)n (2n + 1)(an,i − bn,i )2 α2 , (2) where, an,i = ψn,i (α)ψ ′ n (β ) − mψn (β )ψ ′ n,i (α) ξn,i(α)ψ ′ n (β ) − mψn (β )ξ ′ n,i(α) and mψn,i (α)ψ ′ n (β ) − ψn (β )ψ ′ n,i (α) mξn,i (α)ψ ′ n (β ) − ψn (β )ξ ′ n,i(α) with i = R, L, α = kda, kd being the propagation constant of the radiation in ECF and a bn,i = is the scatterer size. β (= mkda) with m being the relative refractive index of the scatterer (RBC in the present case). Also, ψn (α) = αjn (α) and ξn (α) = αh′ n (α), where jn (α) and hn (α) represent the spherical Bessel and Hankel functions, respectively. For the present experimental setup kd = 7.53µ−1m. The ma jor light scatter with human sub ject is RBC. Hence the average size of the RBC is considered as a = 7µm. In order to obtain the variation in refractive index as a function of glucose concentration, we have experimentally calculated m (=refractive index of glucose solution/refractive index of water) and exhibited in Figure 2. The inset shows the dependence of refractive index with the concentration of electrolytes (NaCl, KCl, Glucose) and Urea in water. It is clearly established that the contribution of glucose is dominating over others. Figure 3 exhibits a variation of DOCP with glucose concentration. The solid line showing a linear increase with glucose concentration is obtained using eqs. 2 and 3 and using the current experimental conditions. Data obtained with intralipid is shown (as filled dots) in 7 x e d n I e v i t c a r f e R 1.42 1.40 1.38 1.36 1.34 1.32 1.30 0 150 100 50 Glucose Concentration (mg/dL) 200 Figure 2. Linear variation in Refractive index has been with increasing glucose concentration. the same figure for comparison. A qualitative agreement is obtained beween the theoretical and experimental observations. IV. RESULTS Armed with the encouraging results obtained for the samples of tissue phantom, we pro- ceed further with human sub jects. In Figure 4, the degree of circular polarization obtainable from (D2/D1) is plotted as a function of glucose concentration for four sub jects. The filled square with error-bars is obtained for voluntary sub jects while the solid curve is a linear fit to the data. The slope obtained for different sub jects are shown in Table 1. It is inter- esting to note that the ratio of DOCP and glucose concentration remains a constant for a particular sub ject. This fact is verified with the same sub ject with different conditions such as day/night, post-meal/pre-meal, etc. This fact is promising that a dedicated glucometer could be designed and fabricated for regular and hand-held usage. In normal sub jects the blood glucose level stabilizes within 120 minutes. It is well known 8 P C O D 0.3 0.2 0.1 0.0 Experiment Theory 40 200 160 120 80 Glucose Concentration(mg/dl) Figure 3. Linear variation in DOCP with increasing glucose concentration has observed theoretical and experimental are shown with black and red symbols,respectively. [16] that the blood sugar level in human sub jects initially increases after taking meal and reaches the normal value after 30-45 minutes. The normal sugar level is reached within 120 minutes [16]. In some of the sub jects, a hypoglycemic dip may also occur. After a glucose drink, we examined the sub ject for variation in blood glucose level in every 10-15 minutes interval. The glucose level variation as a function of time has been plotted in Figure 5. The glucometer measurements are shown as a solid square while DOCP measurements are shown by open squares with error bar. We find an excellent quantitative agreement between these curves. The corresponding data with linear fit is shown as inset. V. CONCLUSIONS We have designed and developed PS-OCT for non-invasive glucose monitoring of blood glucose with human sub jects. The results exhibited in Figures 4-5 shows that a definite linear correlation exists between the glucose concentration and the degree of circular polarization. 9 1.00 80 100 Abhishek Tripathi 120 140 160 120 140 160 Arvind Malviya Mean Linear Fit of Sheet1 A 0.75 0.50 P C O D 1.00 0.75 0.50 0.25 P C O D Lokesh Jain Vipin Kaushik P C O D 1.00 0.75 0.50 1.00 0.75 P C O D 0.50 0.25 80 100 200 180 160 140 120 Glucose Concentration (mg/dl) 70 80 130 120 110 100 90 Glucose Concentration (mg/dl) 140 Figure 4. Simultaneous monitoring of DOCP using our setup and using commercial grade Glu- cometer. The results are promising. On the other hand, we could find that the ratio of degree of circular polarization and glucose concentration is a reproducible constant value for a same person. Hence, if a hand held glucometer is developed, the slope value could be used as a signature for measuring glucose concentration for the sub ject under study. ACKNOWLEDGMENTS The authors thankfully acknowledged the support given by the volunters. The authors thank Prof. P. K. Sen, SGSITS for fruitful discussions. They also acknowledge the financial support received from UGC & DBT, New Delhi, India and MPCOST, Bhopal, India. [1] N. Ghosh, J. Soni, M. F. G. Wood, M. A. Wallenberg, and I. A. Vitkin, Muller matrix po- larimetry for the characterization of complex random medium like biological tissues, Pramana J. Phys., 2010, 75, 1071-1086. 10 180 170 160 150 140 130 120 ) l d / g m ( t n e m e r u s a e M r e t e m o c u l G k n i r D e s o c o l G 1.5 1.0 P O C D 0.5 120 180 160 140 Glucose Concentration (mg/dl) 0 10 20 30 40 Time (min) 50 60 70 80 1.4 1.2 1.0 0.8 0.6 0.4 T C O S P h t i w d e r u s a e M P C O D Figure 5. DOCP with increasing glucose concentration provide linearity both in theoretical and experimental which are shown with black and blue symbols,respectively. Sub ject Name Slope RBC Size Age (mg/dl)−1 (µm) Years 1 2 3 4 A. Tripathi 0.03276 A. Malviya 0.04818 L. Jain 0.05106 V. Kaushik 0.11702 7.52 8.10 7.90 8.20 26 31 30 29 [2] M. R. Hee, D. Huang, E. A. Swanson, and J. G. Fujimoto, Polarization-sensitive low-coherence reflectometer for birefringence characterization and ranging, J. Opt. Soc. Am. B, 1992, 9, 903- 908. [3] M. Kohl, M. Cope, M. Essenpreis, and D. Bocker, Influence of glucose concentration on light scattering in tissue-simulating phantoms, Opt. Letts., 1994, 19, 2170-2172. [4] J. T. Bruulsema, J. E. Hayward, T. J. Farrell, M. S. Patterson, L. Heinemann, M. Berger, T. Koschinsky, J. S. Christiansen, and H. Orskov, M. Essenpreis, G. S. Redeker, and D. Bocker, 11 Correlation between blood glucose concentration in diabetics and noninvasively measured tissue optical scattering coefficient, Opt. Letts., 1997, 22, 190-192. [5] X. Wang, G. Yao, and L. V. Wang, Monte Carlo model and single scattering approximation of the propagation of polarized light in turbid media containing glucose, Appl. Opt., 2002, 41, 792-801. [6] K. V. Larin, M. Motamedi, T. V. Ashitkov, and R. O. Esenaliev, Specificity of noninvasive blood glucose sensing using optical coherence tomography technique: a pilot study, Phys. Med. Biol., 2003, 48, 1371-1390. [7] K. V. Larin, T. Akkin, R. O. Esenaliev, M. Motamodi, and T. E. Milner, Phase-sensitive optical low-coherence reflectometry for the detection of analyte concentrations, 2004, 43, 3408- 3414. [8] B. D. Cameron, Y. Li, Polarization-Based Diffuse Reflectance Imaging for Noninvasive Mea- surement of Glucose, J. Diab. Sci. and Tech., 2007, 1, 873-878. [9] S. W. Lee, J. H. Kang, J. Y. Yoo, and, B. M. Kim, Measurement of Scattering Changes Using Polarization-Sensitive Optical Coherence Tomography, Proceedings of the 29th A. Intern. Confer. of the IEEE EMBS, France, 2007, 3350-3352. [10] W. Cai, X. Ni, S. K. Gayen, and R. R. Alfano, Analytical cumulant solution of the vector radiative transfer equation investigates backscattering of circularly polarized light from turbid media, Phys. Rev. E, 2006, 74, 056605-1-10. [11] X. Ni and R. R. Alfano, Time-resolved backscattering of circularly and linearly polarized light in a turbid medium, Opt. Lett., 2004, 29, 2773. [12] M. Xu and R. R. Alfano, Random Walk of Polarized Light in Turbid Media, Phys. Rev. Lett, 2005, 95, 213901-1-4; Phys. Rev. E, 2005, 72, 065601. [13] I. Tinoco, Jr. and David Keller, Scattering of Circularly polarized radiation, J. phys. Chem, 1983, 87, 2915-2917. [14] M. Born and E. Wolf, Principle of optics: Electromagnetic Theory of propagation Interference and Diffraction of light, Pergamon Press, New York, 1980 pp. 647-661. [15] J. Solanki, P. Sen, J. Andrews and K. Thareja, Cyclic Correlation of Diffuse Reflected Signal with Glucose Concentration and Scatterer Size, Journal of Modern Physics, (2012), 3, 64-68. [16] H. Varley, Practical Clinical Biochemistry: Glucose tolerance tests and tests for investigating hypoglycemics, Gulab Vazirani, Arnold-Heinemann, Manchester Univ. 1969 pp. 97-109. 12
1111.6906	1	1111	2011-11-29T17:15:55	Tug-of-war of molecular motors: the effects of uneven load sharing	[ "physics.bio-ph", "q-bio.SC" ]	We analyze theoretically the problem of cargo transport along microtubules by motors of two species with opposite polarities. We consider two different one-dimensional models previously developed in the literature. On the one hand, a quite widespread model which assumes equal force sharing, here referred to as mean field model (MFM). On the other hand, a stochastic model (SM) which considers individual motor-cargo links. We find that in generic situations the MFM predicts larger cargo mean velocity, smaller mean run time and less frequent reversions than the SM. These phenomena are found to be consequences of the load sharing assumptions and can be interpreted in terms the probabilities of the different motility states. We also explore the influence of the viscosity in both models and the role of the stiffness of the motor-cargo links within the SM. Our results show that the mean cargo velocity is independent of the stiffness while the mean run time decreases with such a parameter. We explore the case of symmetric forward and backward motors considering kinesin- 1 parameters, and the problem of transport by kinesin-1 and cytoplasmic dyneins considering two different sets of parameters previously proposed for dyneins.	physics.bio-ph	physics	Tug-of-war of molecular motors: the effects of uneven load sharing . Sebasti´an Bouzat1 and Fernando Falo2 1 Consejo Nacional de Investigaciones Cient´ıficas y T´ecnicas, Centro At´omico Bariloche (CNEA), (8400) Bariloche, Argentina. 2 Dpto de F´ısica de la Materia Condensada and BIFI, Universidad de Zaragoza, 50009 Zaragoza, Spain. E-mail: [email protected] Abstract. We analyze theoretically the problem of cargo transport along microtubules by motors of two species with opposite polarities. We consider two different one-dimensional models previously developed in the literature. On the one hand, a quite widespread model which assumes equal force sharing, here referred to as mean field model (MFM). On the other hand, a stochastic model (SM) which considers individual motor-cargo links. We find that in generic situations the MFM predicts larger cargo mean velocity, smaller mean run time and less frequent reversions than the SM. These phenomena are found to be consequences of the load sharing assumptions and can be interpreted in terms the probabilities of the different motility states. We also explore the influence of the viscosity in both models and the role of the stiffness of the motor-cargo links within the SM. Our results show that the mean cargo velocity is independent of the stiffness while the mean run time decreases with such a parameter. We explore the case of symmetric forward and backward motors considering kinesin- 1 parameters, and the problem of transport by kinesin-1 and cytoplasmic dyneins considering two different sets of parameters previously proposed for dyneins. PACS numbers: 87.16.A, 87.16.Nn, 87.16.Uv Keywords: molecular motors, tug of war, load sharing Submitted to: Phys. Biol. The effects of uneven load sharing 2 1. Introduction Transport of cargo driven by multiple molecular motors along microtubules has become a very active subject of research because of its relevance for many cellular functions [1, 2, 3, 4]. In recent years, a myriad of experiments and models have attempted to understand the way in which motors work together [4, 5, 6, 7, 8, 9], and, still, there are many fundamental details which remain unclear and deserve further research, most particularly for the case of bidirectional transport by two motor species. The complexity of the multiple motor systems and the difficulties for controlling the experiments are often quite important so that performing the connection between models and experiments must be done carefully. Models involve always many parameters, including for instance detachment and attachment rates, stall forces, motor stiffness and viscosity of the media. Usually, many of these parameters are a priori not well known in the experiments, and even more fundamental features such as the number of motors, or whether more than a single species is participating on the transport, remain unclear. Thus, distinct models may provide different fitting of the experimental data and, consequently, different interpretations. Moreover, recent in vivo experiments [10] have revealed important differences with in vitro systems. In this context, a detailed knowledge of the consequences of specific modeling assumptions as well as the comparison of different kinds of models becomes quite relevant. The aim of this paper is to contribute in these two important aspects. References [11] and [12] have originated a modeling framework that has largely contributed to the understanding of transport by several motors. The model introduced in [11] deals with cargo transport by a single class of motors, while in [12] the formalism is extended to account for bidirectional transport associated to tug of war between two motor types with opposite polarities. Assuming certain force -- velocity relations, and specific attachment and detachment probabilities for individual motors, the model enables the calculation of the probabilities of different motility states characterized by different number of motors, and the reproduction of trajectories and velocity distributions as well. In a series of papers [7, 13, 14, 15] the model was further developed and several effects and transport conditions have been analyzed, providing a deep physical insight on the problem. An important assumption of the model is that all the motors of the same polarity simultaneously engaged to the microtubule share equally the load. In real systems, however, fluctuations of the distances between motor-microtuble binding position and motor-cargo binding position may lead to non- negligible differences between the forces supported by the different motors [8, 16, 17, 18]. Consequently, the model would eventually fail to predict exact quantitative results. In reference [17], the model was referred to as mean field due to the equal sharing of load approximation. We will keep such a name throughout this work. Several models have gone beyond the mean field approach by considering independent motor-cargo links for each motor, and incorporating different degree of detail in their description of individual motor properties [8, 9, 17, 19, 20]. Although The effects of uneven load sharing 3 such models generally provide less instrumental (and less elegant) formulations than the mean field model, and they mostly lack analytical results, they may be more successful in predicting numerical results for multiple motors through simulations based on individual motor parameters. In a different but related context, models in references [21] and [22] consider the load applied only to the leading motor and constitute thus interesting extreme examples of models beyond mean field. Although not directly connected to our approach for processive motors on microtubules, studies on non-processive motors [23, 24] and general ratchet models [25] provide also relevant analysis of bidirectional motion in many motor systems. In this paper we investigate bidirectional cargo transport by two opposing teams of processive motors within two different models. On the one hand, the mean field model. On the other hand, a recently introduced [19] stochastic model which considers independent cargo-motor links for individual motors, allowing for uneven load sharing. In this way, at the same time that we investigate how cargo transport depends on the system parameters, we are able to clearly identify the consequences of the assumption of equal load sharing. Our work follows the spirit of the paper by Kunwar and Mogilner [17]. There, the authors compared results from both kind of models focussing on the case of cargo transport by a single team of motors, and provided also an analysis of the velocity distributions for bidirectional transport. Moreover, they studied the influence of the non linearities of the force-velocity relations of individual motors. Our studies focus on analyzing the dependence of cargo transport on the number of motors of each polarity, the viscous drag, and the stiffness of the motor cargo link, while we do not consider the influence of additional load forces acting on cargo. In Section 2 we present the models. Section 3 studies the case of equal forward and backward motors. The effects of varying the number of motors to each side and the influence of viscous drag are analyzed within both models. In section 4 we present results for bidirectional transport by asymmetric motors considering system parameters compatible with kinesin-1 and cytoplasmic dynein. Section 5 is devoted to the conclusions. 2. Models and methods As indicated in the introduction, we will consider two different models for the analysis of cargo transport by multiple motors. The mean field model (MFM), and our recently proposed stochastic model (SM). We first introduce the characteristics that are common to both. The two models consider the cargo as a point particle which performs a continuous trajectory x(t) in one dimension. The cargo is linked to Nf forward motors and Nb backward motors. The first of them can pull the cargo in the positive direction while the second can pull it in the negative one. At a given time, the number of forward and backward motors engaged to the microtubule are respectively nf (t) ≤ Nf and nb(t) ≤ Nb. Each engaged motor i, detaches from the microtubule with a probability per time unit given by ǫ exp(fi/Fd). Here fi is the instantaneous force exerted by motor i The effects of uneven load sharing 4 on the cargo, ǫ is the reference zero-load detachment rate and Fd > 0 is the detachment force. We will call ǫf , ǫb, Fdf and Fdb the corresponding parameters for forward and backward motors. Conversely, a detached motor engages to the microtubule with rate Πf or Πb, according to its type. When loaded with a force fi > 0 (considered positive if exerted against the polarity of the motor), the motor i advances with velocity vi =( v0(1 − fi/Fs) v1(1 − fi/Fs) forfi ≤ Fs forfi > Fs. (1) Here Fs > 0 is the stall force, v0 the zero-load velocity and v1 a reference backward velocity. Considering both motor species we have the system parameters Fsf , Fsb, v0f , v0b, v1f and v1b. The linear force-velocity relation for single motors of Eq. (1) is a natural choice for comparing the SM and MFM since it is used in most works on the MFM [7, 12, 13] and is also a common assumption in other theoretical models [17, 26, 27]. Studies in [17] suggest that the consideration of a general non-linear relations would lead to no relevant qualitative changes in the results. It is important to mention that, while Eq.(1) is taken as instantaneously exact in the MFM, within the SM it is only valid in terms of time averages, i.e. vi is the mean velocity of a motor subject to a constant force fi. The way to compute the forces fi and the cargo motion depends on the model as we explain in the following subsections. 2.1. Cargo dynamics in the mean field model The MFM [12] assumes that all the motors (backward and forward) move with the same velocity than cargo at any time, and that motors of the same polarity share the force equally. It also assumes the total force acting on the cargo vanishes at almost any time (it has discontinuities at the times at which the number of engaged motors changes). With such hypothesis, by performing a force balance and using the force- velocity relation for single motors of Eq.(1), it is possible to obtain the cargo velocity as a function of the numbers of engaged motors nf and nb [12]. We thus have a discrete set of allowed cargo velocities v(nf , nb), corresponding to the different motility states (nf , nb) considering nf = 0, 1, ..., Nf , nb = 0, 1, ..., Nb. The model can be implemented through two main different methods. First, by means of a master equation which allows to compute stationary probabilities P (nf , nb) and, thus, velocity distributions P (v). And, second, by means of a Gillespie algorithm [12, 28] which allows to compute cargo trajectories. This latter numerical scheme determines the temporal evolution of the system by ruling the transitions between different motility states taking into account the attachment and detachment probabilities. During each time interval between two transitions, the cargo velocity is assumed to be constant and equal to the corresponding value v(nf , nb). For two motors species, the model was first introduced in [12] without considering any external force. Then, in [15] it was generalized to include the cases of viscous The effects of uneven load sharing 5 environments and non vanishing load forces. This is done by modifying appropriately the force balance, but without changing any of the model hypothesis mentioned before. As part of our studies we have implemented both the master equation and Gillespie formulations of the model without external forces (although we will only show results from the latter), and also the Gillespie method considering non negligible viscous drag. We have checked that our results correctly reproduce some selected ones from references [12] and [15]. In all cases we have assumed the linear force-velocity profile of Eq.(1). 2.2. Stochastic model As other models in the literature [8, 17], the stochastic model introduced in [19] considers a Langevin dynamics for cargo motion, a discrete-steps stochastic dynamics for individual motors and cargo-motor links described by non-linear springs. The Langevin equation for cargo is fi + ξ(t). (2) γ xc =Xi Here γ is the viscous drag, fi (i = 1, ..., N = Nf + Nb) the force exerted by the i-th motor and ξ(t) the white thermal noise. The viscous drag is defined through the Stokes relation γ = 6πηr [8, 17], where η is the viscosity of the medium and r the radius of the cargo for which we consider r = 500nm throughout the paper. The thermal noise satisfies hξ(t)i = 0 and the correlation formula hξ(t1)ξ(t2)i = 2Dδ(t1 − t2) [29]. Here h i represents ensemble average, δ(t) is the Dirac Delta and D the diffusion coefficient satisfying the fluctuation-dissipation relation D = kBT /γ [29], with kB the Boltzmann constant and T the temperature. In all our calculations we consider T = 300K. Each motor is modeled as a particle that can occupy discrete positions separated by ∆x = 8nm along the same spatial coordinate used for the cargo. Its dynamics is governed by a Monte Carlo algorithm [19] that rules the elementary processes of step forward, step back, detachment, and attachment. At each time step of duration dt, an engaged forward motor has a probability pjump = dt/τD(F ) of performing an 8nm step, which may be forward (right) with probability Pr(F ) = [R(F )/(1 + R(F ))] or backward (left) with probability Pl(F ) = [1/(1 + R(F ))]. Here τD(F ) is the dwell time [30, 31] and R(F ) is the forward-backward ratio of jumps [30, 31, 32]. The resulting mean velocity for a single forward motor with constant load F is v(F ) = ∆x(Pr(F ) − Pl(F ))/τD(F ). In [19] the model was developed assuming certain specific formulas for R(F ) and τD(F ) based on experimental data for kinesin-1, while v(F ) was left as free. Here, in order to compare results with the MFM, we consider v(F ) as a known relation instead of τD(F ). Then, the value of τD(F ) entering in the algorithm is determined by inverting the corresponding formulas. For the R(F ) we consider the experimentally based [30, 31, 32] formula R(F ) = A exp(− log(A)F /Fs), with A = 1000 and Fs the before mentioned single motor stall force that leads to Pr = Pl. For the backward motors we consider the same single motor model but interchanging right and left. The forces fi are computed assuming the cargo is linked to each motor by a non linear spring [8, 17] which produces The effects of uneven load sharing 6 only attractive interactions, and only for distances larger than a critical one. Let us call xi the position of motor i and ∆i = xi −xc. We define fi = k(∆i −x0) for ∆i ≥ x0, fi = 0 for −x0 < ∆i < x0 and fi = k(∆i + x0) for ∆i ≤ −x0, with x0 = 110 nm [8, 17]. Here, k is the stiffness of the motor for which we consider values kf and kb for forward and backward motors respectively. Note that while in [19] we have included volume excluded interaction between motors, here we consider only interactions mediated through cargo. The detachment and attachment processes occur according to the probabilities per time unit indicated at the beginning of this section. The attachment of detached motors occurs with equal probability in any of the discrete sites xj satisfying xj − xc < x0. 2.3. Relevant quantities and numerical simulations We study the cargo dynamics within MFM and SM by performing numerical simulations of the evolution of the system for different values of the parameters. As initial condition (at time tini ≡ 0) we consider a random number of motors of each species engaged on the microtubule. Each realization finishes when all the motors are detached (at time referred to as tend). The numerical simulations of the SM are performed as explained in [19] using time steps between dt = 2 × 10−5s and dt = 3 × 10−7s depending on the value of γ. For the MFM we use our implementation of the Gillespie algorithm explained in [12, 28]. In order to characterize the long-time properties of cargo dynamics we compute the following quantities. • Cargo mean velocity. Defined as the average over realizations of the ratio (xend − xini)/(tend − tini). • Run length. Defined as the average over realizations of (xend − xini). • Run time. Denoted as τr, equal to the average of (tend − tini). Concerning the analysis of the dynamical properties during forward and backward stages of the motion, we compute the mean forward run length (rf ) and the mean backward run length (rb). We define them as the average distance traveled by the cargo during the time intervals at which nf > nb and nf < nb respectively. Note that the association of forward (backward) motion of the cargo with nf > nb (nf < nb) makes sense only for symmetric motors (i.e. equal parameters for motors of both polarities). For asymmetric motors, the characterization of forward and backward stages of motion demands a signal analysis of the trajectories including filtering and appropriate definitions of switching points and forward and backward runs, as it is usually done in experimental works [33, 34]. Such kind of studies is out of the scope of the present work. Note that, for the MFM with symmetric motors, any definition of forward and backward run lengths based on signal analysis of trajectories would lead exactly to the same results as our definitions based on nf and nb. This is because, in such a case, the condition nf > nb or nf < nb determine the direction of motion. In contrast, within the SM, the exact coincidence between both kind of definitions cannot be ensured, since The effects of uneven load sharing 7 for very short times we could have forward (backward) motion of cargo with nf < nb (nb < nf ). Nevertheless, the differences are expected to be small. In any case, the definitions in terms of nf and nb are relevant by themselves for our theoretical analysis. 3. Results for symmetric motors First we analyze the results for both models considering equal parameters for forward and backward motors. For shortness, we speak of equal or symmetric motors. Except when specially stated, we consider single motor parameters compatible with kinesin-1 [12] for both models: Fs = 6pN, v0 = 1000nm/s, v1 = 6nm/s, ǫ = 1/s, Fd = 3.18pN and Πf = 5/s. We left the numbers of motors Nf and Nb, and the viscosity as free parameters. For the SM, except when indicated, we consider the parameter k = 0.32pN/nm usually taken as reference value for kinesin-1 [17]. 3.1. Trajectories As a first step in our study, we glance at the trajectories within both models. Figure 1.a shows cargo and motors trajectories for a system with Nf = 2 and Nb = 1 computed using the SM. Regions of tug of war leading to pauses and reversions of the cargo motion can be appreciated. In figure 1.b we show MFM and SM cargo trajectories for Nf = Nb = 2. At first glance we see that both models produce similar trajectories for such parameters. Thus, we can expect that this may lead to compatible results for ensemble averaged quantities. In contrast, results in figures 1.c and 1.d indicate us that, in the case Nf = 3, Nb = 2 both models predict very different results even at the level of single trajectories. Thus, depending on the parameters we may expect that the two models give results which may be statistically equivalent or not. 3.2. Results for negligible viscous drag Now we begin our systematic analysis of both models focussing on the behavior of the cargo mean velocity, run length and run time. We analyze first the dynamics for negligible viscous drag. To do so, we consider the MFM without viscous drag [12], and the SM with a very small value of γ, so that the system is essentially at the zero viscosity limit. Actually, we use γ = 9.42 10−6pNs/nm, calculated using the Stokes formula [8, 17] with water viscosity and a radius of the cargo equal to 0.5µm. Note that for such a value of γ, even if we consider a fast cargo velocity of 103nm/s we get a viscous drag of order 10−2pN which is quite small compared to the typical forces on the scale of 1pN involved in motor dynamics. In figure 2 we study the case of a single species of motors considered with forward polarity. We plot the run length and the velocity as functions of the number of motors. The results are already well known from a number of previous works: the velocity is independent of the number of motors (for negligible viscosity), while the run length grows exponentially. Our contribution here is to show that both models agree in their The effects of uneven load sharing 8 numerical results. As the analysis of single trajectories suggest and we will shortly confirm, this is not always the case when we consider two species of motors. In figure 3.a we show results for the cargo velocity as a function of the number of backward motors Nb for fixed Nf = 3 in the symmetrical case. It can be seen that both models coincide only for Nb = 0 and Nb = Nf , while for intermediate values of Nb the MFM predicts considerably larger velocities. These differences are a consequence of the load sharing hypothesis. Note that, while in the MFM all engaged forward motors contributes equally to pulling the cargo, in the SM only those motors which are instantaneously beyond the limit distance x0 from cargo exert non vanishing forces. Hence, each of such pulling motors are more loaded than motors in the MFM and, thus, their velocity is smaller. Clearly, this causes a smaller cargo velocity, since cargo velocity is essentially controlled by such leading motors. It is interesting to realize that for the SM we obtain the simple linear behavior v = v0f (Nf −Nb)/Nf , regardless the value of the motor stiffness k. This demonstrates a certain degree of robustness of the motor team performance independently of the stalk stiffness. However, as we will see, other relevant quantities do depend on k. In figure 3.b we show the run time τr as a function of Nb for the same system as in figure 3.a. We see that, within the SM, τr increases with Nb and decreases with k. Except for vanishing Nb, the SM predicts sensibly larger values of τr than the MFM. Note that, while the cargo velocity is controlled by the pulling motors, the run time is expected to be essentially determined by the total number of engaged motors, regardless its polarity. This is because forward and backward motors contribute equally to linking the cargo to the microtubule (at least for symmetric motors). The relevance of differentiating between engaged and pulling motors was discussed in [19] when analyzing transport by a single species against an external load. The relation between run time and total number of engaged motors becomes evident with the results in figure 3.c, where we show the mean number of engaged motors as function of Nb for the same systems in 3.a and 3.b. The parallelism between curves in 3.b and 3.c is apparent. The total number of engaged motors increases with Nb, decreases when passing from MFM to SM and decreases with k within SM. The causes of these behaviours will be explained later when studying the probabilities of the different motility states. In figure 3.d we show the run length as a function of Nb for the same parameters as those in figure 3.a. and 3.b. As expected, the run length decreases with Nb in both models following the decrease of the mean velocity. Within the SM, the decrease of the run length with k can be associated to that of the run time and to the invariance of the mean velocity. This seems compatible with a factorization of the mean values. Interestingly, the results for the MFM are similar to those for the SM with k = 0.32pN/nm. However, this seems to be due to a compensation between the decrease of the velocity and the increase of τr when passing from the MFM to the SM. Now we study the probabilities of the different states (nf , nb) for fixed Nf and Nb. This is relevant for finding out to what extent forward and backward motors coexist linked to the microtubule, and for evaluating the mean number of engaged motors. Let us take the case Nf = 3, Nb = 1 with k = 0.32pN/nm as an example. Figures 4.a and The effects of uneven load sharing 9 4.b show the probabilities of the different states (nf , nb) for SM and MFM respectively. Note that the state nf = nb = 0 has null probability as it determines the end of the simulation. The results clearly indicate that the states with one engaged backward motor are much more likely in the SM than in the MFM. Note that for the MFM, the states with nb = 0 accumulate around a 85% of the probability. This means that the backward motor is detached most of the time. In particular, the states (3, 0) and (2, 0) alone dominate the dynamics a 80% of the time. In contrast, the SM predicts that the backward motor will be essentially half of the time engaged to the microtubule. Moreover the probabilities for the states (3, 0), (3, 1), (2, 0) and (2, 1) are all similar to each other. States with the backward motor engaged are more likely in the SM than in the MFM due to that, within the SM, the backward motor is unloaded a non negligible part of the time and, when it is loaded, it is in a tug of war only with those forward motors that are beyond the limit of 110nm from cargo. In contrast, in the MFM the engaged backward motor is all the time in a tug of war with Nf motors, each of which is less loaded than the backward motor. Within the SM, the probabilities of states with a backward motor engaged is found to decrease with the stiffness k, as we show in Figure 4.c. This is reasonable, since smaller stiffness lead to lower probability of detachment and results in more permissive of tug of war states. In the case of the system with Nf = 3 and Nb = 2 the results (not shown) are completely analogous to those for Nf = 3 and Nb = 1 in figure 4. Thus, we can state generally that the probabilities of states with engaged backward motors increase when passing from MFM to SM and, within the SM, they increase with decreasing k. This explains the increase of the mean number of engaged motors when changing from MFM to SM and when decreasing k (figure 3.c) as a consequence of the contribution of the tug of war states, which have larger total number of motors than single species states. In figure 5 we show the results for rf and rb corresponding to the same systems analyzed in figure 3. As expected, rf decreases monotonously with Nb in both models, since the probability of reversions grows with Nb. We also see that the SM predicts shorter forward excursions than the MFM. This is because the larger probability of having a backward motor engaged leads to a larger probability of changing from forward to backward motion. The results for rb in figure 5.b are much more intriguing due to the abrupt variations with Nb. Nevertheless, we can give an almost complete explanation for them. Concerning the results for the SM, the counterintuitive fact that rb is larger for Nb = 1 than for Nb = 2 is due to the contribution of the state (nf , nb) = (1, 2) in the latter case. In fact, this state is found to be the one that largely contributes to rb for Nb = 2 (it has probability p = 0.096 while states (0, 2) and (0, 1) have only p = 0.035 and p = 0.017 respectively). Clearly, the state (1, 2) is expected to produce smaller velocities (and thus shorter runs) than state (0, 1) which is the only one contributing to rb for Nb = 1 (with p = 0.019). However, note that, for Nb = 2, the accumulated probability of having nf > nb is larger than for Nb = 1. It thus happens that, for Nb = 2, backward excursions are shorter but more frequent than for Nb = 1. The results for rb in the MFM are less intriguing, except maybe for the abrupt decrease when passing from The effects of uneven load sharing 10 Nb = 3 to Nb = 2. This is simply related to that, due to equal force sharing, leaving the symmetric situation Nb = Nf works largely against the species which results with lower number of motors. 3.3. Influence of the viscous drag Now we analyze bidirectional transport under non negligible viscous drag. Figure 6.a shows the cargo mean velocity as a function of Nb for Nf = 3 considering a viscous drag equal to 1000 times that of water, both for MFM and SM. As expected, the velocities are smaller than those for water viscosity shown in figure 3.a (typically by a factor 1/2). The general behavior of both models is similar to that for water viscosity, with one relevant difference. Now, MFM and SM give different results even for Nb = 0. This is because, while in the MFM all the forward motors share the load coming from the viscous drag, in the SM the load acts essentially only on the pulling motors. The same would occur when considering any other kind of external load force acting against the advance of the cargo, as the results in [17] for a single motor species suggest. Figure 6.b shows the run times for the same systems analyzed in Figure 6.a. As in the case of the velocities, we find that the differences between both models extend to the case Nb = 0. Finally, we complete our analysis of the influence of viscous drag with the results in figure 6.c, which show the dependence of the cargo velocity on γ for systems with and without backward motors. It can be seen that, although the general dependence on γ for all systems and models are similar to each other (τr is constant for γ . 5 × 10−3pNs/nm and decreases exponentially for γ & 5 × 10−3pNs/nm), the predictions of both models coincide only in the case of low viscosity and no backward motors. Moreover, the differences between the results from both models increase with the addition of backward motors. 4. Results for uneven forward and backward motors Now we leave the symmetric case and study the models considering backward motor parameters that can be associated to cytoplasmic dynein. For forward motors we continue using the kinesin-1 parameters considered in the previous section. Since the walking and detachment properties of dyneins are not as well known as for kinesin-1, the parameters for dyneins are not quite clear. Here we consider two different sets of parameter's values (named simply as set A and set B) based on the two proposals in [12, 13, 15]. Set A has been considered within MFM in [12] and with small changes in [15] on the base of previous experimental and theoretical results (see Table 1 in [12], supporting material in [15] and references therein). Set B was obtained by fitting experimental data on Drosophila lipid-droplet transport [12] and is consistent with previous experimental data [35]. According to set A, the main differences between dyneins and kinesin-1 appear in the binding and unbinding rates. In contrast, set B considers also important differences on the typical velocities and stall forces of both It is interesting thus to investigate the bidirectional motion of cargo motor types. The effects of uneven load sharing 11 transported with kinesin-1 motors and both models of dyneins. Note that for the SM, in addition to binding, unbinding and velocity parameters, it is also necessary to specify the forward-backward ratio of jumps as a function of the load force. Since we have not experimental data for dyneins, in this work we consider the same exponential form used for kinesins. The use of any other reasonably formula is not expected to produce relevant changes in the results. In fact in [19] it was shown that even the consideration of no backward steps produces relatively small changes for the case of transport by kinesins. In figure 7 we show results for cargo transport by kinesins-1 and set A dyneins under negligible viscous drag. Since the differences between the parameters for both types of motors are relatively small, the results are similar to those for symmetrical motors. The effects of the asymmetry are mainly notable in the case Nf = Nb = 3, for which we observe a net backward motion. This can be seen both in the velocities in figure 7.a and in the trajectories in figure 7.b. The fact that set A dyneins win the tug of war for Nf = Nb is mainly due to their slightly larger stall force. The differences between the predictions of MFM and SM for the velocities (figure 7.a) and the run times (figure 7.c) are considerably relevant, and they occur in a similar fashion to that observed for symmetrical motors. The same happens with the probabilities of having a backward motor engaged (figure 7.d), which are found to be larger for the SM than for the MFM. As in the case of symmetrical motors, this latter result helps us to understand the differences in the run times from both models. Going back to figure 7.b, we see that the SM trajectories are much more winding than those from MFM. This phenomenon is related to the reduction of rf analyzed in the case of symmetric motors. The larger rate of reversion in the SM is due to that only some of the engaged motors pull the cargo at a given time and, in addition, it is more likely to have opposing motors engaged than in the MFM. Now we consider dynein with parameter set B. In this case, dynein is a motor force, much lower sensibly weaker than kinesin-1, since it has much lower stall detachment force, lower attaching probability and also lower ratio Fs/Fd. Thus, for equal number of kinesin and dyneins, the kinesins win the tug of war. In fact, we find that several dyneins are needed to produce average null velocity for a cargo pulled by only one kinesin. In figure 8.a we show the cargo mean velocity as a function of Nb for systems with Nf = 1 and Nf = 2. It can be seen that for small Nb the velocity is similar for both models (almost coincident for the case Nf = 1). In contrast, for relatively large values of Nb the predictions of both models differ substantially. In particular, the number of dyneins needed to attain zero average velocity for a cargo pulled by one or two kinesins are considerably different. For Nf = 1 we find Nb ∼ 8 for the MFM and Nb ∼ 12 for SM, while for Nf = 2 we find respectively Nb ∼ 14 and Nb ∼ 20. Figure 8.b shows the dependence of the run time on the number of dyneins for a cargo pulled by one kinesin. It can be seen that, even for Nb < 4 for which both models give similar mean velocities, they predict quite different results for τr. In the region nb ∼ 8, the situation is the opposite, both models give similar run times but they The effects of uneven load sharing 12 predict quite different velocities. For nb > 8 the run time in the MFM grows very fast with nb due to that dyneins win the tug of war (the mean velocity is negative). This makes dyneins to remain mostly attached while the only kinesin detaches. Thus, the total number of engaged motors which controls the run time increases. In figure 8.c we show results for the probabilities of the different states for a system with Nf = 1 and Nb = 4. For simplicity, we show only the probabilities of states with nf = 1 (states with nf = 0 have very small contributions in both models). Again, the SM gives a much larger probability of engagement of backward motors than the MFM. Finally, we briefly explore the influence of the viscosity and of possible differences on the stiffness of motors from both species. In figure 8.d we study the effective number of set B dyneins needed to achieve zero mean velocity when pulling against one kinesin. We consider different values of γ and kb. Note that our results provide non integer effective values for Nb which correspond to interpolations leading to zero cargo velocity. We see that for the MFM the results are almost independent of γ at a value close to Nb = 8. Interestingly, this value can be estimated by equating the powers produced by both motor teams considering all the motors attached at stall force. Namely, considering Nb × vsb × fsb = vsf × fsf we get to Nb = 8.39. The behavior within the SM is much more complex. The effective number of dyneins needed to stop a kinesin depends both on the viscosity and the stiffness. Actually, it decreases with γ and increases with kb. The decrease with γ is clearly due to that viscosity helps to stop the cargo. The increase with kb is due to that larger kb lead to less force production by dyneins, due to easier detachment. Note that we have have considered a quite small value for the dynein's stiffness (kb = 0.08pN/nm). The reason for this is twofold. First, it is the only way to reduce to reasonable values the effective number of set B dyneins needed to attain null velocity when pulling against one kinesin. Second, recent experiments [36] for transport mediated by dynein and kinesin-2 reported values of the stiffness in such range. 5. Conclusions Cargo transport along microtubules mediated by two opposing motor species provides interesting challenges both from the experimental and theoretical points of view. With the main aim of understanding the consequences of specific modeling assumptions, in this paper we have theoretically analyzed several aspects of the problem considering two different mathematical models: the mean field model (MFM) [12], and a recently introduced stochastic model (SM) [19] which share some commons with models in [8] and [17]. The main difference between the MFM and the SM stems from the assumptions of force sharing by the different motors. The MFM assumes equal load sharing by all the engaged motors of the same polarity while SM considers individual cargo motor linking allowing for uneven force sharing. Our main results indicate that both models show complete agreement only when there is essentially no load to share, that is, in systems with a single type of motors and with no relevant viscous effects. In other situations, the MFM predicts larger cargo mean The effects of uneven load sharing 13 velocity and smaller mean run time than the SM. We have found that the differences in the velocities are mainly due to the fact that, within the SM (and in agreement with statements in [8, 17, 18]), only some of the engaged motors pull the cargo at a given time. Moreover, the probability of engaged backward motors during forward excursions is larger in the SM tan in the MFM. This leads also to a larger rate of reversions within SM when compared with the MFM, or equivalently, to shorter excursions toward each polarity. The difference between the mean velocities predicted by both models is found to increase with the viscosity. We have also found that the mean run time is essentially controlled by the mean number of engaged motors at a given time, which depends on the probabilities of the different motility states and, ultimately, on the force sharing assumptions. Our results for the SM show that the mean cargo velocity turns out to be rather independent of the stiffness of the motor-cargo link (k). In contrast, the mean run time decreases with k approaching the MFM results for large k. These conclusions were obtained analyzing ideal systems in which motors of In addition, in section 4 we opposite polarities have identical dynamical properties. have provided results for the asymmetric case of transport driven by kinesin-1 and cytoplasmic dyneins, considering two different sets of parameters for dyneins usually found in the literature. Finally, it is interesting to mention the possibility of considering a hybrid-modeling framework taking advantage of the benefits of both kinds of models: the simplicity and computational advantage of the Gillespie formulation of the MFM, and the higher reliability of models including uneven force sharing. Note that the SM provides an alternative way to compute (numerically) the transition rates and velocities of the motility states entering in the Gillespie algorithm considering uneven load sharing. The same could be done with models such as those in references [8, 17]. Thus, when fitting the single motor parameters needed to reproduce experimental trajectories, different intermediate procedures combining computations with both kinds of models could be imagined, depending on the particular problem and on the a priori knowledge of the parameters. Acknowledgments SB acknowledge L. Bruno for useful discussions. Financial support from the Argentinean agencies CONICET (under Grant No. PIP 11220080100076) and CNEA, and from of the Spanish MICINN through project FIS2008-01240, co-financed by FEDER, are also acknowledged. 6. Figures The effects of uneven load sharing Nf=2, Nb=1 cargo forward motors backward motor a 2000 1500 ) m n ( x 1000 500 1 2 0 1500 0 c 1000 500 ) m n ( x 0 -500 0 2 4 time (s) Nf=2, Nb=2 SM MFM 14 2 4 6 8 10 Nf=3, Nb=2 MFM 6 8 2 4 time (s) b 1500 1000 500 0 3 -500 d 0 5000 4000 3000 2000 1000 0 0 Nf=3, Nb=2 SM 6 8 Figure 1. Trajectories. a) Typical cargo and motors trajectories for the SM considering Nf = 2, Nb = 1. b) Cargo trajectories for Nf = Nb = 2 for SM and MFM. c) Several different cargo trajectories for Nf = 3 and Nb = 2 for SM. d) Ibid c for MFM. In all the cases the viscosity considered is 100 times the water viscosity. [1] Howard J 2001 Mechanics of Motor Proteins and the Cytoskeleton (Sunderland, MA: Sinauer Associates). [2] Schliwa M and Woehlke G 2003 Molecular Motors Nature 422 759-65. [3] L.S.B. Goldstein and Z. Yang. Microtubule-based transport systems in neurons: the roles of kinesins and dyneins. Annu. Rev. Neurosci. 23 3971, 2000 [4] Gross SP 2004 Hither and yon: a review of bi-directional microtubule-based transport. Phys. Biol. 1 R1-R11. [5] Welte MA 2004. Bidirectional Transport along Microtubules. Current Biology 14 R525 - R537. [6] Kolomeisky A B, Fisher M E 2007. Molecular Motors: A Theorist's Perspective, Annu. Rev. Chem. 58 675-695. [7] Lipowsky R , Beeg J, Dimova R, Klumpp S, Muller MJI 2010 Cooperative behavior of molecular motors: cargo transport and traffic phenomena. Physica E 42 649-661 [8] Kunwar A, Vershinin M, Xu J, and Gross SP, 2008 Stepping, Strain Gating, and an Unexpected Force-Velocity Curve for Multiple-Motor-Based Transport Current Biology 18 1173. [9] Korn CB, Klupp S, Lipowsky R and Scwarz US, Stochastic simulations of cargo transport by procesive molecular motors. J. Chem. Phys. 131, 245107 (2009). [10] Shubeita GT, Tran SL, Xu J, Vershinin M, Cermelli S, Cotton SL, Welte MA, Gross SP 2008, Consequences of Motor Copy Number on the Intracellular Transport of Kinesin-1-Driven Lipid Droplets. Cell 135 1098-1107. The effects of uneven load sharing 15 ) m n ( h t g n e l n u r 105 104 103 Nb=0 MFM SM 4 5 1 2 3 Nf 4 5 ) s / m n ( y t i c o e v l 1200 1100 1000 900 800 3 Nf 1 2 Figure 2. Results for a single team of motors (Nb = 0). Run length and cargo mean velocity (inset) as functions of Nf for SM and MFM considering water viscosity. [11] Klumpp S, Lipowsky R 2005 Cooperative cargo transport by several molecular motors. Proc Natl Acad Sci USA 102, 17284-17289 [12] Muller MJI, Klumpp S, Lipowski R 2008 Tug-of-war as cooperative mechanism for bidirectional cargo transport by molecular motors. Proc Natl Acad Sci USA 102, 17284-17289 [13] Melanie J.I. Muller, Stefan Klumpp, Reinhard Lipowsky (2008), Motility States of Molecular Motors Engaged in a Stochastic Tug-of-War, J Stat Phys 133: 1059-1081. [14] Zhang, Y (2009) Properties of tug-of-war model for cargo transport by molecular motors, Phys. Rev. E 79, 061918 [15] Muller MJI, Klumpp S, Lipowsky R, 2010 Bidirectional Transport by molecular motors: Enhanced Processivity and response to external forces. Biophys. J. 98 2610-2618. [16] Erickson RP, Jia Z, Gross SP, Yu CC (2011) How Molecular Motors Are Arranged on a Cargo Is Important for Vesicular Transport. PLoS Comput Biol 7 (5): e1002032. [17] Kunwar A and Mogilner A, Robust transport by multiple motors with nonlinear force-velocity relations and stochastic load sharing. Phys. Biol. 7 (2010) 016012. [18] Jamison, DK, Driver, JW Rogers, AR, Constantinou, PE Diehl, MR. (2010) Two Kinesins Implications for Intracellular Transport Cargo Primarily via the Action of One Motor: Transport. Biophysical Journal 99 2967-2977. [19] Bouzat S, Falo F 2010. The influence of direct motor-motor interaction in models for cargo transport by a single team of motors. Phys.Biol. 7 046009. The effects of uneven load sharing a 1000 MFM SM, k=0.1 pN/nm SM, k=0.32pN/nm SM, k=0.75pN/nm Nb 2 3 4 0 1 b ) s ( e m i t n u r 80 70 60 50 40 30 20 10 ) s / m n ( y t i c o e v l 800 600 400 200 0 0 4.0 3.6 3.2 2.8 c t s r o o m d e g a g n e 2.4 0 y t i c o e v l 800 600 400 200 0 Nb 1 2 3 0 0 d 16000 1 2 Nb ) m n ( h t g n e 12000 8000 l n u r 4000 3 0 0 1 2 Nb 1 2 Nb 16 3 3 Figure 3. Cargo mean velocity (a), mean run time (b), mean number of engaged motors (c) and run length (d) as functions of Nb for systems with Nf = 3. Results for mean field and stochastic models under conditions of negligible viscous drag. In all the cases we consider kinesin-1 parameters both for forward and backward motors. The inset in panel (a) shows velocity results for Nf = 4 whose analogy with those for Nf = 3 indicates that the latter case (studied throughout the paper) is typical. [20] Zhang, Y (2011) Cargo transport by several motors, Phys. Rev. E 83 011909 [21] Campas O, Kafri Y, Zeldovich KB, Casademunt J, Joanny JF, 2006 Collective Dynamics of Interacting Molecular Motors Phys. Rev. Lett., 97 038101. [22] Brugu´es J, Casademunt J 2009 Self-Organization and Cooperativity of Weakly Coupled Molecular Motors under Unequal Loading Phys. Rev. Lett., 102 118104. [23] Hexner D, and Yariv Kafri Y, 2009. Tug of war in motility assay experiments Phys. Biol., 6 036016 (15pp). [24] Gur B. and Farago O 2010. Biased Transport of Elastic Cytoskeletal Filaments with Alternating Polarities by Molecular Motors Phys. Rev. Lett. 104, 238101. [25] Gillo D, Gur B, Bernheim-Groswasser A and Farago O 2009 Cooperative molecular motors moving back and forth Phys. Rev. E 80 021929. [26] Nedelec F 2002. Computer simulations reveal motor properties generating stable antiparallel microtubule interactions. J. Cell Biol. 158 1005 [27] Wollman R, Civelekoglu-Scholey G, Scholey J M and Mogilner A 2008. Reverse engineering of The effects of uneven load sharing a ) b n , f n ( e a t t s f o y t i l i b a b o r p 0.6 0.5 0.4 0.3 0.2 0.1 0.0 SM nb =0 nb =1 0 1 2 nf 3 b ) b n , f n ( e a t t s f o y t i l i b a b o r p 0.6 0.5 0.4 0.3 0.2 0.1 0.0 MFM nb =0 nb =1 c y t i l i b a b o r p state (2,1) state (3,1) 17 0.26 0.24 0.22 0.20 0.18 0 1 nf 2 3 0.0 0.4 0.6 0.2 k (pN / nm) 0.8 Figure 4. Probability of the states (nf , nb) for a system with Nf = 3 and Nb = 1 considering SM with k = 0.32pN/nm (panel a) and MFM (panel b). Panel c shows the probabilities of the states (2, 1) and (3, 1) as functions of k within the SM. In all the cases, the normalization of the probabilities is such that their sum over the eight possible states is equal to one. force integration during mitosis in the Drosophila embryo. Mol. Syst. Biol. 4 195. [28] Gillespie DT, 1977. Exact stochastic simulation of coupled chemical reaction. J. Phys. Chem. 81: 2340-2361 [29] Van Kampen N G 1992, Stochastic processes in physics and chemistry (North-Holland. Elsevier Science Publishers B.V.) [30] Carter NJ and Cross RA 2005 Mechanics of the kinesin step Nature, 435 308. [31] Schitzer MJ, Visscher K, Block SM 2000 Force production by single kinesin motors Nature Cell Biology, 2 718-723 [32] Hyeon C, Klumpp S, Onuchic JN 2009, Kinesin's backsteps under mechanical load, Phys. Chem. Chem. Phys., 11, 4899-4910. [33] Gross SP, Welte MA, Block S, Wieschaus EF 2000, Dynein-mediated Cargo Transport In Vivo: A Switch Controls Travel Distance. The Journal of Cell Biology, 148 945. [34] Levi V, Serpinskaya AS, Gratton E, Gelfand V 2006, Organelle Transport along Microtubules in Xenopus Melanophores: Evidence for Cooperation between Multiple Motors, Biphysical Journal 90, 318. [35] Mallik R, Petrov D, Lex SA, King SJ, Gross SP (2005) Nature 427:649652;King SJ, Schroer TA (2000) Nat Cell Biol 2:2024. [36] Bruno L, Salierno M, Wetzler DE, Desposito MA, Levi V (2011) Mechanical Properties of Organelles Driven by Microtubule-Dependent Molecular Motors in Living Cells. PLoS ONE 6(4): e18332 MFM SM, k=0.1 pN/nm SM, k=0.32pN/nm SM, k=0.75pN/nm The effects of uneven load sharing a 10000 ) m n ( f r 1000 b ) m n ( b r 100 250 200 150 100 50 0 0 1 2 Nb 18 3 Figure 5. Results for rf and rb as functions of Nb for the same systems analyzed in figure 3, considering MFM and SM. The effects of uneven load sharing 600 500 400 300 200 100 ) s / m n ( y t i c o e v l 0 0 MFM SM 10 a MFM SM b 1000 Nb=0 Nb=2 ) s ( e m i t n u r 1 2 Nb 3 8 6 4 2 0 0 ) s / m n ( y t i c o e v l 1 2 Nb 3 100 MFM SM 10 1E-5 1E-4 1E-3 0.01 0.1 g (pN s/ nm) 19 c 1 Figure 6. Influence of the viscous drag. Cargo mean velocity (a) and run time (b) as functions of Nb for fixed Nf = 3 considering γ = 9.4210−3pN s/nm for both models. This value of γ corresponds to a thousand times water viscosity and a cargo of 500 nm radio. c) Cargo mean velocity as a function of γ for Nf = 3 and different values of Nb for both models. All calculations within SM are for k = 0.32pN/nm. The effects of uneven load sharing Nf=3 MFM SM 1 2 3 Nb 4 5 Nf=3 MFM SM a ) s / m n ( v 1000 800 600 400 200 0 -200 -400 -600 -800 -1000 0 1000 c ) s ( e m i t n u r 100 10 0 1 2 Nb 3 4 b 5x103 0 -5x103 ) m n ( x -1x104 -2x104 -2x104 -3x104 -3x104 d ) 1 , n ( e a t f t s f o y t i l i b a b o r p 0.18 0.16 0.14 0.12 0.10 0.08 0.06 0.04 0.02 0.00 Nf= Nb=3 SM 20 100 time (s) MFM 0 50 Nf=3, Nb=1 MFM SM 0 1 2 nf 150 200 3 Figure 7. Kinesins and dyneins. Cargo mean velocity (a), cargo trajectories (b), mean run time (c) and probability of states with nb = 1 (d) for a cargo pulled by kinesin-1 and dynein considering set A parameters for dyneins and negligible viscosity conditions for both models. Set A: Fsb = 7.pN, F db = 3.18pN, ǫb = 0.27s−1, Πb = 1.6s−1, v0b = 1000nm/s and v1b = 6nm/s. The effects of uneven load sharing a 1000 Nf=2 Nf kinesins, Nb dyneins g =10-2 pNs/m m (water) MFM SM Nf=1 800 600 400 200 0 ) s / m n ( y t i c o e v l -200 -400 c y t i l i b a b o r p 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0.0 21 Nf=1 (kinesin) Nb (dyniens) viscosity: water MFM SM b 1000 ) s ( e m i t n u r 100 10 1 0 4 8 12 Nb 16 20 Nf=1, Nb=4 MFM, nf=1 SM, nf=1 d i s n e n y d f o r e b m u n e v i t c e f f e 0 1 3 4 2 nb 0 2 4 6 8 Nb 10 12 14 12 11 10 9 8 7 6 5 4 3 2 1 0 0.01 SM, kdyn=0.32pN/nm, kkin=0.32pN/nm SM, kdyn=0.08pN/nm, kkin=0.32pN/nm MFM 0.1 1 g (pN s / nm) 10 100 Figure 8. Kinesins and dyneins. Cargo mean velocity (a), mean run time (b) probability of different states (c) and effective number of dyneins leading to vanishing mean cargo velocity (d) for a cargo pulled by kinesin-1 and dynein considering set B parameters for dyneins and negligible viscosity conditions for both models. Set B: Fsb = 1.1pN, F db = 0.75pN, ǫb = 0.27s−1, Πb = 1.6s−1, v0b = 650nm/s and v1b = 72nm/s.
1606.01392	1	1606	2016-06-04T16:49:58	The Physics of Biofilms -- An Introduction	[ "physics.bio-ph", "cond-mat.soft", "q-bio.CB" ]	Biofilms are complex, self-organized consortia of microorganisms that produce a functional, protective matrix of biomolecules. Physically, the structure of a biofilm can be described as an entangled polymer network which grows and changes under the effect of gradients of nutrients, cell differentiation, quorum sensing, bacterial motion, and interaction with the environment. Its development is complex, and constantly adapting to environmental stimuli. Here, we review the fundamental physical processes the govern the inception, growth and development of a biofilm. Two important mechanisms guide the initial phase in a biofilm life cycle: (\emph{i}) the cell motility near or at a solid interface, and (\emph{ii}) the cellular adhesion. Both processes are crucial for initiating the colony and for ensuring its stability. A mature biofilm behaves as a viscoelastic fluid with a complex, history-dependent dynamics. We discuss progress and challenges in the determination of its physical properties. Experimental and theoretical methods are now available that aim at integrating the biofilm's hierarchy of interactions, and the heterogeneity of composition and spatial structures. We also discuss important directions in which future work should be directed.	physics.bio-ph	physics	The Physics of Biofilms – An Introduction Max Planck Institute for Dynamics and Self-Organization, Am Fassberg 17, 37077 Göttingen, Germany Marco G. Mazza∗ (Dated: June 7, 2016) Biofilms are complex, self-organized consortia of microorganisms that produce a functional, pro- tective matrix of biomolecules. Physically, the structure of a biofilm can be described as an entangled polymer network which grows and changes under the effect of gradients of nutrients, cell differen- tiation, quorum sensing, bacterial motion, and interaction with the environment. Its development is complex, and constantly adapting to environmental stimuli. Here, we review the fundamental physical processes the govern the inception, growth and development of a biofilm. Two impor- tant mechanisms guide the initial phase in a biofilm life cycle: (i) the cell motility near or at a solid interface, and (ii) the cellular adhesion. Both processes are crucial for initiating the colony and for ensuring its stability. A mature biofilm behaves as a viscoelastic fluid with a complex, history-dependent dynamics. We discuss progress and challenges in the determination of its physi- cal properties. Experimental and theoretical methods are now available that aim at integrating the biofilm's hierarchy of interactions, and the heterogeneity of composition and spatial structures. We also discuss important directions in which future work should be directed. I. INTRODUCTION Biofilms are among the most ancient evidence of life on Earth: they appear as fossils of microbial mats and stro- matolites from western Australia dating back to about 3.5 billion years ago [1, 2]. Thus, it should come as no surprise that biofilms are the most widespread form of life [3]. They form extremely diverse and complex structures, are capable of colonizing almost every envi- ronment, and have evolved a vast arsenal of biological re- sponses to environmental stimuli. Today we know that all three domains of life can produce biofilms: bacteria [4], archaea [5], microalgae [6] and fungi [7, 8] all produce biofilms. But the understanding that most microorgan- isms live in aggregates rather than in a planktonic state has only recently emerged. Antonie van Leeuwenhoek observed "animalcules" in the plaque of his own teeth under a microscope of his fabrication. His report in 1676 marks the discovery of biofilms [9]. However, only in the first half of the twentieth century with the work of Henrici [10] and Zobel [11] was the import of biofilms re- ally appreciated by the scientific community, and finally the word biofilm was first used in 1981 [12], which marks the recognition of its function and organization. Biofilms are structured, self-organized communities of microorganisms that synthesize a protective matrix and adhere to each other and/or to an interface [13, 14]. They can be populated by a single species but more often mul- tiple species are present. The level of complexity and specialization present in biofilms brings to mind the anal- ogy to a large, bustling city [15] with different levels of organization [16]. Biofilms can grow on solid surfaces in the presence of water virtually anywhere, for exam- ple on pebbles in a river bed (the periphyton), in deep- sea hydrothermal systems [17] and vents [18], or on boat ∗ [email protected] hulls; biofilms can also grow at the air-water interface of limnic or marine environments [19]. Like the inhabitants of a city, the microorganisms in a biofilm can modify their surroundings. Cellular metabolism can either in- crease the porosity of geologic media by the dissolution of minerals, or reduce the porosity by precipitating sec- ondary minerals or by clogging the pores with biomass. Microbial activity can change the electrochemical prop- erties, surface roughness, elastic moduli, stiffness, as well as seismic and magnetic properties of minerals through the precipitation of bacterial magnetosomes. In general, biofilms play a role in so many geophysical processes that a new discipline is now emerging: biogeophysics [20]. Biofilms play a role also in human activities. Their growth has an enormous impact in biomedical sci- ences [21]. Microbial ecosystems can grow on the surface of teeth and on open wounds; they can infest the res- piratory mucous and the airways of patients with cystic fibrosis pneumonia. Pseudomonas aeruginosa, for exam- ple, lives normally (and harmlessly) on our skin but if it enters the blood circulation of immunocompromised individuals it can infect organs of the urinary and res- piratory system, bones and joints. Medical devices such as catheters, prosthetic heart valves, cardiac pacemakers, skull implants, cerebrospinal fluid shunts and orthopedic devices can harbor pathogens [4, 22], and due to the inti- mate contact of these medical tools with the human body biofilms can grow and trigger virulent infections. Biofilms cause billions of dollars in damage to metal pipes in the oil and gas industry, and metal structures in water treatment plants. Sulfate-reducing bacteria, for example members of the Acidithiobacillus genus, trans- forms molecular hydrogen into hydrogen sulfide which, in turn, produces sulfuric acid that corrodes metal surfaces causing catastrophic failures [23–25]. Biofilms have, however, also found useful employment in environmental biotechnology. They are used in a well-established technique for wastewater treatment [26– 28], or for in situ immobilization of heavy metals in 6 1 0 2 n u J 4 ] h p - o i b . s c i s y h p [ 1 v 2 9 3 1 0 . 6 0 6 1 : v i X r a soil [29, 30]. Biofilms allow to process large volumes of liquids, and they naturally grow by converting organic materials. Furthermore, microorganisms (typically bac- teria and fungi) can be used for microbial leaching, that is, as a way to extract metals from ores [31, 32]. Copper, uranium and gold are examples of metals commercially recovered by microorganisms [33]. Numerous experimental studies have determined that biofilms have a heterogeneous structure and composi- tion [34–40] which vary greatly among different species. But we also know that they share some fundamental el- ements. Microorganisms colonizing an interface produce a mass of biopolymers, collectively termed extracellular polymeric substance (EPS), that provides protection and structural stability to the cells. The EPS also performs numerous functions specific to the biofilm state, such as creating its three-dimensional architecture, protection against physical, chemical and biological agents, provid- ing an external digestive system, facilitating cell-cell sig- naling and horizontal gene transfer [41]. The EPS matrix represents up to 90% of the dry mass of the biofilm (the remaining being the cells) and is com- posed of polysaccharides, proteins, humic acids, DNA, lipids, and remnants of lysed cells [3]. (However, there are exceptions: some strains of Staphylococcus epidermidis and Staphylococcus aureus produce smaller amounts of EPS.) Figure 1 shows a sketch of a biofilm with its pri- mary elements. Below, we briefly describe these compo- nents and highlight their functions in the biofilm. · Polysaccharides are long, Diverse microscopy methods linear or branched, polymeric chains that are ubiquitously found in biofilms, and represent their largest component [42, 43]. The num- ber and diversity of these molecules is stunning. Very little can be said in general about them as the num- ber of chemical and structural possibilities is virtually infinite [44, 45]. Gram-negative bacteria typically pro- duce polyanionic polysaccharides, or in some cases neu- tral ones, due to the presence of uronic acids or ketal- linked pyruvates; some chemical linkages (1, 3- or 1, 4-β- linked hexose) provide backbone rigidity to the polymers, while other linkages produce more flexible polymers [46]. show that polysaccharides form a complex network of fine strands linking cells to each other and to the substratum. Polysaccharides are involved in most processes that take place in a biofilm [3]: (i) adhesion; for example, the polysaccharides Pel and Psl (named after the operons involved in pellicle formation and polysaccharide synthe- sis locus, respectively [49]) are essential for the adhesion of Pseudomonas aeruginosa to a variety of substrata and they provide redundant mechanisms for this task [50–52]. (ii) Structural stability and cohesion; the polysaccharides of algae form rigid structures because the binding (chela- tion) of Ca2+ or Sr2+ ions forms cross-linked polymer networks [46, 53]. Because bacterial polysaccharides are often acetylated, the interaction between polymers is re- duced and so is the resulting network-forming ability [46]; for example, Pseudomonas fluorescens requires an acety- [47, 48] 2 lated form of cellulose for the effective colonization of the air-liquid interface [54], and N-acetylglucosamines are crucial for intercellular adhesion in the biofilm of Staphy- lococcus epidermidis [55, 56], a major cause of nosocomial infections. (iii) Hydration; polysaccharides can bind wa- ter molecules (for example, hyaluronic acid can bind the considerable amount of 1 kg of water per 1 g of sac- charide [46]; however, most molecules will probably bind less). Because the polysaccharides form an entangled polymer network they are subject to an osmotic pressure Π such that will swell with water until the shear modulus of the network G ≈ Π [57–59]. A hydraulic decoupling between unsaturated soil bacteria and the vadose zone has been hypothesized to protect the cells from cycles of wetting and drying events [60]. (iv) Storage of nutrients; both the polysaccharide network itself and absorbed or- ganic matter serve as a source of carbon during periods of nutrient deprivation [61]. (v) Protection from toxic ions; the exopolysaccharides can adsorb toxic ions such as Cd, Zn, Pb, Cu, and Sr [47, 62]; for example, Pseudomonas aeruginosa in a biofilm is up to 600 times more resistant to Zn, CU and Pb than in the planktonic state [63]. (vi) Antibiotic protection; Pseudomonas aeruginosa infecting cystic fibrosis patients can change to a mucoid phenotype characterized by increased production of the polysaccha- ride alginate, but the biofilm contains also Pel and Psl. All three components have been found to have a pro- tective role against antibiotics [64–66]. (vii) Reservoir of enzymes; enzymes are stored, accumulated and sta- bilized by the polysaccharide network [67]; for example, it has been suggested [68] that epilithic biofilms of mi- croorganisms growing on river beds (a highly changeable environment) benefit from the accumulation of enzymes because, firstly, the products of enzymatic activity are readily available to the cells, secondly, they may addi- tionally trigger new enzymatic activity in adjacent cells, and, thirdly, new generations do not need to spend en- ergy on enzyme synthesis. (viii) Sink for excess carbon; when supplied with excess C, the production of polysac- charides is greatly enhanced [69] · Extracellular DNA has been found not to be merely debris left from lysed cells but an actively pro- duced component of the EPS that plays a crucial role in the formation and structural integrity of the biofilms of Variovorax paradoxus and Rhodococcus erythropolis [70]. Extracellular DNA functions as an intercellular connec- tor in Pseudomonas aeruginosa's biofilms [71]; also, if these biofilms are younger than 60 hours they can be dissolved by enzymatic degradation of the extracellular DNA [72]. The Gram-positive Bacillus cereus employs extracellular DNA as an adhesin [73]. Recently, a novel function of extracellular DNA has been recognized: an- timicrobial activity. By binding and sequestering cations, extracellular DNA induces physical alterations in the bacterial outer membrane [74]. · Lipids or biosurfactants modify the hydrophobic character of microbial cells and therefore can modify their ease of adhesion to surfaces. For example, lipopeptides 3 FIG. 1. (a) Pictorial representation of a biofilm produced by a community of microorganisms. Different cellular shapes represent different species of microorganisms coexisting in the same biofilm, while different colors represent phenotypic variations within the same species. (b) Zoomed-in view of the main components of a biofilm, underlying the complexity of its physico-chemical properties: the extracellular polymeric substance and cells. We highlight (i) the cellular membrane, (ii) proteins embedded in the membrane, (iii) pili, (iv) extracellular DNA, (v) polysaccharides, (vi) proteins, (vii) lipids. Inspired by a sketch in Ref. [3]. produced by Bacillus subtilis change the hydrophobicity of the cells depending on the initial level of hydropho- bicity [75]. These functions are also used to protect the biofilm from invading microorganisms. In the late stages of its biofilm, Pseudomonas aeruginosa produces rham- nolipids that can disrupt biofilms formed by Bordetella bronchiseptica [76] or inhibit the attachment of a variety of bacteria and yeasts [77]. Bacillus subtilis and licheni- formis produce biorsurfactants that selectively prevent the adhesion of pathogenic cells [78]. · Proteins are a large component of the EPS and have diverse functions. As enzymes they form an external digestive system for the microorganisms in the biofilm as they break down biopolymers to low-molecular-mass products that can be utilized by the cells [3]. Enzymes can degrade the EPS matrix [79, 80] or protect against oxidizers [81]. Finally, non-enzymatic proteins contribute to the structural stability of the EPS network by attaching the cellular surface to the polysaccharides. Lectins are one such class [82]. For example, lectins contribute to thicker biofilms in different environmental conditions in Pseu- domonas aeruginosa [83]; the presence of a plant lectin can greatly influence the biofilm of a rhizobium [84]. In addition to their composition, also the temporal de- velopment of biofilms is structured. Five stages have been recognized in the life-cycle of a biofilm [39]: (i) the ini- tial attachment of cells to a surface or substratum; (ii) production of EPS and irreversible attachment; (iii) de- velopment of the biofilm architecture; (iv) maturation of the structure and composition of the biofilm; (v) dis- persal of individual cells from the biofilm that revert to the planktonic phenotype to find a new niche and that close in this way the life-cycle of the biofilm. During dispersal the EPS matrix immobilizing the cells is bro- ken down. This last stage of the biofilm's life-cycle al- lows the colonization of new interfaces and it may be necessary when the microenvironment becomes unfavor- able to sessile microorganisms. Dispersal is a highly com- plex process [53, 85] involving the production of enzymes to degrade the EPS network [86], production of surfac- tants [87], and induction of motility [88]. For more details regarding the important phase of dispersal we refer the reader to a useful review [89]. The possibility that the vast majority of microorgan- isms can form biofilms points to a strong evolutionary incentive for this cooperative behavior. Biofilms grow- ing in different environments show strikingly similar re- sponses to similar physical stimuli. As Fig. 2 shows, fast-moving waters produce similar morphologies called "streamers" in very different environments. Quiescent waters induce instead biofilms with more rounded mor- phologies [4]. These facts point to convergent survival strategies within a biofilm. Indeed, social interactions in microorganisms have been recognized [90]. The advan- tages of a biofilm mode of life consists not only of the protection from toxins, dehydration, antibiotics, starva- tion, but also of a synergetic consortium that enhances immensely the chances of survival by sharing resources and allowing differentiation, for example. For all the reasons mentioned above, it is not surprising that the study of biofilms has attracted enormous interest in microbiology, pharmacology, medicine, industry, and in more recent years also in physics. In fact, the forma- 4 FIG. 2. Examples of biofilms growing in different environments. (a-c) Biofilms growing in hydrothermal hot springs from the Biscuit Basin thermal area, Yellowstone National Park, USA; (d) Gardener River, Yellowstone National Park, USA; (e) Hyalite Creek, Bozeman, Montana, USA; (f-g) Pseudomonas aeruginosa PAN067 biofilm grown in a flow cell with a flow of 1 m/s. Biofilms growing in stagnant waters tend to form round, mushroom-like structures, as in (a) and (d), while biofilms growing under hydrodynamic flow form ripples, as in (c) and (f), and streamers, as in (b), (e) and (g). Images courtesy of Dr. P. Stoodley. Reprinted with permission from Ref. [4]. tion, growth, organization and structure of a biofilm are all processes that need to be understood from a physical point of view. In this Review we explore a selection (far from being exhaustive) of key physical processes taking place in a biofilm. The first important step is the motil- ity. Different kinds of motilities occur in two phases of the biofilm life-cycle: stage (i), the attachment, and stage (v), the dispersal. Next, the process of adhesion is de- cisive for the formation and growth of a stable biofilm. Finally, the mature biofilm resulting from myriad inter- actions and history-dependent dynamical responses can be viewed, on a macroscopic level, as a biomaterial, that is susceptible of physical characterization. The topic we address in this review has a multitude of aspects and spans such a broad range of disciplines (from biochemistry and genetics to hydrodynamics and rheology) that we cannot hope to give a complete depic- tion of the current knowledge about biofilms. However, we have made efforts to provide a rather general picture that at least points to the most eminent physical aspects. At times, this work might appear as pedagogical. This is intended. We hope in fact that experts in one dis- cipline will find the discussions of other disciplines con- nected to biofilms helpful. For further study we direct to more detailed reviews, such as Ref. [91] for the concept of multicellularity, and Ref. [92] for the self-organization of microorganisms. One glaring aspect of biofilm that we neglect is the genetic and regulatory network of signals that controls the biofilm's formation, maturation and dis- persal. For these topics we recommend the specialized reviews [13, 85, 89, 93]. This work is organized as follows. In Sec. II we discuss various mechanisms of cellular motility, from the prob- lem of swimming at low Reynolds numbers to surface motility. Section III describes the fundamental physical processes that allow cells to adhere to solid surfaces and thus initiate a biofilm. In Sec. IV we discuss the vis- coelastic behavior of biofilms as materials and the main physical concepts used in the literature. In Sec. V we describe modern advances in the experimental investiga- tion of biofilms propelled by the fields of microfluidics and nanofabrication. In Sec. VI we review the theoretical and computational models of biofilm growth and dynamics. Finally in Sec. VII we summarize our conclusions. II. MOTILITY Why is motility relevant to biofilms? The first, obvious answer is that the approach of microorganisms to a sur- face is the first step to the constitution of a biofilm. This corresponds to stage (i) in the life-cycle of a biofilm. For example, prior to attachment and initiation of a biofilm Vibrio cholerae scans a surface by swimming in its close proximity, using two strategies called roaming and orbit- ing, which differ in the radius of gyration of the trajecto- ries [94]. Swimming is also relevant in the last stage of a biofilm's life-cycle: the dispersal. Cells of Pseudomonas 5 aeruginosa swim away from the internal regions of the biofilm into the bulk liquid to regain access to fresh nu- trients [95]. It is often claimed that once formed biofilms are communities of sessile microorganisms. A new pic- ture has emerged where motility, at least in some species, plays a significant role in the biofilm's structure forma- tion [96, 97], with spatial and temporal heterogeneities in phenotypic differentiation and organization [96, 98]. It must be said, however, that living organisms exhibit a stunning variety of behaviors that refuse to be easily cat- egorized. In fact, non-swimming species, such as Staphy- lococcus epidermidis and Staphylococcus aureus can also form biofilms. Thus, swimming motility is by all means not a necessary condition for biofilm formation. The in- evitable generalizations sometimes present in this work should be understood from a physical point of view and not biological: that the physical processes discussed be- low have been shown to take place in some species does not imply that all biofilm-forming microorganisms dis- play the same characteristics, but rather that they be- long to the "toolbox" of physical mechanisms available to some species and/or phenotypes, and, as our knowl- edge of biofilms expands, one should consider that these physical processes might be at play. Six different kinds of motility have been identified among 40 bacterial species belonging to 18 different gen- era [99]: swimming, swarming, gliding, twitching, sliding and darting (see Fig. 3 for a sketch of the most important motility modes). Except for swimming, all other modes are associated to the presence of a surface. Swimming at or near a surface is actuated through one or more flagella. Let us consider one of the most studied motile bacteria: Escherichia coli. It exhibits on average six flagella, 5 to 10 µm long and 20 nm thick, randomly distributed around its body. The flagellum is composed of a basal body (embedded in the cellular membrane), a hook and a filament, that is, a left-handed helix with a pitch of about 2.5 µm and a diameter of about 0.5 µm [100]. When the flagellar motor, driven by the proton motive force, rotates counterclockwise (CCW) a helical wave travels along the filament. When all flagella rotate CCW mechanical and hydrodynamical forces group the flagella in a single bun- dle that propels the bacterium in roughly straight tra- jectories with speeds up to 40 µm/s. While the bundle rotates CCW at about 100 Hz the cell body rotates clock- wise (CW) at about 25 Hz [101] because of conservation of angular momentum. When one or more flagellar mo- tors rotate CW those filaments undergo a polymorphic transition to a right-handed helix. The bundle of flag- ella divides and a random reorientational motion called "tumbling phase" results. The tumbling phase allows the cells to change swimming direction, and it is regulated by the chemotactic signal transduction network [102]. Escherichia coli's flagellum has a torsional spring con- stant kθ ≈ 4 × 10−12 dyne cm rad−1 [103], but only up to twist angles of 100◦; for larger twist angles a sec- ond regime appears with an order of magnitude stiffer response [103]. One possible explanation for these two A. Swimming at low Reynolds numbers 6 We now discuss the general physical mechanism that produces propulsion in microorganisms that posses flag- ella or cilia. On general grounds, the motion of a body immersed in an aqueous environment, that is, a Newto- nian fluid, is described by the Navier–Stokes equations (cid:18) ∂~v ∂t ρ (cid:19) + ~v · ∇~v = −∇p + η∇2~v , (1) ∇ · ~v = 0 , (2) where ρ is the fluid density, ~v the flow velocity, p the pres- sure, η the viscosity, t is time, and "·" is the symbol of scalar product. Physically, Eq. (1) represents conserva- tion of linear momentum in the fluid, and Eq. (2) the con- dition of incompressibiliy of the fluid. The presence of a body immersed in the fluid is incorporated by the bound- ary conditions necessary to solve Eq. (1)-(2). The no-slip boundary condition (valid for viscous fluids) states that at the boundary S of the body ~vfluid(S) = ~vsolid(S). Solv- ing the Navier–Stokes equations means obtaining expres- sions for ~v and p that satisfy Eq. (1)-(2) and the boundary conditions. From this knowledge, the stress tensor σ can be cal- culated. For a fluid the stress tensor can be decomposed as σ = −pI + τ, where I is the unit tensor, that is, into a term corresponding to the pressure stresses, which are isotropic, and a second term called deviatoric stress ten- sor τ that includes viscous (shear) or other stresses. For a Newtonian fluid the deviatoric stress depends linearly on the instantaneous values of the velocity gradient, so that one can write σ = −pI + η(cid:2)∇~v + ∇~v T(cid:3) , (3) where the superscript 'T' indicates matrix transposition. Once σ is found, the total hydrodynamic force ~F and torque ~τ acting on the swimming body can be derived from integrals over its surface ~F = σ · ~n dS , ~τ = ~r × σ · ~n dS (4) S S where ~n is a unit vector normal to the surface. Equation (1) represent a balance of inertial forces (the left-hand side) and viscous forces (the term η∇2~v), thus it is natural to find a way to quantify the relative impor- tance of inertial to viscous forces. The Reynolds number R = ρ U L/η, where U is the typical velocity and L the characteristic length scale of the swimmer, gives a dimen- sionless measure of this balance. For a typical microor- ganism such as Escherichia coli (L ≈ 2 µm) swimming in water (ρ ≈ 103 kg/m3, η ≈ 10−3 Pa·s) with a typical speed U ≈ 20 µm/s the Reynolds number R ≈ 10−5. It is then clear that to study the swimming of microorgan- isms we can neglect the inertial term in Eq. (1). In this limit the flow obeys the Stokes equations ∇ · ~v = 0 . ∇p = η∇2~v , (5) Z Z FIG. 3. Pictorial representation of different motility modes of microorganisms. (a) Swimming in a bulk fluid by means of single polar flagellum; (b) bacteria in the hyperflagellated state swarming on a solid surface; (c) twitching by means of type IV pili; (d) gliding by means of focal adhesion complexes. regimes of the flagellar compliance is that the hook ex- hibit a soft component, which dominates the initial phase of the twist, and the filament has a stiffer compliance, which dominates larger twist angles. This mechanism has the advantage of removing discontinuities in the motion of the motor, whose dynamics is intrinsically stochastic because it is powered by the passage of protons, which is a Poisson process. Hence the waiting times between events is an exponentially distributed random quantity [104]. A microorganism cannot swim like a fish for a sim- ple physical reason: the viscous drag dominates the mo- tion [105]. Motile cells must instead adopt a different strategy that uses drag forces to their advantage. In the next section we discuss the hydrodynamic fundamentals of bodies swimming in a viscous fluid. This is a fas- cinatingly vast topic. For more details we recommend specialized works on the general problem of swimming [105–107], or swarming [108]. Because the Stokes equation is linear and because of Eq. (4) the forces acting on the swimming body are di- rectly proportional to the flow velocities [109]. Because the displacement of a solid body can be decomposed at any instant of time as the superposition of a translation with velocity ~V (t) and a rotation with angular velocity ~Ω(t) (this is the Mozzi–Chasles theorem), the instanta- neous local velocity of a point can be written as ~v = ~V + ~Ω × ~r . (6) The hydrodynamic force and torque then obey the equa- tions Fi = −η(AijVj + BijΩj) , τi = −η(CijVj + DijΩj) . (7) (8) Physically, the matrices Aij and Dij represent the cou- pling of forces to translations, and of torques to rotations, respectively, while the matrices Bij and Cij represent cross-term couplings between forces and rotations, and between torques and rotations. Dimensionally, the coef- ficients of Aij have dimensions of length, Bij and Cij of an area, and Dij of a volume. Because of very general arguments [109], which do not depend on the shape of the moving body, these matrices obey the relations Aij = Aji , Dij = Dji , Bij = Cji , (9) thus, the matrices Bij and Cij are always the transpose of each other. This means that there is a reciprocity principle for the motion in viscous fluids: the force on a rotating body and the torque on a translating body have the same coupling coefficients. A moving body with the symmetry of a cube or a sphere will have Bij = 0, and Aij = λδij, where δij is the unit matrix. Physically, the drag force on the body will be collinear with the direction of motion, ~V . But, for a general shape, the matrix Aij will not be proportional to the unit matrix; thus, there will be drag anisotropy: ~F · ~V < 0, that is, force and velocity are not collinear and the sign indicates the energy loss due to viscous dissipa- tion [109]. For example, for a prolate ellipsoid of revolu- tion with one axis much larger than the other, '1 (cid:29) '2, the matrix A = Ake⊗ e+ A⊥(I− e⊗ e), with A⊥ ≈ 2Ak, where e is the unit vector (versor) in the direction of the long axis of the ellipsoid, ⊗ the tensor product. Physi- cally, this means that the drag force normal to the long axis is twice as large than along it. Drag anisotropy is the key physical effect that makes locomotion possible at low Reynolds numbers. Consider for example a thin filament, like a bacterial flagellum, deformed by a periodic wave-form. For simplicity we as- sume that the cell moves along the x axis, and each point on the flagellum moves only in the normal direction to the x axis. A short segment of the filament can be approxi- mated by a straight, thin rod that experiences a viscous drag ~f = ~fk + ~f⊥ = −ξk~vk − ξ⊥~v⊥, where ξ⊥ ≈ 2ξk, 7 and ~vk and ~v⊥ are the components of the velocity par- allel and perpendicular to the segment tangent, that is, vk = v cos θ, v⊥ = v sin θ. The drag forces have compo- nents along the x axis: fkx = fk sin θ, f⊥x = f⊥ cos θ. The total force along the x axis, or the propulsion force then results fprop = (ξk − ξ⊥)u sin θ cos θ [105]. Thus, a periodic change in shape and in the direction of beat- ing can produce a net propulsive force in the direction perpendicular to the beating. Similar calculations show that a helical filament rotating about its axis generates a net propulsive force proportional to the rotational ve- locity [109]. The Stokes equation (Eq. (5)) is linear and time- independent. An important consequence of this fact is that if the velocity of motion is reversed, the propulsive force also changes sign. Thus, a scallop moving at vanish- ing Reynolds numbers cannot have any net displacement. This is Purcell's famous scallop theorem [110]. Note that the theorem is valid only if the sequences of deformations of the moving body is time reversible, nor does it apply close to another surface. Another consequence of a vanishing Reynolds number is that the dynamics is overdamped: the rate of change of momentum, that is, the acceleration is negligible. New- ton's laws then become simple balances of forces and torques, ~Ftot = 0, ~τtot = 0. Thus, the swimming problem of a body immersed in a viscous fluid must be described by two opposite forces, a force dipole. The flow field at position ~r created by a swimmer located at the origin of a coordinate system is (cid:2)3(r · e)2 − 1(cid:3) r , ~v(~r) = 'f 8πηr2 (10) where e is the versor in the swimming direction, r = ~r, r ≡ ~r/r, ' the dipole length, and f its force. Depending on the sign of f the swimmer is called "pusher" (f > 0), such as Escherichia coli, or "puller" (f < 0), such as algae of the genus Chlamydomonas, which swim with characteristic "breast stroke" movements of the flagella. If we now consider two microorganisms swimming in the same environment, each will be affected by the flow field generated by the other. This interaction transmitted by the fluid is called a hydrodynamic interaction. Equa- tion (10) has a number of consequences for the mutual interaction of two swimmers: (i) two side by side push- ers attract each other, while two side by side pullers repel each other; (ii) two pushers aligned along the swimming direction repel, while two similarly aligned puller attract each other. Calculations up to the octupole order [111] show that to leading order the hydrodynamic forces and torques between two swimmers moving in the directions e1, e2 exhibit a nematic symmetry, that is the interac- tions are invariant for changes e1 → −e1, e2 → −e2. Dense suspensions of swimmers are unstable to long wavelength fluctuations [112, 113], leading to the so- called "bacterial turbulence" [114, 115]. There is how- ever an ongoing debate regarding the role of hydrody- namic interactions [116–118]. On the one hand, a simple model that includes only deterministic, short-ranged in- teractions can reproduce the velocity structure functions of dense suspensions of Bacillus subtilis [119], and experi- ments on Escherichia coli have shown that hydrodynamic interactions are washed out by the stochasticity intrinsic in the bacterial motion [120]. On the other hand, exper- iments on confined suspensions of B. subtilis show that long-range hydrodynamic interactions drive the bacterial collective behavior [121]. As a swimming microorganism approaches a solid sur- face it will be increasingly affected by the hydrodynamic interactions with the boundary. Four effects have been identified [105]: (i) the swimming speed increases near a solid boundary [122, 123] because the drag anisotropy increases as the distance to the surface decreases. (ii) The swimming trajectory will change shape. For exam- ple, the flow field generated by Escherichia coli away from any boundary has a cylindrical symmetry, but close to a solid boundary it loses this property. The rotation of the flagellum produces a force normal to it, while an equal and opposite force is produced also on the cell body. The resulting torque is the physical origin of the circu- lar, clockwise motion of Escherichia coli (when viewed from the same side of the cell with respect to the sur- face) [124]. Circular trajectories were also seen in bacte- ria with a single flagellum (monotrichous) such as Vibrio alginolyticus [125, 126] and Caulobacter crescentus [127]. (iii) Solid walls attract and reorient cells. Mathemati- cally, a solid wall with no-slip boundary conditions can be replaced by a virtual, image cell on the opposite side of the wall, and the two cells will interact hydrodynam- ically. Thus, a pusher aligns parallel to the wall, and once aligned is attracted to the wall. The higher den- sity of cells close to solid surfaces is experimentally ob- served [128–130]. Pullers instead align normal to the walls, and therefore tend to swim against them. (iv) The range of the hydrodynamic interactions decreases. Because the image cells represent an opposite flow field, this decays faster in space; for example, the flow field de- scribed by Eq. (10) decays as r−2 in an infinite fluid but in the presence of a wall it can decay as r−3 or r−4 [131]. Effectively, walls screen the hydrodynamic interactions among swimmers. We have already discussed (see Sec. II) the relevance of swimming motility to biofilms. However, we note that more work is certainly necessary to firmly establish the connection of the swimming state with the chemical, bi- ological and physical forces acting within a biofilm. B. Swarming Swarming is the multicellular motion of bacteria on a surface powered by flagella [132]. Under the appropriate circumstances, some microorganisms undergo a pheno- typic change and their motion is characterized by spatial and temporal correlations that include billions of individ- uals [133]. At present only three families of bacteria are 8 known to exhibit swarming: the phylum Firmicutes, and the classes Alphaproteobacteria and Gammaproteobac- teria [132]. However, it is likely that the laboratory con- ditions select against swarming behavior, so that it might be more widespread than currently thought. The concen- tration of agar on a plate, for example, is important to induce or inhibit swarming. A concentration above 0.3% inhibits swimming and forces the cells to move on the surface, but concentrations above 1% stop the swarm- ing of most species. A commonly used concentration is 1.5%. Swarming is a complex phenomenon that requires many concurrent mechanisms. The most important is the increase of the number of flagella. Although some species with a single polar flagellum can both swim and swarm, in most species contact with a surface induces the synthesis of lateral flagella randomly distributed on their bodies [134], which are exclusively used for swarm- ing [135, 136]. The bacteria are said to become hyper- flagellated. The polar and lateral flagellar systems are driven by separate motors, encoded by different genes, and subject to different regulatory networks [136–138]. Swarming behavior also requires a rich medium be- cause the synthesis of flagella and the motion in a highly viscous environment have a high metabolic cost [136], and therefore a medium which supports high growth rates is necessary [139]. Another (almost) necessary condition for swarming is the production of surfactants by the bacteria that fa- cilitate the spreading of the colony on the solid surface and protects the cells against desiccation [140]. Muta- tions inhibiting the synthesis of surfactant stop swarm- ing, but it can be resumed if exogenous surfactants are added [141, 142]. Synthesis of surfactants is regulated via quorum sensing [143]. Because surfactants are effec- tive in large quantities, it is reasonable to conclude that this quorum sensing mechanism evolved to avoid waste- ful synthesis by isolated cells. Interestingly, Escherichia coli swarms without any surfactant and it is not known what substance promotes its surface motility [132]. The collective character of swarming is evident in the formation of rafts, dynamical groups of closely associated bacteria with strong orientational correlations and that move together [141, 144]. Interactions among flagella may be responsible for the formation and maintenance of the rafts [145]. For example, in swarming colonies of Escherichia coli the flagellar bundles may even align at 90◦ with respect to the cellular body while remaining intact, and they remain cohesive even during collisions with neighboring cells [145]; the cells of Proteus mirabilis in the swarming state grow twenty times larger than nor- mally, and move only when they are in contact with other individuals [146]; in a raft their flagella are interwoven into helical structures that stabilize the collective motion of adjacent cells [147]. The dynamics of cells in the rafts is out of thermal equilibrium. Clusters exhibit persistent reorganization, they split and merge with other clusters. Individuals constantly leave one raft to join another. Clusters of wild-type Bacillus subtilis have many different sizes, and the larger the colony, the more likely it is to find a large cluster. The probability distribution function of the cluster size P(n) obtained from the observations is well described by a power law with an exponential trun- cation, P(n) ∼ n−αe−n/nc, where α = 1.85 indepen- dently of the bacterial density, while nc does depend on density [148]. That the same dependence of P(n) is predicted by theoretical models and found in fish and African buffaloes [149] points to a general mechanism in the collective motion of living organisms. The fluc- tuations of the bacterial density h∆Ni in a given re- gion grows with the number of cells N. However, while h∆Ni ∼ √ for a system in thermal equilibrium it is expected that in the swarming colony h∆Ni ∼ N β, where β ≈ 0.75 [148]. Density fluctuations with a scaling stronger than the N 1/2 law are termed "giant number fluctuations" [150, 151]. N, A spectacular feature of many swarming bacterial species is the formation of patterns. As the colony spreads on the solid substratum a pattern of concentric rings, dendrites or vortices emerges on a scale orders of magnitude larger than the cell size. Different patterns are largely a result of environmental conditions [152, 153]. Physical models of the pattern formation must inevitably consider a simplified representation of the system, how- ever, they capture the main dynamic of the relevant degrees of freedom. For example, a model using two advection-diffusion-reaction equations that represent the phenotypic cycle of differentiation-dedifferentiation from swimming to swarming state and vice versa can re- produce both the formation of concentric rings and branched structures [154]; other reaction-diffusion equa- tions predict the phase diagram of continuous and pe- riodic expansion of the colony [155]; an approach that includes lubrication and chemotactic signaling captures many aspects of the branching pattern [156]. More de- tailed equations that include nutrient dynamics, nucle- ation theory concepts for the cells, and individual-based motion and growth reproduce the knotted-branching pattern of Bacillus circulans [157]. A particle-based model reproduces vortices in Bacillus circulans by in- cluding attractive or repulsive rotational chemotactic re- ~vi~vi × (~vi × ∇C), where C is sponse through the term 1 the concentration of chemotactic substance [158]; how- ever, the role of chemotaxis in swarming remains un- clear [132]. The "communicating walkers model" cou- ples a simple diffusion-reaction equation with a random walk and reproduces a variety of achiral [159] and chi- ral [160] branched patterns. A different approach is used in Ref. [161] where a simple model of self-propelled rod- like particles interacting only through steric repulsions is used. This model reproduces swarming for large as- pect ratios of the particles, and exhibits giant number fluctuations. How is the swarming motility relevant to biofilms? The role of swarming in the early phases of biofilm formation has been recognized only recently. For example, stud- 9 ies of Pseudomonas aeruginosa show that surface motil- ity does influence the morphology of the ensuing biofilm. Large surface motility produces flat biofilms, while lim- ited surface motility produces more corrugated biofilms [162]. The stators 'MotAB' and 'MotCD' involved in the flagellar mechanism of Pseudomonas aeruginosa are both involved in the initial, reversible attachment to a substra- tum [163]. The connection between swarming and biofilm formation runs deeply, at the level of genetic regulation and quorum sensing; however, its understanding is still in its infancy and more work is required [164]. C. Twitching Twitching motility is a bacterial, surface-related mo- tion actuated by the extension and retraction of type IV pili. These pili are the major virulence factor of Pseudomonas aeruginosa [165] and their twitching motil- ity allows the opportunistic infection of wounds; they are required for the so-called "social gliding motility" of Myxococcus xanthus and the twitching of Neisseria gonorrhoeae. All Gram-negative bacteria have type IV pili [166]. Type IV pili are semiflexible homopolymers of the protein pilin, 6 − 9 nm in diameter and several mi- crons in length, with a persistent length of about 5 µm and helical conformation [167, 168]. They extend from one or both poles of the bacterium [169]. The tip of a type IV pilus contains proteins that can adhere to or- ganic (like mammalian or plant cells) or inorganic sur- faces. The twitching motion is produced by the exten- sion, adhesion and retraction of the type IV pili, each tugging the cell body in different directions, which re- sults in the characteristic "jerky" motion of the cell. By hydrolyzing ATP within molecular motors the cells dy- namically polymerize and depolymerize the pili, actu- ating in this way the extension and retraction, respec- tively [170, 171]. Type IV pili can retract with a speed of about 0.5 − 1 µm/s [167, 170] and exert a force up to 110 pN for Neisseria gonorrhoeae [172], or 150 pN for Myxococcus xanthus [173, 174]. A number of experimental facts suggest a single sce- nario that explains the twitching motion. Cells crawl forward by retracting a number of pili at the forward end and unbinding those at the rear end. In Neisse- ria gonorrhoeae the velocity of the crawling motion is about 1.6 µm/s and is lower than the retraction speed of a pilus (2 µm/s) [175]. The random motion of cells is not a Brownian motion but has a persistence length of several pili, and this persistence length increases with the number of pili per cell [175]. A scenario where a "tug of war" takes place between the front and rear pili can explain this situation: the pili are under load during the motion, and thus the crawling speed is lower than the retraction speed; the motion is biased towards the direc- tion where several pili are attached close together, and produces a correlation length of several pili [175, 176]. A one-dimensional theoretical model, which assumes only mechanical interactions and no biochemical regulation, is consistent with this scenario [177]. A more precise two-dimensional, stochastic model predicts the existence of directional memory in the retraction and bundling of pili, and is corroborated by experiments [178]. The crawling of Pseudomonas aeruginosa alternates between two different modes: a slow, linear translation of long duration (0.3−20 s), and a fast roto-translation of short duration (less than 0.1 s) and 20 times as fast as the first mode [179]. The second, fast mode results from the sudden release of a single, taut type IV pilus that creates a "slingshot" effect with the cell body, and whose large velocity allows the cell to move through shear-thinning fluids such as the EPS of the ensuing biofilm [176, 179]. Type IV pili of Pseudomonas aeruginosa also allow a "walking" motion where the cell body is oriented per- pendicular to the solid surface [180]. The persistence p ≈ 2 µm is smaller than length of the walking mode 'w p ≈ 6 µm), the persistence length in the crawling mode ('c suggesting an uncorrelated motion of the pili [180]. The concerted action of the pili in the walking mode is also evident in the dependence on observation time τ of the bacterial mean square displacement h(∆r)2(τ)i ∼ τ α. While for the walking mode α ≈ 1.1, so a nearly dif- fusive motion, α ≈ 1.4 in the crawling mode, indicating a super-diffusive motion that explores space more effi- ciently [176, 181]. The twitching motion can be physically described within the broad context of continuous time random walks and Lévy flights [182]. Consider particles that move in a homogeneous, one-dimensional medium and undergo a stochastic motion. The quantity of interest is the probability density of finding the particle at po- sition x at time t, P(x, t). The particle can jump a distance x with probability density function g(x), where −∞ g(x)dx = 1, and then waits a time τ before mak- ing another jump; the waiting times follow a distribution 0 f(τ)dτ = 1. This is the definition of a continuous time random walk. If we consider the survival 0 f(τ)dτ, that is, the probabil- ity of not jumping in the interval [0, t], and given the initial distribution of particles, P0(x), then we can write the formal solution to the problem: the Montroll–Weiss equation [183] R +∞ f(τ), such thatR ∞ probability F(t) = 1 −R t F(s) P0(k) 1 − f(s)g(k) , (11) P(k, s) = f(s) ≡ L{f(t)} = R ∞ is 0 e−stf(t) dt the where R +∞ Laplace transform of the waiting time distribution, F(s) = L{F(t)} = [1 − f(s)]/s, g(k) ≡ F{g(x)} = −∞ g(x)e−ikx dx is the Fourier transform of the jump 1 2π distribution, and P0(k) = F{P0(x)}. To proceed fur- ther one must specify the waiting time distribution f(τ) and the jump length distribution g(x). If, in general, the first moment of f(τ), hτi, and the second moment of g(x), hx2i, exist, then one recovers the standard diffusion equation ∂P(x, t) ∂t = D∇2P(x, t) whose solution is the Gaussian distribution e−x2/4Dt , P(x, t) = 1√ 4πDt 10 (12) (13) with D = hx2i 2hτi. Assume now that for large jump lengths and long waiting times the probability density functions scale as f(τ) ∼ τ−(γ+1) . g(x) ∼ x−(2β+1) , (14) If β > 1 and γ > 1 then the moments hx2i, hτi exist and we have again standard diffusion. For β < 1 or γ < 1, instead, one or both moments diverge. If hx2i is finite but hτi diverges then the process is anomalously slow, and the mean square displacement is subdiffusive ¯x = ph(x − hxi)2i ∼ tγ/2, with γ < 1. If the waiting times are finite but the jumps have no upper bound, then the process is superdiffusive, ¯x ∼ t1/2β, with β < 1 [182, 184]. Under conditions of scarcity of resources Lévy flights are considered the most efficient way of exploring space. The question whether heavy tail distributions (as in Eq. (14) with β < 1 or γ < 1) and true Lévy walks are realized by living organisms is passionately debated and source of many controversies. We refer the interested reader to more specialized reviews as Ref. [182, 185, 186]. However, the run-and-tumble motion of Escherichia coli can be described as a Lévy walk [182, 187]: the cell moves in a specific direction with constant speed, and then changes direction after a random waiting time. Re- cent experiments on individual cells show that the intrin- sic molecular noise emerges in a power-law distribution of run times [188]. How is twitching motility relevant to biofilms? In gen- eral type IV pili are important adhesins, that is, they can actuate the initial attachment to a surface. Fur- thermore, the twitching motility that they generate helps in the repositioning of cells with respect to one another and leads to the biofilm differentiation [189]. Complex morphogenetic structures may be directed by regulated, cellular twitching motility (among other factors); for example, the formation of mushroom-like structures in the biofilm of Pseudomonas aeruginosa where non-motile mutants form the mushroom stalks, while motile individ- uals, driven by type IV pili, climb the stalks and aggre- gate on top to form the mushroom caps [190]. There is evidence that in Pseudomonas aeruginosa type IV pili and type IV pili-mediated twitching motil- ity play a role in the microcolonies that seed the biofilm. Type IV pili play a direct role in stabilizing interactions with the surface and/or in the cell-to-cell interactions required to form a microcolony. Type IV pili-mediated twitching motility may also be necessary for cells to mi- grate along the surface to form the multicellular aggre- gates characteristic of the normal biofilm [191]. Xylella fastidiosa (a notorious plant pathogen) has two classes of pili: type I and type IV pili that are implicated in surface attachment and in the formation of biofilm [192], whose density seems to be greatly influenced by the presence of type I pili [193]. Studies of Vibrio cholerae El Tor shows that MSHA type IV pili and flagella accelerate attach- ment to abiotic surfaces [194]. D. Gliding and Sliding Gliding is a surface-associated movement that does not involve flagella or pili, but instead uses macromolecular structures, known as focal adhesion complexes, that con- nect the cellular surface to external molecules (for ex- ample in the EPS) or surfaces [132]. Tens of proteins participate in the dynamic assembly and disassembly of these complexes that do not just mediate adhesion but also form a mechanosensing system, since they are cou- pled to the signal-transduction network of the cell [195], and they are linked to the cytoskeleton [196], and the actin-myosin network [197]. The term gliding is used with different meanings in the literature; for example, the "social gliding motility" of Myxococcus xanthus is actuated by type IV pili as they collectively spread over a surface, and small rafts of cells move beyond the boundaries of the expanding colony. A second motility mechanism termed "adventurous gliding motility" does not require pili [198], but rather focal ad- hesion complexes that assemble at the leading cell pole and disperse at the rear of the cells [199]. As the gliding motion does not require appendages and hydrodynamic interactions are negligible simple models that include only steric interactions and active motion are amenable to describe gliding motility. Experiments on the adventurous gliding motion of Myxococcus xan- thus show that collective motion with cluster formation appears for packing fractions φ >∼ 0.17 and at the tran- sition the cluster size distribution is a power law [200]. Additionally, the clusters exhibit giant number fluctua- tions h∆Ni ∼ N β with β ' 0.8 for φ >∼ 0.17 [200]. A similar exponent was found in a simple model of point particles moving with constant speed and aligning ne- matically with each other [201]. A similar transition to the clustered state was found in models of self-propelled rods [202, 203]. Microorganisms can also move without any active ap- pendages or specialized molecular motors but simply un- der the influence of the expansive force of a growing colony, and facilitated by the secretion of biosurfactants that reduce the surface tension between cells and sur- face, such as lipopeptides, lipopolysaccharides and gly- colipids [99, 140]. The fact that groups of cells move as a single unit indicates that this is not an active form of movement. For example, the Gram-positive Mycobac- terium smegmatis and Mycobacterium avium (a human opportunistic pathogen) spread by forming a monolayer of cells arranged in pseudo-filaments, where the cells pas- 11 sively move along their longitudinal axes [204]. Recently, it was discovered that beside the simple concept of cells pushing each other another physical mechanism affect the sliding motility. The secretion of EPS in the biofilm of Bacillus subtilis generates a gradient of osmotic pres- sure, which causes an uptake of water from the surround- ing environment that in turn produces a swelling of the biofilm [57]. Theoretical calculations show that the os- motically driven biofilm undergoes a transition from an initial vertical swelling of the biofilm to a subsequent hor- izontal expansion [57]. How are gliding and sliding rel- evant to biofilms? Work on Mycobacterium smegmatis and other members of this genus shows that these motil- ity modes are important for the colonization of surfaces, such as during biofilm dispersal, and the restructuring of the biofilm [205]. The glycopeptidolipids synthesized by these microorganisms lower the surface tension and facilitate their surface motility, and there is a strong cor- relation between the presence of lipids, surface motility and biofilm morphology [205–208]. III. ADHESION The adhesion of cells to a solid substratum is the first, necessary step for the formation of a biofilm. Contact with a substratum corresponds to the moment when the regulatory genetic network initiates the production of EPS, induces the genetic expression of polysaccha- rides [209] and progressively deactivates flagellar mo- tion [140]. inert or reactive, Microorganisms can adhere to organic or inorganic living or devitalized [210, surfaces, 211]. The process of adhesion can be divided into two steps [212]: (i) docking, that is, the cell comes in close proximity of the substratum thanks to motility or Brow- nian motion; this phase of attachment can be easily reverted and is governed by basic physical interactions based on the Coulomb and van der Waals forces. (ii) locking, that is, the irreversible anchoring of the cell by means of chemical bonding mediated by adhesins [213]. A. Basic Theoretical Framework A microbial cell, for example a bacterium, is a micron- sized particle, thus, in the colloidal regime. Most bac- teria and natural or artificial surfaces are negatively charged [213–215]. Consequently, the docking phase of adhesion is possible because of the existence of an equi- librium point between the electrostatic repulsion and the attractive van der Waals interaction. The mechanism of adhesion of charged colloids is described by the Der- jaguin, Landau, Verwey, and Overbeek (DLVO) theory of colloidal stability [216–218], which quantitatively de- scribes the adhesion of cells to solid surfaces. We now discuss the fundamental concepts of the DLVO In its classical formulation the DLVO theory theory. includes only the effects of van der Waals and electro- static interactions, while neglecting other effects, such as steric, solvation, and depletion forces [219, 220]. Thus, the DLVO interaction energy between two charged ma- terials separated by a distance d is UDLVO(d) = UvdW(d) + UC(d) , (15) where UvdW(d) includes the interaction between a per- manent dipole and an induced dipole (Debye energy), the interaction between two permanent dipoles (Keesom en- ergy), and the interaction between instantaneous dipoles (London dispersion energy), and UC(d) is the Coulom- bic potential energy. While for atoms or molecules UvdW(d) = −CvdW/d6, for larger objects one finds a power law with a smaller exponent; for example, the van der Waals energy between a sphere of radius R and a flat surface is UvdW(d) = −RA/6d, while between two flat surfaces is UvdW(d) = −A/(12πd2) [221], where A = π2ρ1ρ2C is the Hamaker constant, ρ1 and ρ2 are the number densities of the two interacting materials, and C is the coefficient in the interatomic pair potential which is proportional to the square of the polarizability. In gen- eral van der Waals energies (and hence forces) scale as power laws, which means that they can be long ranged, and therefore the structure and composition of the solid substratum can have surprising effects [222]. Surfaces in an aqueous milieu become charged because of the dissociation of surface chemical groups (releasing H+ or Na+) or because of the adsorption of ions from the solvent. The negatively charged surface is balanced by counter-ions which are attracted to it by the electrostatic force. The closest counter-ions are bound to the surface but further away there is a diffuse layer of unbound ions in thermal equilibrium that form the so-called electric double layer. We can derive the equation governing the electrostatic potential ψ(x) for a simple system of only counter-ions in the presence of the surface. The chemical potential µ of the counter-ions at position x is µ = zeψ(x) + kBT ln ρ(x) , (16) where z is the valence, e the absolute value of the elec- tron charge, kB the Boltzmann constant, and T the tem- perature of the system. By imposing that in equilibrium the chemical potential is constant one finds the (number) density distribution ρ = ρ0 exp(− zeψ(x) kBT ) , (17) which is just the Boltzmann distribution of the ions. In electrostatics the charge distribution obeys the Poisson equation ∇2ψ = − zeρ ε (18) where ε is the permittivity of the solvent. Insert- ing Eq. (17) into Eq. (18) one arrives at the Poisson– Boltzmann equation , (cid:18) (cid:19) ∇2ψ = − zeρ0 ε exp − zeψ kBT . (19) 12 For a system with more than one species of ions in so- lution Eq. (19) generalizes in a straightforward manner to a sum on the right-hand side over the different ionic species. The Poisson–Boltzmann equation is nonlinear, and therefore very challenging to solve. In the limit of dilute solutions we can expand in Taylor series the expo- nential in Eq. (19) (Debye–Hückel approximation). The zeroth order term vanishes because of the electroneutral- i zieρ0,i = 0). Then, if we stop at the ity condition (P first order, we obtain for the ionic species i ∇2ψi = κ2ψi , κ2 ≡ e2 εkBT ρ0,iz2 i (20) X i where λD = κ−1 is the Debye length, which physically represents the thickness of the electric double layer. Typ- ical solutions of Eq. (20) are of the form ψi(x) = ce−κx/x, with c a constant [221]. Thus, putting all together, UDLVO(d) is the sum of an attractive term UvdW(d) ∼ −1/dn and a repulsive term UC(d) ∼ e−κd. At small enough distances the attraction will always prevail. It exhibits a deep (primary) minimum at d = 0, which cor- responds to adhesion of the surfaces. At slightly larger distances there is an energy barrier due to the overlap of the electric double layers; the energy barrier increases as the ionic strength decreases. Further away, at typical distances of 5 nm, there is a small (secondary) minimum if the ionic strength is large enough. We can now describe what happens to a bacterium, in the idealized situation described above, when it approaches a surface. As a bacterium moves closer to a surface it will experience an energy barrier a few nm away from the surface that cannot be overcome by active motility or simple Brownian motion. However, depending on the ionic strength the cell will find the secondary minimum and it will be reversibly bound to the surface. In the following step (locking) the biosynthesized pili or EPS surmount the barrier and allow the irreversible adhesion to the surface [211, 223]. B. Complex Role of Biopolymers and Appendages The simple picture given by the DLVO theory above describes an idealized situation of homogeneous, charged surfaces interacting. However, it leaves out the true com- plexity of the adhesion process of microorganisms. In general terms, there are three main classes of structures that cause drastic deviations from the predictions of the DLVO theory in real measurements of adhesion forces: (i) biopolymers associated with the cellular membrane, (ii) the EPS that modifies the physico-chemical proper- ties of surfaces, (iii) appendages, such as pili. The main features of their roles in adhesion will be reviewed below. The cellular surface is a physically and chemically het- erogeneous structure [224]. For example, the outer mem- brane of Gram-negative bacteria contains lipopolysaccha- rides (LPS) which provide protection from many antibi- otics. LPS are composed of a lipid A (anchored to the outer membrane), negatively charged core polysaccha- rides, and the O-antigen, which can extend 30 nm or more from the outer membrane. Lipopolysaccharides have an inhomogeneous spatial distribution and have been shown to play a key role in the adhesion process [225–227]. A portion of the O-antigen has been shown to form hy- drogen bonds with different mineral substrata [228], and 1000 or fewer of these bonds can firmly anchor the cell to a surface [223, 228]. The characteristics of the LPS in Pseudomonas aeruginosa are crucial to determine their adhesion to hydrophilic or hydrophobic surfaces, and the phenotypic variations in the expression of LPS can regu- late adhesion in their survival strategy [229]. Mutations in Escherichia coli's genes involved in LPS biosynthe- sis showed decreased adhesion [230]; removal of up to 80% of LPS molecules also drastically decreases adhe- sion affinity [231]. Recent work on Staphylococcus aureus shows that adhesion initiates already at distances of 50 to 100 nm (where the DLVO theory predicts negligible forces), and furthermore the fluctuations of protein den- sity and structure are more important than the details of the binding potential [232]. Another important factor in the adhesion process is the EPS. The polymeric macromolecules form nonco- valent bonds, such as electrostatic or hydrogen bonds, with the substratum that lead to the adhesion of the biofilm [233]; EPS contains many polar portions that anchor to a hydrophilic glass and can turn it from hy- drophilic to hydrophobic and therefore renders adhesion thermodynamically favorable [234, 235]. The EPS also produces a "polymeric interaction" because the macro- molecules bound to the cell membrane effectively increase its size and boost the (attractive) van der Waals interac- tion with the surface. Additionally, EPS macromolecules around the cell can bind with macromolecules adsorbed on the solid surface, thus bridging the electrostatic bar- rier [236, 237]. The hydrophobic components of the cell membrane also favor adhesion as has been repeatedly shown: hy- drophobic cells adhere better than hydrophilic cells [238– 240]. The hydrophobic effect is a consequence of a fun- damental aspect of water's molecular structure: the ex- tensive presence of a hydrogen-bond (H-bond) network. When an apolar solute is introduced in bulk water, it will locally disrupt the H-bond network because water molecules cannot bind to it. To minimize this disruption the water molecules surrounding the solute will form a cage-like structure: a clathrate. Because of the clathrates the water molecules have smaller translational and orien- tational entropy; thus this change is unfavorable. When two apolar moieties are close to each other, it is thermo- dynamically more favorable to reduce their separation until they are in contact with each other because in this way the surface area of the clathrates is reduced and hence the entropy is maximized [241–243]. Different kinds of pili are also involved in the adhesion 13 process. Type IV pili can bind to inert surfaces bac- terial and eukaryotic cells because of adhesins at their tips [169]. The Gram-negative bacterium Caulobacter crescentus is among the first to colonize submerged sur- faces and in the nonmotile stage of its life cycle pro- duces a stalk tipped by a polysaccharide adhesin that can exert forces in the micronewton range [244]. Type I (one in Roman numerals) and type P pili are short ad- hesive structures present in Gram-negative bacteria such as Escherichia coli and Salmonella species. They are peritrichate and are not associated with twitching motil- ity [223]. Type I pili mediate adhesion to mucosal epithe- lia and other cells [245, 246]. P pili are composite struc- tures that mediate the adhesion to urinary tract epithe- lium [247]. Gram-positive bacteria, such as Corynebacte- ria and Streptococci, have also pili, but they are very thin (2 to 3 nm) and have been only recently observed [248]. We now know that Streptococci use pili to penetrate mu- cus (in a human host, for example) or a conditioned sur- face, and then employ multiple adhesins to build stronger bonds [249]. Staphylococcus epidermidis exhibits two kinds of proteins on fimbria-like appendages protruding from its cell surface that are required for the adhesion function [250]. Furthermore, polymeric molecules present on the cell wall of Staphylococcus have been shown to play a key role in its adhesion capacity and pathogenic- ity [251, 252]. Loeb and Neihof were the first to report that surfaces immersed in seawater very quickly were coated with organic molecules dissolved in the water before the colonization of microorganisms [253]. We now know that this is a very general process. These substances form a so-called "conditioning layer" that alters substantially the physicochemical properties of the substratum and its interaction with microorganisms [211, 254–258]. A surface may become selectively attractive to some cells. The adhesion of microorganisms might be influenced by hydrophilic and hydrophobic features of the conditioned surface, charge density and exposed chemical groups. The presence of a conditioning layer causes significant deviations from the DLVO theory. In fact, exopolymers may play a dual role: first, by coating the biofilm substratum which strengthen adhesion, and second, by forming polymeric bridges between the EPS-encased cells and the coated substratum [235]. IV. VISCOELASTIC PROPERTIES We now take a larger, almost macroscopic, view of the biofilm and ask the question: how does a biofilm respond to a mechanical perturbation? Clearly a biofilm does not flow as a simple liquid; it requires some structural stability to allow the colony to grow and prosper. It is equally clear that a biofilm cannot be as rigid as a crys- tal because it must allow expansion as cells multiply, and it must deform as a response to changing environmental 14 form links with each others. The molecular relaxation time is now much larger, and can be matched by realistic shear rates. One is naturally led to define a dimension- less number comparing these time scales, the so-called Weissenberg number W ≡ τ0 γ, where γ is the shear rate. When W (cid:28) 1 one recovers a Newtonian fluid; but for W >∼ 1 the shear rate interferes with the molecular re- laxation, and therefore one should expect nonlinear or history-dependent terms in the constitutive equation for the stress tensor. We now discuss the basic physical concepts neces- sary to describe the viscoelastic behavior of biofilms (see [271, 272] for more details). Any material responds with a deformation γ to an applied stress σ (force per unit area). For an elastic solid σ = Gγ, where G is the elastic modulus; for a liquid σ = η ∂γ ∂t , where η is the viscosity. A viscoelastic material must combine both elastic and viscous response. The simplest physical model thereof is the Maxwell element where an elastic spring and a vis- cous dashpot are combined in series (see Fig. 4(a)). For simplicity we consider at first a one-dimensional problem. For this configuration the stresses on the two elements are equal σ = σ1 = σ2 and the deformations are addi- tive, hence ∂γ ∂t + σ η . The constitutive equation for the Maxwell model is then ∂t = ∂γ1 ∂t + ∂γ2 ∂t = 1 ∂σ G (21) where λ ≡ η/G is a relaxation time. The formal solution of Eq. (21) is + σ = ∂γ ∂t ∂σ ∂t λ , σ(t) = 1 λ e− t−t0 λ η ∂γ ∂t0 dt0 . (22) Z t −∞ While for time scales much shorter than the relaxation (λ → ∞) one finds solid-like behavior (σ = Gγ), for ob- servation times much larger than the relaxation (λ → 0) one finds viscous behavior (σ = η ∂γ ∂t ). The Maxwell model then behaves as a fluid, and therefore is not suit- able for a real viscoelastic material. Another, historically important, model is the Kelvin– Voigt element, where the spring and dashpot are ar- ranged in parallel (see Fig. 4(b)). The Kelvin–Voigt con- stitutive equation is ∂γ ∂t . σ = Gγ + η (23) At long observation times the Kelvin–Voigt material be- haves solid-like because it always returns to its equilib- rium configuration. We can improve our model of viscoelastic materials by adding a dashpot in series with a Kelvin–Voigt ele- ment, producing the Jeffreys model (see Fig. 4(c)). The strains in the dashpot and in the Kelvin–Voigt element are additive, ∂γ ∂t + ∂γ2 ∂t , and the stresses are σ = η1 ∂γ1 ∂t . Solving for the total strain γ and total stress σ, one finds the constitutive equation ∂t = ∂γ1 ∂t = Gγ2 + η2 ∂γ2 λ1 ∂σ ∂t + σ = η1 + λ2 ∂2γ ∂t2 , (24) (cid:18) ∂γ ∂t (cid:19) FIG. 4. Sketch of some mechanical models of viscoelastic fluids that are built from elastic elements (denoted with their Young's modulus G) and viscous elements (denoted with their viscosity η) combined in series or in parallel, and to which a stress σ is applied. (b) Kelvin–Voigt model. (c) Jeffreys model. (d) Burgers model. (a) Maxwell model. stresses or stimuli. A biofilm must then be a substance with properties intermediate between a liquid and a solid; thus, it must exhibit both the short-time elastic response of a solid and the long-time viscous response of a liquid. This class of materials is termed viscoelastic [259]. In fact, measurements confirm that biofilms are viscoelastic polymeric materials [260–268]. This is not surprising as a biofilm consists of a hydrated environment of complex, entangled polymeric molecules and bacteria that are sub- ject to multiple adhesion and cohesion forces. In general, a biofilm can be seen as an entangled polymer network where the bonds are non-permanent and physical, rather than chemical [269, 270]. Viscoelastic materials are not-Newtonian fluids, that is, they do not obey Eq. (3). The Newtonian expres- sion works remarkably well for simple liquids composed of small, roughly isotropic molecules. This happens be- cause of a separation of times scales: the molecular re- laxation time is comparable to the time to diffuse one molecular length, which is τ0 ≈ '2/D ≈ 10−13 s. The flow velocity can interfere with these relaxation times only in exceptional cases. Consider now a complex fluid comprised of elongated, flexible molecules that possibly σσσσσσηη1η2(a)(b)(c)ηη1η2σσG1G2(d)where λ1 ≡(cid:0) η1+η2 (cid:1) is a relaxation time and λ2 ≡ η2 G G a retardation time. We can obtain a more general equa- tion by considering the deformation field ξ(x, t) and by realizing that ∂γ ∂x, that is the velocity gradient or rate-of-strain tensor. In three dimensions, it is ∇~v. Remembering that σ = −pI + τ (see Sec. II A), the Jeffreys constitutive equation then becomes ∂ξ(x,t) ∂x = ∂v ∂t = ∂ ∂t ∂D[~v] λ1 ∂τ ∂t D[~v] + λ2 + τ = 2η1 (cid:0)∇~v + ∇~vT(cid:1) is the symmetric part of where D[~v] ≡ 1 2 the rate-of-strain tensor. The formal solution of Jeffreys constitutive equation is (25) ∂t , (cid:18) (cid:19) (cid:18) (cid:19)Z t −∞ τ = 2η1 λ2 λ1 D[~v] + 2 η1 λ1 1 − λ2 λ1 t−t0 λ1 D[~v]dt0. (26) e The Maxwell model is recovered when λ2 = 0, and the Newtonian fluid when λ1 = λ2. Another way of dealing with the rheology of non- Newtonian fluids is to assume that only the rate of dis- sipation in the fluid changes but not the structure of the stress tensor. One then writes σ = −pI + η(D[~v])D[~v] . (27) Because the viscosity cannot change for a change of co- ordinate system (must be invariant), η(D[~v]) must be a function of the tensorial invariants build with D[~v]. The lowest order term is γ2 ≡ 1 2D : D. The rheology of the material is then governed by the properties of the func- ∂ γ > 0 the fluid is called shear-thickening, tion η( γ). If ∂η that is, it becomes increasingly stiffer as the shear-rate increases; if ∂η ∂ γ < 0 the fluid is called shear-thinning, that is the material becomes less viscous as the shear-rate in- creases. Experimentally, a common rheological technique to characterize viscoelastic materials is the measurement of the stress-strain curve. The sample is subject to a shear stress which increases monotonically with time up to a maximum value; this corresponds to the "loading" phase. Subsequently, the shear stress is decreased; this is the "unloading" phase. The presence of hysteresis is characteristic of viscous dissipation in the material. An- other common rheological technique is the creep test, where a stress is applied for a prolonged interval of time and then suddenly released. The temporal evolution of the strain exhibits characteristic regions corresponding to different processes taking place: (i) an instantaneous elastic stretching, (ii) a viscous flow, (iii) after the exter- nal stress ceases there is an instantaneous elastic recoil, (iv) a time-dependent creep recovery [273]. The experimental investigation of biofilms' material properties is still in an early stage, and different mod- els are proposed to explain the experimental data. For example, the biofilms of different Pseudomonas aerugi- nosa strains were subject to strain and creep tests and the results were fitted with a Jeffreys model that includes 15 a slow nonlinearity to account for the shear-thickening behavior [260]. That theory correctly predicts the relax- ation time at high shear rates. However, the biofilm of Streptococcus mutans (common in dental plaque) showed creep compliance consistent with a Burgers model [262] (see Fig. 4(d)). Biofilms vary greatly in their composition and structure. The biofilm of Streptococcus mutans ap- pears to be shear-thinning [262, 274]; the biofilms of the microalga Chlorella vulgaris also shows shear-thinning behavior [275]. Pseudomonas aeruginosa builds instead a shear-thickening biofilm [260]. Therefore, it should not be surprising that measurements based on creep tests of 44 biofilms determined values of G and η spanning eight decades [263]. However, they also found a remarkably small range of variability for the elastic relaxation time λ ≈ η/G, with an average of 18 min [263]. It is argued that this common relaxation time can hardly be a coin- cidence, but rather a sign of convergent evolution. The time scale λ separates the solid-like from the liquid-like behavior of a biofilm. Short mechanical stresses can be absorbed by an elastic response, but a sustained stress can be deleterious as it could lead to structural failure, The biofilm reacts instead as a viscous fluid for long time- scale stresses. The time scale of 18 min coincides with the time required to elicit a phenotypic response at the cellular level [276, 277], which is however expensive in terms of cellular resources. Thus, the elastic relaxation time needs to be large enough to avoid unnecessary re- sponse to intermittent stresses, but smaller or approxi- mately equal to the biological time to initiate expensive genetically-regulated responses [263]. A biofilm growing under flow conditions develops char- acteristic elongated, filamentous structures called stream- ers (see Fig. 2(b), (e) and (g)). They are ubiqui- tous in natural environments and strongly influence flow through porous materials, medical and industrial de- vices [278–280]. As the EPS network of streamers perco- lates through the channels or pores more and more plank- tonic cells are caught within it. Eventually, the stream- ers lead to a catastrophic clogging of the pores that stops the flow [280]. Biofilms grown up to Reynolds numbers of R = 3600 produced streamers that behaved as viscoelas- tic solids for stresses lower than the value at which they were grown, but behaved like viscoelastic fluids at larger stresses [38, 273], that is similar to a Bingham fluid. In the linear regime streamers exhibited a shear modulus of 27 N/m2 and a Young modulus in the range of 17 to 40 N/m2 [38]. Strong flow conditions, in the turbulent regime, pro- duce ripple structures in mixed species biofilms that mi- grate with a speed of 800 µm/h [281]. At R = 4200 the ripples have a wavelength of about 75 µm, but it in- creases to about 200 µm at R = 1200. The ripples moves downstream, which points at an important effect for sur- face colonization. Furthermore the morphology of the ripples responded within minutes to changes in the flow velocity. This fact indicates that the formation of rip- ples must have a hydrodynamical origin. Indeed, recent work [282] on fossilized microbial mats called Kinneyia confirms this. Kinneyia are sedimentary fossils character- ized by ripples with wavelength between 2 and 20 mm. Theory and experiments using an artificial biofilm shows that the rippled structures can be explained by the an- cient flowing of water above the mats that produced a Kelvin–Helmholtz instability [282]. This hydrodynamic instability occurs at the interface between two fluids with different viscosities and flowing with different velocities and produces ripples with wavelength proportional to the thickness of the biofilm. This theory predicts morpholo- gies, wavelengths and amplitudes consonant with the fos- sil samples [282]. Two important challenges have been identified for fu- ture investigations of biofilms from the point of view of rheology and complex fluids [268]: (i) the structure and morphology of biofilms depend on their history of growth and on the substratum; for example, the biofilm of Pseu- domonas aeruginosa growing in human lungs and airways incorporates DNA and the F-actin protein [283], which modify the material properties of the biofilm with re- spect to a lab-grown sample. The thickness of the in vivo biofilm of the fungus Candida albicans is up to four times larger than the in vitro biofilm [284]. (ii) An integrated model of the biofilm's material properties should guide the understanding of its development and also of its dis- solution, which is crucial in clinical settings. Helpful in this regard can be investigations of hydrogels (which are cross-linked networks of hydrophilic polymers) because these well-studied systems can illuminate different as- pects of the biofilm's physics. For example, physical mod- els for the diffusion of macromolecules within gels [285– 287] can instruct the understanding of biomolecules reg- ulating the biofilm growth and the mechanism of drug delivery to destroy it. Self-healing polymeric materi- als [288, 289] can be a guide to biofilms as adaptive ma- terials. More progress is expected from the comparison with highly controllable artificial systems [290]. V. INTEGRATED EXPERIMENTAL METHODS Biofilms are complex systems as they have characteris- tic length and time scales spanning many orders of mag- nitude but that crucially interact with each other in a hierarchical way, and produce robust order [291]. At the nanometer scale molecular processes such as gene expres- sion, biosynthesis and the activity of the signal transduc- tion networks take place. At the micrometer scale flagel- lar and ciliar motion governs the motility of the cells; dif- fusion of nutrients, and gradients in the chemoattractants and quorum sensing cues shape the average behavior of a group of cells. At larger scales hundreds of microor- ganisms form aggregates held together by pili and EPS; they form ripples and streamers under the influence of hydrodynamic forces. The biofilm undergoes stages of maturation and dispersal, while some cells revert to the motile phenotype and colonize new surfaces [292]. An 16 understanding of the biofilm state must then integrate these different length and time scales involved. Different "integrated", experimental approaches have already emerged. Microfluidic and lab-on-a-chip devices offer the possibility to include several functions in a mi- croscopic environment, offer high throughput, and length and time scales compatible with the ecological conditions of biofilms [293, 294]. For example, microfluidic devices which mimics xylem vessels were used to measure the adhesion forces of type I and type IV pili for Xylella fastidiosa [295], or to characterize the adhesion, motil- ity and biofilm formation of Acidovorax citrulli [296]. A nanoporous microfluidic platform was used to determine the amount of acetic acid necessary to impair the motility of Listeria monocytogenes [297]. The biofilm growth and dispersal dynamics of Staphylococcus epidermidis was in- vestigated with a microfluidic device that allows to con- trol the local hydrodynamic conditions [298]. A microflu- idic assay was used to develop an in situ analytical system to assay the susceptibility of biofilms to antibiotics [299]. Microsensors and microfluidics were employed to moni- tor phenotypic responses of fungal biofilms to changes in shear stress and antifungal drugs [300]. A microfluidic biofilm signaling circuit that uses a population-driven quorum sensing switch can form or displace consortia of fungal biofilms [301]. This platform allows for non- invasive, robust continuous monitoring of the biofilm con- ditions. Microfabricated elastomer chips allow to moni- tor and study the evolution of bacterial and yeast colonies in a chemically controlled, dynamic environment [302]. The architecture of the device resembles the complex structures naturally produced by biofilms, that is, clus- ters of cells separated by channels along which nutrients and waste products flow. A microfluidic porous device demonstrated that the formation of streamers takes place only in a certain flow rate range, and that streamers act as precursors to structures found in biofilm growing in porous media [303]. A microfluidic devices that incorpo- rates a microbial fuel cell was used to generate electric- ity and may find application as a small-scale biosensing device that offers great flexibility because of its lab-on- a-chip technology [304]. The bacteria used to produce power were the exoelectrogenic Geobacter sulfurreducens and Shewanella oneidensis. These Gram-negative bacte- ria have the capability to exchange electrons with soluble and insoluble electron acceptors as part of their respira- tory chain [305, 306]. Recently, a novel method that com- bines surface enhanced Raman spectroscopy and confo- cal Raman spectroscopy (SECRaM) was developed [307]. This method allows the in situ spatial and temporal, analytic investigation of Shewanella oneidensis biofilms without disturbing it. Nanofabrication and microfluidics are excellent tools for manipulating the cells' microenvironment and mon- itor their responses [308–310]. By using surfaces with a pattern of micropillars it was shown that Escherichia coli bacteria use their flagella to reach inside the crevices between pillars and adhere to them when the spacing of the pillars is smaller than the cell size [311]. Thus, it was shown that flagella in addition to swimming motility create a dense fibrous network that gives structural sta- bility to biofilms [311]. Microfluidic methods revealed that there is an underlying hydrodynamic mechanism that produces streamers, that is, the secondary vortical motion that is very pronounced around corners [278–280]. Nanofabricated devices have the potential to unravel the complex structural organization of bacterial popula- tions in natural environments, which are chemically and physically heterogeneous. Biofilms commonly contain more than one species, and therefore both cooperation and competition shape the community in dynamic states of organization. Aspects of biofilm sociobiology [312] are now within the grasp of experimental testing. Es- cherichia coli colonizing an array of microhabitat patches formed a "metapopulation", that is a population of pop- ulations, that adapts quickly to habitats with high niche diversity [313]. Well-defined microscale spatial structures are a necessary and sufficient condition for the stability of bacterial communities engaged in "reciprocal syntro- phy", that is a partition of functions all necessary for the survival of the whole community [314]. Thus, the local topography is a crucial factor to account for in the biofilm's growth and viability. There are also efforts to move to three dimensions. A three-dimensional (3D) printing technique was used to study the cell-cell inter- actions in structured 3D environments that model bac- terial aggregates in the human body [315], for example. It was found that Staphylococcus aureus, whose mono- species colonies are susceptible to ampicillin, is granted resistance to it by a community of Pseudomonas aerug- inosa forming a shell around it [315]. However, much still remains to be understood, as it appears that subtle details often determine the fate of the colony [316]. These bottom-up approaches aimed at recreating the complex ecological landscape of bacterial communities in a controlled manner hold the promise to illuminate the behavior of single cells and the dynamical organization occurring in a biofilm. Similarly to the revolution in in- formation technology brought forth by the miniaturiza- tion of semiconductor electronics, the fields of microflu- idics and nanofabrication may usher a new level of un- derstanding of microbiology. VI. THEORETICAL AND COMPUTATIONAL METHODS Theoretical models of biofilm growth and dynamics have started since the 1970s with simple one-dimensional differential equations describing a flat layer. In the fol- lowing we will start by reviewing the basic concepts of that early era that originated from models of bacterial cultures [317, 318] because, although our understanding of biofilms has advanced since then, these ideas and equa- tions are still a part of the more complex models and methods considered in the modern scientific debate. 17 Consider a colony of microorganisms in a well-mixed bath of substrate (for example in a continuous culture apparatus). The specific growth rate of the colony µ is written as µ = 1 dt ln c where c is the concentration of cells. Monod [319] famously related µ to the concen- tration s of an essential growth substrate dt = d dc c (cid:18) s (cid:19) Ks + s µ = µm , (28) where µm is the growth rate constant and Ks is the sat- uration constant, that is, the substrate concentration at which µ = 1 2 µm. The growth rate is proportional to the substrate concentration when s (cid:28) Ks, but it saturates at large values of s, s (cid:29) Ks. A second fundamental law of Monod relates the growth rate to the substrate consumption rate dc dt = −Y ds dt , (29) where Y is the yield constant, the ratio of mass of cells grown and mass of substrate used. Combining Eq. (29) with the definition of µ yields the basic equations ds dt = −kc s Ks + s , dc dt = µmc s Ks + s (30) where k ≡ µm/Y is the maximum specific rate of sub- strate consumption. Consider now a biofilm immersed in a bulk liquid. For simplicity we assume a biofilm growing on a flat surface that we take as the xy-plane and with a z-axis normal to it. The substrate, that is, the energy source diffuses from the surface of the biofilm to the deeper layers, and its concentration decays with depth. Thus, there is an addi- tional term in the balance equation for s due to molecular diffusion ds dt = −k cs Ks + s + Ds ∂2s ∂z2 , (31) where Ds is the diffusivity of substrate within the biofilm. A biofilm is considered in a steady state if the total biomass is equal to what can be sustained by the flux of substrate [320]. It is a statement of conservation of energy. In a steady state, the net biofilm growth rate of cells is [320] (cid:18) s (cid:19) dc dt = µmc Ks + s − bc , (32) where b is the specific decay coefficient. This and similar approaches [321–323] viewed biofilms as simple flat structures, governed by energy balance considerations. However, the development of techniques such as confocal scanning laser microscopy revealed that biofilms have a heterogeneous spatial structure inter- spersed with channels and chambers [34, 324]. Thus, the next step in the mathematical modeling of biofilm was the fully 3D description and the inclusion of a dif- fusive mechanism for biomass spreading with a diffusion coefficient that depends nonlinearly on the biomass con- centration. Eberl et al. [325] proposed the equations ∇ · ~v = 0 , ∂~v ∂t + ~v · ∇~v = −1 ρ ∇p + ν∇2~v , ∂s ∂t + ~v · ∇s = ∇ · (D1(c)∇s) − f(s, c) , ∂c ∂t = ∇ · (D2(c)∇c) + g(s, c) , f(s, c) = k1cs k2 + s , g(s, c) = k3(f(s, c) − k4c), (33) (34) (35) (36) where ~v is the flow velocity, s the nutrient concentra- tion, c the biomass density, and k1 . . . k4 are model pa- rameters. Equations (33) are the Navier–Stokes equa- tions for an incompressible fluid with density ρ and kine- matic viscosity ν. Nutrients spread through convection and diffusion and are consumed with a Monod reaction rate f(s, c). The biomass is produced with a rate g(s, c) and spreads through a diffusive flux D2∇c (Eq. (35)), where D2(c) = ( cmax−c)acb. This functional form pro- duces a biomass with a maximum density cmax, vanish- ing diffusion for c → 0, and significant biomass spreading only when c → cmax. This set of equations are studied for different values of an important dimensionless num- ber G that quantifies the ratio of the biomass growth rate and substrate transport rate [326], and produces heterogeneous, mushroom-like structures similar to what is observed in experiments [325, 327]. This approach was generalized to multispecies biofilm with different so- lutes [328]. Experimental results in the mechanics of biofilms [273, 329] prompted a phenomenological description of the biofilm mechanics in terms of a multi-fluid model [330]. Consider a biofilm with volume fraction φb immersed in a solvent with volume fraction φs, where φs + φb = 1. The fundamental quantity of interest is the cohesion energy functional for the biofilm E[φb] = f(φb) + κ (37) Z (cid:16) 2∇φb2(cid:17) dV, where f(φb) is the energy density of the mixing of biomass and solvent, and the gradient term penalizes short-length scale variations by means of a surface en- ergy. The shape of f(φb) can be deduced from general physical arguments. The function f(φb) should have a minimum at some volume fraction because the cohesive forces of the biomass are balanced by the short-range re- pulsions. Because attractive forces are weak at long dis- tance f(φb) → 0 as φb → 0, and because some amount of solvent is necessary, f(φb) must grow at large φb. Using mass conservation for the biomass dφbdt + ∇ · (φb~ub) = 0 one can write the time derivative of the energy (φb∇ · Π) · ~ubdV , (38) Z dE dt = − 18 where Π = −[f0(φb) − κ∇2φb]I is the cohesive stress tensor, and ~ub the velocity of the biomass. Neglecting the inertial, viscous and viscoelastic terms one finds the evolution equation for φb [330–332] + ∇ · (φb~ub) = ∇ · [a(φb)f00(φb)∇φb]− ∂φb ∂t κ∇ ·(cid:2)a(φb)∇∇2φb (cid:3) , (39) 0 φb(1−φb), and ζ0 is a friction constant. where a(φb) = ζ−1 This model was studied in one dimension with ~ub = 0 (so that Eq. (39) becomes a modified Cahn–Hilliard equa- tion) and a free surface film state is found to form, where spinodal decomposition is a possible mechanism [330]. This approach has been extended to include EPS with an additional phase field, which reproduces the cohesive failure of biofilm under shear flow [333], and with the inclusion of a viscoelastic model of the EPS [334]. In contrast to the models based on Eq. (37), a different paradigm of biofilm growth has been proposed, where the biofilm is modeled with stochastic rules that evolve in a discrete way [335], as opposed to the continuous changes described by partial differential equations. In these early computational attempts the growth of a bacterial colony was studied [159, 336, 337]. For example, experiments on Bacillus subtilis growing on agar plates were compared with three models of stochastic growth: (i) diffusion lim- ited aggregation, (ii) the Eden model, and (iii) the dense branching morphology. Diffusion limited aggregation de- scribes the growth process of small particles in a dilute solution that aggregate when their Brownian trajectories bring them in close contact [338]. Coagulated aerosols or zinc ions in solution, for example, exhibit open, ran- domly branched structures that can be modeled by diffu- sion limited aggregation. A simple method to reproduce computationally diffusion limited aggregation is as fol- lows. Consider a 2D square lattice at whose center the first seed is planted; next, a new particle is placed far away from the seed; it then undergoes a random walk on the lattice until it reaches a lattice site neighboring the initial seed, at which point they stick together. The pro- cedure is repeated and new particles stick to the cluster whenever they reach one of its neighboring sites. Alterna- tively, in the Eden growth model new particles are added next to the perimeter of the cluster [339]. When starting from a seed in a planar geometry, compact, disk-like clus- ters are obtained. Finally, an addition of a kinetic term to the growth process leads to the increase of branching tips as the radius of the agglomerate increases. This pro- cess produces the dense branching morphology, which is characterized by compact branches with a neat spherical envelope [340–342]. Fujikawa [337] found that at rela- tive low nutrient concentrations the colony formed open fractal branches, similar to the simulations of diffusion limited aggregation, and measured a fractal dimension Df = 1.73 ± 0.02 which compares well with known the- oretical values [343]. For higher nutrient concentrations, the colony grew compact and round, similarly to an Eden cluster, as the cells do not have to rely on the diffusion of nutrients from the surrounding. At low agar concen- trations, instead, patterns similar to the dense branching morphology were found, in which case the reduced vis- cosity allows a growth driven by the flagellar motion. An alternative class of models considers the individual microbial cell, or small groups thereof, as the elementary computational unit [355–360]. They differ from cellular automata models as there is no underlying lattice par- titioning space. They are called individual-based mod- These stochastic, discrete approaches eventually pro- duced the widely-used class of cellular automata mod- els [326, 344–349]. Generally, cellular automata models consider a 2D space partitioned with a lattice. Every lattice location may be occupied by a parcel of biofilm consisting of cells, EPS, and other molecules. Addition- ally, fields representing nutrients or waste products can also be distributed on the lattice. Each lattice location is updated at discrete time steps according to simple rules. Given a cell in a specific location, the neighboring com- partments are searched for empty space, and if the nutri- ents in that location are sufficient the cell multiplies and occupies that empty location. Nutrients diffuse stochas- tically through the lattice. More specific rules informed by biological interactions can be implemented. Although cellular automata models appear as a drastic simplifi- cation, they can produce complex structures because of their nonlinear nature [350]. A more recent effort at integrating the experimental knowledge about biofilm growth tackles the challenge of modeling its heterogeneous structure and composi- tion [351, 352]. They consider three kinds of solid com- ponents (the active biomass of microorganisms, the EPS, and the inert biomass) and three kinds of soluble organic compounds (the bacteria's limiting donor substrate, and two types of soluble microbial products, which are ex- creted by cells or released during cell lysis) [353, 354]. The concentrations of all these components evolve in time and space according to partial differential equations that include diffusion, Monod kinetics for the donor substrate, cellular utilization and production, decay of the active biomass and detachment forces. Importantly, biofilm consolidation is included, that is, over time the deeper layers of the biofilm become denser. Coupled to the dynamics of the concentrations of the different biofilm components, the mechanical spreading of active biomass and EPS is implemented through a cellular automata scheme. Once the total density in a compartment ex- ceeds a threshold value, the "mother" cell divides into two "daughter" cells, where one of them moves to a neighbor- ing compartment. Were all neighboring compartments occupied, the daughter cell would push their contents to an unoccupied compartment along the shortest path identified by the algorithm. The threshold density for each compartment grows with time. This fact models the consolidation process due to physical forces acting on the biofilm, for example, hydrodynamic forces tend to induce the solid components to pack more densely. 19 els, particle-based models, or more generally agent-based models. The same basic rules as the cellular automata models for metabolism, substrate uptake, cell multiplica- tion and death are used but no global rules on the growth are imposed. The advantages of agent based simulations are multiple. First, the absence of a lattice grid avoids artificial results such as fixed growth directions. Second, no global laws (for the growth, e.g.) are prescribed, but must emerge from the microscopic dynamics. Third, cells are modeled as hard spheres, and their overlap produces repulsive forces that induces naturally the spreading of the biofilm. Furthermore, since individual agents are re- solved, it is also natural to introduce genetic variabil- ity in the cell's parameters, cell motility [360], coupling with hydrodynamic flow [361], inclusion of metabolic re- actions [362], or a viscous fluid model for the EPS [363]. However, there are also drawbacks [364]: for example, the computational cost of this class of models is typi- cally larger than cellular automata's, and the large-scale structure of biofilm, such as porosity, are less faithfully represented. VII. CONCLUSIONS Biofilms are among the most ancient and most com- mon forms of multicellular organization. They exhibit astounding complexity both at microscopic and meso- scopic length scales. Their importance has been recog- nized only in recent decades and research in this field is growing tremendously, fueled by the possibility of treat- ing infectious diseases or of benefiting from their appetite for organic materials in wastewater applications. We have described the main physical processes in- volved in the inception, growth and dispersal of biofilms. A hydrodynamic description is necessary for the initial transition of microbial cells from the planktonic to the sessile state. Details of the physico-chemical nature of the cellular membrane and its appendages are key for the process of adhesion. Biofilms are macroscopically vis- coelastic materials with a very complex, history depen- dent behavior. Progress in the rheological and mechani- cal characterization will come from studies in conditions mimicking the natural environment. We have also reviewed modern experimental tech- niques of investigation such as microfluidics and lab-on- a-chip technologies that use an integrated approach to biofilms with the aim at capturing their heterogeneous and complex nature. These methods can experimen- tally "simulate" the natural environment of a biofilm and test different hypotheses. Realistic environments are however essential to understand and control the disper- sal of biofilms, for example, which has tremendous con- sequences for ecology and in the context of infectious diseases. Studies of biofilm dispersal, however, are still largely conducted in vitro [93]. Finally, we have reviewed the theoretical and compu- tational modeling of biofilms. Computational approaches at simulating a biofilm are now finally coming of age [365] and some predictions quantitatively match the experi- mental data, but much remains to be done. Ahead lies a particularly daunting challenge when considering all the chemical, biological and physical processes involved in a biofilm, which evolve and interact from the nanometer to the millimeter length scale, and from the nanosecond time scale to hours or days. Future lines of develop- ment should include realistic models for the EPS, either microscopically with a polymer network theory, or meso- scopically with a viscoelastic model. Some attempts are already available [366] where the biofilm is modeled with breakable springs in an immersed boundary method, or through a nonlinear, continuum description of worm-like chains [367]. Once integrated theoretical and computa- tional models will be developed, they will allow to test and manipulate biofilms under multiple physical, chemi- cal and biological cues, both intrinsic and extrinsic. We now want to enter a more speculative realm to dis- cuss some mechanisms at play in the biofilm that could be recognized in the future. The important fact that biofilms represent the most complex effort of simple mi- croorganisms at multicellular organization points to the existence of 'communications' channels among the indi- vidual cells. We know from the biology of multicellular eukaryotes how a complex, well-regulated and special- ized structure may exist. Biofilms are more dynamical and history-dependent in their structure and properties. However, interactions that lead to specialization must be there also in biofilms. Obvious candidates are quorum sensing and cell signaling. Although the role of quorum sensing has been linked multiple times with the growth, 20 structure and function of biofilms, controversial results abound (see Ref. [164] for a critical review of quorum sensing in relation to biofilms). The reason for the con- troversy is probably the inadequacy of our experimental tools of investigation to the task: more subtle tools for in vivo and in situ analysis are required; another factor that complicates the picture is that the study of mutants elic- its the risk of pleiotropy. We speculate that quorum sens- ing and cell signaling provide the communication chan- nels required to generate complex organization. A second physical aspect that is underrepresented in the study of biofilms is the role of motility. We have reviewed the available evidence on the function of swimming, swarm- ing and twitching within a biofilm. We speculate, how- ever, that these motility modes should have a more im- portant role than currently known. While under some physical forces, such as hydrodynamic shear, a biofilm might respond by forming more compact structures with a sessile population, consortia of different species coexist- ing, possibly in syntrophic interaction, may utilize motil- ity to assist the spatial segregation often associated with phenotypic differentiation. Biofilms represent still largely unexplored microcosmoses, that will certainly provide numerous surprises in the future. In this Review, we have omitted the discussion of some crucial aspects in the life-cycle of a biofilm: its ecology and genetic regulation. There is certainly ample space for physical theories of these different levels of interaction. We gratefully acknowledge Gal Schkolnik for helpful conversations and for critically reading the manuscript. We thank Paul Stoodley for kindly providing the images in Fig. 2. [1] Noffke N, Christian D, Wacey D and Hazen R M 2013 [13] O'Toole G, Kaplan H B and Kolter R 2000 Annu. Rev. Astrobiology 13 1103–1124 [2] Van Kranendonk M J, Philippot P, Lepot K, Bodorkos S and Pirajno F 2008 Precambrian Res. 167 93–124 [3] Flemming H C and Wingender J 2010 Nat. Rev. Micro- biol. 8 623–633 Microbiol. 54 49–79 [14] Vert M, Hellwich K H, Hess M, Hodge P, Kubisa P, Rinaudo M and Schué F 2012 Pure Appl. Chem. 84 377–410 [15] Watnick P and Kolter R 2000 J. Bacteriol. 182 2675– [4] Hall-Stoodley L, Costerton J W and Stoodley P 2004 2679 Nat. Rev. Microbiol. 2 95–108 [5] Orell A, Fröls S and Albers S V 2013 Annu. Rev. Mi- crobiol. 67 337–354 [6] Leadbeater B S C and Callow M E 1992 Formation, composition and physiology of algal biofilms Biofilms – Science and Technology (NATO ASI Series vol 223) ed Melo L F, Bott T R, Fletcher M and Capdeville B (Springer Netherlands) pp 149–162 [7] Fanning S and Mitchell A P 2012 PLoS Pathog. 8 e1002585 [8] Ramage G, Mowat E, Jones B, Williams C and Lopez- Ribot J 2009 Crit. Rev. Microbiol. 35 340–355 [9] Porter J R 1976 Bacteriol. Rev. 40 260 [10] Henrici A T 1933 J. Bacteriol. 25 277 [11] Zobell C E 1943 J. Bacteriol. 46 39 [12] McCoy W F, Bryers J D, Robbins J and Costerton J W 1981 Can. J. Microbiol. 27 910–917 [16] Berk V, Fong J C N, Dempsey G T, Develioglu O N, Zhuang X, Liphardt J, Yildiz F H and Chu S 2012 Sci- ence 337 236–239 [17] Reysenbach A L and Cady S L 2001 Trends Microbiol. 9 79–86 145 1254 [18] Taylor C D, Wirsen C O and Gaill F 1999 Appl. Environ. Microbiol. 65 2253–2255 [19] Wotton R S and Preston T M 2005 BioScience 55 137– [20] Atekwana E A and Slater L D 2009 Rev. Geophys. 47 [21] Sihorkar V and Vyas S P 2001 Pharm. Res. 18 1247– [22] Pavithra D and Doble M 2008 Biomed. Mater. 3 034003 [23] Hao O J, Chen J M, Huang L and Buglass R L 1996 Crit. Rev. Env. Sci. Tec. 26 155–187 [24] Hamilton W A 1985 Annu. Rev. Microbiol. 39 195–217 [25] Muyzer G and Stams A J M 2008 Nat. Rev. Microbiol. 6 441–454 eng. [26] Nicolella C, Van Loosdrecht M C M and Heijnen J J 2000 J. Biotechnol. 80 1–33 [27] Du Z, Li H and Gu T 2007 Biotechnol. Adv. 25 464–482 [28] De Beer D and Stoodley P 2013 Microbial biofilms The Prokaryotes ed Rosenberg E, DeLong E, Lory S, Stacke- brandt E and Thompson F (Springer Berlin Heidelberg) pp 343–372 [29] Diels L, van der Lelie N and Bastiaens L 2002 Rev. Env. Sci. Biotechnol. 1 75–82 [30] Gadd G M 2004 Geoderma 122 109–119 [31] Sand W, Gerke T, Hallmann R and Schippers A 1995 Appl. Microbiol. Biot. 43 961–966 [32] Suzuki I 2001 Biotechnol. Adv. 19 119–132 [33] Bosecker K 1997 FEMS Microbiol. Rev. 20 591–604 [34] Lawrence J R, Korber D R, Hoyle B D, Costerton J W and Caldwell D E 1991 J. Bacteriol. 173 6558–6567 [35] Caldwell D E, Korber D R and Lawrence J R 1993 J. Appl. Bacteriol. 74 52S–66S [36] Costerton J W, Lewandowski Z, DeBeer D, Caldwell D, Korber D and James G 1994 J. Bacteriol. 176 2137 [37] Gjaltema A, Arts P A M, Van Loosdrecht M C M, Kue- nen J G and Heijnen J J 1994 Biotechnol. Bioeng. 44 194–204 [38] Stoodley P, Lewandowski Z, Boyle J D and Lappin- Scott H M 1999 Biotechnol. Bioeng. 65 83–92 [39] Stoodley P, Sauer K, Davies D G and Costerton J W 2002 Annu. Rev. Microbiol. 56 187–209 [40] Blauert F, Horn H and Wagner M 2015 Biotechnol. Bio- [41] Madsen J S, Burmølle M, Hansen L H and Sørensen S J 2012 FEMS Immunol. Med. Microbiol. 65 183–195 [42] Wingender J, Strathmann M, Rode A, Leis A and Flem- ming H C 2001 Methods Enzymol. 302–14 [43] Flemming H C, Neu T R and Wozniak D J 2007 J. Bacteriol. 189 7945–7947 [44] Christensen B E 1989 J. Biotechnol. 10 181–202 [45] Sutherland I W and Suthe I 1977 Surface Carbohydrates of the Prokaryotic Cell (Academic Press London) [46] Sutherland I W 2001 Microbiology 147 3–9 [47] Wingender J, Neu T R and Flemming H C 2012 Micro- bial Extracellular Polymeric Substances: Characteriza- tion, Structure and Function (Springer Science & Busi- ness Media) [48] Lawrence J R, Swerhone G D W, Leppard G G, Araki T, Zhang X, West M M and Hitchcock A P 2003 Appl. Environ. Microb. 69 5543–5554 [49] Ryder C, Byrd M and Wozniak D J 2007 Curr. Opin. Microbiol. 10 644–648 [50] Colvin K M, Irie Y, Tart C S, Urbano R, Whitney J C, Ryder C, Howell P L, Wozniak D J and Parsek M R 2012 Environ. Microbiol. 14 1913–1928 [51] Ma L, Jackson K D, Landry R M, Parsek M R and Wozniak D J 2006 J. Bacteriol. 188 8213–8221 [52] Cooley B J, Thatcher T W, Hashmi S M, L'Her G, Le H H, Hurwitz D A, Provenzano D, Touhami A and Gordon V D 2013 Soft Matter 9 3871–3876 [53] Donlan R M 2002 Emerg. Infect. Dis. 8 881–890 [54] Spiers A J, Bohannon J, Gehrig S M and Rainey P B 2003 Mol. Microbiol. 50 15–27 [55] Götz F 2002 Mol. Microbiol. 43 1367–1378 [56] Mack D, Fischer W, Krokotsch A, Leopold K, Hartmann R, Egge H and Laufs R 1996 J. Bacteriol. 178 175–183 [57] Seminara A, Angelini T E, Wilking J N, Vlamakis H, 21 Ebrahim S, Kolter R, Weitz D A and Brenner M P 2012 Proc. Natl. Acad. Sci. USA 109 1116–1121 [58] Wilking J N, Angelini T E, Seminara A, Brenner M P and Weitz D A 2011 MRS Bull. 36 385–391 [59] Rubinstein M and Colby R H 2003 Polymer Physics (Oxford University Press, Oxford) [60] Or D, Phutane S and Dechesne A 2007 Vadose Zone J. [61] Freeman C and Lock M A 1995 Limnol. Oceanogr. 40 [62] Norberg A B and Persson H 1984 Biotechnol. Bioeng. 6 298–305 273–278 26 239–246 [63] Teitzel G M and Parsek M R 2003 Appl. Environ. Mi- crobiol. 69 2313–2320 [64] Hentzer M, Teitzel G M, Balzer G J, Heydorn A, Molin S, Givskov M and Parsek M R 2001 J Bacteriol. 183 5395–5401 [65] Colvin K M, Gordon V D, Murakami K, Borlee B R, Wozniak D J, Wong G C and Parsek M R 2011 PLoS Pathog. 7 e1001264–e1001264 [66] Billings N, Ramirez Millan M, Caldara M, Rusconi R, Tarasova Y, Stocker R and Ribbeck K 2013 PLoS Pathog. 9 e1003526 [67] Wingender J, Jaeger K E and Flemming H C 1999 Interaction between extracellular polysaccharides and enzymes Microbial Extracellular Polymeric Substances (Springer) pp 231–251 [68] Lock M A, Wallace R R, Costerton J W, Ventullo R M and Charlton S E 1984 Oikos 42 10–22 [69] Otero A and Vincenzini M 2004 J. Phycol. 40 74–81 [70] Steinberger R and Holden P 2005 Appl. Environ. Mi- crobiol. 71 5404–5410 [71] Yang L, Barken K B, Skindersoe M E, Christensen A B, Givskov M and Tolker-Nielsen T 2007 Microbiology 153 1318–1328 [72] Whitchurch C B, Tolker-Nielsen T, Ragas P C and Mattick J S 2002 Science 295 1487–1487 [73] Vilain S, Pretorius J M, Theron J and Brözel V S 2009 Appl. Environ. Microbiol. 75 2861–2868 [74] Mulcahy H, Charron-Mazenod L and Lewenza S 2008 [75] Ahimou F, Jacques P and Deleu M 2000 Enzyme Mi- [76] Irie Y, O'toole G A and Yuk M H 2005 FEMS Microbiol. PLoS Pathog. 4 e1000213 crob. Tech. 27 749–754 Lett. 250 237–243 [77] Rodrigues L R, Banat I M, Mei H C, Teixeira J A and Oliveira R 2006 J. Appl. Microbiol. 100 470–480 [78] Rivardo F, Turner R J, Allegrone G, Ceri H and Mar- tinotti M G 2009 Appl. Microbiol. Biot. 83 541–553 [79] Romaní A M, Fund K, Artigas J, Schwartz T, Sabater S and Obst U 2008 Microb. Ecol. 56 427–436 [80] Zhang X and Bishop P L 2003 Chemosphere 50 63–69 [81] Elkins J G, Hassett D J, Stewart P S, Schweizer H P and McDermott T R 1999 Appl. Environ. Microb. 65 4594–4600 [82] Barondes S H 1984 Science 223 1259–1264 [83] Diggle S P, Stacey R E, Dodd C, Cámara M, Williams P and Winzer K 2006 Environ. Microb. 8 1095–1104 [84] Pérez-Giménez J, Mongiardini E J, Althabegoiti M J, Covelli J, Quelas J I, López-García S L and Lodeiro A R 2009 Int. J. Microbiol. 2009 719367 [85] Karatan E and Watnick P 2009 Microbiol. Mol. Biol. R. 73 310–347 [86] Sauer K, Cullen M C, Rickard A H, Zeef L A H, Davies D G and Gilbert P 2004 J Bacteriol. 186 7312–7326 [87] Boles B R, Thoendel M and Singh P K 2005 Mol. Mi- crobiol. 57 1210–1223 [88] Jackson D W, Suzuki K, Oakford L, Simecka J W, Hart M E and Romeo T 2002 Journal of bacteriology 184 290–301 [89] McDougald D, Rice S A, Barraud N, Steinberg P D and Kjelleberg S 2012 Nat. Rev. Microbiol. 10 39–50 [90] Crespi B J 2001 Trends Ecol. Evol. 16 178–183 [91] Shapiro J A 1998 Annu. Rev. Microbiol. 52 81–104 [92] Ben-Jacob E, Cohen I and Levine H 2000 Adv. Phys. 49 395–554 [93] Kaplan J B 2010 J. Dent. Res. 89 205–218 [94] Teschler J K, Zamorano-Sánchez D, Utada A S, Warner C J A, Wong G C L, Linington R G and Yildiz F H 2015 Nat. Rev. Microbiol. 13 255–268 [95] Sauer K, Camper A K, Ehrlich G D, Costerton J W and Davies D G 2002 J. Bacteriol. 184 1140–1154 [96] Tolker-Nielsen T, Brinch U C, Ragas P C, Andersen J B, Jacobsen C S and Molin S 2000 J. Bacteriol. 182 6482–6489 [97] Barken K B, Pamp S J, Yang L, Gjermansen M, Bertrand J J, Klausen M, Givskov M, Whitchurch C B, Engel J N and Tolker-Nielsen T 2008 Environ. Micro- biol. 10 2331–2343 [98] Vlamakis H, Aguilar C, Losick R and Kolter R 2008 Genes Dev. 22 945–953 [99] Henrichsen J 1972 Bacteriol Rev. 36 478 [100] Turner L, Ryu W S and Berg H C 2000 J. Bacteriol. 182 2793–2801 [101] Berg H C 2008 Curr. Biol. 18 R689–R691 [102] Parkinson J S 1993 Cell 73 857–871 [103] Block S M, Blair D F and Berg H C 1989 Nature 338 [104] Berg H C 2003 Ann. Rev. Biochem. 72 19 [105] Lauga E and Powers T R 2009 Rep. Prog. Phys. 72 514–518 096601 [106] Koch D L and Subramanian G 2011 Annu. Rev. Fluid Mech. 43 637–659 [107] Happel J and Brenner H 2012 Low Reynolds Number Hydrodynamics: with special applications to particulate media (Springer Science & Business Media) [108] Copeland M F and Weibel D B 2009 Soft Matter 5 1174– [109] Guyon E, Hulin J P, Petit L and Mitescu C D 2001 Physical Hydrodynamics (Springer, Berlin Heidelberg, New York) [110] Purcell E M 1977 Am. J. Phys 45 3–11 [111] Baskaran A and Marchetti M C 2009 Proc. Natl. Acad. Sci. USA 106 15567–15572 [112] Simha R A and Ramaswamy S 2002 Phys. Rev. Lett. [113] Saintillan D and Shelley M J 2007 Phys. Rev. Lett. 99(5) 89(5) 058101 058102 [114] Kessler J O and Wojciechowski M F 1997 Collective behavior and dynamics of swimming bacteria Bacteria as multicellular organisms ed Shapiro J A and Dworkin M (Oxford University Press) pp 417–450 [115] Dombrowski C, Cisneros L, Chatkaew S, Goldstein R E and Kessler J O 2004 Phys. Rev. Lett. 93(9) 098103 [116] Sokolov A and Aranson I S 2012 Phys. Rev. Lett. 109 1187 248109 [117] Dunkel J, Heidenreich S, Drescher K, Wensink H H, Bär M and Goldstein R E 2013 Phys. Rev. Lett. 110 228102 317 105–108 [151] Das D, Das D and Prasad A 2012 J. Theor. Biol. 308 22 9 339–398 [118] Zöttl A and Stark H 2014 Phys. Rev. Lett. 112 118101 [119] Wensink H H, Dunkel J, Heidenreich S, Drescher K, Goldstein R E, Löwen H and Yeomans J M 2012 Proc. Natl. Acad. Sci. USA 109 14308–14313 [120] Drescher K, Dunkel J, Cisneros L H, Ganguly S and Goldstein R E 2011 Proc. Natl. Acad. Sci. USA 108 10940–10945 [121] Lushi E, Wioland H and Goldstein R E 2014 Proc. Natl. Acad. Sci. USA 201405698 [122] Katz D F 1974 J. Fluid Mech. 64 33–49 [123] Brennen C and Winet H 1977 Annu. Rev. Fluid Mech. [124] DiLuzio W R, Turner L, Mayer M, Garstecki P, Weibel D B, Berg H C and Whitesides G M 2005 Nature 435 1271–1274 [125] Kudo S, Imai N, Nishitoba M, Sugiyama S and Maga- riyama Y 2005 FEMS Microbiol. Lett. 242 221–225 [126] Magariyama Y, Ichiba M, Nakata K, Baba K, Ohtani T, Kudo S and Goto T 2005 Biophys. J. 88 3648–3658 [127] Li G, Tam L K and Tang J X 2008 Proc. Natl. Acad. Sci. USA 105 18355–18359 [128] Rothschild L 1963 Nature 198 1221 [129] Cosson J, Huitorel P and Gagnon C 2003 Cell Motil. Cytoskel. 54 56–63 Rev. Lett. 101 038102 [130] Berke A P, Turner L, Berg H C and Lauga E 2008 Phys. [131] Blake J R and Chwang A T 1974 J. Eng. Math. 8 23–29 [132] Kearns D B 2010 Nature Rev. Microbiol. 8 634–644 [133] Darnton N C, Turner L, Rojevsky S and Berg H C 2010 Biophys. J. 98 2082–2090 [134] Shinoda S and Okamoto K 1977 Journal of bacteriology 129 1266–1271 [135] McCarter L 1999 J Mol. Microbiol. Biotechnol. 1 51–57 [136] Merino S, Shaw J G and Tomás J M 2006 FEMS Mi- crobiol. Lett. 263 127–135 [137] Gavin R, Rabaan A A, Merino S, Tomás J M, Gryllos I and Shaw J G 2002 Mol Microbiol. 43 383–397 [138] Kirov S M, Tassell B C, Semmler A B T, O'Donovan L A, Rabaan A A and Shaw J G 2002 J. Bacteriol. 184 547–555 [139] Jones H E and Park R W A 1967 J. Gen. Microbiol. 47 [140] Harshey R M 2003 Ann. Rev. Microbiol. 57 249–273 [141] Kearns D B and Losick R 2003 Mol. Microbiol. 49 581– 369–378 590 [142] Julkowska D, Obuchowski M, Holland I B and Séror S J 2005 Journal of bacteriology 187 65–76 [143] Lindum P W, Anthoni U, Christophersen C, Eberl L, Molin S and Givskov M 1998 J. Bacteriol. 180 6384– 6388 [144] Eberl L, Molin S and Givskov M 1999 J. Bacteriol. 181 [145] Copeland M F, Flickinger S T, Tuson H H and Weibel D B 2010 Appl. Environ. Microbiol. 76 1241–1250 [146] Morrison R B and Scott A 1966 Nature 211 255–257 [147] Jones B V, Young R, Mahenthiralingam E and Stickler D J 2004 Infect. Immun. 72 3941–3950 [148] Zhang H P, Be'er A, Florin E L and Swinney H L 2010 Proc. Natl. Acad. Sci. USA 107 13626–13630 [149] Bonabeau E, Dagorn L and Fréon P 1999 Proc. Natl. Acad. Sci. USA 96 4472–4477 [150] Narayan V, Ramaswamy S and Menon N 2007 Science 1703–1712 96–104 [152] Shimada H, Ikeda T, ichi Wakita J, Itoh H, Kurosu S, Hiramatsu F, Nakatsuchi M, Yamazaki Y, Matsuyama T and Matsushita M 2004 J. Phys. Soc. Japan 73 1082– 1089 [153] Hiramatsu F, Wakita J i, Kobayashi N, Yamazaki Y, Matsushita M and Matsuyama T 2005 Microbes Envi- ron. 20 120–125 [154] Arouh S 2001 Phys. Rev. E 63 031908 [155] Czirók A, Matsushita M and Vicsek T 2001 Phys. Rev. [156] Golding I, Kozlovsky Y, Cohen I and Ben-Jacob E 1998 E 63 031915 Physica A 260 510–554 [157] Wakano J Y, Maenosono S, Komoto A, Eiha N and Yamaguchi Y 2003 Phys. Rev. Lett. 90 258102–258102 [158] Ben-Jacob E, Cohen I, Czirók A, Vicsek T and Gutnick D L 1997 Physica A 238 181–197 [159] Ben-Jacob E, Schochet O, Tenenbaum A, Cohen I, Czirok A and Vicsek T 1994 Nature 368 46–49 [160] Ben-Jacob E, Cohen I, Shochet O, Tenenbaum A, Czirók A and Vicsek T 1995 Phys. Rev. Lett. 75 2899 [161] Wensink H H and Löwen H 2012 J. Phys.: Condens. Matter 24 464130 [162] Shrout J D, Chopp D L, Just C L, Hentzer M, Givskov M and Parsek M R 2006 Mol. Microbiol. 62 1264–1277 [163] Verstraeten N, Braeken K, Debkumari B, Fauvart M, Fransaer J, Vermant J and Michiels J 2008 Trends Mi- crobiol. 16 496–506 [164] Parsek M R and Greenberg E P 2005 Trends Microbiol. [165] Hahn H P 1997 Gene 192 99–108 [166] Craig L, Pique M E and Tainer J A 2004 Nat. Rev. Microbiol. 2 363–378 USA 98 6901–6904 [167] Skerker J M and Berg H C 2001 Proc. Natl. Acad. Sci. [168] Craig L and Li J 2008 Curr. Opin. Struc. Biol. 18 267– [169] Mattick J S 2002 Annu. Rev. Microbiol. 56 289–314 [170] Merz A J, So M and Sheetz M P 2000 Nature 407 98– 13 27–33 277 102 1143–1146 [172] Maier B, Potter L, So M, Seifert H S and Sheetz M P 2002 Proc. Natl. Acad. Sci. USA 99 16012–16017 [173] Clausen M, Koomey M and Maier B 2009 Biophys. J. 96 1169–1177 Maier B 2009 J. Bacteriol. 191 4633–4638 [175] Holz C, Opitz D, Greune L, Kurre R, Koomey M, Schmidt M A and Maier B 2010 Phys. Rev. Lett. 104 178104 [176] Conrad J C 2012 Res. Microbiol. 163 619–629 [177] Müller M J I, Klumpp S and Lipowsky R 2008 Proc. Natl. Acad. Sci. USA 105 4609–4614 [178] Marathe R, Meel C, Schmidt N C, Dewenter L, Kurre R, Greune L, Schmidt M A, Müller M J I, Lipowsky R, Maier B and Klumpp S 2014 Nat. Comm. 5 [179] Jin F, Conrad J C, Gibiansky M L and Wong G C 2011 Proc. Natl. Acad. Sci. USA 108 12617–12622 [180] Gibiansky M L, Conrad J C, Jin F, Gordon V D, Motto D A, Mathewson M A, Stopka W G, Zelasko D C, Shrout J D and Wong G C L 2010 Science 330 197– 197 [181] Conrad J C, Gibiansky M L, Jin F, Gordon V D, Motto 23 D A, Mathewson M A, Stopka W G, Zelasko D C, Shrout J D and Wong G C L 2011 Biophys. J. 100 1608–1616 [182] Zaburdaev V, Denisov S and Klafter J 2015 Rev. Mod. [183] Montroll E W and Weiss G H 1965 J. Math. Phys. 6 [184] Bouchaud J P and Georges A 1990 Phys. Rep. 195 127– Phys. 87 483 167–181 293 [185] Viswanathan G M, Da Luz M G E, Raposo E P and Stanley H E 2011 The physics of foraging: an intro- duction to random searches and biological encounters (Cambridge University Press) [186] Méndez V, Campos D and Bartumeus F 2013 Stochastic foundations in movement ecology: anomalous diffusion, front propagation and random searches (Springer Sci- ence & Business Media) [187] Taktikos J, Stark H and Zaburdaev V 2013 PLoS ONE [188] Korobkova E, Emonet T, Vilar J M G, Shimizu T S and Cluzel P 2004 Nature 428 574–578 [189] Burrows L L 2012 Annu. Rev. Microbiol. 66 493–520 [190] Klausen M, Aaes-Jørgensen A, Molin S and Tolker- Nielsen T 2003 Mol. Microbiol. 50 61–68 [191] O'Toole G A and Kolter R 1998 Mol. Microbiol. 30 295– 8 e81936 [192] Meng Y, Li Y, Galvani C D, Hao G, Turner J N, Burr T J and Hoch H C 2005 J. Bacteriol. 187 5560–5567 [193] Li Y, Hao G, Galvani C D, Meng Y, De La Fuente L, Hoch H and Burr T J 2007 Microbiology 153 719–726 [194] Watnick P I and Kolter R 1999 Mol. Microbiol. 34 586– 304 595 [195] Geiger B, Bershadsky A, Pankov R and Yamada K M 2001 Nat. Rev. Mol. Cell Bio. 2 793–805 [196] Nan B, Chen J, Neu J C, Berry R M, Oster G and Zusman D R 2011 Proc. Natl. Acad. Sci. USA 108 2498– 2503 [197] Zamir E and Geiger B 2001 J. Cell Sci. 114 3583–3590 [198] McBride M J 2001 Annu. Rev. Microbiol. 55 49–75 [199] Mignot T, Shaevitz J W, Hartzell P L and Zusman D R [200] Peruani F, Starruss J, Jakovljevic V, Søgaard-Andersen L, Deutsch A and Bär M 2012 Phys. Rev. Lett. 108 098102 [201] Ginelli F, Peruani F, Bär M and Chaté H 2010 Phys. Rev. Lett. 104(18) 184502 74(3) 030904 E 82(3) 031904 181 7331–7338 [203] Yang Y, Marceau V and Gompper G 2010 Phys. Rev. [204] Martínez A, Torello S and Kolter R 1999 J. Bacteriol. [205] Recht J and Kolter R 2001 J. Bacteriol. 183 5718–5724 [206] Chen J M, German G J, Alexander D C, Ren H, Tan T and Liu J 2006 J. Bacteriol. 188 633–641 [207] Wu C w, Schmoller S K, Bannantine J P, Eckstein T M, Inamine J M, Livesey M, Albrecht R and Talaat A M 2009 Microb. Pathogenesis 46 222–230 [208] Maya-Hoyos M, Leguizamón J, Mariño-Ramírez L and Soto C Y 2015 BioMed Research International 2015 419549 [209] Davies D G, Chakrabarty A M and Geesey G G 1993 Appl. Environ. Microbiol. 59 1181–1186 [210] Christensen G D, Simpson W A and Beachey E H 1985 [171] Sun H, Zusman D R and Shi W 2000 Curr. Biol. 10 2007 Science 315 853–856 [174] Clausen M, Jakovljevic V, Søgaard-Andersen L and [202] Peruani F, Deutsch A and Bär M 2006 Phys. Rev. E Adhesion of bacteria to animal tissues Bacterial adhe- sion: mechanisms and physiological significance ed Sav- age D and Fletcher M (Springer US) pp 279–305 [211] Gristina A G 1987 Science 237 1588–1595 [212] Marshall K C, Stout R and Mitchell R 1971 J. Gen. Microbiol. 68 337–348 [213] Dunne W M 2002 Clin. Microbiol. Rev. 15 155–166 [214] Carpentier B and Cerf O 1993 J. Appl. Bacteriol. 75 [215] Jucker B A, Harms H and Zehnder A J 1996 J. Bacte- 499–511 riol. 178 5472–5479 14 331–354 59 [216] Derjaguin B and Landau L 1941 Acta Phys-Chim USSR Nature 356 252–255 [217] Derjaguin B and Landau L 1993 Prog Surf. Sci. 43 30– 2007 Science 318 1625–1628 [218] Verwey E J W and Overbeek J T G 1948 Theory of the stability of lyophobic colloids (Elsevier) [219] Israelachvili J 1987 Accounts Chem. Res. 20 415–421 [220] Chandler D 2005 Nature 437 640–647 [221] Israelachvili J N 2011 Intermolecular and surface forces (Academic Press) [222] Loskill P, Hähl H, Thewes N, Kreis C T, Bischoff M, Herrmann M and Jacobs K 2012 Langmuir 28 7242– 7248 [223] Hori K and Matsumoto S 2010 Biochem. Eng. J. 48 [224] Dufrêne Y F 2008 Nat. Rev. Microbiol. 6 674–680 [225] Kotra L P, Golemi D, Amro N A, Liu G Y and Mobash- ery S 1999 J. Am. Chem. Soc. 121 8707–8711 [226] Walker S L, Redman J A and Elimelech M 2004 Lang- 424–434 [228] Jucker B A, Harms H, Hug S J and Zehnder A J B 1997 Colloids Surf. B 9 331–343 [229] Makin S A and Beveridge T J 1996 Microbiology 142 muir 20 7736–7746 8503–8509 299–307 Arch. Microbiol. 172 1–8 Technol. 37 2173–2183 24 2006 Phys. Rev. E 73(4) 041604 [243] Giovambattista N, Lopez C F, Rossky P J and Debenedetti P G 2008 Proc. Natl. Acad. Sci. USA 105 2274–2279 [244] Tsang P H, Li G, Brun Y V, Freund L B and Tang J X 2006 Proc. Natl. Acad. Sci. USA 103 5764–5768 [245] Ofek I, Mirelman D and Sharon N 1977 Nature 265 [246] Hanson M S and Brinton Jr C C 1988 Nature 332 265– 623–625 268 [247] Kuehn M J, Heuser J, Normark S and Hultgren S J 1992 [248] Kang H J, Coulibaly F, Clow F, Proft T and Baker E N [249] Nobbs A H, Lamont R J and Jenkinson H F 2009 Mi- crobiol. Mol. Biol. Rev. 73 407–450 [250] Veenstra G J C, Cremers F F M, van Dijk H and Fleer A 1996 J. Bacteriol. 178 537–541 [251] Heilmann C, Hussain M, Peters G and Gotz F 1997 Mol. Microbiol. 24 1013–1024 [252] Heilmann C, Thumm G, Chhatwal G S, Hartleib J, Uekötter A and Peters G 2003 Microbiology 149 2769– 2778 [253] Loeb G I and Neihof R A 1975 Adv. Chem. 145 319–335 [254] Bos R, Van der Mei H C and Busscher H J 1999 FEMS Microbiol. Rev. 23 179–230 [255] Schneider R P and Marshall K C 1994 Colloids Surf. B [256] Schneider R P 1996 J. Colloid Interface Sci. 182 204– 2 387–396 213 [257] de Kerchove A J and Elimelech M 2007 Appl. Environ. [258] Lorite G S, Rodrigues C M, De Souza A A, Kranz C, Mizaikoff B and Cotta M A 2011 J. Colloid Interface Sci. 359 289–295 [259] Larson R G 1999 The structure and rheology of complex fluids (Oxford University Press New York) P 2002 Biotechnol. Bioeng. 80 289–296 ley P 2003 Biofouling 19 279–285 [262] Vinogradov A M, Winston M, Rupp C J and Stoodley P 2004 Biofilms 1 49–56 [263] Shaw T, Winston M, Rupp C J, Klapper I and Stoodley P 2004 Phys. Rev. Lett. 93(9) 098102 [264] Rupp C J, Fux C A and Stoodley P 2005 Appl. Environ. Microbiol. 71 2175–2178 [265] Lau P C Y, Dutcher J R, Beveridge T J and Lam J S 2009 Biophys. J. 96 2935–2948 [266] Hohne D N, Younger J G and Solomon M J 2009 Lang- muir 25 7743–7751 Soft Matter 7 3307–3314 [227] Atabek A and Camesano T A 2007 J. Bacteriol. 189 Microbiol. 73 5227–5234 [230] Genevaux P, Bauda P, DuBow M S and Oudega B 1999 [260] Klapper I, Rupp C, Cargo R, Purvedorj B and Stoodley [231] Abu-Lail N I and Camesano T A 2003 Environ. Sci. & [261] Towler B W, Rupp C J, Cunningham A L B and Stood- [232] Thewes N, Thewes A, Loskill P, Peisker H, Bischoff M, Herrmann M, Santen L and Jacobs K 2015 Soft Matter XXX XXX [233] Tsuneda S, Aikawa H, Hayashi H, Yuasa A and Hirata A 2003 FEMS Microbiol. Lett. 223 287–292 [234] van Loosdrecht M C M, Norde W, Lyklema J and Zehn- der A J B 1990 Aquat. Sci. 52 103–114 [235] Azeredo J and Oliveira R 2000 Biofouling 16 59–67 [236] van Loosdrecht M C M and Zehnder A J B 1990 Expe- rientia 46 817–822 B 14 141–148 [237] Azeredo J, Visser J and Oliveira R 1999 Colloids Surf. [267] Lieleg O, Caldara M, Baumgärtel R and Ribbeck K 2011 [238] Rosenberg M and Kjelleberg S 1986 Hydrophobic inter- actions: role in bacterial adhesion Advances in microbial ecology (Springer US) pp 353–393 [239] van Loosdrecht M C M, Lyklema J, Norde W, Schraa G and Zehnder A J B 1987 Appl. Environ. Microbiol. 53 1893–1897 [240] Thewes N, Loskill P, Jung P, Peisker H, Bischoff M, Her- rmann M and Jacobs K 2014 Beilstein J. Nanotechnol. 5 1501–1512 [241] Stillinger F H 1973 J. Solution Chem. 2 141–158 [242] Giovambattista N, Rossky P J and Debenedetti P G [268] Billings N, Birjiniuk A, Samad T S, Doyle P S and Ribbeck K 2015 Rep. Prog. Phys. 78 036601 [269] Körstgens V, Flemming H C, Wingender J and Bor- chard W 2001 J. Microbiol. Meth. 46 9–17 [270] Wloka M, Rehage H, Flemming H C and Wingender J 2004 Colloid Polym. Sci. 282 1067–1076 [271] Joseph D D 2013 Fluid dynamics of viscoelastic liquids (Springer Science & Business Media) [272] Morozov A and Spagnolie S E 2015 Introduction to complex fluids Complex Fluids in Biological Systems (Springer) pp 3–52 [273] Stoodley P, Cargo R, Rupp C J, Wilson S and Klapper I 2002 J. Ind. Microbiol. Biot. 29 361–367 [274] Cheong F C, Duarte S, Lee S H and Grier D G 2009 Rheol. Acta 48 109–115 [302] Groisman A, Lobo C, Cho H, Campbell J K, Dufour Y S, Stevens A M and Levchenko A 2005 Nat. Methods 2 685–689 [303] Valiei A, Kumar A, Mukherjee P P, Liu Y and Thundat [275] Wileman A, Ozkan A and Berberoglu H 2012 Biore- T 2012 Lab Chip 12 5133–5137 25 source Technol. 104 432–439 [276] Van Dyk T K, Majarian W R, Konstantinov K B, Young R M, Dhurjati P S and LaRossa R A 1994 Appl. Envi- ron. Microbiol. 60 1414–1420 [277] Ptitsyn L R, Horneck G, Komova O, Kozubek S, Krasavin E A, Bonev M and Rettberg P 1997 Appl En- viron. Microbiol. 63 4377–4384 [278] Rusconi R, Lecuyer S, Guglielmini L and Stone H A 2010 J. R. Soc. Interface 7 1293–1299 [279] Rusconi R, Lecuyer S, Autrusson N, Guglielmini L and Stone H A 2011 Biophys. J. 100 1392–1399 [280] Drescher K, Shen Y, Bassler B L and Stone H A 2013 Proc. Natl. Acad. Sci. USA 110 4345–4350 [281] Stoodley P, Lewandowski Z, Boyle J D and Lappin- Scott H M 1999 Environ Microbiol. 1 447–455 [282] Thomas K, Herminghaus S, Porada H and Goehring L 2013 Phil. Trans. R. Soc. A 371 20120362 [283] Walker T S, Tomlin K L, Worthen G S, Poch K R, Lieber J G, Saavedra M T, Fessler M B, Malcolm K C, Vasil M L and Nick J A 2005 Infect. Immun. 73 3693– 3701 [284] Andes D, Nett J, Oschel P, Albrecht R, Marchillo K and Pitula A 2004 Infect. Immun. 72 6023–6031 [285] Phillips R J, Deen W M and Brady J F 1989 AIChE J. 35 1761–1769 775 [286] Masaro L and Zhu X X 1999 Prog. Polym. Sci. 24 731– [287] Amsden B 1998 Macromolecules 31 8382–8395 [288] Wool R P 2008 Soft Matter 4 400–418 [289] Wu D Y, Meure S and Solomon D 2008 Prog. Polym. Sci. 33 479–522 M J 2015 Sci. Rep. 5 [291] Simon H A 1991 The architecture of complexity Facets of Systems Science (International Federation for Sys- tems Research International Series on Systems Science and Engineering vol 7) (Springer US) pp 457–476 [292] Morgenroth E and Milferstedt K 2009 Rev. Environ. Sci. Biotechnol. 8 203–208 [304] Li Z, Zhang Y, LeDuc P R and Gregory K B 2011 Biotechnol. Bioeng. 108 2061–2069 [305] Nealson K H and Saffarini D 1994 Annu. Rev. Microbiol. 48 311–343 [306] Gorby Y A, Yanina S, McLean J S, Rosso K M, Moyles D, Dohnalkova A, Beveridge T J, Chang I S, Kim B H, Kim K S, Culley D E, Reed S B, Romine M F, Saffarini D A, Hill E A, Shi L, Elias D A, Kennedy D W, Pinchuk G, Watanabe K, Ishii S, Logan B, Nealson K H and Fredrickson J K 2006 Proc. Natl. Acad. Sci. USA 103 11358–11363 [307] Schkolnik G, Schmidt M, Mazza M G, Harnisch F and Musat N 2015 PLoS One 10 e0145871 [308] Weibel D B, DiLuzio W R and Whitesides G M 2007 Nat. Rev. Microbiol. 5 209–218 [309] Hol F J H and Dekker C 2014 Science 346 1251821 [310] Rusconi R, Garren M and Stocker R 2014 Annu. Rev. Biophys. 43 65 [311] Friedlander R S, Vlamakis H, Kim P, Khan M, Kolter R and Aizenberg J 2013 Proc. Natl. Acad. Sci. USA 110 5624–5629 [312] Nadell C D, Xavier J B and Foster K R 2009 FEMS Microbiol. Rev. 33 206–224 [313] Keymer J E, Galajda P, Muldoon C, Park S and Austin R H 2006 Proc. Natl. Acad. Sci. USA 103 17290–17295 [314] Kim H J, Boedicker J Q, Choi J W and Ismagilov R F 2008 Proc. Natl. Acad. Sci. USA 105 18188–18193 [315] Connell J L, Ritschdorff E T, Whiteley M and Shear J B 2013 Proc. Natl. Acad. Sci. USA 110 18380–18385 [316] Hol F J, Galajda P, Woolthuis R G, Dekker C and Keymer J E 2015 BMC Res. Notes 8 245 Microbiol. 14 601–622 [318] Van Uden N 1967 Arch. Mikrobiol. 58 155–168 [319] Monod J 1942 Recherches sur la croissance des cultures bacteriennes (Paris, Hermann et Cie) [320] Rittmann B E and McCarty P L 1980 Biotechnol. Bio- eng. 22 2343–2357 [321] Kissel J C, McCarty P L and Street R L 1984 J. Envi- [322] Wanner O and Gujer W 1986 Biotechnol. Bioeng. 28 [323] Chaudhry M A S and Beg S A 1998 Chem. Eng. Tech- nol. 21 701–710 [324] Costerton J W, Lewandowski Z, Caldwell D E, Korber D R and Lappin-Scott H M 1995 Annu. Rev. Microbiol. 49 711–745 [325] Eberl H J, Parker D F and Van Loosdrecht M C M 2001 [290] Stewart E J, Ganesan M, Younger J G and Solomon [317] Herbert D, Elsworth R and Telling R C 1956 J. Gen. [293] Vertes A, Hitchins V and Phillips K S 2012 Anal Chem. ron. Eng. 110 393–411 [294] Karimi A, Karig D, Kumar A and Ardekani A M 2015 314–328 84 3858–3866 Lab Chip 15 23–42 [295] De La Fuente L, Montanes E, Meng Y, Li Y, Burr T J, Hoch H C and Wu M 2007 Appl. Environ. Microbiol. 73 2690–2696 [296] Bahar O, De La Fuente L and Burdman S 2010 FEMS Microbiol. Lett. 312 33–39 [297] Wright E, Neethirajan S, Warriner K, Retterer S and J. Theor. Med. 3 161–175 Srijanto B 2014 Lab Chip 14 938–946 [326] Picioreanu C, Van Loosdrecht M C M and Heijnen J J [298] Lee J H, Kaplan J B and Lee W Y 2008 Biomed. Mi- 1998 Biotechnol. Bioeng. 58 101–116 crodevices 10 489–498 [299] Kim K P, Kim Y G, Choi C H, Kim H E, Lee S H, Chang W S and Lee C S 2010 Lab Chip 10 3296–3299 [300] Richter L, Stepper C, Mak A, Reinthaler A, Heer R, Kast M, Brückl H and Ertl P 2007 Lab Chip 7 1723– 1731 [301] Hong S H, Hegde M, Kim J, Wang X, Jayaraman A and Wood T K 2012 Nature Comm. 3 613 [327] Eberl H J, Picioreanu C, Heijnen J J and Van Loos- drecht M C M 2000 Chem. Eng. Sci. 55 6209–6222 [328] Noguera D R, Pizarfo G, Stahl D A and Rittmann B E 1999 Water Sci. Technol. 39 123–130 [329] Fux C A, Wilson S and Stoodley P 2004 J. Bacteriol. 186 4486–4491 [330] Klapper I and Dockery J 2006 Phys. Rev. E 74 031902 [331] Klapper I and Dockery J 2010 SIAM Rev. 52 221–265 [332] Wang Q and Zhang T 2010 Solid State Commun. 150 [333] Tierra G, Pavissich J P, Nerenberg R, Xu Z and Alber M S 2015 J. R. Soc. Interface 12 20150045 [334] Lindley B, Wang Q and Zhang T 2012 Phys. Rev. E [335] Fujikawa H and Matsushita M 1989 J. Phys. Soc. Japan 1009–1022 85(3) 031908 58 3875–3878 506 1400–1403 [336] Matsushita M and Fujikawa H 1990 Physica A 168 498– biology 144 3275–3287 26 [352] Laspidou C S and Rittmann B E 2004 Water Res. 38 [353] Laspidou C S and Rittmann B E 2002 Water Res. 36 [354] Laspidou C S and Rittmann B E 2002 Water Res. 36 3349–3361 3362–3372 2711–2720 1983–1992 [355] Kreft J U, Booth G and Wimpenny J W T 1998 Micro- [356] Kreft J U, Picioreanu C, Wimpenny J W T and van Loosdrecht M C M 2001 Microbiology 147 2897–2912 [357] Kreft J U and Wimpenny J W T 2001 Water Sci. Tech- nol. 43 135–142 [358] Picioreanu C, Kreft J U and van Loosdrecht M C M 2004 Appl. Environ. Microbiol. 70 3024–3040 [359] Xavier J B, Picioreanu C and Van Loosdrecht M 2005 Environ. Microbiol. 7 1085–1103 [360] Picioreanu C, Kreft J U, Klausen M, Haagensen J A J, Tolker-Nielsen T and Molin S 2007 Water Sci. Technol. 55 337–343 [361] Graf von der Schulenburg D A, Pintelon T R R, Pi- cioreanu C, Van Loosdrecht M C M and Johns M L 2009 AIChE J. 55 494–504 ISSN 1547-5905 [362] Lardon L A, Merkey B V, Martins S, Dötsch A, Pi- cioreanu C, Kreft J U and Smets B F 2011 Environ. Microbiol. 13 2416–2434 [363] Alpkvist E, Picioreanu C, van Loosdrecht M and Hey- den A 2006 Biotechnol. Bioeng. 94 961–979 nation 250 390–394 [365] Kreft J U, Plugge C M, Grimm V, Prats C, Leveau J H J, Banitz T, Baines S, Clark J, Ros A, Klapper I, Topping C J, Field A J, Schuler A, Litchman E and Hellweger F L 2013 Proc. Natl. Acad. Sci. USA 110 18027–18028 [366] Alpkvist E and Klapper I 2007 Water Sci. Technol. 55 [367] Ehret A E and Böl M 2013 J. R. Soc. Interface 10 265–273 20120676 [337] Fujikawa H 1994 FEMS Microbiol. Ecol. 13 159–168 [338] Witten T A and Sander L M 1981 Phys. Rev. Lett. 47 [339] Eden M 1961 A two-dimensional growth process Pro- ceedings of the Fourth Berkeley Symposium on Mathe- matical Statistics and Probability, Volume 4: Contribu- tions to Biology and Problems of Medicine (Berkeley, Calif.: University of California Press) pp 223–239 [340] Ben-Jacob E, Deutscher G, Garik P, Goldenfeld N D and Lareah Y 1986 Phys. Rev. Lett. 57 1903–1906 [341] Meakin P 1998 Fractals, scaling and growth far from equilibrium vol 5 (Cambridge University Press) [342] Stanley H E and Ostrowsky N 2012 Random fluctua- tions and pattern growth: experiments and models vol 157 (Springer Science & Business Media) [343] Hayakawa Y and Sato S 1997 Phys. Rev. Lett. 79 95–98 [344] Halley J M, Comins H N, Lawton J H and Hassell M P 1994 Oikos 70 435–442 biol. Ecol. 22 1–16 [346] Hermanowicz S W 1999 Water Sci. Technol. 39 107–114 [347] Picioreanu C, Van Loosdrecht M C M and Heijnen J J 2000 Biotechnol. Bioeng. 69 504–515 [348] Pizarro G, Griffeath D and Noguera D R 2001 J. Envi- ron. Eng. 127 782–789 [349] Tang Y and Valocchi A J 2013 Water Research 47 5729– [350] Wolfram S 1984 Nature 311 419–424 [351] Laspidou C S and Rittmann B E 2004 Water Res. 38 5742 [345] Wimpenny J W T and Colasanti R 1997 FEMS Micro- [364] Laspidou C S, Kungolos A and Samaras P 2010 Desali-
1611.09820	1	1611	2016-11-29T20:16:11	Exploring Strategies for Classification of External Stimuli Using Statistical Features of the Plant Electrical Response	[ "physics.bio-ph", "q-bio.QM", "stat.AP", "stat.ML" ]	Plants sense their environment by producing electrical signals which in essence represent changes in underlying physiological processes. These electrical signals, when monitored, show both stochastic and deterministic dynamics. In this paper, we compute 11 statistical features from the raw non-stationary plant electrical signal time series to classify the stimulus applied (causing the electrical signal). By using different discriminant analysis based classification techniques, we successfully establish that there is enough information in the raw electrical signal to classify the stimuli. In the process, we also propose two standard features which consistently give good classification results for three types of stimuli - Sodium Chloride (NaCl), Sulphuric Acid (H2SO4) and Ozone (O3). This may facilitate reduction in the complexity involved in computing all the features for online classification of similar external stimuli in future.	physics.bio-ph	physics	Journal of the Royal Society Interface Exploring Strategies for Classification of External Stimuli Using Statistical Features of the Plant Electrical Response Shre Kumar Chatterjeea, Saptarshi Dasa,, Koushik Maharatnaa, Elisa Masib, Luisa Santopolob, Stefano Mancusob and Andrea Vitalettic,d a) School of Electronics and Computer Science, University of Southampton, Southampton SO17 1BJ, United Kingdom b) Department of Agri-food Production and Environmental Science (DISPAA), University of Florence, viale delle Idee 30, 50019 Sesto Fiorentino, FIorence, Italy c) WLAB S.r.L., via Adolfo Ravà 124, 00142 Rome, Italy d) DIAG, SAPIENZA Università di Roma, via Ariosto 25, 00185 Rome, Italy Author's Emails: [email protected] (S.K. Chatterjee) [email protected], [email protected] (S. Das) [email protected] (K. Maharatna) [email protected] (E. Masi) [email protected] (L. Santopolo) [email protected] (S. Mancuso) [email protected] (A. Vitaletti) Abstract: Plants sense their environment by producing electrical signals which in essence represent changes in underlying physiological processes. These electrical signals, when monitored, show both stochastic and deterministic dynamics. In this paper, we compute 11 statistical features from the raw non-stationary plant electrical signal time series to classify the stimulus applied (causing the electrical signal). By using different discriminant analysis based classification techniques, we successfully establish that there is enough information in the raw electrical signal to classify the stimuli. In the process, we also propose two standard features which consistently give good classification results for three types of stimuli - Sodium Chloride (NaCl), Sulphuric Acid (H2SO4) and Ozone (O3). This may facilitate reduction in the complexity involved in computing all the features for online classification of similar external stimuli in future. Keywords: Plant electrical signal, classification, discriminant analysis, statistical feature, time series analysis 1. Introduction Plants produce electrical signals, when subjected to various environmental stimuli [1– 7]. These electrical signals in essence represent changes in underlying physiological processes influenced by the external stimuli. Thus, analysing such plant electrical signals may uncover possible signatures of the external stimuli embedded within the signal. The stimuli 1 Journal of the Royal Society Interface may vary from different light conditions, burning, cutting, wounding, gas or liquid [8] etc. This opens up the possibility to use such analysis techniques to turn a green plant into a multiple-stimuli sensing biological sensor device [9]. If such an association between the external stimuli and the resulting plant electrical signal could be made, then it may serve the purpose of holistic monitoring of environmental constituents at a much cheaper cost (because of abundance of plants) thereby eliminating the need to install multiple individual sensors to monitor the same external stimuli. In this work, we attempt to explore the possibility of classifying three external stimuli - Sodium Chloride (NaCl), Sulphuric Acid (H2SO4) and Ozone (O3), from the electrical signal response of plants as the first step towards that goal. Here, we chose heterogeneous stimuli that reproduce some of the possible environmental pollutants e.g. H2SO4 is a major component of acid rain. Ozone is a tropospheric air pollutant and is the main component of smog. Salinization often results from irrigation management practices or treatment of roads with salt as de-icing agent and can be linked to environmental soil pollution. These three stimuli – NaCl, H2SO4, O3 are specifically chosen to study the change in plant physiological response to represent the effect of environmental pollution. Electrical signals were collected from a number of tomato (Solanum lycopersicum) and cucumber (Cucumis sativus) plants using NaCl, H2SO4 and O3 as stimulus in controlled settings. Multiple experiments were conducted for each stimulus to ensure the repeatability of the electrical signal response each time. We then extracted 11 statistical features from these plant signal time series in order to investigate the possibility of accurate detection of external stimulus through a combination of these features and simple discriminant analysis classifiers. We believe this work will not only form the backbone of using plants as environmental biosensors [9], [10] but also open up a new field of further exploration in plant signal behaviours with meaningful feature extraction and classification similar to the studies done using other human body (ECG), Electroencephalogram (EEG) and Electromyogram (EMG) [11]. like Electrocardiogram electrical responses Although there have been few recent attempts on signal processing, feature extraction and statistical analysis using plant electrical responses [12–18], there has been no attempt to associate features extracted from plant electrical signals to different external stimuli. The focus of our work is to address this gap. Here, we analysed the statistical behaviour of raw electrical signals from plants similar to previous studies on raw non-stationary biological signals which exhibit random fluctuations such as EMG/EEG, adopting a similar approach to develop a classification system [19–22]. The present paper reports the first exploration of its kind, aiming at finding meaningful statistical feature(s) from segmented plant electrical signals which may contain some signature of the stimulus hidden in them, in different extents. As a first exploration, this work focussed on the ability to classify the stimuli by only looking at a small segment of raw plant electrical response. The questions which arise in order to explore this possibility of classification are: 1) which features give a good discrimination between the stimuli, and 2) which type of simple classifier will give a consistently good result. The simplicity of the classifier is important issue here because our ambition is to run it on resource constrained embedded systems, such as sensor nodes in future. In order to tackle the first question, we start by using 11 statistical features which have been used in other biological signals as well (e.g. EEG, ECG and EMG) [11]. We here explore which feature alone (univariate analysis) or feature combinations (bivariate analysis) 2 Journal of the Royal Society Interface consistently indicate towards that particular signature of the stimulus. In order to answer the second question, we will start with a simple discriminant analysis classifier and then its other variants to observe the average classification rate. 2. Material and methods (a) Stimulus and experimental details Here we try to develop a classification strategy to detect three different stimuli viz. O3, H2SO4, NaCl. Four set of experiments were conducted with H2SO4, NaCl 5ml and 10ml each as stimuli, as shown in Table 1. For each stimuli mentioned above, a between subjects design for experiments were setup where four different tomato plants (similar age, growing conditions and heights) were used with each plant being exposed to the stimulus only once. Thus for 12 experiments, 12 tomato plants were used. For Ozone as stimuli, six cucumber plants and two tomato plants were used for eight experiments with each plant being subjected to only one experiment but multiple application of the stimulus. Table 1: Different stimulus, plants species and number of data-points (each capturing 11 statistical measures of 1000 samples) used for the present study For each plant, we used three stainless steel needle electrodes - one at the base (reference for background noise subtraction), one in the middle and the other on top of the stem as shown in Figure 1. The electrodes were 0.35 mm in diameter and 15 mm in length, similar to those used in EMG from Bionen S.A.S. and were inserted around 5 – 7 mm into the plant stem so that the sensitive active part of the electrodes (2mm) are in contact with the plant cells [8]. The electrodes were connected to the amplifier-Data Acquisition (DAQ) system in a same way previously studied in [8]. Plants were then enclosed in a plastic transparent box with proper openings to allow the presence of cables and inlet/outlet tubes, and exposed to artificial light conditions (LED lights responding to plant's photosynthetic needs, mimicking a day/night cycle of 12 hours). Each experiment was conducted in a dark room to avoid external light interferences. The whole setup was then placed inside a Faraday cage to limit the effect of electromagnetic interference as shown in Figure 1. After the insertion of the electrodes into the plant, we waited for about 45 minutes to allow the plant(s) to recover before starting the stimulations. Electrical signals acquired by the electrodes were provided as input to a 2-channel high impedance (1015 Ω) electrometer (DUO 773, WPI, USA) while data recording was carried out through 4-Channel DAQ 3 Journal of the Royal Society Interface (LabTrax, WPI) and its dedicated software LabScribe (WPI) [23]. The sampling frequency was set as 10 samples per second for all the recordings. For the treatments with liquid, sulphuric acid (5 ml H2SO4, 0.05M) or sodium chloride (5 or 10 ml NaCl 3M), a syringe placed outside of the Faraday cage and connected to a silicone tube inserted into the plant soil, was used to inject the solution as shown in Figure 2(a). O3, produced by a commercial ozone generator (mod. STERIL, OZONIS, Italy), [24] was injected into the box through a silicone tube (1 minute spray every 2 hours, 16 ppm), while a second outlet tube threw the Ozone from the box to the chemical hood as shown in Figure 2(b). The concentration of Ozone inside the box was monitored using a suitable sensor. Figure 1: Experimental setup showing a tomato plant inside a plastic transparent box, kept inside a Faraday cage. The placement of the electrodes on the stem is also shown. Figure 2: Tubes system for introducing pollutants inside the box. (a) For the treatments with H2SO4 or NaCl, a syringe placed outside of the Faraday cage and connected to a silicone tube inserted into the plant soil was used to inject the solution at various concentrations. (b) Ozone was injected into the box through a silicone tube, while a second outlet tube threw the ozone from the box to the chemical hood. (b) Data processing and segmentation 4 Journal of the Royal Society Interface Each dataset was obtained after one (H2SO4, NaCl 5 ml and 10 ml) or multiple (O3) application of that particular stimulus. This is illustrated in Figure 3 where the application of stimulus is marked by a vertical dotted line with the post stimulus part of the time series on the right side and the background or pre-stimulus part indicated on the left side of the line. In the case of O3, multiple application of the stimulus is shown by multiple markers. Figure 3: (Top) The vertical dotted lines mark the application time of the four stimuli. (Bottom) Separating the plant electrical signal into background and post-stimulus parts and then dividing them into smaller blocks of 1000 samples, as shown by dashed circles). As a general observation, from Figure 3 we can see that there are sudden spiking changes in the signal after the application of H2SO4 and Ozone as stimulus. However for the NaCl 5ml and 10 ml stimuli, the changes in the electrical signal response are relatively slow. Thereafter, for each experiment, we divided the data such that we have a post stimulus part of the signal as well as the background (pre-stimulus) part. In case of O3 where multiple stimuli were applied, we divided the data such that the signal duration between consecutive applications of the stimuli is a separate post-stimulus response. This way, we ended up having several post and pre-stimuli datasets for all the four stimuli. Next, each of these datasets was segmented into blocks of fixed window length of 1000 samples (100 seconds) which is shown in Figure 3. 5 Journal of the Royal Society Interface The reason for this data segmentation is to facilitate batch processing of large volumes of data acquired during continuous monitoring. We extracted 11 statistical features from these small chunks of 1000 samples and wanted to explore if the features from such small chunks give enough information to the classifier to discriminate which stimulus that particular data chunk (time period) belonged to. Again, a successful classification of the stimulus from the features of such a small signal block will enable a fast decision time. This is due to smaller buffer-size for batch processing compared to the whole length of the signal acquisition thereby making it easier for possible online implementation in future. Since this is the first exploration of its kind, we stick to 1000 samples only, for extracting statistical features which give sufficiently good classification accuracy but there is scope for further exploration of an optimum window length to classify the stimulus. The classifier was trained using only the blocks of samples belonging to the post-stimulus part of the plant signal. The pre-stimulus part was also divided into similar segments in order to study the effect of the background for different plants under different experimental condition. The stimulus induced plant signals have both deterministic and random dynamics i.e. local and global variations in amplitudes and different statistical measures of smaller data segments [6], [9], [10], [25], [26]. The research question which we try to answer through the present exploration is – is it possible to identify the stimulus by only looking at the statistical behaviour of small segments of the plant electrical response? A successful answer to this question would pave the way of conceptualising an electronic sensor module in future for classifying the environmental stimulus. This sensor module can be fitted on the plant for batch processing of segmented plant signals, statistical feature extraction and classification, without much memory requirement in future applications. (c) Statistical feature extraction from segmented time series Here, we started with 11 features which are predominantly used in the analysis of other biological signals [27]. Different descriptive statistical features like mean ( m ), variance 2s ), skewness (g ), kurtosis ( b ) as given in (1) and Interquartile range ( ( , i.e. the difference between the 1st and 3rd Quartile) were calculated. IQR Q Q 1 = 3 m = [ E x i ] s , 2 = [ E x i m g ] 2 , = ( m s  E x  i ) b = , 3   ( m s E x i   ) 4   (1) In the calculation of four basic moments in (1), xi is the segmented raw electrical signals each of them containing 1000 samples and E[.] is the mathematical expectation operator. Apart from these five, the remaining six features taken are – Hjorth mobility, Hjorth complexity, detrended fluctuation analysis (DFA), Hurst exponent, wavelet packet entropy and average spectral power which are briefly described below. Hjorth's parameters The Hjorth mobility and complexity, described in [28], quantify a signal from its mean slope and curvature by using the variances of the deflection of the curve and the variances of their first and second derivatives. Let the signal amplitudes at discrete time instants be na at time nt . The measures of the complexity of the signal is based on the second moments in time domain of the signal and the signal's first and second derivatives. The finite differences of the signal or time derivatives can be viewed in (2). 6 - - - - Journal of the Royal Society Interface ¢= a n d n = a n d n = = a + 1 n a n , where n a + 1 n a n , where n = = 1, 2, 1, 2, ( ( -⋯ N , -⋯ N , ) 1 and ) 2 (2) The variances are then computed as (3) [29]. s 2 a = ∑ , 2 a n 1 N N = 1 n s 2 d = ( N 1 N 1 - ∑ ) 1 = 1 n ( d n ) 2 d n 1 s , 2 dd = ( N 2 N 1 - ∑ 2 ) = 1 n ( d n ) 2 d n 1 (3) These variances (3) are used to calculate the Hjorth mobility ( ( Hc ) [29] as shown in (4). ( ( ) s s s s s s = = ) and Hm ) and the Hjorth complexity (4) 2 dd 2 d 2 d 2 a m H d a c H Detrended fluctuation analysis (DFA) DFA has been introduced in identifying long range correlations in non-stationary time series data. By using a scaling exponent (a ), one can describe the significant autocorrelation properties of signals with a provision of capturing the non-stationary behaviour as well [30], [31]. The different values of a represents certain auto-correlation properties of the signal [30], [31]. For a value of less than 0.5, the signal is described as anti-correlated. A value of exactly 0.5 indicates uncorrelated (white noise) signal, whereas a value greater than 0.5 indicates positive autocorrelation in the signal. When a =1, the signal is indicated to be 1 f noise and a value of 1.5 indicates the signal to be random walk or Brownian noise [30], [31]. Hurst exponent The Hurst exponent ( H ), a dimensionless estimator similar to DFA, is used as a measure of the long term memory of a time series data xi [32], [33]. The value of the Hurst exponent lies between 0 and 1, with a value between 0 – 0.5 indicating anti-persistent behaviour. This denotes that a decrease in the value of an element will be followed by an increase and vice versa. This characteristic is also known as mean reversion, which is explained as the tendency of future values to return to longer term mean value. The mean reversion phenomenon gets stronger for a series with exponent value closer to zero [32], [33]. When the value is close to 0.5, a random walk (e.g. a Brownian time series) is indicated. In such a time series, there is no correlation between any element and predictability of future elements is difficult [32], [33]. Lastly, when the value of the exponent is between 0.5 and 1, the time series exhibits persistent behaviour. This means the series has a trend or there is a significant autocorrelation in the signal. The more closer the exponent value gets towards unity, a stronger trend is indicated for the time series [32], [33]. Wavelet entropy (Wentropy) The time series may be represented in frequency and/or time-frequency domains by decomposing the signal in terms of basis functions such as harmonic functions (as in Fourier analysis) or wavelet basis functions (with consideration of non-stationary behaviour), respectively. Given such decomposition, it is possible to consider the distribution of the expansion coefficients in this basis. Quantification of the degree of variability of the signal 7 - ¢ ¢ ¢ ¢ ¢ - - - - - - - - Journal of the Royal Society Interface could be done using the entropy measure, where high values indicate less ordered distributions. The wavelet packet transform based entropy (WE) measures the degree of disorder (or order) in a signal [34–36]. A very ordered underlying process of a dynamical system may be visualized as a periodic single frequency signal (with a narrow band spectrum). Now the wavelet transformation of such a signal, will be resolved in one unique level with value nearing one, and all other relative wavelet energies being minimal (almost equal to zero) [34–36]. On the other hand, a disordered system represented by a random signal will portray significant wavelet energies from all frequency bands. The wavelet (Shannon) entropy gives an estimate of the measure of information of the probability distributions. This is calculated is of the ith wavelet by converting the squared absolute values of the wavelet coefficients decomposition level as shown in (5). WE = - ∑ 2 s i log i ( 2 s i ) Average spectral power (5) The average spectral power ( P ) is the measure of the variance of signal power, distributed across various frequencies [37]. It is given by the integral of the power spectral density (PSD) curve of the signal within a chosen frequency band of interest ( X e w j ) 2 ( ) x t (bounded by the low and high frequency – ωl, ωh respectively) as shown in (6). P (d) Adopted classification scheme ( X e w j ) 2 w d (6) w h = ∫ w l All of the above mentioned extracted features are first normalized to scale them within a maximum (1) and minimum (0) value and to avoid any unnecessary emphasis of some of the features on the classifier weights due to their larger magnitude than the others. Amongst all the 11 features, their relative importance in each of the binary classification set has been obtained by computing the Fisher's Discriminant Ratio (FDR) [38]. The FDR is a measure to explore the discriminating power of a particular feature to separate two classes and are computed as( s s and 2 are the standard deviation of the features in the two classes respectively and therefore should not be confused with that of the raw signal in (1). Higher ranking, based on FDR, will be assigned to those features which have higher difference in the mean values and small standard deviation implying compact distantly located clusters. Due to the application of multiple-stimulus, the FDR based feature ranking is applied for each of the stimulus pairs, in the present work [38]. m are the mean and 1 [38], where, m and 2 1 m m ( s + 2 1 2 s ) 2 2 ) 1 2 The classifiers implement algorithms which help in distinguishing between two or more different groups or classes of data. Different classification algorithms are obtained by first training the class labels (stimulus applied in this case) of a certain portion of the known (training) groups and then using the trained model to predict the class labels for a group of unknown (test) dataset. Once it is found that the testing phase is successful (high accuracy in 8 - Journal of the Royal Society Interface identifying the stimulus) using the trained model, the algorithm can be used to identify which class an unknown data belongs to. In cases, where the distinction is easily achievable, discriminant analysis classifiers such as Linear Discriminant Analysis (LDA) could be effective. Where such distinctions are not that straightforward, nonlinear classifiers such as kernel based techniques like support vector machine (SVM) can be applied. Cases where only two groups need to be identified, binary classification are generally carried out. This is a much simpler process than multiple class classification. The choice of a classifier (discriminant or complex kernelized SVM) may be determined sometimes by looking at the distribution plot of the features of the two groups. If the distribution plots show two well separated means, we can conclude that a simple linear or other discriminant analysis based classifiers should be able to classify the data to a sufficient extent. Unnecessarily involving a complex nonlinear classification technique often gives high classification accuracy on the training dataset, but is prone to over-fitting. In the present study we focus on five different discriminant analysis classifiers which are based on least square method for training the classifier weights compared to the computationally heavy optimization process involved in SVM. Amongst five discriminant analysis variants the QDA uses a quadratic kernel with the feature vectors. The Diaglinear and Diagquadratic classifiers are also known as Naïve Bayes classifier using a simple linear and quadratic kernel and use the diagonal estimate of the covariance matrix (neglecting the cross-terms or feature correlations). The Mahalanobis classifier uses a different distance measure than the standard Euclidean distance [38]. We used different discriminant analysis classifiers due to their simplicity to see the characteristic changes traced in the features due to these stimuli. Two types of approach could be taken in classification – 1) choice of meaningful statistical features followed by simple classifier, 2) simple features followed by a complex classifier. The former case is preferable in the present case since it may help in understanding the change in statistical behaviour of the signal which might be indicative towards some consistent modification of the underlying biological process. Cross validation schemes are often employed to avoid the introduction of any possible bias due to the training data-set [38]. Here we use the leave one out cross validation (LOOCV) where, if there are N data-points, then (N-1) number of samples is used for training the classifier and the one held-out sample is used to test the trained structure. Thereafter, the single test sample will be included in the next training set, and again a new sample from the previous training set will be set aside as the new test data. This loop will go for N times, till all the samples have been tested and the average classification accuracy for all the N instances are calculated [38]. 3. Results The classification results of 5 discriminant analysis classifier variants, using 11 statistical features from plant electrical signal response to four different stimuli viz. H2SO4, NaCl of 5 ml and 10 ml and O3 are presented in this section. We also investigate which stimuli are best detected by looking at the classification accuracy, thereby suggesting the ability of the plant to detect few particular stimuli better than the others. a) Need for subtracting the background information of individual features Figure 3, shows the four plant electrical signal responses to four different stimuli beginning at different amplitude levels. This means the background signal (even before the 9 Journal of the Royal Society Interface application of the stimuli) is different in all four different cases. This may bias the final classification result due to the already separated background information within the multiple features considered. Due to the effect of different backgrounds, we can see a clear separation between the stimuli for some features such as Hjorth mobility, Hjorth complexity and skewness, in Figure 4, where histogram plots for each of the features for each stimulus are plotted without any background subtraction. Figure 4: Normalized histogram plots for 11 individual features showing stimuli separability (no background subtraction). Figure 5: Univariate histograms of each of the 11 features for four different stimuli (with background subtraction) 10 Journal of the Royal Society Interface This encourages us to look at only the incremental values of the features under different stimuli. The incremental values are obtained by subtracting the mean of every feature extracted from the background from the corresponding feature extracted from the post-stimulus part of the signal. The histogram plots of the incremental values of the individual features, after the background is subtracted, are given in Figure 5 which shows a lesser separability in the stimuli which were as expected. We now used these incremental values of the features to see how good they are in providing a successful classification (using five different discriminant analysis classifier variants) between any two stimuli (six binary combinations of four stimuli). As an example, although the histogram plots in Figure 5 shows clear separation of the distributions for NaCl and O3 using skewness as feature due to their peaky nature, the frequency of occurrence of the histograms show that the distributions have wider spread which has been reflected by the moderate rate of classification reported in the next subsections using that particular feature. b) Correlation of features to avoid redundancy Table 2: Correlation coefficient between 11 statistical features extracted from plant electrical signals (after subtracting the mean of the pre-stimulus features from the post-stimulus ones) Between all the features, a correlation test was carried out to find out their inter- dependence. The result of this test, given in Table 2, is obtained by checking the Pearson correlation coefficient values between all feature pairs. A correlation value of (~ +1/-1) indicates a strong positive/negative correlation between a pair, whereas a value closer to zero indicates the feature pairs are independent and are thus more informative about the underlying process. A good classification strategy should ideally involve uncorrelated features, in order to avoid redundancy in training the classifier. In this work, we proceeded by initially taking all features into account and then ignored the ones with high correlation. c) Classification using univariate and bivariate features 11 Journal of the Royal Society Interface The classification results were obtained in two ways – using univariate and bivariate features, to make the analysis intuitive and simple to infer. That is, instead of taking all the features together to get a multivariate classification (which may give good classification accuracy but are less intuitive and reliable due to increase in complexity and dimension of the problem), we just explored the results with 11 individual features and 55 possible feature pairs. Table 3 presents the results, obtained using individual features, averaged across all the six stimuli combinations and all the five different classifier variants. We have also presented the relative multi-class separability score given by the scatter matrix (S) in (7) for each feature in terms of the within-class (Sw) and between-class (Sb) scatter matrix [38]. = S ( 1 trace S S w b = S w c ∑ = 1 i PS S i , i b ) = ( P i c ∑ = 1 i )( m m m m i 0 ) T m 0 m = 0 , i c ∑ = 1 i P i = , i i 1, 2, , c ⋯ (7) c = ) and has been Here, Pi is the a-priori probability for the present four class problem ( considered as ¼. Also, the mean and covariance matrices for each of the classes are denoted by { m is the global mean vector. The scatter matrix extends the concept of class-separability using FDR from binary classification to multi-class problems. and the 0 },i iSm 4 Table 3: Average accuracy (averaged across all six binary stimuli combinations and all five classifier variants) and best accuracy (averaged across four one vs. rest stimuli combinations) for classification using individual features The scatter matrices value in Table 3 provide an insight about how good the separation between all the four classes (stimuli) are using the individual features. From Table 3 we can see that the signal 'mean' on its own has the best classification result for all the six 12 - - - Journal of the Royal Society Interface binary combinations of four stimuli. However since we have extracted the features from the raw non-stationary plant electrical signals, mean is not a very reliable feature to base any conclusions on, because it can be influenced by various artefacts and noise during measurement or from various environmental factors (e.g. sudden gust of breeze could shake the electrodes connected to the plant body etc.). The next five best features (best average accuracies given in Table 3), when taken individually, are wavelet packet entropy, Hjorth complexity, interquartile range, variance and average spectral power respectively. From now on, we will only consider these features as the top five features. In Table 3, we also report the best achievable accuracy along with the best classifier using each of the single features to discriminate the four stimulus within a 'one vs. rest' strategy. This highlights the possibility of isolating one particular class from the other classes using a single feature, with a certain degree of confidence. So far we have seen the averaged results of classification for six binary stimuli combinations using individual features. We next find the best classified stimuli combination using only the top five individual features and using the five variants of the discriminant analysis classifiers, as mentioned above. As a result, we obtained five classification accuracies (for five individual features) for every classifier for each of the six binary stimuli combination. That results in 25 classification accuracies for each of the six binary stimuli combinations. All these 25 results were averaged for each stimuli combination and given in Table 4 which shows the best discrimination possible is for H2SO4 and O3 with classification accuracy over 73%. Additionally, discrimination between NaCl (both concentrations) and O3/H2SO4 also shows promising result with accuracy over 65% and 63% respectively. Table 4: Accuracy using top five individual (univariate) features (F2 through F6) and averaged across five classifiers (average separability between different stimulus combinations) The average classification results presented in Table 4 encourages us to look at the best results achieved using individual features, for each stimuli combination, so that we can see if there is any consistent feature giving good classification results. This is shown in Table 5 from where it is evident that F3 (Hjorth complexity) gives the best result for three different binary stimuli combinations with an accuracy over 74%. Overall, the best accuracy is achieved for classification between H2SO4 and O3, with an accuracy of >94% using F2 (Wentropy) and QDA classifier. Although in Table 4 the discrimination between NaCl and O3/H2SO4 are shown in terms of the average accuracy which might seem to be relatively low (63% or 65%), the best cases for such a discrimination can be found in Table 5 (accuracies of >78% and >72% respectively) between the same set of stimuli. Also from Figure 5 we can see that though skewness shows a good discrimination between H2SO4 and other stimuli, from Table 3 we can see that the average classification accuracy using skewness as an individual feature is very low. This is due to the fact that skewness on its own did not give good classification results between other remaining stimuli combinations. 13 Journal of the Royal Society Interface Table 5: Best accuracy taking individual features for each stimulus combinations (best separability between different stimulus combinations) d) Classification using feature pairs Table 6: Average accuracy obtained using top five feature pairs (bivariate) and five classifiers (average separability between different stimulus combinations) Figure 6: (top) Classification accuracy for different feature combinations with background information removed; (bottom) deterioration in accuracy for the features without background information removed. 14 Journal of the Royal Society Interface Next, we looked at the effect of all possible feature pairs using 11 individual features (totalling 55 independent feature pairs) on the classification results between six different stimuli combinations. These classification accuracies are shown in Figure 6 along with the difference in accuracy (error) when the background is not subtracted as discussed in earlier section. The features mentioned as{ ,11⋯ in Figure 6 are the features designated by{ respectively in Table 2. Since we ignored mean as a feature in the previous section, we explored the effect of taking binary combinations of the next five individual features (F2 through F6, as mentioned in Table 3) on the classification accuracy. The results obtained using each of these bivariate features (pairs), using all the five classifier variants were averaged and given in Table 6 which are found to be better than the averaged results obtained using just univariate features as given in Table 4. f⋯ 1, 2, f 1 , } f 2 , } 11 , Table 7: Best accuracy for each stimulus combination using two features (best separability between different stimulus combinations) By this exploration, we wanted to find out if there is any improvement on the classification accuracy when a feature pair is used rather than just individual feature. From Table 6, we can see that classification accuracy is improved for all stimuli combination except NaCl 5ml vs. H2SO4. We can also observe from Table 6 that the top two best accuracies are obtained for stimuli combinations of NaCl 10ml vs. O3 and H2SO4 vs. O3. Now, let us look at the best feature pair(s), among all 55 bivariate feature pairs, as given in Table 7. We notice that a combination of F4 (IQR) and F5 (variance) results in the best classification accuracies for four out of six different stimuli combinations. For the remaining two stimuli combinations, a feature pair of F4 and F6 (average spectral power) gives the best classification accuracies. e) Finding the most reliable combination of feature or feature pair and classifier variant So far we have found that individual features F2, F3, F8 and F9 and feature pairs F4-F5 and F4-F6 produced the best classification results for one or more (out of the six) stimuli combinations. We now explore these features and feature pairs for all stimuli combinations. Table 8 gives the results of classification when we used just F2, F3, F8 and F9 as an individual feature using all classifier variants for all binary stimuli combinations. Similarly, Table 9 gives the results using the feature pairs F4-F5 and F4-F6 for all the six stimuli combinations, 15 Journal of the Royal Society Interface using all the five classifier variants. These results will help us choosing the right classifier and deciding the feature or feature-pair which provides the best average accuracy for all the binary combinations of stimuli. Table 8: Accuracy of different classifiers for six stimuli combinations (in %) using the best individual features From Table 8, we note that using just F2 or F3 provides consistently better average classification accuracies than using F8 or F9. It is also noticed that although F2 provides a better classification for the stimuli combinations NaCl 10ml vs. O3 and O3 vs. H2SO4, F3 provides much consistent and better result for the remaining stimuli combinations. While considering a single feature for discriminating the four stimuli, the best average result (73%) could be obtained using the F2 (Wentropy) as feature and Mahalanobis classifier, although it is highly correlated with the signal mean (F1) as shown in Table 2. Since mean as a feature was ignored due to its susceptibility to artefacts, therefore, we also ignore Wentropy and instead propose Hjorth complexity as the best individual feature for achieving good average classification accuracy. ng the individual features. From REF _Ref386557847 m Table 9 that the top two classification accuracies (>73%) are obtained using F4-F5 combination and Diagquadratic and QDA as classifiers respectively. Both these top bivariate classification results (average accuracy of 69.65% across all stimuli and classifiers) are better than that obtained in univariate case in Table 8 (average accuracy of 62.98% across all stimuli and classifiers). Although again from Table 2, we realize the IQR and variance are highly correlated with each other but since we are achieving a good result in terms of classification using these two 16 Journal of the Royal Society Interface features, we note that calculating IQR and variance from a block of 1000 samples of raw non- stationary plant electrical signal, along with QDA or Diagquadratic classifier will provide consistently good results in terms of classifying which external stimuli caused the particular signature in the plant electrical signal. Table 9: Accuracy of different classifiers for six stimuli combinations (in %) using the best feature pairs We next explore some pairs of uncorrelated features for classification by looking at the 12 next best average classification accuracies (obtained across all stimuli combinations and using all five different classifiers) as shown through a 2D normalized histogram (volume being unity) plots showing the separation of the four stimuli in Figure 7. Average accuracy obtained (over all stimuli combinations and classifiers) using particular feature pairs (denoted by 1 f⋯ , as described in Table 2) are also mentioned in the title of each subplot in f Figure 7. It is observed that the second best average classification accuracy is achieved using variance and skewness as features which are almost uncorrelated (correlation of ~0.01 in Table 2). 11 f , , 2 , 3 f In Figure 7, except the first subplot with 2 f , all the rest combinations are almost uncorrelated and still give a good classification performance. Thus as a reliable measure of analysis, it has been found that the variance and skewness calculated from a block of 1000 samples of plant electrical signal will be able to give an average (over all six stimuli combinations and using all five discriminant classifiers) accuracy of 70% during binary classification of the stimuli. It is to be noted that in the bivariate classification scheme, the mean (F1) has not been considered as one of the features. Also, the best bivariate accuracies were achieved involving the variance (F2) along with all the other features (F4…F11) in Figure 7, while ignoring the F2-F3 combination due to their high inter-dependence. As a summary, a better reliable classification scheme is expected (~67%-70%) involving bivariate features as shown in Figure 7, with respect to the univariate features as given in Table 3 (<67%, ignoring mean and Wentropy). 17 - Journal of the Royal Society Interface Figure 7: Bivariate histograms of top feature pairs with highest classification accuracy for all the four stimuli (accuracy mentioned in title of each subplot). The results presented in this work only takes into account the experimental data for individual stimulus under controlled environment (laboratory). The next step can be to setup experiments where multiple stimuli could be applied together on the plant and its electrical signal response could be extracted for further analysis and classification of the most influential stimuli. Also, the robustness of the statistical features due to possible artefact (e.g. movement of the leaf due to wind, rainfall etc.) are to be explored in future in a more naturalistic environment, outside the controlled laboratory set-up. 4. Discussion In our exploration, the data from two channels per plant (per experiment) were used to record the electrical response, and then statistical features were calculated from both the channels and pooled together. Here, the location on the plant body for the data extraction was ignored, as the work was primarily focussed more on the possibility of classification of applied external stimuli from the extracted plant electrical signal. Similarly the effect of a different species of plants to study all the four stimuli have also been ignored, except the introduction of an additional species (cucumber) for Ozone stimulus. The idea behind developing an external stimuli classification scheme, based on plant electrical response is focussed on generic plant signal behaviour and not of a specific species. However such isolation forms a very good study and could be taken up as future scope of work. There might be some possibility of confounding effects based on the position of the electrodes and plant species in any classification scheme. But such confounding effects will be minimal due to the large number of data samples as shown in Table 1 and the use of cross-validation scheme to test the performance of the discriminant analysis classifiers. Also we did not use kernel based nonlinear classifiers like SVM which could over-fit these plant specific characteristics and still give good classification result, rendering the loss of generalizing capability of the classifier. Moreover, the present classification scheme is based on the raw non-stationary plant signal. In bio-signal processing literature [11], using of high-pass filter is recommended to 18 Journal of the Royal Society Interface make a bio-signal stationary instead of extracting features from the raw non-stationary signal. But there is also a possibility with an ad-hoc filtering that some useful information in the data may get lost since the cut-off frequency for plant signal processing is not yet known. That is why we considered the features from the post-stimulus signal to train the classifier by removing any possible bias of the channel or plant using incremental features i.e. using the mean of the features in the pre-stimulus part. The segmentation of the signal in a block of 1000 samples also disregards the temporal information of the stimuli, since we primarily tried to answer the question if classification is indeed possible by looking at any segment of the post-stimulus part of the signal. Also, in a realistic scenario, we would not know when the response to a particular stimulus started. So we need to base our classification on the in- coming stream of live data. 5. Conclusion Our exploration using raw electrical signals from plants provides a platform for realizing a plant signal based bio-sensor to classify the environmental stimuli. The classification scheme was based on 11 statistical features extracted from segmented plant electrical signals, followed by feature ranking and rigorous univariate and bivariate feature based classification using five different discriminant analysis classifiers. External stimuli like H2SO4, O3 and NaCl in two different amounts (5 ml and 10 ml) have been classified using the adopted machine learning approach with 11 statistical features, capturing both the stationary and non-stationary behaviour of the signal. The classification has yielded a best average accuracy of 70% (across all stimuli and five classifier variants using variance and skewness as feature pairs) and the best individual accuracy of 73.67% (across all stimuli and using variance and IQR as feature pairs in Diagquadratic classifier). The very fact that, by looking at the statistical features of plant electrical response, we can successfully detect which stimuli caused the signal is quite promising. This will not only open the possibility of remotely monitoring the environment of a large geographical area, but will also help in taking timely preventive measures for natural or man-made disasters. Acknowledgements The work reported in this paper was supported by project PLants Employed As SEnsor Devices (PLEASED), EC grant agreement number 296582, URL: http://pleased-fp7.eu/. Data accessibility The experimental data fp7.eu/?page_id=253. Reference is available in the PLEASED website at http://pleased- [1] J. B. Sanderson, "Note on the electrical phenomena which accompany irritation of the leaf of Dionaea muscipula," Proceedings of the Royal Society of London, vol. 21, no. 139–147, pp. 495–496, 1872. [2] C. Darwin and S. F. Darwin, Insectivorous plants. J. Murray, 1888. [3] J. C. Bose, The physiology of photosynthesis. Long Mans, Green And Co; London, 1924. 19 Journal of the Royal Society Interface [4] B. G. Pickard, "Action potentials in higher plants," The Botanical Review, vol. 39, no. 2, pp. 172–201, 1973. [5] E. Davies, "Electrical signals in plants: facts and hypotheses," Plant electrophysiology, pp. 407–422, 2006. [6] A. G. Volkov, Plant electrophysiology: theory and methods. Springer, 2006. [7] J. Fromm and S. Lautner, "Electrical signals and their physiological significance in plants," Plant, cell & environment, vol. 30, no. 3, pp. 249–257, 2007. [8] S. K. Chatterjee et al., "Forward and Inverse Modelling Approaches for Prediction of Light Stimulus from Electrophysiological Response in Plants," Measurement, vol. 53, pp. 101–116, 2014. [9] V. Manzella et al., "Plants as sensing devices: the PLEASED experience," Proceedings of the 11th ACM Conference on Embedded Networked Sensor Systems, p. 76, 2013. [10] A. G. Volkov and D. R. A. Ranatunga, "Plants as environmental biosensors," Plant signaling & behavior, vol. 1, no. 3, pp. 105–115, 2006. [11] L. Sӧrnmo and P. Laguna, Bioelectrical signal processing in cardiac and neurological applications. Academic Press, 2005. [12] J. Lu and W. Ding, "The feature extraction of plant electrical signal based on wavelet packet and neural network," Automatic Control and Artificial Intelligence (ACAI 2012), International Conference on, pp. 2119–2122, 2012. [13] Y. Liu, Z. Junmei, L. Xiaoli, K. Jiangming, and Y. Kai, "The Research of Plants' Water Stress Acoustic Emission Signal Processing Methods," Measuring Technology and Mechatronics Automation (ICMTMA), 2011 Third International Conference on, vol. 3, pp. 922–925, 2011. [14] W. Lan-zhou, L. Hai-xia, and L. Qiao, "Studies on the plant electric wave signal by the wavelet analysis," Journal of Physics: Conference Series, vol. 48, no. 1, p. 1367, 2007. [15] L. Jingxia and D. Weimin, "Study and evaluation of plant electrical signal processing method," Image and Signal Processing (CISP), 2011 4th International Congress on, vol. 5, pp. 2788–2791, 2011. [16] L. Wang and Q. Li, "Weak electrical signals of the jasmine processed by RBF neural networks forecast," Biomedical Engineering and Informatics (BMEI), 2010 3rd International Conference on, vol. 7, pp. 3095–3099, 2010. [17] L. Wang and J. Ding, "Processing on information fusion of weak electrical signals in plants," Information and Computing (ICIC), 2010 Third International Conference on, vol. 2, pp. 21–24, 2010. [18] L. Huang et al., "Electrical signal measurement in plants using blind source separation with independent component analysis," Computers and Electronics in Agriculture, vol. 20 Journal of the Royal Society Interface 71, pp. S54–S59, 2010. [19] A. Khasnobish, S. Bhattacharyya, A. Konar, D. Tibarewala, and A. K. Nagar, "A Two- fold classification for composite decision about localized arm movement from EEG by SVM and QDA techniques," Neural Networks (IJCNN), The 2011 International Joint Conference on, pp. 1344–1351, 2011. [20] A. Gupta, S. Parameswaran, and C.-H. Lee, "Classification of electroencephalography (EEG) signals for different mental activities using Kullback Leibler (KL) divergence," Acoustics, Speech and Signal Processing, 2009. ICASSP 2009. IEEE International Conference on, pp. 1697–1700, 2009. [21] G. R. Scolaro and F. M. de Azevedo, "Classification of epileptiform events in raw EEG signals using neural classifier," Computer Science and Information Technology (ICCSIT), 2010 3rd IEEE International Conference on, vol. 5, pp. 368–372, 2010. [22] W. Atsma, B. Hudgins, and D. Lovely, "Classification of raw myoelectric signals using finite impulse response neural networks," Engineering in Medicine and Biology Society, 1996. Bridging Disciplines for Biomedicine. Proceedings of the 18th Annual International Conference of the IEEE, vol. 4, pp. 1474–1475, 1996. [23] "http://www.wpiinc.com/blog/2013/05/01/product-information/data-trax-software-for- labscribe/." [24] "http://www.sepra.it/products-linea-generatori-serie-steril250mgo3h-da-aria-6.html." [25] X. Yan et al., "Research progress on electrical signals in higher plants," Progress in Natural Science, vol. 19, no. 5, pp. 531–541, 2009. [26] Z. Y. Wang et al., "Monitoring system for electrical signals in plants in the greenhouse and its applications," Biosystems Engineering, vol. 103, no. 1, pp. 1–11, 2009. [27] D. Kugiumtzis and A. Tsimpiris, "Measures of analysis of time series (MATS): a MATLAB toolkit for computation of multiple measures on time series data bases," Journal of Statistical Software, vol. 33, 2010. [28] B. Hjorth, "EEG analysis based on time domain properties," Electroencephalography and clinical neurophysiology, vol. 29, no. 3, pp. 306–310, 1970. [29] B. Hjorth, "Time domain descriptors and their relation to a particular model for generation of EEG activity," CEAN-Computerized EEG analysis, pp. 3–8, 1975. [30] J.-M. Lee, D.-J. Kim, I.-Y. Kim, K.-S. Park, and S. I. Kim, "Detrended fluctuation analysis of EEG in sleep apnea using MIT/BIH polysomnography data," Computers in Biology and Medicine, vol. 32, no. 1, pp. 37–47, 2002. [31] A. K. Goli'nska, "Detrended Fluctuation Analysis (DFA) in biomedical signal processing: selected examples," Logical, Statistical and Computer Methods in Medicine, vol. 29(42), 2012. 21 Journal of the Royal Society Interface [32] N. Kannathal, U. R. Acharya, C. Lim, and P. Sadasivan, "Characterization of EEG-A comparative study," Computer methods and Programs in Biomedicine, vol. 80, no. 1, pp. 17–23, 2005. [33] J. Mielniczuk and P. Wojdyllo, "Estimation of Hurst exponent revisited," Computational Statistics & Data Analysis, vol. 51, no. 9, pp. 4510–4525, 2007. [34] R. Q. Quiroga, O. A. Rosso, E. Bacsar, and M. Schürmann, "Wavelet entropy in event- related potentials: a new method shows ordering of EEG oscillations," Biological Cybernetics, vol. 84, no. 4, pp. 291–299, 2001. [35] O. A. Rosso et al., "Wavelet entropy: a new tool for analysis of short duration brain electrical signals," Journal of neuroscience methods, vol. 105, no. 1, pp. 65–75, 2001. [36] L. Zunino, D. Perez, M. Garavaglia, and O. Rosso, "Wavelet entropy of stochastic processes," Physica A: Statistical Mechanics and its Applications, vol. 379, no. 2, pp. 503–512, 2007. [37] R. H. Shumway and D. S. Stoffer, Time series analysis and its applications: with R examples. Springer, 2010. [38] S. Theodoridis and K. Koutroumbas, Pattern recognition, 4th ed. Academic Press, 2009. 22
1605.03901	2	1605	2016-05-31T03:15:24	Analysis of a stochastic model for bacterial growth and the lognormality of the cell-size distribution	[ "physics.bio-ph", "cond-mat.stat-mech", "physics.data-an" ]	This paper theoretically analyzes a phenomenological stochastic model for bacterial growth. This model comprises cell division and the linear growth of cells, where growth rates and cell cycles are drawn from lognormal distributions. We find that the cell size is expressed as a sum of independent lognormal variables. We show numerically that the quality of the lognormal approximation greatly depends on the distributions of the growth rate and cell cycle. Furthermore, we show that actual parameters of the growth rate and cell cycle take values that give a good lognormal approximation; thus, the experimental cell-size distribution is in good agreement with a lognormal distribution.	physics.bio-ph	physics	Analysis of a Stochastic Model for Bacterial Growth and the Lognormality of the Cell-Size Distribution Ken Yamamoto and Jun-ichi Wakita Department of Physics, Faculty of Science and Engineering, Chuo University, Kasuga, Bunkyo, Tokyo 112 -- 8551, Japan This paper theoretically analyzes a phenomenological stochastic model for bacterial growth. This model comprises cell division and the linear growth of cells, where growth rates and cell cycles are drawn from lognormal distributions. We find that the cell size is expressed as a sum of independent lognormal variables. We show numerically that the quality of the lognormal approximation greatly depends on the distributions of the growth rate and cell cycle. Furthermore, we show that actual parameters of the growth rate and cell cycle take values that give a good lognormal approximation; thus, the experimental cell-size distribution is in good agreement with a lognormal distribution. I. INTRODUCTION A. General overview In the study of complex phenomena, we can extract statistically useful information from the size distribu- tion of observed elements. Along with the power-law distribution[1, 2], which is closely associated with criti- cal phenomena in statistical physics, the lognormal dis- tribution [3] is found in a wide range of complex sys- tems. [4, 5] It appears in natural phenomena such as fragment and particle sizes[6, 7], the fluctuation of X-ray bursts[8], and genetic expression in bacteria[9], and in so- cial phenomena such as population[10], citations of scien- tific papers[11], and proportional elections[12]. The com- monness of the lognormal behavior is helpful for studying various complex systems from a unified viewpoint based on statistical physics. A random variable X is said to follow a lognormal dis- tribution lnN (µ, σ2) if its logarithm ln X is normally dis- tributed with mean µ and variance σ2. The probability density of lnN (µ, σ2) is 1 fµ,σ(x) = √2πσ2 x exp(cid:18)− [ln x − µ]2 2σ2 (cid:19) , and the (upper) cumulative distribution function is Fµ,σ(x) =Z ∞ x 1 fµ,σ(y)dy 1 2 erfc(cid:18) ln x − µ √2 σ (cid:19) , (1) √2 σ (cid:19)(cid:21) = = 2(cid:20)1 − erf(cid:18) ln x − µ where erf(x) = (2/√π)R x 0 exp(−y2)dy is the Gauss er- ror function and erfc(x) = 1 − erf(x) is the complemen- tary error function[13]. The lognormal distribution is a heavy-tailed distribution; that is, the tail of Fµ,σ(x) decays slower than any exponential function. Thus, a phenomenon obeying a lognormal distribution easily pro- duces values much larger than the mean. The mean of the distribution lnN (µ, σ2) is exp(µ + σ2/2) and the median is exp(µ). The mean is always larger than the median, and this is a consequence of the heavy-tailed property of the lognormal distribution.[14] The lognormal distribution is typically generated by a multiplicative stochastic process.[15] Consider a stochas- tic process X1, X2, . . . given by Xn+1 = MnXn, where Mn is a positive random variable. If the growth rates M1, M2, . . . are independently and identically dis- tributed, and E[ln Mn] = µ and V [ln Mn] = σ2 are fi- nite, the central limit theorem implies that the distribu- tion of (ln Xn − nµ)/(√nσ) converges to the standard normal distribution (with zero mean and unit variance). Roughly speaking, the distribution of Xn for sufficiently large n is reasonably approximated by the lognormal dis- tribution lnN (nµ, nσ2). The lognormal parameters nµ and nσ2 diverge as n → ∞; thus, Xn itself does not have a stationary distribution exactly. Many systems have multiplicativity in some sense; thus, the lognormal distribution is a natural distribution for their statistical properties[4]. It should be checked carefully whether a probability distribution obtained from actual data is truly lognor- mal or not, even if the lognormal fitting is visually suc- cessful. Many researchers have pointed out that a log- normal distribution can be confused with other prob- ability distributions such as power-law[16], normal[17], and stretched exponential[18] distributions. Further- more, the multiplicative nature does not necessarily lead to a lognormal distribution. In fact, lognormal behavior can easily change into other distributions if additional rules are assigned to the multiplicative stochastic pro- cess. For instance, multiplicative processes with addi- tive noise[19], reset events[20], random stopping[21] and successive sum[22] produce power-law distributions. It is difficult to examine whether an experimental data set fol- lows a genuine lognormal distribution or a lognormal-like distribution. In this paper, we perform a theoretical analysis of a stochastic model of bacterial cell size. Wakita et al.[23] reported that the cell-size distribution of the bacterial species Bacillus (B.) subtilis is well described by a log- normal distribution. They also proposed a phenomeno- 2 logical stochastic model for the growth of bacterial cells and confirmed that the numerical result quantitatively reproduces the actual cell-size distribution. In contrast to these results, we show in this paper that the cell-size distribution generated by this model is not a genuine log- normal distribution in reality. In order to elucidate the situation and the problem, we give an outline of the ex- periment and model in the next subsection. Cell A B C D E h t g n e l l l e C 1 g r o w t h r a t e v A l A C e l cell cycle τA C v D v E τC v C v B τB Time B. Experiment and modeling of bacterial cell size As a background of the present paper, we concisely review the above-mentioned experimental study [23] re- garding the cell-size distribution of B. subtilis, which is a rod-shaped bacteria of 0.5 -- 1.0 µm in diameter and 2 -- 5 µm in length. The morphology of a B. subtilis colony spreading on an agar surface changes with the nutrient and agar concentrations and is classified into five types: diffusion-limited aggregation-like, Eden-like, concentric ring-like, homogeneously spreading disk-like, and dense branching morphology-like patterns. The relation be- tween each pattern and the nutrient and agar concen- trations is summarized in the form of a morphological diagram[24]. Bacterial colonies have been extensively investigated from the standpoint of physics as a typi- cal pattern-formation phenomenon in which the different patterns appear by changing external conditions.[25] According to the measurement of bacterial cells in ho- mogeneously spreading disk-like colonies (formed in soft agar and rich nutrient), the cell size at the beginning of the expanding phase follows the lognormal distribution lnN (ln 2.7, 0.242) = lnN (0.99, 0.242), whose median is exp(0.99) = 2.7 µm. In addition, the cell length in the lag phase has been reported to increase linearly with time up to cell division. The cell cycle is represented by the log- normal distribution lnN (ln 22, 0.202) = lnN (3.1, 0.202) (the median is 22 min), and the growth rate also ex- hibits lognormal behavior with lnN (ln 0.078, 0.302) = lnN (−2.6, 0.302) (the median is 0.078 µm/min). On the basis of these experimental results, a phe- nomenological model has been proposed. The procedure of this model is schematically shown in Fig. 1. The model starts with a single cell A of unit length, and assigns cell cycle τA and growth rate vA drawn from lognormal distri- butions. Cell A grows linearly with time, i.e., the size of cell A at time t < τA is 1 + vAt. At t = τA, cell A divides equally into two cells B and C. Growth rates vB and vC and cell cycles τB and τC are newly assigned. For simplic- ity, the growth rate and cell cycle of each cell are assumed to be independently distributed. The cells continue their linear growth and cell division in the same way. Accord- ing to a numerical calculation[23], the cell-size distribu- tion becomes stationary after a long time. By choosing realistic parameters in the lognormal distributions for the growth rate and cell cycle [namely, lnN (3.1, 0.202) for the cell cycle and lnN (−2.6, 0.302) for the growth rate], the resultant stationary distribution is described FIG. 1: (Color online) Illustration of the growth model for bacterial cells. An initial cell A of unit length grows linearly with growth rate vA, and it divides equally into two cells B and C at t = τA. Cells B and C respectively grow linearly with rates vB and vC until the next cell division, and so forth. The arrangements of the cells at time t = 0 (initial state), t = τA (immediately after the first cell division), and t = τA + τB (immediately after the second division) are schematically shown in the balloons. by lnN (0.92, 0.282). This is in good agreement with the actual cell-size distribution lnN (0.99, 0.242) of B. sub- tilis. Furthermore, it has been reported that the long- normal behavior of the numerical cell-size distribution disappears if the lognormal parameter σ of the growth rate or cell cycle is too small or too large. We theoretically analyze this phenomenological model in the present paper. We derive the cell size at the onset of the cell cycle in Sect. II, and the stationary cell size of the model in Sect. III. They are expressed by sums of lognormal variables, and the corresponding cell-size distributions are not exactly lognormal. We propose a quantity that evaluates how far the cell size is different from the lognormal behavior. We numerically show that the deviation from lognormality depends on the product of the growth rate and cell cycle, and derive its scaling property. From these results, the cell-size distribution of B. subtilis is suggested to be only a lognormal-like distri- bution, but it can be approximated well by a lognormal distribution owing to the parameter values of the growth rate and cell cycle, which give a good lognormal approx- imation. II. ANALYSIS 1: CELL SIZE IMMEDIATELY AFTER CELL DIVISION The aim of this study is to obtain the stationary cell- size distribution of the model, but it is complicated to de- rive it without preparation. At each moment, there exist cells that have just divided, divided some earlier, and are about to divide. We need to consider the variation of the time elapsed from the previous division. Before study- ing this cell-size distribution, we start with the cell-size distribution limited to the cells that have just divided, which is a simpler problem. We focus on the cell size at the onset of the cell cycle n as the initial cell size and introduce a random variable X ′ 100 10−2 10−4 n o i t u b i r t s i d e v i t a l u m u C 10−6 10−1 n = 2 n = 5 n = 10 n = 100 100 culated as E[X ′ n] = E[Lk] 2n−k eµL+σ2 L/2 − 1 2n−1 , 1 2n−1 + n−1 Xk=1 L/2 − V [Lk] 4n−k = eµL+σ2 n−1 Xk=1 101 102 Size V [X ′ n] = = 3 . FIG. 2: (Color online) Cumulative distributions of X ′ n for n = 2 (squares), 5 (circles), 10 (diamonds), and 100 (crosses). Each distribution is made up of 106 independent samples, where µL = 0 and σL = 1. at the nth cell cycle. By setting Vn as the nth growth rate and Tn as the nth cell cycle, the growth increment during the nth cycle is expressed as VnTn. The cell size at the end of this cycle is X ′ n + VnTn. This length is split in half at the cell division, so X ′ n+1 is given by X ′ n+1 = 1 2 X ′ n + 1 2 VnTn. We easily obtain its solution as X ′ n = 1 2n−1 + VkTk 2n−k n−1 Xk=1 (2) (3) by applying Eq. (2) recursively and using the initial con- dition X ′ 1 = 1. Note that the effect of the initial length X ′ 1 vanishes exponentially as n increases. The random variables Vk and Tk always appear in the form of the product VkTk throughout this paper, and we set a new random variable Lk := VkTk. In light of the ex- periment, Vk and Tk in the model are lognormal variables. The product Lk is therefore a lognormal variable, because the product of two independent lognormal variables again follows a lognormal distribution [3]. We set lnN (µL, σ2 L) for the distribution of Lk and investigate the properties of X ′ n in terms of µL and σL. The model assumes that {Vk} and {Tk} are independent, and hence {Lk} are in- dependently and identically distributed. Hence, X ′ n is given by the sum of independent lognormal variables. By using the mean and variance E[L1] = E[L2] = ··· = eµL+σ2 L/2, L (eσ2 V [L1] = V [L2] = ··· = e2µL+σ2 L − 1), (4a) (4b) e2µL+σ2 L (eσ2 3 L − 1) − e2µL+σ2 L − 1) L (eσ2 3 · 4n−1 They converge exponentially as n → ∞. As shown in Fig. 2, the distribution of X ′ n also rapidly becomes sta- tionary; the distributions for n = 10 and 100 appear to be almost identical. In the numerical calculations be- low, therefore, we use the distribution for n = 10 as a substitute for the stationary distribution. The variable X ′ n is given by the weighted sum of in- dependent and identical lognormal variables Lk. A sum of lognormal variables does not exactly follow a lognor- mal distribution. (In contrast, a sum of normal vari- ables again follows a normal distribution.) However, such a sum of variables can be approximated by a sin- gle lognormal distribution. Among the many techniques to approximate a lognormal sum by a single lognormal distribution [26 -- 28], the simplest and fastest method is the Fenton-Wilkinson (FW) approximation [29], in which the lognormal parameters are estimated by match- ing the first and second moments. When sample values {x1, . . . , xN} of X ′ n are given, the best lognormal distri- bution lnN (µ, σ2) in the FW sense is determined by eµ+σ2/2 = e2(µ+σ2) = x1 + ··· + xN 1 + ··· + x2 x2 N N N =: m1, =: m2. The left-hand sides are the first and second moments of lnN (µ, σ2). The estimates µ and σ are explicitly written as µ = 2 ln m1 − 1 2 ln m2, σ2 = ln m2 − 2 ln m1. (5) In addition to its simplicity, the FW method is known to closely approximate the tail of the cumulative distribution[30]. Since we mainly focus on the cumu- lative distribution, the FW method is suitable for this study. We numerically investigated whether X ′ n is approxi- mated by a lognormal distribution. For each µL and σL, we generated independent samples {x1, . . . , xN} of X ′ n by using Eq. (3) and estimated µ and σ by the FW approx- imation (5). To evaluate the validity of the lognormal approximation, we calculated the difference between the estimated lognormal distribution Fµ,σ and the empirical cumulative distribution of {x1, . . . , xN} defined by number of elements ≥ x . N of Lk, the mean and variance of X ′ n are respectively cal- G(x) = 4 D = 0.4 D = 0.2 D = 0.1 D = 0.05 D = 0.02 0 1 3 2 m L 4 5 0 2 6 4 m L 8 s L = 1 s L = 0.5 s L = 0.2 10 m L = 5 m L = 2 m L = 0 1 0.8 0.6 0.4 0.2 0 8 7 6 5 4 3 2 1 0 (a) L s (b) (c) 0.6 0.5 0.4 0.3 0.2 0.1 0 We define the difference as D =(cid:20)Z ∞ 0 (Fµ,σ(x) − G(x))2dx(cid:21)1/2 . When the set of samples {x1, . . . , xN} closely fits the lognormal distribution, D becomes small. We use cu- mulative distributions Fµ,σ and G in the definition of D, because we are primarily interested in whether the tail of G(x) exhibits lognormal decay. For simplicity, we as- sume that {x1, . . . , xN} is arranged in descending order (x1 ≥ ··· ≥ xN ), which yields G(xi) = i/N . We replace the integral of D with the Riemann sum 1/2 ∆ ="N −1 Xi=1 ="N −1 Xi=1 (cid:18)Fµ,σ(xi) − (Fµ,σ(xi) − G(xi))2 (xi − xi+1)# (xi − xi+1)# N(cid:19)2 i 1/2 . This is a discrete form of D. In Fig. 3, we illustrate the numerical result of how close the stationary distribution of X ′ n is to the lognormal dis- tribution. We generated 104 independent samples of X ′ 10 using Eq. (3) (recall that the distribution of X ′ n appears to become stationary even at n = 10), estimated µ and σ by the FW method, and computed ∆. Figure 3(a) is a contour plot of ∆ in the (µL, σL) plane. The distribution of X ′ 10 deviates from the lognormal distribution (i.e., ∆ becomes large) as σL and µL increase. Figures 3(b) and 3(c) show profile curves at fixed σL (σL = 0.2, 0.5, and 1) and µL (µL = 0, 2, and 5), respectively. The dependence of ∆ on µL [Fig. 3(b)] can be ex- If we multiply {x1, . . . , xN} plained briefly as follows. by a factor eα to obtain {eαx1, . . . , eαxN}, the lognor- mal estimates become µ + α and σ2. Since the scaling relation Fµ+α,σ(eαxi) = Fµ,σ(xi) holds from Eq. (1), ∆ for the sample {eαxi} is expressed as 1/2 i ∆({eαxi}) =" N (eαxi+1 − eαxi)# N(cid:19)2 Xi=1(cid:18)Fµ+α,σ(eαxi) − (xi+1 − xi)# ="eα N(cid:19)2 Xi=1(cid:18)Fµ,σ(xi) − = eα/2∆({xi}). 1/2 N i Hence, multiplying {xi} by eα causes the factor exp(α/2) to ∆. Meanwhile, multiplying {xi} by eα corresponds to replacing Lk(= VkTk) in Eq. (3) with eαLk. Since eαLk follows lnN (µL + α, σ2 L), using eαLk instead of Lk is equivalent to adding α to µL. Finally, ∆ ∝ exp(µL/2) is derived. We numerically confirmed this relation (see Fig. 4). Figure 4(a) shows that the graphs in Fig. 3(b) grow exponentially. Thus, the three graphs in Fig. 3(c) collapse into a single curve by using the scaled value 0 0.2 0.4 s L 0.6 0.8 1 FIG. 3: (Color online) Quality of the FW approximation to the distribution of X ′ n (n = 10). (a) Contour plot of ∆ in the (µL, σL) plane. Contours of ∆ = 0.02, 0.05, 0.1, 0.2, and 0.4 are shown by the solid curves. (b) Graphs of ∆ vs µL at σL = 0.2, 0.5, and 1. (c) Graphs of ∆ vs σL at µL = 0, 2, and 5. ∆/ exp(µL/2), as shown in Fig. 4(b). In short, the in- crease in ∆ against µL is simply due to the scaling effect of ∆. III. ANALYSIS 2: STATIONARY CELL-SIZE DISTRIBUTION IN THE MODEL In this section, we investigate the cell-size distribution in the model after a long time, which also corresponds to the experimental cell-size distribution. Note that this distribution involves all cells, not only cells just after di- D D (a) 101 100 10−1 10−2 10−3 2 / L m e D(cid:181) s L = 1 s L = 0.5 s L = 0.2 0 2 6 4 m L 8 10 m L = 0 m L = 2 m L = 5 (b) 0.05 ) 2 / L m ( p x e / 0.04 0.03 0.02 0.01 0 0 0.2 0.4 s L 0.6 0.8 1 FIG. 4: (Color online) Scaling property of ∆. (a) ∆ grows exponentially with µL. The relation ∆ ∝ exp(µL/2) is shown by the solid line. (b) The graphs in Fig. 3(b) collapse to a single curve by scaling ∆ by the factor exp(µL/2). Different points are alternately aligned to avoid overlapping. vision. Thus, the distribution is different from X ′ in the previous section. n stated After a long time, during which the cells undergo di- vision many times, the cell size just after the division is written as X ′ ∞ = Lk 2k . ∞ Xk=1 (6) This expression is obtained by taking the limit n → ∞ in Eq. (3). In this stationary state, the cell size just before the division is given by L0 +X ′ ∞, where L0 is a lognormal variable with lnN (µL, σ2 L). At an arbitrary time, a ran- domly chosen cell takes a uniformly random size between X ′ ∞ and L0 + X ′ ∞. The size of a cell randomly chosen at an arbitrary time is therefore written as X = rL0 + Lk 2k , ∞ Xk=1 (7) where r is a uniform random number on the unit inter- val [0, 1] and L0, L1, . . . are independently and identically distributed with the lognormal distribution lnN (µL, σ2 L). The first term rL0 on the right-hand side represents the fluctuation of the time elapsed from the previous cell di- vision. Note that r = 0 and r = 1 respectively corre- 5 n o i t u b i r t s i d e v i t a l u m u C 100 10−2 10−4 10−6 Model X X′n 100 Size [µm] 101 FIG. 5: (Color online) Cell-size distributions obtained by di- rect simulation using the model (squares), and of X (dia- monds) and X ′ ∞ (triangles). The distribution of X is close to that obtained by direct simulation, but the distribution of X ′ ∞ is not. spond to the cell sizes just after division and just before division. We numerically verified that Eq. (7) gives a cell-size distribution corresponding to the model (see Fig. 5). The square plots show the cell-size distribution obtained by directly simulating the time evolution of cells along the procedure of the model (as in Fig. 1). The cell cycles and growth rates were drawn from the lognormal dis- tributions lnN (3.1, 0.202) and lnN (−2.6, 0.302), respec- tively. We computed the cell size distribution when the total number of cells became 106. The diamonds show the distribution of X obtained from 106 samples, in which we replaced the infinite sum in Eq. (7) with the first ten terms. These two graphs are very close to each other, which indicates that Eq. (7) is a valid expression. The distribution of X ′ n (n = 10) is shown as the triangles, but is obviously different from the other distributions. It is not clear whether the distribution of X is approx- imated by a single lognormal distribution. We show in Fig. 6 the validity of the lognormal approximation of X by the FW method. As with Fig. 3, we calculated ∆ by using 104 samples of X, in which the infinite sum in Eq. (7) was replaced with the first ten terms. Figure 6(a) is a contour plot of ∆ in the (µL, σL) plane. The shape of the contour lines is more intricate than that in Fig. 3(a). Figures 6(b) and 6(c) respectively illustrate profile curves at fixed σL (σL = 0.2, 0.5, and 1) and µL (µL = 0, 2, and 5). The curves in Fig. 6(c) are not monotonic and they all take the minimum values at σL = 0.43 -- 0.44. That is, the distribution of X is greatly inconsistent with a lognormal distribution when σL is too small or too large. Upon using the scaled value ∆/ exp(µL/2), the curves in Fig. 6(c) collapse into a single curve [see Fig. 6(d)]. In Fig. 7, the distributions of X with µL = 0 and σL = 0.2 (lower triangles), 0.4 (circles), and 0.8 (upper triangles) are presented. The lognormal distributions ob- tained by the FW approximation are shown by the solid curves. The distribution of σL = 0.4 [near the bottom of the curve in Fig. 6(d)] perfectly fits the lognormal dis- D D D =0.02 (a) L s 1 0.8 0.6 0.4 0.2 0 (b) D = 0.4 D = 0.2 D = 0.1 D = 0.05 0 1 3 2 m L 4 5 8 7 6 5 4 3 2 1 0 (c) 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 m L = 5 m L = 2 m L = 0 0 0.2 0.4 s L 0.6 0.8 1 (d) 0.05 ) 2 / L m ( p x e / 0.04 0.03 0.02 0.01 6 s L = 1 s L = 0.2 s L = 0.5 0 2 6 4 m L 8 10 m L = 0 m L = 2 m L = 5 0 0 0.2 0.4 s L 0.6 0.8 1 FIG. 6: (Color online) Properties of ∆ for the distribution of X. (a) Contour plot of ∆ in the (µL, σL) plane. (b), (c) Profile curves at fixed σL (σL = 0.2, 0.5, and 1) and µL (µL = 0, 2, and 5), respectively. (d) The curves in (c) overlap each other upon using the scaled value ∆/ exp(µL/2). As with Fig. 4(b), different points are aligned alternately. 100 10−2 10−4 n o i t u b i r t s i d e v i t a l u m u C 10−6 10−1 s L s = 0 . 8 L = 0 . 4 s L = 0 . 2 100 Size 101 FIG. 7: (Color online) Comparison of distributions of X with µL = 0 and σL = 0.2 (lower triangles), 0.4 (circles), and 0.8 (upper triangles). The solid curves are the lognormal approx- imation results obtained by the FW method. tribution. In contrast, the distribution of σL = 0.2 de- cays faster than the lognormal distribution, and that of σL = 0.8 decays slower than the lognormal. In the actual growth of B. subtilis, the growth rate Vk and cell cycle Tk respectively obey lnN (ln 22, 0.202) and lnN (ln 0.078, 0.302). The distribution of their product Lk = VkTk is then given by lnN (ln 22 + ln 0.078, 0.202 + 0.302) = lnN (0.54, 0.362). Note that σL = 0.36 is near the bottom of the curve in Fig. 6(d), at which ∆ becomes small. This is a reason why the actual cell-size distribution of B. subtilis agrees very well with a lognormal distribution. The mean of X is easily calculated as E[X] = E[r]E[L0] + E[Lk] 2k = 3 2 eµL+σ2 L/2, ∞ Xk=1 where we used Eq. (4a) and E[r] = 1/2 (mean of the uniform random number on [0, 1]). The variance of X is calculated as V [X] = V [rL0] + V [Lk] 4k = 2 3 e2µL+2σ2 L − 7 12 e2µL+σ2 L , ∞ Xk=1 where we used Eq. (4b) and the formula[31] V [rL0] = V [r]V [L0] + E[r]2V [L0] + V [r]E[r]2 for the variance of the product, with V [r] = 1/12. The square mean of X is then given by E[X 2] = V [X] + E[X]2 = e2µL+2σ2 L + 2 3 e2µL+σ2 L . 5 3 The parameter µ of the FW approximation is µ = 2 ln E[X] − 1 2 ln E[X 2], D D D where we substitute E[X] and E[X 2] for m1 and m2 in Eq. (5), respectively. We then obtain the median of X as eµ = E[X]2 E[X 2]1/2 = 9√3 4 eµL (2 + 5e−σ2 L)1/2 . With this result, we can compute the median of the sta- tionary cell size X from two parameters, µL and σL. By using the values µL = 0.54 and σL = 0.36 from the ex- periment, we have eµ ≈ 2.65 µm. This estimate correctly gives the experimental median cell size of 2.7 µm. IV. DISCUSSION In this paper, we have solved the phenomenological stochastic model for bacterial growth and have shown that the stationary cell size (7) is expressed by using lognormal variables and a uniform variable. This re- sult suggests that the cell size of B. subtilis does not fol- low a genuine lognormal distribution but a lognormal-like distribution. The parameter σL is an important indica- tor of the quality of the lognormal approximation. This study has elucidated that the value σL = 0.36 for actual bacterial growth, which is near the bottom of Fig. 6(d) (σL ≈ 0.43), is the reason why the experimental cell-size distribution can be approximated well by a lognormal distribution. The value of σL can be varied by changing external conditions or by using other bacterial species such as Escherichia coli[32] and Proteus mirabilis[33]. We consider that the model and analysis in this study should be quantitatively inspected by estimating σL and ∆ from experiments under various conditions. T ) and lnN (µV , σ2 We make a further comparison between our result and a numerical result of Wakita et al.[23] in which the cell cy- cle Tk and growth rate Vk are varied separately. Here we set lnN (µT , σ2 V ) for the distributions of Tk and Vk, respectively. Wakita et al.[23] investigated the lognormality of the stationary cell-size distribution of the model in the (σT , σV ) plane, instead of σL, with fixed medians exp(µT ) = 20 min and exp(µV ) = 0.1 µm/min. This result is shown by the circles (the lognormal ap- proximation is valid) and upper and lower triangles (the lognormal approximation is not good) in Fig. 8. The size distribution deviates from the lognormal distribution for small σT and σV or for large σT and σV . In this fig- ure, the contours of ∆ = 0.01, 0.02, and 0.04 are also shown by the solid curves. The relation σ2 T + σ2 V is obtained from Lk = VkTk, and hence, the curve of σL = const. forms an arc centered at the origin in the (σT , σV ) plane. The value of ∆ depends on σL and not on σT and σV ; thus, the contours of ∆ are concentric cir- cles. In contrast, the boundary between the circles and the lower triangles seems to be a straight line. We de- duce that this difference arises because the judgment of L = σ2 7 D = 0.02 D = 0.01 0.4 0.3 V s 0.2 0.1 0 D =0.04 D D = 0 . 0 2 = 0 . 0 1 0 0.1 0.2 s T 0.3 0.4 0.5 FIG. 8: (Color online) Quality of the lognormality of the sta- tionary cell-size distribution in the (σT , σV ) plane. The circles (showing that the lognormal approximation is valid) and up- per and lower triangles (lognormal approximation is invalid) are results of Wakita et al.[23]. Contours of ∆ = 0.01, 0.02, and 0.04 are shown by solid curves. a circle or triangle relied on human eyes. Nevertheless, we consider our result to be consistent with the previous study as a whole. In fact, the region of ∆ < 0.01 contains only circles, and that of ∆ > 0.02 contains only lower tri- angles. Moreover, circles and lower triangles both exist in the narrow range 0.01 < ∆ < 0.02. The model assumes the lognormality of the growth rate Vk and cell cycle Tk. (Theoretically, the lognormality of their product Lk = VkTk is essential in our analysis.) As a matter of fact, however, it is unclear why they follow lognormal distributions. At the same time, we cannot ex- clude the possibility that they actually follow lognormal- like distributions. The stochastic properties of Vk and Tk may involve dynamics inside a cell, and this is a problem for future research. Increasing the amount of experimen- tal data can also help to investigate Vk and Tk in more detail. ∞ ≡ exp(µL)] and X ′ The difference ∆ between a cell-size distribution and its FW approximate becomes large when σL increases in both Figs. 3(a) and 6(a). This behavior has also been found for a finite sum of lognormal variables[26]. For small σL, on the other hand, ∆ for X [Fig. 6(a)] is greatly different from that for X ′ ∞ [Fig. 3(a)]. Let us consider the limiting case σL = 0, that is, the random variable Lk is constant [Lk ≡ exp(µL)]. X ′ ∞ in Eq. (6) is also con- stant [X ′ ∞ is distributed lognormally with lnN (µL, 0). Thus, ∆ = 0 for X ′ ∞. [We can observe ∆ → 0 as σL → 0 in Fig. 3(c).] On the other hand, the random variable rL0 in Eq. (7) is uniformly distributed in the interval [0, exp(µL)], and X(= rL0 + X ′ ∞) is uni- formly distributed on [exp(µL), 2 exp(µL)]. The uniform distribution is different from a lognormal distribution, so ∆ for X does not become zero as σ → 0. Thus, the property of ∆ for small σL is affected by the term rL0 in Eq. (7). We conjecture that the balance between the increase in ∆ for large σL and the existence of the term rL0 corresponds to the minimum point in Fig. 6(d), but we have not yet carried out a detailed analysis, especially of why the minimum is at σL ≈ 0.43. In order to comprehend complex systems, we usually describe a phenomenon broadly at first, neglecting its details, then gradually increase the accuracy. In this pa- per, we have investigated a measure of how much an ac- tual distribution deviates from the approximate lognor- mal distribution, which is an in-depth expression of the result that the cell-size distribution is close to the lognor- mal distribution. Lognormal behavior has been reported in many systems as a first approximation, and we hope that our analysis will help promote further research on such systems. Acknowledgments 8 The authors are very grateful to H. R. Brand for the fruitful discussion of the manuscript. This work was supported by a Grant-in-Aid for Young Scientists (B) (25870743) from the Japan Society for the Promotion of Science (KY), and a Chuo University Grant for Special Research and a Grant-in-Aid for Exploratory Research (15K13537) from JSPS (JW). [1] M. E. J. Newman, Contemp. Phys. 46, 323 (2005). [2] M. Buchanan, Ubiquity: Why Catastrophes Happen Phys. Soc. Jpn. 78, 125001 (2009). [18] M. P. H. Stumpf and P. J. Ingram, Europhys. Lett. 71, (Three Rivers Press, New York, 2000). 152 (2005). [3] E. L. Crow and K. Shimizu, Lognormal Distributions [19] H. Takayasu, A. Sato, and M. Takayasu, Phys. Rev. Lett. (Marcel Dekker, New York, 1988). 79, 966 (1999). [4] N. Kobayashi, H. Kuninaka, J. Wakita, and M. [20] S. C. Manrubia and D. H. Zanette, Phys. Rev. E 59, 4945 Matsushita, J. Phys. Soc. Jpn. 80, 072001 (2011). (1999). [5] E. Limpert, W. A. Stahel, and M. Abbt, BioScience 51, [21] K. Yamamoto and Y. Yamazaki, Phys. Rev. E 85, 011145 341 (2001). (2012). [6] H. Katsuragi, D. Sugino, and H. Honjo, Phys. Rev. E 70, 065103(R) (2004). [22] K. Yamamoto, Phys. Rev. E 89, 042115 (2014). [23] J. Wakita, H. Kuninaka, T. Matsuyama, and M. [7] M. Bittelli, G. S. Campbell, and M. Flury, Soil Sci. Soc. Matsushita, J. Phys. Soc. Jpn. 79, 094002 (2010). Am. J. 63, 782 (1998). [8] P. Uttley, I. M. McHardy, and S. Vaughan, Mon. Not. R. Astron. Soc. 359, 345 (2005). [24] M. Matsushita, F. Hiramatsu, N. Kobayashi, T. Ozawa, Y. Yamazaki, and T. Matsuyama, Biofilms 1, 305 (2004). [25] T. Vicsek, Fluctuations and Scaling in Biology (Oxford [9] C. Furusawa, T. Suzuki, A. Kashiwagi, T. Yomo, and K. University Press, New York, 2001). Kaneko, Biophysics 1, 25 (2005). [26] S. C. Schwartz and Y. S. Yeh, Bell Syst. Tech. J. 61, 1441 [10] Y. Sasaki, H. Kuninaka, and M. Matsushita, J. Phys. (1982). Soc. Jpn. 76, 074801 (2007). [11] S. Redner, Phys. Today 58, 49 (2005). [12] S. Fortunato and C. Castellano, Phys. Rev. Lett. 99, 138701 (2007). [13] M. Abramowitz and I. A. Stegun, Handbook of Mathe- [27] C.-L. Ho, IEEE Trans. Veh. Technol. 44, 756 (1995). [28] N. B. Mehta, J. Wu, A. F. Molisch, and J. Zhang, IEEE Trans. Wireless Commun. 6, 2690 (2007). [29] L. F. Fenton, IRE Trans. Commun. Syst. 8, 57 (1960). [30] A. A. Abu-Dayya and N. C. Beaulieu, IEEE Trans. Veh. matical Functions (Dover, New York, 1965). Technol. 43, 163 (1994). [14] D. Sornette, Critical Phenomena in Natural Sciences (Springer, Berlin, 2006). [15] M. Mitzenmacher, Internet Math. 1, 226 (2004), 2nd ed. [16] Y. Malevergne, V. Pisarenko, and D. Sornette, Phys. Rev. E 83, 036111 (2011). [17] H. Kuninaka, Y. Mitsuhashi, and M. Matsushita, J. [31] L. A. Goodman, J. Am. Stat. Assoc. 55, 708 (1960). [32] M. Fujihara, J. Wakita, D. Kondoh, M. Matsushita, and R. Harasawa, Afr. J. Microbiol. Res. 7, 1780 (2013). [33] T. Matsuyama, Y. Takagi, Y. Nakagawa, H. Itoh, J. Wakita, and M. Matsushita, J. Bacteriol. 182, 385 (2000).
1206.4815	1	1206	2012-06-21T09:23:25	Conformational Transitions in Silver Nanoparticals: DNA and Photoirradiation	[ "physics.bio-ph" ]	Photoirradiation of silver nanoparticles in water solution of NaNO3 (0.01M) and in dissolved DNA was investigated by spectrophotometric and thermodynamic kinetic approaches. It is shown that only the irradiated complexes AgNPs-DNA have distinctly expressed isosbestic point. The test with the free AgNPs demonstrates that as a result of photo-irradiation desorption of silver atoms and their oxidation takes place. We have also observed that at photo-irradiation of the complexes by AgNPs-DNA desorption of silver atoms from the surface of AgNPs takes place. Kinetic study of photo-desorption of silver atoms has allowed to estimate desorption rate constant and the heat of adsorption for free AgNPs and AgNPs bound with DNA. On DNA example toxicity of AgNPs at their application to photo-chemo and photo-thermo therapy of cancer is discussed.	physics.bio-ph	physics	1 Conformational Transitions in Silver Nanoparticals: DNA and Photoirradiation Vasil G. Bregadze1, Zaza G. Melikishvili2, Tamar G. Giorgadze1 Abstract. Photoirradiation of silver nanoparticles in water solution of NaNO3 (0.01M) and in dissolved DNA was investigated by spectrophotometric and thermodynamic kinetic approaches. It is shown that only the irradiated complexes AgNPs -DNA have distinctly expressed isosbestic point. The test with the free AgNPs demonstrates that as a result of photo-irradiation desorption of silver atoms and their oxidation takes place. We have also observed that at photo-irradiation of the complexes by AgNPs- DNA desorption of silver atoms from the surface of AgNPs takes place. Kinetic study of photo-desorption of silver atoms has allowed to estimate desorption rate constant and the heat of adsorption for free AgNPs and AgNPs bound with DNA. On DNA example toxicity of AgNPs at their application to photo-chemo and photo-thermo therapy of cancer is discussed. 1 Ivane Javakhishvili Tbilisi State University, Andronikashvili Institute of Physics, 6 Tamarashvili Str. 0177 Tbilisi, Georgia; 2Georgian Technical University, Institute of Cybernetics. 5, Euli Str., 0186, Tbilisi, Georgia. Keywords: Silver nanoparticles, silver ions, DNA-intercalator complex, optical spectroscopy of intramolecular interactions. Corresponding author: Vasil G. Bregadze [email protected], [email protected] Introduction One of interesting tendencies in nano-technology is the application of metal nanoparticles in cancer photo-chemo and photo-thermotherapy. For the purpose it is important to choose materials and sizes of nanoparticles that are capable for: 1) conformational changes; 2) diffusion and 3) chemical transformations in the presence of DNA, light, temperature, ionic strength, redox agents and agents changing environment polarity. A fine example of such material can serve silver nanoparticles (AgNPs) having the size not more than 10 nm [1]. It is hard to overstate the role of metal ions, especially transition ones, in vital activity of organisms. Particular interest causes the interaction of metal ions such as Pt(II), Ag(I), Cu(I) with DNA because metal induced point defects in DNA [2-4] can lead to point mutations and can participate in formation of cross-links between the chains of DNA. One of the interesting examples is the application of cis-diamine -dichlorine- platinium and so-called photo-cis- platinium for tumor treating. More and more articles have been published lately where the usage 2 of metal nanoparticles particularly gold [5-9] and carbon [10,11] ones for tumor photo-thermo- therapy is discussed. Silver nanoparticles (AgNPs) are considerably rarely used for the purpose because one of their substantial properties is stability in solutions. The goal of the present investigation is to study the interaction of AgNPs with thymus DNA by spectroscopic and thermodynamic methods in 1) darkness; 2) under photo-irradiation and 3) temperature exposure. Materials and Methods  Colloidal silver suspension in distilled water was prepared of silver nanoparticles (AgNPs) of 1-2 nm size (DDS Inc., D/B/A/, Amino Acid & Botanical Supply).  Calf thymus DNA produced by Sigma was dissolved in 0.01 M NaNO3 solution – background electrolyte, pH~6.0. DNA concentration was evaluated by UV absorption spectrophotometer. Molecular extinction factor έ = 6600 cm-1 M-1 (P), λ = 260 nm.  Registration of absorption spectra was carried out by optical fiber spectrometer AvaSpec ULS 2048 – USB 2.  Photoirradiation was carried out in reactor with the fixed light beam in 1cm rectangular fluorescent quartz cell. In the same cell with the interval of 5 min absorption spectra of irradiated solutions were registered by Avantes spectrometer. Before each absorption registration the cell was shut with a shutter that protected the solution from photo irradiation. Registration time was 8 msec. As a source of radiation xenon arch discharge lamp with rating of 35 W in glass balloon was used. At solution irradiation water filter and light filter with light wave transmission λ=436 nm were used. Radiation power in the cell was 300 mW for water filter and 15 mW for water filter matched with light filter (λ=436 nm). Results and Discussion Fig.1 shows the spectra of AgNPs, AgNPs-DNA complexes. Fig. 1 Absorbtion spectra of AgNPs and AgNPs-DNA complexes. [AgNPs]-0.72·10-4 M (Ag0), [DNA]-1.6·10-4 M(P), [NaN03]=10-2M 3 The analysis of spectra given in Fig.1 demonstrates that DNA at interaction with AgNPs causes hypsochromic shift of absorption band on 6 nm and 20% hypochromic effect. Blue shift points out that a kind of disintegration (loosening) i.e. attenuation of interactions between silver atoms has taken place. And the decay of intensity means partial corrosion of AgNPs [1]. Kinetics of AgNPs photoirradiation has been studied. Fig. 2 and 3 show superposed absorption spectra of AgNPs and AgNPs in complex with DNA before and after irradiation using water filter. Fig.2 Absorbtion spectra of AgNPs before and after irradiation(5min interval) [AgNPs]-1.94·10-4 M (Ag0), [NaN03]=10-2M Fig.3 Absorbtion spectra of AgNPs-DNA before and after irradiation (5min interval) [AgNPs]- 1.94·10-4 M (Ag0), [DNA]-1.6·10-4 M(P), [NaN03]=10-2M The analysis of the spectra on Fig. 2 and 3 demonstrate that only the irradiated complexes AgNPs-DNA have distinctly expressed isosbestic point. The test with the free AgNPs shows that as a result of photoirradiation desorption of silver atoms and their oxidation by H3O+ atoms to Ag+ ions takes place . The presence of isosbestic points in the absorption spectra of irradiated AgNPs-DNA complexes proves that the system has two states, i.e. AgNPs-DNA complex has two forms of existence joint by structural photodiffusive transition from one form, e.g. spherical one, to extended long and probably one-dimentional form along DNA double helix. The analysis of the spectra really shows with good correlation ≤5% that the space under the spectra is preserved which means that there are no changes in chromophore electron structure . Besides, half width of absorption spectra Δλ1/2 is changed from 140 nm to 360 mn. Red shift and widening of AgNPs absorption band points out to the typical for molecular systems increase of electron conjugation (linear and cyclic conjugated systems [12]). 4 The company nanoComposix [13,14] gives sample absorption spectra for spherical AgNPs with particle sizes of 10, 20, 30, 40, 50, 60, 70, 80, 90 and 100 nm at the same mass concentration 0.02 mg/ml. The given data show that with the increase of the size of the particles widening of the red shift of absorption band can be observed. It is notable that despite the growth of the particle size, i.e. decrease of their total number in the solution, the intensity of absorption bands for nanoparticles with the sizes from 10 to 40 nm is not practically changed, only a small shift of absorption band maximums can be observed. The above explicitly points out that chromophore units are silver atoms and not nanoparticles. Thus, we can draw a conclusion that silver atoms in AgNPs are sufficiently isolated and bound together by dispersion interaction (induced dipole-induced dipole). As these interactions are performed in water surrounding they should be considerably amplified at the expense of so called hydrophobic effect [15] , that means compaction and then minimization of the surface (decrease of the system entropy). We especially point out that in nanoparticle, which consists of one kind of atoms, along with the mentioned dispersial interaction, the so-called resonance interaction should take place [16]. Such types of interactions are typical for molecular crystals and they usually lead to exiton splitting of the principal absorption band. Inevitable condition for exiton splitting is the presence of a system consisting of identical groups and having hard structure [17]. The absence of splitting can mean that AgNPs under investigation (fig.1) have liquid structure resembling a drop which under definite conditions (such as temperature, photo-irradiation, variations in dielectric constant of the environment) should be characterized by conformational transitions. For the estimation of the point we have carried out thermodynamic kinetic analysic of absorbtion spectra of AgNPs (see fig.2 and 3). Fig.4 Kinetic curves of photodesorbtion in Mt/Me and t1/2 coordinates for free AgNPs and AgNPs bound with DNA Fig.5 Kinetic curves of photodesorbtion in ln[Me/(Me–Mt)] and t coordinates for free AgNPs and AgNPs bound with DNA 5 Along with spectroscopic analysis of the curves given on Fig. 2 and 3 thermodynamic kinetic analysis of the mechanism of AgNPs photo diffusion on DNA double helix has been carried out. The mechanism consists of: 1 – desorption of silver atoms from the surface of nanoparticles bound to DNA; 2 – adsorption of silver atoms on DNA surface and 3 – creation of cross-links between silver atoms and DNA chains. Let’s consider the changes in absorption spectra for photo-irradiated free AgNPs and AgNPs–DNA complex given in Fig. 2 and 3 versus the duration of irradiation in M t/Me and t1/2 (see Fig.4). Me is the number of silver atoms in nanoparticles at the beginning (adsorption A t when λ=430 nm at t=0), Mt is molar quantity of silver atoms desorbed by the time moment t (difference between absorption At=0 –At at λ=430 nm). As it can be seen the curves in Fig.4 have S-shape form both for photo desorption kinetics of silver atoms from the surface of free AgNPs and AgNPs-DNA complexes. S –shape appearance of the curves denotes that photo-induced desorption of atoms is a complex and multiphase process [18]; it is diffusion of silver atoms from the inner part of a nanoparticle to its surface, conformational changes in the particles especially in those ones that are adsorbed on DNA surface. Next we consider the results given in Fig. 2 and 3 in ln[Me/(Me- Mt)] and t coordinates, which are given in Fig 5. The analysis of the curves shows that only initial stage of the given curves of desorption kinetics obey linear law of first-order equation ln[Me/(Me–Mt)]=kt. The constants of desorption rate of silver atoms from the surface of AgNPs has been evaluated from the slopes of the curves and the date are: for free AgNPs kd = (5.50.2)10-4s-1. The values allow us to evaluate desorption reaction activation energy Ed using the equation kd = υ0exp (-Ed/RT), where υ0 is pre- exponential factor assumed as υ0 =~10-10s-1 (reciprocal quantity to silver atom oscillation time in nanoparticles). In this case we have got the following values for Ed at T= 300o K: Ed=~18.2 kcal/mol Ag0 for free AgNPs and Ed =~19.3 kcal/mol Ag0 for AgNPs bound with DNA. As Ed=Ea+Qa, where Ea and Qa are activation energies for starting activation and heating of nanoparticles, so Q1=18.2 kcal/mol and Q2=19.3 kcal/mol under the condition that formation of nanoparticles is not an activated value. The values of heat are specific for cluster nanostructures [19,20]. Next we give evaluation of life time for Ag0 complex with DNA. As far as in 1996 one of the authors [21] proposed a thermodynamic model of interaction between small ligands and DNA. Based on the example of interaction between the ions of transition metals and DNA it was shown that the life time of the complexes τ is connected to equilibrium characteristic by stability constant K and equation τ=τ0xK, where τ0 is the duration of the fluctuation excitation of the adsorbing ligands or molecules interacting with a solid surface and it lyes between 10-11 and 10-10 6 sec. Silver atom photodesorption from the surface of nanoparticles, which are in DNA complex, has desorption activation energy of 19.3 kcal/mol Ag0 and, consequently, we can assume that photoindused diffusion of AgNPs on DNA double helix takes place along with activation of silver atoms desorbtion energy equal to 19.3 kcal/mol Ag0 and thus we can state that the energy of Ag0 interaction with DNA double helix is not less than 19.3 kcal/mol. In accordance with the equation ΔG = - RTlnK and with the application of the evaluated energy 19.3 kcal/mol we assume that the stability constant of the complex is not less than 1014 and, consequently, life time of the complexes is equal to 104 sec. Life-times of 1 sec. order and more are characteristic for inter-strand interactions with the partipipation of transition metal ions, so called cross-links [20]. As far as in 1969 Wilhelm and Daune [22] showed that Ag+ ions form cross-links between DNA chains thus releasing protons bound with N1 guanine and N3 thymine into the solution. We have estimated stability constants of Cu+ and Ag+ ions with DNA which are equal to 10.8 for Ag+ and 14.9 for Cu+. Thereafter the change of free energy for Ag+ is 15kcal/M and 20.5 kcal/M for Cu+ and lifetimes are 0.63 sec for Ag+ and 8.6x103 sec for Cu+ [4]. Conclusions Using spectrophotometry and thermodinamic approaches we have shown that 1) at interaction with DNA-AgNPs are adsorbed on it and only partial corrosion of nanoparticles at the level of Ag+ ions is observed; 2) at photoirradiation (λ=436 nm or full spectrum of visible band) desorption of silver atoms from the surface of AgNPs takes place. The atoms are first adsorbed on the surface of DNA and then penetrate inside the double helix (cross links between complementary DNA base pairs) making prolate stretched structure (AgNPs absorption spectrum width is changed from 140 nm to 360 nm at half-height); 3) Kinetic study of photo-desorption has made it possible to determine desorption rate constant Kd and adsorption heat Qa that are equal to Kd = (5.5± 0.2) x 10-4s-1 ; Qa ≥ 18.2 kcal/M Ag0 for free AgNPs and Kd = (8.9± 0.5) x 10-5s -1; Qa ≥ 19.3 kcal/M Ag0 for AgNPs bound with DNA. 4) On DNA example toxicity level of AgNPs is shown that can be application to for medical purposes, e.g for photo-chemo and photo-thermo terapy of cancer. Acknowledgments: The work was partly supported by the Grant No GNSF/ST09_508_2-230. References: 7 1. V. Bregadze, S. Melikishvili, Z. Melikishvili, G. Petriashvili, Nanochemistry and Nanotechnologies Proceedings of Papers of the First International Conference March 23 - 24, 2010, Tbilisi Georgia, Publishing House “UNIVERSAL” Tbilisi 2011, 136-140. 2. Natile G., Cannito F., In: Metale Complex-DNA Interactious. N. Hadjiliadis and E. Sletten, eds. Chichuster; Blackwell Publishing Ltd, 2009, 135-173. 3. Morchan V., Grandas A., In: Metale Complex-DNA Interactious. N. Hadjiliadis and E. Sletten, eds. Chichuster; Blackwell Publishing Ltd, 2009, 273-300. 4. Bregadze V.G., Khutsishvili I.G., Chkhaberidze J.G., Sologashvili K. Inorganica Chimica Acta, 2002, V. 339, 145-159. 5. Niidome T, Shiotani A, Akiyame Y, Ohga A, Nose K, Pissuwan D, Niidome Y. Yakugaku Zasshi. 2010 Dec; 130(12):1671-7. Review. Japanese. 6. Khlebtson N, Dykman L. Chem Soc Rev. 2011 Mar; 40(3):1647-71. Epub 2010 Nov 16. Review. 7. Sekhon BS, Kamboj SR. Nanomedicine. 2010 Oct; 6(5):612-8. Epub 2010 Apr 22. Review. 8. Ratto F, Matteini P, Centi S, Rossi F, Pini R. J Biophotonics. 2011 Jan;4(1-2):64-73. Doi: 10.1002/jbio.201000002.Epub 2010 Mar 1. Review. 9. Pissuwan D, Valenzuela S, Cortie MB. Biotechnol Genet Eng Rev. 2008;25:93-112. Review. 10. Whitney JR, Sarkar S, Zhang J, Do T, Young T, Manson MK, Campbell TA, Puretzky AA, Rouleau CM, More KL, Geohegan DB, Rylander CG, Dorm HC, Rylander MN. Lasers Surg Med. 2011 Jan;43(1):43-51. Doi:10.1002/lsm.21025. 11. Fisher JW, Sarkar S, Buchanan CF, Szot CS, Whitney J, Hatcher HC, Torti SV, Rylander CG, Rylander MN. Cancer Res. 2010 Dec 1;70(23):9855-64. Epub 2010 Nov 23. 12. Keniti Higasi, Hiroaki Baba and Alan Rembaum, Quantum Organic Chemistry (in Russian), "Mir", Moscow, (1967) 379p. 13. The effect of aggregation on optical properties: Data of firm nanoComposix, Silver Nanoparticles: Optical Properties. www.nanocomposix.com. 14. The effect of size on optical properties: Data of firm nanoComposix, Silver Nanoparticles: Optical Properties. www.nanocomposix.com. 15. Volkenshtein M. V., Molecular Biophysics (in Russian), “Nauka” Moscow. (1975), 616p. 16. Davidov A.S. Theory of Molecular Excitons, McGrow-Hill Book Company, 1962 8 17. Ignasio Tinoco, Jr., Anita Halpern, and William T. Simpson, Polyaminoacids, Polypeptides and Proteins, Univ. of Wesconsion Press, 147 (1962). 18. Rogers C. Solubility and Diffusion in,”Physical and Chemical Problems of Organic Compounds in Solid State (in Russian), "Mir", Moscow (1968) p. 229-328. 19. Physical Encyclopaedia (in Russian), Cluster ions, Vol 2. Moscow. (1990). p.372-373. 20. Chemical Encyclopaedia (in Russian), Clusters, Vol 2. Moscow. (1990). p.400-403. 21. Bregadze, V.G. In: Metal Ions in Biological Systems. A. Sigel and H. Sigel, eds. New York: Marcel Dekker, 1996, Vol.32 , 453-474. 22. Wilhelm F.X. and Daune M., Biopolymers, 8, 121 (1969).
1902.01730	1	1902	2019-02-05T15:08:47	Crawling in a fluid	[ "physics.bio-ph", "cond-mat.soft" ]	There is increasing evidence that mammalian cells not only crawl on substrates but can also swim in fluids. To elucidate the mechanisms of the onset of motility of cells in suspension, a model which couples actin and myosin kinetics to fluid flow is proposed and solved for a spherical shape. The swimming speed is extracted in terms of key parameters. We analytically find super- and subcritical bifurcations from a non-motile to a motile state and also spontaneous polarity oscillations that arise from a Hopf bifurcation. Relaxing the spherical assumption, the obtained shapes show appealing trends.	physics.bio-ph	physics	Crawling in a fluid Alexander Farutin,1, ∗ Jocelyn ´Etienne,1 Chaouqi Misbah,1 and Pierre Recho1 1Univ. Grenoble Alpes, CNRS, LIPhy, F-38000 Grenoble, France (Dated: Received: February 6, 2019/ Revised version: (date)) There is increasing evidence that mammalian cells not only crawl on substrates but can also swim in fluids. To elucidate the mechanisms of the onset of motility of cells in suspension, a model which couples actin and myosin kinetics to fluid flow is proposed and solved for a spherical shape. The swimming speed is extracted in terms of key parameters. We analytically find super- and subcritical bifurcations from a non-motile to a motile state and also spontaneous polarity oscillations that arise from a Hopf bifurcation. Relaxing the spherical assumption, the obtained shapes show appealing trends. Introduction Cell motility plays an essential role dur- ing embryogenesis, tissue renewal and repair, response of the immune system to infection as well as patholog- ical processes such as cancer cell migration. It is also important from an engineering point of view to conceive biomimetic robots. A longstanding dogma in biology is that mammalian cells need mechanical interaction with a solid substrate in order to move forward. The commonly adopted idea about the source of motion is that actin polymeriza- tion and actin contraction assisted by molecular motors (myosin) create a tangential flow of the cell cortex, the retrograde flow. Thrust is then achieved by the trans- mission of momentum to the outer environment by a dynamic adhesive coupling through transmembrane pro- teins such as integrins [1]. Several modes of such crawl- ing motion are distinguished in the literature, most of which involve ample shape changes [2 -- 5] but crawling in a shape-preserving manner is also possible, as notably observed for fish keratocytes [6 -- 9]. A major step forward has been achieved by identifying that leukocytes can mi- grate in an extracellular matrix without the assistance of integrins [10], raising the important question about the role of specific adhesion in crawling. Indeed, it is reported that crawling can be mediated by friction only [11] when the cell is confined by a 3D environment. Increasing at- tention has been recently paid to the understanding of such adhesion-independent crawling [12 -- 14]. However, it is also reported that some cells can swim in an unconfined fluid [15 -- 17]. Swimming relies on periodic distortions of the cell surface which generate a reaction force from the ambient fluid [18]. Beside the special case of ciliated squirmers [19], the mainstream notion of swim- ming relies on ample shape changes, the ones of flagellas [18] or of the cell body [16, 20 -- 23]. In this letter, we show that swimming may be generi- cally possible also for mammalian cells through the same growth and contraction driven cortical flows operating during crawling and in the absence shape changes. Me- chanical interaction with a substrate or extracellular ma- trix, and ample shape changes are therefore not neces- sary for crawling. This actuation mechanism necessitates spatial symmetry breaking in order to generate directed motion, in the same way as in the case of crawling on a solid substrate [24 -- 26]. Recently, symmetry breaking in cortex dynamics and the resulting cell shapes have been analyzed for quasi-spherical cells [27] but the direct rela- tion between the symmetry breaking and the swimming motion has not been described. By coupling the acto- myosin dynamics to the internal, external, and cortex flow, we provide analytically the functional dependence of the swimming speed on the relevant parameters. A second finding is the identification of a rich panel of instabilities leading to cell polarization. Depending on three key parameters of the model (actin stiffness, rate of actin turn-over, myosin contractility) we find ei- ther a supercritical bifurcation, leading to a continuous transition between static and motile configurations, or a subcritical bifurcation implying a metastable coexistence between the static and motile configurations for a finite range of parameters. For a small enough actin turn-over rate, the system exhibits a Hopf bifurcation resulting in a permanent oscillation of the polarization. Such be- haviour is reminiscent of recent experimental evidences of actomyosin oscillations [28, 29]. Finally, we show the shapes of polarized cells if the fixed shape assumption is relaxed. The shapes are obtained in a quasi-spherical limit and show reasonable trends. Model We consider a neutrally buoyant cell in a fluid environment. Swimming is achieved by transmission of shear stress by the cortex flow to the surrounding fluid. We leave the question of the exact nature of the trans- mission mechanism and its efficiency outside the scope of this study, and assume perfect no-slip conditions between the cortex and the fluids inside and outside the cell. We first focus on a fixed spherical shape and show how spontaneous polarization of acto-myosin dynamics and the resulting cortex flow can drive the deformation- independent swimming of the cell. The cortex of the cell is modeled as a compressible two-dimensional (2D) New- tonian fluid along the surface of the sphere. The surface density of actin meshwork is denoted ca. The distribu- tion of myosin that crosslinks actin is characterized by the concentration cµ. The velocity field (in the labora- tory frame) of the cortex is denoted as uc. The surface strain rate tensor reads ǫs = ∇s ⊗ uc · Is + Is · (∇s ⊗ uc)T where Is is the surface projection operator and ∇s ≡ Is ·∇ is the surface gradient. In the framework of the active gel theory [30], we use the following constitutive equation for the surface stress σs = ηsǫs + ηbIs(∇s · uc) + Is(χcµ − αca), (1) where ηs and ηb are 2D shear and bulk viscosities, χ is the myosin-induced contractility and α is the 2D bulk modulus of the actin meshwork. The surface forces are calculated as f = ∇s · σs − f nn, where n is the outward normal and f n is a Lagrange multiplier that enforces the fixed shape [31]. It satisfies the zero total force condition H f nndA = 0, where dA is the area element on the boundary of the cell. The forces f are balanced by viscous stresses of the cytoplasm (the fluid occupying the cell interior) and the suspending fluid at the cell boundary. Both are considered incompressible Newtonian fluids of viscosities ηin and ηout, respectively. With this assumption and at zero Reynolds number, 3D fluid dynamics is governed by Stokes equations, which can be combined into a boundary integral equation [32] ηin + ηout 2 uc i (x) =I Gij (x, x′)fj(x′)dA(x′) +(ηout − ηin)I Kijk(x, x′)uc j(x′)nk(x′)dA(x′) (2) using the boundary conditions at infinity, as well as stress balance and continuity of velocity at the cell surface. Here x and x′ are points on the cell surface. The Green's kernels Gij and Kijk are listed in [33]. The swimming speed is defined as the volume-averaged velocity of the cytoplasm, which can be converted ino a surface integral vs = ( 4 the cell radius. The fixed shape assumption implies that the cortex velocity must be tangential to the cell surface in the comoving frame: (uc − vs) · n = 0. This condition fixes f n. 3 πR3)−1H x(uc ·n)dA, where R is To close the description we express the conservation equations on the actin and myosin fields ca and cµ: ca + ∇s · [ca(uc − vs)] = β(ca 0 − ca) cµ + ∇s · [cµ(uc − vs)] = Dµ∆scµ, (3) (4) where dot denotes time derivative. The term β(ca 0 − ca) represents actin turnover, with β being the turnover rate and ca 0 the equilibrium concentration in the cortex. The term Dµ∆scµ represents the surface diffusion of myosin, Dµ is the diffusion coefficient. The average myosin con- centration cµ 0 is conserved by eq. (4). This model contains two different active drivings of cell motility: molecular motors are pulling agents generating positive stresses in (1) while actin turnover in (3) can be associated with pushing agents generating negative stress in (1). The interplay between these two types of agents in cell crawling was analyzed in [34, 35]. 2 Results System (1)-(4) constitutes a closed set of equations for determining the cortex velocity field and the actomyosin dynamics. The solution strategy is as follows: It can be shown (see [33]) that for a spherical cell the cortex flow is potential, uc = ∇sU + vs, where U is the flow potential. The mechanical part of the prob- lem (1)-(2) is then solved analytically by expansion in scalar or vector spherical harmonics [33]. This allows us to express U for a spherical shape as a linear combination of the harmonic coefficients of ca and cµ and compute the swimming velocity vs = 2 3 R αca 1 − χcµ 1 2(ηs + ηb) + R(3ηin + 2ηout) , (5) 1 and cµ where ca myosin concentration, defined precisely below. 1 are the first harmonics of the actin and Equations (3) and (4) are nonlinear but still can be tackled analytically by perturbation expansion. We vali- dated the analytical results by numerical solution of eqs. (3) and (4). The details of the numerical method are given in [33]. An important observation is that the dissipation mech- anism highlighted in Eq. (26) arises from a combination of cortex, internal and external fluid viscosities. They act as dashpots in parallel, as shown by viscosity ad- ditivity (up to numerical prefactors). Interestingly, the speed is finite even if ηout = 0: while the suspending fluid is essential for swimming, its viscosity is not deci- sive for setting the value of the swimming speed. We therefore consider the limit of ηin = ηout = 0 in the rest of this study, motivated by the fact that in physi- ological conditions most of the dissipation occurs in the cortex. We also set ηb = 0 for simplicity. Along with these assumptions, we use from now on as characteristic scales R for space, R2/Dµ for time, Dµηs/R2 for surface stresses, ca a for actin and myosin concentrations but keep the same symbols to avoid new notations. Three non-dimensional parameters fully define the problem, the motor contractility ¯χ ≡ χcµ 0 R2/(Dµηs), the compressibil- 0R2/(Dµηs) and the turnover rate of ity of actin ¯α ≡ αca actin ¯β ≡ R2β/Dµ. 0 and cµ A homogeneous solution for actin and myosin (ca = cµ = 1) always exits. This solution yields a zero cortical flow field and the cell is at rest. This solution can become unstable leading to a polarized cortex dynamics. Such symmetry breaking results in a cortical flow and, in turn, in spontaneous cell motion. To quantify such process, we expand scalar fields in axisymmetric spherical harmonics as ca,µ(θ) = 1 + ca,µ l Pl(cos θ), ∞ Xl=1 where Pl are Legendre polynomials and θ is the polar an- gle between the swimming direction and the vector po- sition on the sphere counted from the center of mass. 3 We get a steady bifurcation for ¯χ1 < ¯χ2 and a Hopf one otherwise. Once the contractility exceeds the criti- cal value given by (7) the perturbations grow exponen- tially with time and nonlinear terms can no longer be neglected. In general resorting to numerical resolution is necessary. However, in the vicinity of the bifurcation point a perturbation analysis is possible. First, let us consider the case of a steady bifurcation. A nonzero first harmonic (concentration polarity) induces a spontaneous cell motion (See (26)). This propulsion mode is driven by the appearance of a retrograde flow of actin, directed from a front region to a rear region and compensated by actin creation at the front ( ¯β(1 − ca) > 0, polymerization) and degradation at the rear ( ¯β(1 − ca) < 0, depolymerization). Figure 1 shows the acto-myosin and the cortex flow patterns. The nature of this instability is similar to the one presented in [25] in a simplified picture. The unstable eigenmode is given by C1 = 2(¯α + ¯β)cµ 1 − 2 ¯αca 1. To obtain the velocity close to the bifurcation threshold, we must also consider the second harmonic. We set ca 2 = 0, because relaxation times for these modes are much smaller than that of C1. This yields the expressions of ca 2 as a function of C1, which upon substitution into the expression of 2 and cµ C1 leads to 2 = cµ ¯β2 C1 = ν = − ¯β2 + ¯α ¯β − 2 ¯α(cid:2)( ¯χ − ¯χ1)C1 + νC3 1(cid:3) , 6 ¯α2 ¯β − 12 ¯α2 + ¯α ¯β3 + 6 ¯α ¯β2 − 24 ¯α ¯β + ¯β4 ¯β(¯α + 2 ¯β)( ¯β2 + ¯α ¯β − 2 ¯α)2 (8) . The sign of ν dictates the nature of the steady bifurca- tion with ν < 0 corresponding to a supercritical bifur- cation (nonlinearity saturates linear growth) and ν > 0 corresponding to a subcritical one (nonlinearity amplifies linear growth). In the case of a supercritical bifurcation, (8) has a sta- ble fixed point given by C1 = p( ¯χ − ¯χ1)/ν. The swim- ming speed then reads FIG. 1. (Color on-line) The actin and myosin distribution (color fields) and cortex flow (arrows). Cell swims to the left. 5 4 3 vs 2 1 0 A B - 3 s v 5 4 3 2 1 0 - 1 11 12 13 14 - analytical stable unstable 1.5 C 1 x 0.5 -0.5 -1.5 - 1 8 9 10 -1.5 -0.5 0.5 1.5 x3 11 - 12 13 14 FIG. 2. (Color on-line) A and B: Swimming velocity as a func- tion of contractility. Stable (unstable) branches are shown by continuous (dashed) lines. A: Supercritical case ¯α = 10.0, ¯β = 3.0. B: Subcritical case ¯α = 10.0, ¯β = 2.0. Inset C shows a typical shape (retrograde flow is from left to right). Linearizing eqs. (3) and (4) about the homogeneous so- lution, yields l + J µµ l cµ l , (6) vs = ¯β 3( ¯β2 + ¯α ¯β − 2 ¯α)r ¯χ − ¯χ1 −ν . (9) l ca l + J aµ cµ l = J µa l = J aa ca l ca l = −l(l + 1)¯α/λl − ¯β, J aµ l cµ l , where J aa l = l(l + 1) ¯χ/λl, J µa l = −l(l + 1)¯α/λl, J µµ l = l(l + 1) ¯χ/λl − l(l + 1), and λl = 2(l2 + l − 1). The growth rate ω of a perturbation in (6) therefore obeys a quadratic equation. When the real part of ω vanishes (onset of instability), the imaginary part can either be zero or non zero. The first case defines a steady bifurcation, while the second one defines a Hopf bifurcation. A simple analysis shows that the minimum contractility at which the instability occurs is controlled by the first harmonic l = 1. The instability takes place if ¯χ > ¯χc = min( ¯χ1, ¯χ2), where ¯χ1 = 2 + 2 ¯α ¯β , ¯χ2 = 2 + ¯α + ¯β. (7) The bifurcation diagram in Fig. 2A shows both the ana- lytical results and the full numerical simulations. When the steady bifurcation is subcritical (i.e. first order transition), velocity abruptly switches to a finite value. Due to the finite jump, the perturbative expan- sion leading to eq.(8) is illegitimate. One has thus to resort to a numerical solution. Figure 2B shows the re- sults. There is a region of metastable coexistence be- tween a non-polarized static state and a polarized motile one between a turning point located at ¯χ = ¯χ3 and the linear stability threshold ¯χ = ¯χ1. A finite perturbation at a constant contractility can therefore be sufficient to initiate or arrest motion in this range. 5 4 steady branches stable unstable oscillation amplitude stable (numeric) stable (analytical) ) s v ( x a m , s v 3 2 - 3 - 4 1 0 13 - 2 13.5 14 - 14.5 15 15.5 - 1 102 102 101 101 0 100 - ig. 2A Fig. 2B Fig. 3 2 PF 1 H2 -1 10-1 1 10-2 10-2 10-1 10-1 100 100 102 102 103 103 104 104 101 101 - 4 105 104 103 102 101 2 - c - 100 10-1 10-2 10-3 FIG. 3. (Color on-line) Amplitude of the swimming veloc- ity as a function of contractility, supercritical Hopf case. ¯α = 10.0, ¯β = 1.5. Stable (unstable) branches are shown by continuous (dashed) lines. Dotted line shows maximum velocity for the limit cycle. FIG. 4. (Color on-line) Phase diagram of bifurcation types as a function of ¯α and ¯β. Color code shows (in log scale) the critical contractility ¯χc, for which the uniform solution loses stability. The bifurcation types are supercritical (PF2) and subcritical (PF1) pitchfork and supercritical Hopf (H2). Rectangle shows typical estimates of realistic parameters. When the bifurcation is of Hopf type, it is always su- percritical with periodic oscillations of velocity emerging continuously from the static trivial solution at the crit- ical value of contractility ¯χ2. A second critical contrac- tility ¯χ4 exists at which the periodic oscillations disap- pear and the system abruptly jumps to a stable motile steady-state. The steady-state branch can be continued for ¯χ ≥ ¯χ3. An unstable branch bent backwards emerges continuously from the uniform solution at critical con- tractility ¯χ1, and annihilates with the stable steady-state branch. The full bifurcation diagram shown in Fig.3 is reminiscent of a heteroclinic bifurcation. In the range [ ¯χ3, ¯χ2], we again observe the coexistence of static and motile solutions while in the interval [ ¯χ3, ¯χ4], oscillatory and motile solutions are metastable potentially giving rise to complex cell gaits in the presence of biological noise. We summarize these results on the bifurcation nature in a phase diagram in Fig. 4 where the critical value of ¯χc is shown in color code. Relaxing the fixed shape assumption, we have deter- mined the trend of the cell shape. For that purpose we have assumed the shape of the cell to be preserved by sur- face tension (see [33]). A typical shape is shown in Fig.2C having a satisfactory tendency (mushroom-like flattened shape) compared to some real cell shapes [36]. Discussion A rich panel of patterns and dynamics due to a spontaneous symmetry breaking of the acto- myosin kinetics is revealed. The cell thus swims thanks to a retrograde flow of the cortex grasping on the external fluid. Orders of magnitudes of the three key parameters ¯α, ¯β and ¯χ can be estimated from available experimental data ¯α ∼ 10−1 − 103, ¯β ∼ 1 − 102 and ¯χ ∼ 10−2 − 103 (see [33]). The wide ranges are due to disparate val- ues of ηs in the literature [37]. The rectangle in Fig. 4 shows the ranges of estimated parameters. It is inter- esting to see that this corresponds to a rich region of dynamics including the various bifurcations encountered in this study. The swimming speed (9) yields (in the pa- rameter range shown by rectangle in Fig. 4) in physical units about vs ∼ 1 − 10Dµ/R ∼ 0.1 − 1 µm/min, which is a reasonable value for many mammalian cells [38]. How the cortex flow is transmitted to the external fluid is still unclear. Three pathways are possible (i) through the membrane of the cell via recirculation of phospho- lipids, (ii) through transmembrane proteins (iii) through bumps that are advected backward by cortex in the form of waves (or shape changes). For axisymmetric flow (as considered here) the first scenario is impossible (it can be shown that the membrane is at rest in the swimming frame), except if endo/exocytosis is operating as reported recently [39]. The second scenario has been recently suc- cessfully applied to the T-lymphocyte swimming [17]. The third scenario would require a cooperative motion of bumps along the membrane but numerical evaluation seems to rule out this possibility [17]. Our results show that shape-invariant actomyosin- based motility, a hallmark of crawling, can occur away from any solid substrate, provided that the retrograde flow of actin can transmit momentum to the surround- ing fluid. This is consistent with recent experimental observations [17] and a theoretical model of active ne- matic droplets [40]. The rich bifurcation structure that allows actomyosin to break symmetry is also reminiscent of the one known for cells crawling on a solid substrate [8, 25, 28] pointing to the interpretation that the interac- tion of the actomyosin contractility with its own viscous resistance to deformation and turn-over is sufficient to propel the cell robustly on a solid and in a fluid. Acknowledgments We are grateful to Karin John for her role in the initial discussions leading to this work. AF and CM thank CNES (Centre National d'Etudes Spa- tiales) and the French-German University Programme "Living Fluids" (Grant CFDA-Q1-14) for financial sup- port. JE and PR are supported by a CNRS Momentum grant, ANR-11-LABX-0030 "Tec21" and IRS "Aniso- Tiss" of Idex Univ. Grenoble Alpes. All are members of GDR 3570 MecaBio and GDR 3070 CellTiss of CNRS. The computations were performed using the Cactus clus- ter of the CIMENT infrastructure, supported by the Rhone-Alpes region (GRANT CPER07 13 CIRA) and the authors thank Philippe Beys who manages the clus- ter. ∗ [email protected] [1] A. Mogilner, Journal of mathematical biology 58, 105 (2009). [2] R. J. Petrie and K. M. Yamada, Trends in cell biology 25, 666 (2015). [3] J. Zhu and A. Mogilner, Interface focus 6, 20160040 (2016). [4] D. R.-B. Aroush, N. Ofer, E. Abu-Shah, J. Allard, O. Krichevsky, A. Mogilner, and K. Keren, Current Bi- ology 27, 2963 (2017). [5] E. K. Paluch and E. Raz, Current opinion in cell biology 25, 582 (2013). [6] J. Lee, M. Leonard, T. Oliver, A. Ishihara, and K. Ja- cobson, The Journal of Cell Biology 127, 1957 (1994), ISSN 0021-9525. [7] B. Rubinstein, M. F. Fournier, K. Jacobson, A. Verkhovsky, and A. Mogilner, Biophys. J. 97, 1853 (2009). [8] E. Tjhung, A. Tiribocchi, D. Marenduzzo, and M. E. Cates, Nat Commun 6, 462 (2015). [9] F. Ziebert, S. Swaminathan, and I. S. Aranson, Journal of The Royal Society Interface p. rsif20110433 (2011). [10] T. Lammermann, B. L. Bader, S. J. Monkley, T. Worbs, R. Wedlich-Soldner, K. Hirsch, M. Keller, R. Forster, D. R. Critchley, R. Fassler, et al., Nature 453, 51 (2008). [11] M. Bergert, E. Anna, D. R. A., I. M. Aspalter, A. C. Oates, G. Charras, G. Salbreux, and E. K. Paluch, Nat. Cel. Biol. 17, 524 (2015). [12] E. K. Paluch, I. M. Aspalter, and M. Sixt, Annual Review of Cell and Developmental Biology 32, 469 (2016), pMID: 27501447. [13] J. Renkawitz, K. Schumann, M. Weber, T. Laemmer- mann, H. Pflicke, M. Piel, J. Polleux, J. P. Spatz, and M. Sixt, Nature Cell Biology 11, 1438 (2009). [14] R. J. Hawkins, M. Piel, G. Faure-Andre, A. M. Lennon- Dumenil, J. F. Joanny, J. Prost, and R. Voituriez, Phys. Rev. Lett. 102, 058103 (2009). [15] N. P. Barry and M. S. Bretscher, Proceedings of the Na- tional Academy of Sciences 107, 11376 (2010). [16] M. Arroyo, L. Heltai, D. Mill´an, and A. DeSimone, Pro- ceedings of the National Academy of Sciences 109, 17874 (2012). [17] L. Aoun, P. Negre, A. Farutin, N. Garcia-Seyda, M. S. Rivzi, R. Galland, A. Michelot, X. Luo, M. Biarnes- Pelicot, C. Hivroz, et al., bioRxiv (2019). [18] E. Lauga and T. R. Powers, Reports on Progress in Physics 72, 096601 (2009). [19] H. A. Stone and A. D. T. Samuel, Physical review letters 5 77, 4102 (1996). [20] C. Brennen and H. Winet, Annual Review of Fluid Me- chanics 9, 339 (1977). [21] O. T. Fackler and R. Grosse, The Journal of cell biology 181, 879 (2008). [22] A. Farutin, S. Rafaı, D. K. Dysthe, A. Duperray, P. Peyla, and C. Misbah, Phys. Rev. Lett. 111, 228102 (2013). [23] E. J. Campbell and P. Bagchi, Physics of Fluids 29, 101902 (2017). [24] A. B. Verkhovsky, T. M. Svitkina, and G. G. Borisy, Current Biology 9, 11 (1999). [25] P. Recho, T. Putelat, and L. Truskinovsky, Physical re- view letters 111, 108102 (2013). [26] E. Barnhart, K.-C. Lee, G. M. Allen, J. A. Theriot, and A. Mogilner, Proceedings of the National Academy of Sciences p. 201417257 (2015). [27] A. C. Callan-Jones, V. Ruprecht, S. Wieser, C. P. Heisen- berg, and R. Voituriez, Phys. Rev. Lett. 116, 028102 (2016). [28] I. Lavi, M. Piel, A.-M. Lennon-Dum´enil, R. Voituriez, and N. S. Gov, Nature Physics 12, 1146 (2016). [29] A. Godeau, Ph.D. thesis, Strasbourg (2016). [30] K. Kruse, J.-F. Joanny, F. Julicher, J. Prost, and K. Seki- moto, Phys. Rev. Lett. 92, 078101 (2004). [31] A. Farutin, O. Aouane, and C. Misbah, Phys. Rev. E 85, 061922 (2012). [32] C. Pozrikidis, Boundary Integral and Singularity Meth- ods for Linearized Viscous Flow (Cambridge University Press, Cambridge, UK, 1992). [33] AF, JE, CM, and PR, Supplementary Information (2019). [34] A. E. Carlsson, New Journal of Physics 13 (2011), ISSN 1367-2630. [35] P. Recho and L. Truskinovsky, Mathematics and Mechan- ics of Solids (2015). [36] V. Ruprecht, S. Wieser, A. Callan-Jones, M. Smutny, H. Morita, K. Sako, V. Barone, M. Ritsch-Marte, M. Sixt, R. Voituriez, et al., CELL 160 (2015). [37] P.-H. Wu, D. R.-B. Aroush, A. Asnacios, W.-C. Chen, M. E. Dokukin, B. L. Doss, P. Durand-Smet, A. Ekpeny- ong, J. Guck, N. V. Guz, et al., Nat Methods 15, 491 (2018). [38] A. Jilkine and L. Edelstein-Keshet, PLoS Comput Biol 7, e1001121 (2011). [39] M. C. Jones, P. T. Caswell, and J. C. Norman, Current opinion in cell biology 18, 549 (2006). [40] L. Giomi and A. DeSimone, Physical review letters 112, 147802 (2014). [41] A. G. Clark, K. Dierkes, and E. K. Paluch, Biophysical journal 105, 570 (2013). [42] H. Turlier, B. Audoly, J. Prost, and J.-F. Joanny, Bio- physical journal 106, 114 (2014). [43] M. Bergert, A. Erzberger, R. A. Desai, I. M. Aspalter, A. C. Oates, G. Charras, G. Salbreux, and E. K. Paluch, Nature cell biology 17, 524 (2015). [44] R. J. Hawkins, R. Poincloux, O. B´enichou, M. Piel, P. Chavrier, and R. Voituriez, Biophysical journal 101, 1041 (2011). [45] R. Uehara, G. Goshima, I. Mabuchi, R. D. Vale, J. A. Spudich, and E. R. Griffis, Current biology 20, 1080 (2010). [46] M. Fritzsche, A. Lewalle, T. Duke, K. Kruse, and G. Charras, Molecular biology of the cell 24, 757 (2013). SUPPLEMENTARY INFORMATION 6 The supplementary information contains the explicit expressions for the Green's kernels of the Stokes equations, the calculation of the cortex flow in terms of the concentration fields, the description of the numerical procedure, the technique of shape reconstruction in the quasi-spherical limit and the physical values used to estimate non-dimensional parameters. The following kernels are to be used in eq. (2) of the main text: GREEN'S KERNELS Gij(x, x′) = Kijk(x, x′) = 1 8π (cid:20) 3 4π δij (x − x′)i(x − x′)j + x − x′) x − x′3 (x − x′)i(x − x′)j(x − x′)k (cid:21) , x − x′5 . (10) The results in this section are presented in the most general form, without any assumptions about the relative values of ηin, ηout, ηs, and ηb. FULL SOLUTION All fields on the cell surface are expanded in spherical harmonics of the vector pointing from the center of the cell to a given point of its surface. The spherical harmonics are defined as Spherical Harmonics Yl,m(x) =s 2l + 1 4π (l − m)! (l + m)! P m l (cid:16) x3 x (cid:17)(cid:18) x1 + ix2 x1 + ix2(cid:19)m where P m l are associated Legendre polynomials. The following expansions are used ∞ ca l (x) ca l,mYl,m(x) ca(x) = ca l (x) = cµ(x) = cµ l (x) = f n(x) = f n l (x) = l Xl=0 Xm=−l ∞ l Xl=0 Xm=−l ∞ l Xl=2 Xm=−l cµ l (x) cµ l,mYl,m(x) f n l (x) f n l,mYl,m(x) , (11) (12) (13) (14) U (x) = Ul(x) = Ul(x) ∞ Xl=1 Xm=−l l Ul,mYl,m(x) 7 (15) Note that f n 1 = 0 is zero, as required by the condition of the total force acting on the cell being equal to zero, f n 0 is irrelevant because it just shifts the osmotic pressure drop across the membrane, and U0 is irrelevant because only gradients of U enter equations. We define the vector spherical harmonics as The following expansions are used Y 1,l,m(x) = [∇s − (l + 1)x]Yl,m(x) Y 2,l,m(x) = [∇s + lx]Yl,m(x) Y 3,l,m(x) = x × ∇sYl,m(x) uc = 3 ∞ l Xj=1 Xl=0 Xm=−l f = 3 ∞ l Xj=1 Xl=0 Xm=−l uc j,l,mY j,l,m(x) fj,l,mY j,l,m(x) Force calculation The force f can be represented as a sum of two contributions fj,l,m = f v j,l,m + f e j,l,m, where f v 1,l,m = − f v 2,l,m = − (l + 2)[2ηs(l2 + l + 1) + ηb(l + 1)(l + 2)] 2l + 1 (l + 1)(l + 2)(l − 1)(2ηs + ηb) 1,l,m uc R2 − l(l − 1)(l + 2)(2ηs + ηb) 2l + 1 1,l,m uc R2 − (l − 1)[2ηs(l2 + l + 1) + ηb(l − 1)l] 2l + 1 2l + 1 uc R2 , 3,l,m f v 3,l,m = −ηs(l + 2)(l − 1) f e 1,l,m = f e 2,l,m = l + 2 2l + 1 l − 1 2l + 1 f e 3,l,m = 0. [χcµ − αca]l,m R [χcµ − αca]l,m R + − f n l,m 2l + 1 f n l,m 2l + 1 , , (16) (17) (18) (19) (20) (21) 2,l,m uc R2 , uc R2 , 2,l,m The amplitudes f v amplitudes f e j,l,m contain the contribution of the surface viscosity terms in eq. (1) of the main text, while the j,l,m contain contributions of myosin contractility, cortex compressibility, and Lagrange multiplier f nn. Fluid dynamics The integrals in eq. (2) of the main text can be calculated analytically for a spherical cell. The results are ηin + ηout 2 j,l,m = Rgj,lfj,l,m + (ηout − ηin)kj,luc uc j,l,m, (22) where the coefficients gj,l and kj,l are listed in table I. 8 j gj,l kj,l 1 l 2 l+1 3 1 (2l+1)(2l+3) (2l−1)(2l+1) 3 2(2l+1)(2l+3) − 3 2(2l−1)(2l+1) − 2l+1 3 2(2l+1) TABLE I. Integrals of Green's kernels for a spherical cell. Explicit solution We note that the Green's kernels (10) are diagonal in the basis of vector spherical harmonics for spherical shape. Furthermore, we see that the amplitudes of the surface viscosity force f v 3,l,m. This implies that the vector spherical harmonics with first index 3 are completely decoupled from the two other types. Since f e 3,l,m = 0 for all l and m, we conclude that uc 3,l,m = 0 for all l and m as well. Adding the fixed shape condition (uc − vs) · n = 0 yields that uc − vs can be written as a surface gradient of some surface potential U , as used in the main text. Or, in spherical harmonics, 3,l,m depend only on uc uc 1,l,m = Ul,m l R 2l + 1 , uc 2,l,m = Ul,m R l + 1 2l + 1 , uc 3,l,m = 0 for l > 1. Solving the equations (21), (22), and (23) for U and f n yields Ul,m = R(χcµ − αca)l,m λl . λl =( 3ηinR + 2ηoutR + 2ηs + 2ηb, for l = 1, (2l + 1)(ηin + ηout)R + l(l + 1)ηb + 2(l2 + l − 1)ηs, for l > 1. vs = − 2 3 ∇U1. f n l,m = − [(l + 2)ηoutR + (l − 1)ηinR + 2(l + 2)(l − 1)ηs] Ul,m R2 for l > 1. (23) (24) (25) (26) (27) Using the expressions (24) and (26), the retrograde flow and the swimming velocity can be expressed as a function of the concentration fields. The time evolution equations of the concentration fields are obtained by substituting U into eqs. (3) and (4) of the main text. The following equation can be used to reduce all calculations to scalar spherical harmonics ∇s · [c(uc − vs)] = ∇s · (c∇sU ) = ∆s(cU ) + c∆sU − U ∆sc 2 . (28) NUMERICAL PROCEDURE The numerical procedure consists in representing ca, cµ and U by the amplitudes of the spherical harmonics for all values of l < lmax and m ≤ l, where lmax is the cut-off value. We take lmax = 64 for such calculations. We observed that regardless of the initial conditions, the dynamics relaxed to an axisymmetric solution. We therefore also performed simulations with the shape assumed axisymmetric from the beginning, which is achieved by setting all amplitudes for m 6= 0 to zero. With this assumption, lmax = 1024 was used, which proved to be necessary for strongly polarized cells. The eqs. (3) and (4) of the main text were solved by an explicit Euler scheme by truncating the harmonic expansion of the advection terms to l < lmax. The time step was chosen small enough to avoid the instability due to the stiffness of the diffusion equation (typically 10−4 in non-dimensional units). In some cases, a small diffusion of actin (diffusion coefficient 10−3 in non-dimensional units) was added to enhance the stability of the actin advection equation. The steady-state branches in Figs. 2 and 3 of the main text were obtained by solving eqs. (3) and (4) of the main text with ca = cµ = 0 using Newton's method. MODEL FOR THE CELL SHAPE 9 The calculation of the shape follows the method used in Ref. [28] of the main text. Spherical shape of a cell in suspension can be physically achieved by a combination of high osmotic pressure ∆P inside the cell and the inextensibility of the membrane. Further in this section, we allow the shape of the cell to deviate from a sphere, albeit weakly, taking the leading terms in the small-deformation expansion. We parametrize the cell shape by a shape function ρ x = R0 [1 + ρ(x)] , (29) where x is the distance from the center of the cell to a given point on its boundary. ρ0 = 0 to the leading order in deformation because of the conservation of the membrane area. ρ1 = 0 because this term corresponds to a translation of the cell to the leading order. We show below that ρ scales as ∆P −1, which justifies an expansion in powers of ρ. We consider the quasi-spherical limit, taking the leading terms in such expansions. The function ρ(x) is expanded in spherical harmonics of x to be used below ∞ ρ(x) = ρl(x) Xl=2 Xm=−l l ρl(x) = ρl,mYl,m(x). (30) Assuming the tension of the membrane ζ0 to be homogeneous (unaffected by the cortex flow) we can write for the tension force f m f m = −Hζ0n = −ζ0H0n − (H − H0)nζ0, (31) where H is the mean curvature of the membrane (sum of the principal curvatures) and H0 = 2/R is the value of H for a perfectly spherical cell. The term −ζ0H0n in eq. (31) corresponds to an isotropic compression of the fluid inside the cell, which is balanced by the osmotic pressure. This relates the tension of the membrane ζ0 to the pressure difference by the Laplace law: ζ0 = R∆P 2 . (32) The term (H − H0)nζ0 in eq. (31) corresponds to a position-dependent normal force, which we identify with the Lagrange multiplier f nn used to maintain the shape of the cell. This justifies that H −H0 scales as ζ −1 0 or, equivalently, as ∆P −1 for fixed f n. Since f n is governed by the actomyosin dynamics, as follows from eq. (27), we obtain that the shape of the cell can be indeed made as close to a sphere as necessary by choosing ∆P large enough. This shows that all calculations made for perfectly spherical cells remain valid to the leading order in the limit of large ∆P even if the spherical-shape condition is relaxed. The mean curvature can be related to the shape function by This yields the final relation between the shape harmonics ρl,m and the Lagrange multiplier f n: H(x) = 2 R + 1 R ∞ Xl=2 (l − 1)(l + 2)ρl(x). ρl,m = − 2f n l,m (l − 1)(l + 2)∆P . PHYSICAL PARAMETERS (33) (34) We list in table. II the physical data we have considered to obtain rough estimates of the three non-dimensional parameters entering in the model. 10 name cortical thickness cortical viscosity myosin contractility F-actin compressibility symbol h ηs χcµ αca 0 0 myosin diffusion coefficient Dµ typical value 10−7 m [41, 42] h × (103 h × (102 h × 103 Pa.m [44] 10−13 10−5 m 10−2 10−5 m 102 − 103 s − 106) Pa m s [42, 43] − 103) Pa m [42, 43] − 10−12 m2s−1 [44, 45] − 10−1 s−1 [42, 44, 46] cell size F-actin turnover characteristic length R β l0 = R t0 = R2/Dµ characteristic time characteristic surface stress σ0 = Dµηs/R2 1 − 104 Pa m contractility parameter compressibility parameter turnover parameter 0 /σ0 0 /σ0 ¯χ = χcµ ¯α = αca ¯β = βt0 − 103 − 103 10−2 10−1 1 − 102 TABLE II. Estimates of material coefficients and non dimensional parameters definitions.
1905.07675	1	1905	2019-05-19T02:36:35	Physical Models of Collective Cell Migration	[ "physics.bio-ph", "q-bio.TO" ]	Collective cell migration is a key driver of embryonic development, wound healing, and some types of cancer invasion. Here we provide a physical perspective of the mechanisms underlying collective cell migration. We begin with a catalogue of the cell-cell and cell-substrate interactions that govern cell migration, which we classify into positional and orientational interactions. We then review the physical models that have been developed to explain how these interactions give rise to collective cellular movement. These models span the sub-cellular to the supracellular scales, and they include lattice models, phase fields models, active network models, particle models, and continuum models. For each type of model, we discuss its formulation, its limitations, and the main emergent phenomena that it has successfully explained. These phenomena include flocking and fluid-solid transitions, as well as wetting, fingering, and mechanical waves in spreading epithelial monolayers. We close by outlining remaining challenges and future directions in the physics of collective cell migration.	physics.bio-ph	physics	Physical Models of Collective Cell Migration Ricard Alert∗1,2 and Xavier Trepat†3,4,5,6 1Princeton Center for Theoretical Science, Princeton University, Princeton, USA, NJ 08544 2Lewis-Sigler Institute for Integrative Genomics, Princeton University, Princeton, USA, NJ 3Institute for Bioengineering of Catalonia, The Barcelona Institute for Science and Technology 08544 4Facultat de Medicina, University of Barcelona, Barcelona, Spain, 08036 (BIST), Barcelona, Spain, 08028 5Institució Catalana de Recerca i Estudis Avançats (ICREA), Barcelona, Spain, 08028 6Centro de Investigación Biomédica en Red en Bioingeniería, Biomateriales y Nanomedicina, Barcelona, Spain, 08028 May 21, 2019 Abstract Collective cell migration is a key driver of embryonic development, wound healing, and some types of cancer invasion. Here we provide a physical perspective of the mechanisms underlying collective cell migration. We begin with a catalogue of the cell-cell and cell-substrate interactions that govern cell migration, which we classify into positional and orientational interactions. We then review the physi- cal models that have been developed to explain how these interactions give rise to collective cellular movement. These models span the sub-cellular to the supracel- lular scales, and they include lattice models, phase fields models, active network models, particle models, and continuum models. For each type of model, we dis- cuss its formulation, its limitations, and the main emergent phenomena that it has successfully explained. These phenomena include flocking and fluid-solid transi- tions, as well as wetting, fingering, and mechanical waves in spreading epithelial monolayers. We close by outlining remaining challenges and future directions in the physics of collective cell migration. ∗[email protected] †[email protected] 1 Contents Contents 1 Introduction 2 Forces and interactions of migrating cells 2.1 Positional cell-substrate interactions . . . . . . . . . . . . . . . . . . 2.2 Positional cell-cell interactions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3 Orientational cell-cell interactions 2.4 Orientational cell-substrate interactions . . . . . . . . . . . . . . . . 3 Physical models, from sub-cellular to supracellular scales 3.1 Lattice models: The cellular Potts model . . . . . . . . . . . . . . . . 3.2 Phase-field models . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3 Active network models . . . . . . . . . . . . . . . . . . . . . . . . . 3.4 Particle models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.5 Continuum models . 4 Conclusions and outlook References 1 Introduction 2 2 4 4 5 7 8 8 9 10 13 15 18 23 24 It has long been recognized that cells move as collectives during development, regen- eration and wound healing. Reports from the late XIXth century already agreed that these processes involve collective movements of cells but mechanisms remained con- troversial [1 -- 4]. Some authors proposed that cell movements were driven by pressure, either pre-existing in the tissue or generated de-novo by cell division (see Ref. [3] and references therein). Others claimed that cells would move by spreading their volume to occupy the largest possible surface [1]. Yet, others defended that cell sheets advanced by the active pulling force generated by leader cells at the tissue margin [2]. In those early days of cell biology, proposed mechanisms were largely physical in origin but not even the sign of the tissue stress, i.e. tension vs compression, was agreed upon. Later on, the discovery of genes and proteins shifted the attention from the whole to the parts, and the search for a global physical understanding of collective migration was largely abandoned. This trend has been reversed in the past decade due to groundbreaking technical [5] and conceptual [6 -- 8] advances together with a progressive questioning of reductionist approaches [9]. Time-lapse imaging and fluorescence microscopy have become stan- dard tools in life-science laboratories, and technologies such as particle imaging ve- locimetry (borrowed from fluid mechanics) now enable a detailed mapping of velocity fields and strain tensors in the tissue [10, 11]. Moreover, a range of new technologies such as traction microscopy have enabled the direct mapping of the forces that cells exert on their surroundings as they migrate [12, 13]. All mechanical variables rele- vant to the problem of collective cell migration have thus become available in time and space (Fig. 1). This technological revolution has coincided with the development of the theory of active matter [6 -- 8], which provides an ideal framework to rationalize the 2 Figure 1 Mechanical measurements during collective migration of cell monolayers. a, Vectorial representation of traction forces in LifeAct-GFP MDCK cells closing a wound. Color coding indicates the value of the radial component, with positive forces pointing away from the wound. For clarity, values between 100 and -100 Pa were not plotted. Panels labeled as i and ii show a close-up of the regions indicated by arrows in panel a. Scale bar, 20 µm. Adapted from [15]. b, Velocity vectors (green) and monolayer stress ellipses (red) indicating the maximum and minimum principal stresses in an expanding colony of MDCK cells (phase contrast). Adapted from [11]. collective movement of cells. The traditional view that physics should serve to illu- minate biological function is shifting towards the idea that biological systems inspire new physical theories and allow us to test them; the concept of 'physics for biology' is now paralleled by the emergent notion of 'biology for physics'. In this sense, from the perspective of condensed matter physics, collective cell migration is interesting as a prominent example of the emergence of collective mechanical phenomena in a system of soft active entities with complex interactions. Finally, life scientists have recognized that collective cell migration is not only key to development, regeneration and wound healing, but also to devastating diseases such cancer [14]. The coincidence in time of these different technological and conceptual advances has placed collective cell mi- gration back to the center stage of research at the interface between life and physical sciences. Collective cell migration comes in different flavors depending on the biological tissue and process [16]. During epithelial morphogenesis, wound healing and regener- ation, cells generally move as sheets adhered on an inert hydrogel called the extracel- 3 aiiiiiiT┴(Pa)blular matrix (ECM). In some forms of cancer invasion, cells also invade as sheets at the interface between tissues. In general, however, both in development and in tumor in- vasion, cells invade as strands or clusters within a complex 3D environment composed of ECM and different cell types [16 -- 18]. Despite recent advances [19], we remain far from accessing physical forces in 3D, so we will focus this review on cell sheets migrating in 2D (Fig. 1). In this mode of migration, central to epithelial function, all relevant cellular forces have been accurately mapped in vitro and can therefore be used to test physical models. For other perspectives on this subject, the reader is referred to excellent recent reviews [20 -- 22]. 2 Forces and interactions of migrating cells In this section, we propose a classification of the forces and interactions that a migrat- ing cell exerts on and experiences from other cells and the substrate. In addition to distinguishing cell-cell and cell-substrate interactions, we separate the interactions that directly act on a cell's position from those that affect cell orientation. Even though any categorization may suffer from some degree of oversimplification, we think that it nev- ertheless provides some unifying principles over a large body of somehow fragmented literature. 2.1 Positional cell-substrate interactions The field of continuum mechanics defines a traction as a force per unit area applied at any point of the surface of a body. In cell mechanics, traction is usually understood as the stress applied by a cell on its underlying inert substrate, typically a polymeric gel known as extracellular matrix (ECM). Cell-substrate traction can be interpreted as the sum of two contributions: the force that drives cell motion, which we call active traction, and the passive friction between the cell and its substrate. 2.1.1 Active traction forces Active traction forces stem from the cell's actomyosin cytoskeleton, where the action of myosin molecular motors on actin filaments generates contractile forces. These forces are then transmitted to the substrate through cell-substrate adhesion complexes called focal adhesions, which physically connect the cytoskeleton to the extracellular matrix (Fig. 2). For traction forces to lead to cell motion, the cell must break symmetry and polarize to define a front and a rear. To do so, the cell often develops frontal actin-based protrusions such as lamellipodia and filopodia. These structures generate an inwards- pointing active traction that is most prominent at the cell's leading edge (Fig. 2). The resulting force propels the cell forward in the direction of its polarity, defined by the position of the protrusion with respect to the cell's center of mass.1 Thus, in models, 1The front-rear polarity of a cell is a morphological, dynamical, and biochemical asymmetry between the cell's leading and trailing edges. The polarity direction of an isolated cell can be unambiguously identified from the direction of its migration, regardless of its sub-cellular origin. However, this may no longer be true in a tissue, where cell motion is also affected by intercellular forces. In this case, to distinguish polarity from velocity, one must identify a sub-cellular observable that defines cell polarity. Morphological features such as protrusions are often not apparent in cells inside tissues, which may feature cryptic lamellipodia that extend beneath neighboring cells. Thus, an alternative approach is to identify cell polarity from traction forces. This is valid as long as tractions are dominated by active forces, with negligible contributions from passive friction forces. Identifying sub-cellular features that appropriately account for cell polarity remains an open challenge. 4 Figure 2 Forces and interactions of migrating cells. Schematic representation of a migrating epithelial monolayer. Side view (left), top view (right). We illustrate a subset of the interactions discussed in Section 2. In particular, we sketch some of the biological structures that generate and transmit cell-substrate and cell-cell forces. We also indicate contact interactions that regulate and coordinate cell migration, as well as physical variables used to describe collective cell migration. the force that drives cell migration is usually assumed proportional to a cell polarity variable, with a coefficient that depends both on cell-substrate adhesion and on the active force-generating processes in the cytoskeleton. 2.1.2 Cell-substrate friction forces Cell motion takes place at very low Reynolds numbers, which implies that inertial forces are negligible and, hence, that the resultant force on the cell's center of mass must vanish. Indeed, the active traction applied by the cell on the substrate is balanced by friction forces. Friction with the surrounding fluid medium is usually negligible in front of cell-substrate friction forces, which are mediated by the attachment and de- tachment of proteins at focal adhesions (Fig. 2). On time-scales larger than the inverse binding rates, this protein-mediated friction is expected to be proportional to the veloc- ity of the cell relative to the substrate [23]. Thus, in a first approximation, cell-substrate friction is often modeled as a viscous damping force akin to Stokes' drag, with a co- efficient that reflects cell-substrate adhesion. However, the kinetics of focal adhesion proteins under force are extremely non-linear and involve reinforcement feedbacks that can be accounted for in more detailed models of cellular friction [24]. 2.2 Positional cell-cell interactions 2.2.1 Cell-cell adhesion A characteristic feature of epithelial cells is that they establish stable cell-cell adhe- sions, whereas mesenchymal cells tend to form transient and weaker adhesions. Cell- cell adhesion is mediated by specific transmembrane protein complexes, which build cell-cell junctions that physically link the actomyosin cortices of the adhering cells, thereby enabling force transmission between cells (Fig. 2). Cell-cell junctions endow 5 Gap junctionsAdherenscell-cell junctionsActomyosin cortexFocal adhesionsCell-cell tensionCRLActomyosin stress fibersCell-cell frictionContact inhibition of locomotionContact following of locomotionCell velocityCell polarityBiological structuresForcesContact regulation of locomotion (CRL)VariablesActive tractionCell-substrate frictiontissues with cohesion energy and surface tension, as well as with a bulk modulus that quantifies their resistance to rapid isotropic expansions, which is mainly due to cy- toskeletal elasticity [22,25]. Thus, different modeling frameworks account for cell-cell adhesion by either an interfacial energy contribution, a short-range attraction that op- poses cell-cell detachment, or a tissue bulk modulus. In addition, by enabling the transmission of active cytoskeletal tension, cell-cell adhesion is also an implicit factor in tissue-scale active stress terms. The influence of cell-cell adhesion on tissue mechanical properties does not end here. On the one hand, decreasing cell-cell adhesion leads to less elongated cell shapes that can induce a jamming transition whereby the tissue acquires a finite shear modulus, thus becoming a solid material [26, 27]. Therefore, cell-cell adhesion may not only provide bulk but also shear elasticity to epithelial tissues. On the other hand, cell-cell adhesion proteins turn over and, hence, cell-cell junctions are remodeled. Junction remodeling is a source of dissipation that can relax stress, possibly contributing to the long-time viscous response of fluid tissues [22, 25, 28, 29]. 2.2.2 Cell-cell friction Cell-cell adhesion also entails cell-cell friction forces when cells slide past each other (Fig. 2). Similar to cell-substrate friction, cell-cell friction is based on the sliding, turnover, and attachment kinetics of cell-cell junction proteins [30 -- 32]. Usually, cell- cell friction is modeled as a shear force proportional to the relative velocity between the cells or, in tissue-level descriptions, as shear viscous stresses. 2.2.3 Cell-cell repulsion In addition to the attractive interactions mediated by cell-cell adhesion, adhered cells also experience a soft repulsion from other cells. At short times, cell compression is re- sisted by cytoskeletal elasticity [22,25], which, in epithelial monolayers, gives rise to an area compressibility. At longer times, however, both the cytoskeleton and cell-substrate adhesions may reorganize to enable cell shape changes. If their volume is conserved, cells under compression can lose area and gain height, at least until their nuclei become tightly packed. However, cells can actually change their volume by exchanging fluid both with the surrounding medium through water channels and with other cells through cell-cell channels called gap junctions [33, 34] (Fig. 2). Finally, under sufficient com- pression, epithelial cells can be extruded from a monolayer [35 -- 37], thus enabling monolayer area reduction. The opposite process, cell insertion in a monolayer, also occurs in spreading cell aggregates [38]. Altogether, this means that epithelial mono- layers can change their area via dissipative processes, with an associated viscosity. 2.2.4 Active cell-cell forces A last type of cell-cell forces are active forces generated by myosin molecular motors in the cytoskeleton and transmitted through cell-cell junctions (Fig. 2). Cytoskeletal structures such as the cell cortex and the apical actin belt generate a roughly isotropic tension at the cell scale, thus giving rise to isotropic active stress at the tissue level. However, migrating cells are polarized and, hence, their cytoskeleton exhibits highly anisotropic structures such as stress fibers (Fig. 2). Stress fibers generate anisotropic tension that, in addition to being transmitted to the substrate as traction forces (Sec- tion 2.1.1), can also be transmitted to cell-cell junctions through the cell cortex, thus 6 giving rise to anisotropic active stresses at the tissue scale. Given that they are simi- larly generated, traction forces and anisotropic cell-cell active stresses are interdepen- dent [39]. Yet, they are distinct since their respective transmission to the ECM and neighboring cells relies on different adhesion complexes (Fig. 2). 2.3 Orientational cell-cell interactions 2.3.1 Polarity alignment One of the most prominent orientational interactions between cells is the tendency to align their polarities. Alignment might simply rely on the elongated and deformable shape of migrating cells, but it may also involve biochemical regulation of cell migra- tion. Polarity alignment is often explicitly implemented either via Vicsek-like rules or via torques on cell polarity in discrete models, and via an orientational stiffness of the polarity field in continuum models. 2.3.2 Contact regulation of locomotion Here, we propose the term contact regulation of locomotion (CRL) to subsume a num- ber of processes whereby cells tune their migration direction upon contact interactions with other cells (Fig. 2). At least three such processes have been described. First, con- tact inhibition of locomotion (CIL) refers to the process whereby, upon head-to-head collision, many cell types retract their lamellipodia and repolarize in a different direc- tion, thus migrating away from cell-cell contacts [40, 41]. Second, contact following of locomotion (CFL) refers to the tendency of cells to follow others upon head-to-tail contact [42, 43]. Finally, in addition to altering the migration direction, head-to-tail collisions have also been found to increase the persistence of cell motion -- a ten- dency known as contact enhancement of locomotion (CEL) [44]. In general, CRL depends strongly on the cell-cell collision angle [45, 46], thus making orientational cell-cell interactions highly anisotropic. The mechanisms underlying CRL may be di- verse and, given that cells polarize in response to tension transmitted through cell-cell junctions [47 -- 53], they could rely on mechanotransduction of cell-cell forces [54 -- 57]. Although a number of phenomenological models of CRL have been proposed, a coher- ent theoretical picture of contact regulation of locomotion is lacking. 2.3.3 Polarity-flow alignment Inhomogeneous tissue flows may produce shear. Similar to molecules in liquid crystals, elongated cells subject to shear should experience a torque that tends to minimize shear stress. Indeed, shear tissue flows reorient cell polarity in the fly wing [58] as well as the cell division axis in epithelial monolayers [59]. Moreover, cells in epithelial mono- layers tend to migrate in the local direction of lowest shear stress -- a behavior known as plithotaxis [60 -- 63]. However, unlike in ordinary liquid crystals, cell reorientation may not entirely stem from cell shape but it likely involves an active mechanosensitive response. Regardless of its yet-unclear mechanism, this feedback between polarity and flow mediates another type of orientational cell-cell interactions. Even though some recent continuum models have probed the effects of flow-polarity coupling [64, 65], further research is needed to clarify its role in collective cell migration. 7 2.3.4 Polarity-shape alignment Almost by definition, cell polarity and cell shape asymmetry are interdependent and, hence, they are often assumed to align. Then, given that cell-cell interactions mod- ify cell shape, cell-autonomous polarity-shape alignment can give rise to intercellular alignment interactions [66 -- 68]. 2.4 Orientational cell-substrate interactions 2.4.1 Polarity-velocity alignment Through their interaction with the substrate, cells may be able to align their polarity to their velocity, thus tending to align self-propulsion with drag cell-substrate forces [69]. Such a polarity-velocity coupling is a generic property of active polar systems interacting with a substrate [70 -- 73]. For these systems, the polarity not only reorients in flow gradients but also in uniform flows, like a weathercock in the wind. Even though its cellular mechanism is not yet well understood, polarity-velocity alignment has been introduced in a number of models. However, in some situations, polarity and velocity are strongly misaligned in epithelial monolayers [74, 75] (Fig. 2), possibly due to the dominance of cell-cell interactions [75,76]. How such possibly conflicting polarization cues coexist and cooperate remains poorly understood [77]. 2.4.2 Substrate-induced polarization In addition to polarizing in response to cell-cell forces (Section 2.3.2), cells can also polarize in response to asymmetric forces at the cell-substrate interface [52,78]. In par- ticular, given that cells exert larger tractions on more adhesive and/or stiffer substrates, gradients of substrate adhesivity and/or stiffness can polarize cells [79 -- 81]. The en- suing migrations towards regions of higher adhesivity and/or stiffness are known as haptotaxis and durotaxis, respectively. Moreover, even changes in uniform substrate properties may lead to cell polarization. Specifically, increasing substrate stiffness trig- gers an isotropic-nematic transition in the actomyosin cytoskeleton [52, 82 -- 84]. This transition results in cell elongation, which, in turn, might promote a spontaneous po- larization [85]. 3 Physical models, from sub-cellular to supracellular scales In this section, we review the different physical descriptions that have been used to model collective cell migration. These descriptions cover different levels of coarse- graining; we start from those describing sub-cellular detail and move up to continuum models that only describe supracellular features. Complementary presentations have been provided in recent reviews [20, 86, 87]. Here, we emphasize how the cellular forces and interactions reviewed in the previous section can be accounted for by each of the modeling approaches. We focus on two-dimensional models that explicitly include cell migration. 8 3.1 Lattice models: The cellular Potts model In the spirit of classical models of statistical mechanics, such as the paradigmatic Ising model, lattice models describe individual cells as domains on a lattice, thus resolving sub-cellular details of cell shape (Fig. 3a). In particular, this description is based on the Potts model, and hence it is known as the Cellular Potts Model (CPM) [88]. Each lattice site i = 1, . . . , N is assigned a state variable σi = 1, . . . , m corresponding to one of m − 1 cells. The state of each lattice site is then updated using a state-exchange Monte Carlo scheme with Metropolis dynamics at a sufficiently low temperature to ensure that cells remain as compact domains. 3.1.1 Effective Hamiltonian The dynamics minimizes the effective Hamiltonian H =(cid:88)(cid:104)i,j(cid:105) J(σi, σj) + λ m−1(cid:88)σ=1 (Aσ − A0)2 − P m−1(cid:88)σ=1 (cid:126)Rσ · (cid:126)pσ. (1) The original CPM only included the first two terms. The first term, whose sum runs over neighboring sites (cid:104)i, j(cid:105), accounts for the interfacial tension between neighboring cells as well as between cells and the medium (state σ = m), which are encoded in the interaction matrix J. The simplest choice is J(σi, σj) = α(1 − δσi,σj ), where α is the interfacial energy that controls the amplitude of cell shape fluctuations (Fig. 3). Thus, this energy captures the combined effects of cell-cell adhesion and cortical tension (Sections 2.2.1 and 2.2.4). The second term penalizes changes in cell area around a preferred value A0, with an area modulus λ > 0 (Fig. 3a and Section 2.2.3). The area of cell σ, i.e. its number of lattice sites, is simply given by Aσ =(cid:80)N i=1 δσi,σ. 3.1.2 Cell migration Later, the third term was added to implement cell motility by decreasing the energy of those configurations in which a cell's center of mass (cid:126)Rσ = A−1 i=1(xi, yi) δσi,σ has advanced toward the direction of its polarity (cid:126)pσ [90]. This term corresponds to an active polar force of magnitude P > 0 on each cell: (cid:126)Fσ = −(cid:126)∇ (cid:126)RσH = P (cid:126)pσ (Fig. 3a and Section 2.1.1). The model does not specifically include friction forces. Rather, an effective damping of cell motion arises from the Metropolis dynamics itself, which is dissipative in nature. In fact, the mean cell speed is linear in P/α over a wide range of parameter values [91]. Thus, the cell-cell coupling strength α, which controls the diffusion coefficient of a cell in the absence of motility, is proportional to the effective viscous friction coefficient, consistent with the Stokes-Einstein relation. σ (cid:80)N 3.1.3 Polarity dynamics The cellular polarity (cid:126)pσ was proposed to align with the velocity over some time-scale [91, 92] (Section 2.4.1) or, alternatively, to simply undergo rotational diffusion [89]. In a variant of the CPM, cell motion was dictated by the gradient of a self-secreted chemoattractant, whose concentration evolves with its own dynamics [93]. CPMs with alternative polarity dynamics should be explored in the future. Along these lines, Coburn et al. have recently proposed a hybrid CPM that accounts for CIL [94] (Sec- tion 2.3.2). 9 Figure 3 Lattice Models: The Cellular Potts Model. a, Lattice sites corresponding to two different cells are shown in different colors. Cell-cell and cell-medium interfaces have an interfacial energy α (black). Cell migration in the direction of the polarity is favored by a self-propulsion magnitude P . b, Snapshots of the system and cell trajectories at the fluid (α = 1) and solid (α = 4) regimes. Neighboring cells are colored differently, with arbitrary colors. Cell boundaries are rougher and longer for smaller interfacial energy. Adapted from [89] with permission from EPL. 3.1.4 Collective phenomena Initially, the self-propelled CPM was primarily used to study velocity correlations of complex flows in cell monolayers [30, 91, 92]. More recently, it has also been used to study fluid-solid transitions and glassy dynamics in cell monolayers [89, 91] (Fig. 3b), collective rotations [95], gap closure [94], and tissue spreading [96], including the fingering instability of the tissue front [93, 96]. 3.1.5 Discussion The CPM is based on an explicit and detailed description of cell shape and cell-cell adhesion which, by means of intensive simulations, enables close investigation of cell- scale mechanisms of cell rearrangements. However, the Metropolis dynamics yields somewhat artificial cell shape fluctuations that depend on a temperature parameter not directly related to experimental measurements. Moreover, the model is not readily suited to incorporate some kinds of cellular interactions relevant for collective cell migration. In particular, how to distinguish cell-cell and cell-substrate friction and how to appropriately capture the active nature of some cellular forces with the relaxational algorithm of the CPM remains unclear. 3.2 Phase-field models With their origins in interface dynamics [97], phase-field models also describe cell shape in sub-cellular detail. However, unlike the CPM, they do not rely on a lattice. Rather, each cell i = 1, . . . , N is described by a phase field φi((cid:126)r, t), which is 1 inside 10 Interfacial energySelf-propulsion magnitudeM.ChiangandD.MarenduzzoFig.1:(Colouronline)(A)AsnapshotofthecellsnearthecentreofthelatticeinthecellularPottsmodelsimulations,for(i)α=1and(ii)α=4.Differentneighbouringcellsarecoloureddifferently.Notethattheboundariesofthecellsarerougherandlongerforthesmallerαcase.(B)Trajectoriesofthecentreofmassofasampleofcellsnearthecentreofthelattice,againfor(i)α=1and(ii)α=4.simulatingthemotionofcloselypackedcellsinamono-layer.IntheCPM,acellmonolayerisrepresentedonatwo-dimensionallattice.WeconsiderasquarelatticeofsizeL2withL=200andwithperiodicboundarycondi-tions.EachlatticesiteisassignedanintegerPottsspin0≤σ≤N(the"cellindex"),whereNisthetotalnum-berofcells,whichwesetto1000.Twolatticesiteswhichshareasideoracornerareconsideredasneighbours.Abi-ologicalcellisdenotedbythesetofsiteswiththesamespin(seefig.1(A)forsnapshotsofthesimulations).Notethattheparameterchoiceensuresthatcellsareconnectedsetsoflatticesites.Thespinσ=0isusuallyreservedforfluidsites:thesearenotpresentinourmodel,aswesimulateaconfluentmonolayer(withcellpackingfractionequaltounity).ThedynamicsofthecellsevolveviaakineticMonteCarloalgorithmbasedonaneffectiveHamiltonian(orenergyfunction)H,givenbyH=!⟨x,x′⟩J(σ(x),σ(x′))+λN!i=1(ai−Ai)2−PN!i=1ni·ri,(1)wherexandx′denotelatticesites,aiandAiarethecurrentandtargetareaofthei-thcell(whichwesetto40latticesites),whereasriandnirepresentitscentreofmassanddirectionofmotility,respectively.Thethreetermsineq.(1)respectivelyaccountforinterfacialef-fects,approximatecellareaconservation,andcellmotility.Furthermore,inthefirstterm,theparameterJcreatesaneffectiveinterfacialtensionbetweencellsinthebulk[16].InlinewithcommonpracticeinCPMsimulations[13 -- 15],theparameterJischosenasfollows:J(σ,σ′)="0,forσ=σ′,α,forσ=σ′,(2)whereα>0determinestheinterfacialenergybetweentwocells.Previousstudieshavefoundthatitsmagni-tudesaffectboththecellshapeandtheflexibilityofthecellboundary:smallmagnitudesproducelongandroughboundarieswithmoreactivecelldynamics,whilelarge-magnitudesresultinstraightboundarieswithlesscellularmotion[13,17].Inthesecondterm,λsetsthestrengthoftheapproximatecellconservationconstraintandgov-ernsthedegreeoffluctuationsofeachcell'sarea.Finally,inthethirdterm,Pisafreeparameterassociatedwiththestrengthofthemotility(takentobethesameforallcells).Wenotethattheenergyweusediffersfromthatusedinthevertexmodelin[11],wheretherewasatar-getperimeter(aswellasatargetarea)forcellsinthemonolayers.The(standard)CPMMonteCarlo(MC)dynamicsweconsideristhefollowing(see[13,14]formoredetailsontheCPM,and[18]foracomparisonbetweenvertexmod-elsandtheCPM).Eachelementarymoveisanattempttochangethespinofarandomlatticesiteatothatofaran-domlychosenneighboursiteb,whereσ(a)=σ(b).Theprobabilityofacceptingaspinchange,p(σ(a)→σ(b)),isdeterminedbytheMetropolistestaccordingtotheassociatedchangeinenergy(insimulationswefixboththeBoltzmannconstant,kB,andthetemperature,T,to1).TimeismeasuredinMCsweeps,whereaMCsweep(MCS)isaseriesofL2successiveattemptedMCelemen-tarymoves.Simulationswereinitialisedbyconstructingasquarelatticeofcellsandbylaterallowingthemtoequili-bratefor1000MCSbeforestartingtotakemeasurements.Tomodelmotility,wealsoneedtowritedownadynam-icalequationforthepolarityvector,ni,whichdeterminesthecelldirection.Thespecificmechanismsofhowacellgeneratesapolarizedmotionandhowthepolaritychangesovertimeremainpoorlyunderstoodandaresubjecttoin-tenseinvestigations[19,20].WechoosethesimplemodelsuggestedbyBietal.[11],whereweassumethatthepo-larityvectorundergoesrotationaldiffusionovertime.Informulas,ifonedefinesthepolarityangleofthei-thcell,θi,astheanglebetweenthecell'spolarityvector,ni,andareferenceaxis(takentobethex-axisherein),therota-tionaldiffusionprocessisgovernedbytheequation∂tθi(t)=ηi(t),(3)whereηiisaGaussiannoise,withzeromeanandavari-ance⟨ηi(t)ηj(t′)⟩=2Drδ(t−t′)δij(seefootnote1),withDrtherotationaldiffusioncoefficient.1Notethatδ(t−t′)andδijindicatetheDiracandKroneckerdelta,respectively.28009-p2FluidSolidInterfacial energySnapshotTrajectoriesCell 2Cell 1the cell and 0 outside (Fig. 4a). A similar approach relies on describing cell shape via a contour function [66]. Some models describe even intracellular structures, such as the nucleus, using additional phase fields [98]. 3.2.1 Phase-field free energy Cell-cell interactions are built into a free energy functional of the phase field. Although formulations vary [98 -- 100], a possible form is F = FCH + Farea + Fcell-cell with FCH = Farea = Fcell-cell = γ (cid:90)A(cid:104)4φ2 i (1 − φi)2 + 2(cid:126)∇φi2(cid:105) d2(cid:126)r, N(cid:88)i=1 µ(cid:18)1 − πR2(cid:90)A N(cid:88)i=1 (cid:90)A(cid:104)φ2 N(cid:88)i=1(cid:88)j(cid:54)=i i d2(cid:126)r(cid:19)2 j − τ 4(cid:126)∇φi2(cid:126)∇φj2(cid:105) d2(cid:126)r. i φ2 φ2 κ 1 , (2a) (2b) (2c) The first contribution is a Cahn-Hilliard free energy that stabilizes the phase-field inter- face. The first term is a double-well potential with minima at the cell interior (φi = 1) and exterior (φi = 0), which are connected by an interface of width and tension γ that delineates the cell boundary. Here, we have neglected the bending rigidity of the inter- face [98]. The second contribution penalizes departures of cell area from its preferred value πR2, with area modulus µ (Section 2.2.3). The third contribution accounts for cell-cell interactions. It includes a repulsive term that penalizes cell overlapping (Sec- tion 2.2.3), with strength κ, and an attractive interaction between cell boundaries that models cell-cell adhesion (Section 2.2.1), with strength κτ [98]. 3.2.2 Phase-field dynamics and force balance The dynamics of cell shape reads ∂tφi + (cid:126)vi · (cid:126)∇φi = − δF δφi . (3) Here, (cid:126)vi is a cell velocity defined as [99] 1 ξ(cid:90)A δF δφi 1 ξ(cid:90)A δFcell-cell (cid:126)vi = (cid:126)∇φi d2(cid:126)r = (cid:126)∇φi d2(cid:126)r = where ξ is a friction coefficient (Section 2.1.2), and (cid:126)F int is an interaction force on the i interface of cell i due to overlaps with neighboring cells. This relationship can be generalized to include cell motility in the form of an active polar contribution Ta(cid:126)pi to the force balance [69] (Section 2.1.1): δφi (4) (cid:126)F int i , 1 ξ ξ(cid:126)vi = (cid:126)F int i + Ta(cid:126)pi, (5) The interaction force F int i can also be generalized to account for additional interactions. In continuum mechanics, short-range interaction forces are described in terms of the stress tensor field σ((cid:126)r, t). In the phase-field formulation, this corresponds to [100] (cid:126)F int i =(cid:90)A φi (cid:126)∇ · σ d2(cid:126)r = −(cid:90)A σ · (cid:126)∇φi d2(cid:126)r. (6) 11 Figure 4 Phase-Field Models. a, Phase field of a cell. Adapted from [99]. b, The overlap between phase fields,(cid:80)i(cid:54)=j φiφj, identifies cell-cell interfaces (white). The inset shows cell contours, φi = 1/2, along with a sketch of extensile stress along the principal axis of the cell deformation tensor (see text). Adapted with permission from [100]. Copyright (2019) by the American Physical Society. c-d, Collisions between deformable cells lead to velocity alignment (c) and collective motion (d). Adapted from [67]. In addition to the the phase-field interactions, which give a pressure term, the stress tensor may also include other contributions such as viscous and active stresses. In this case, combining the approaches of Refs. [69, 100], the stress tensor could read σ((cid:126)r, t) = −P ((cid:126)r, t) I + ξc(cid:88)j(cid:54)=i ((cid:126)vi − (cid:126)vj) (cid:126)∇φj((cid:126)r, t) − ζQ((cid:126)r, t), (7) where the second term accounts for cell-cell friction with coefficient ξc (Section 2.2.2), and the third term describes anisotropic active stresses proportional to the nematic or- der parameter tensor field Q((cid:126)r, t) =(cid:80)N i=1 φi((cid:126)r, t) Si (Section 2.2.4). Here, Si is the orientation tensor of cell i, which may be based either on cells' polarities, Si = (cid:126)pi(cid:126)pi − 1/2(cid:126)pi2 I, or on cells' shapes as proposed in [100], Si = −(cid:82)A(cid:104)((cid:126)∇φi)T (cid:126)∇φi − (cid:126)∇φi2 I(cid:105) d2(cid:126)r (Fig. 4b). 3.2.3 Polarity dynamics Regarding the polarity dynamics, interactions such as CIL and CFL (Section 2.3.2), polarity alignment (Section 2.3.1), and polarity-velocity alignment (Section 2.4.1) have been explored [98], as well as couplings to chemotactic fields [101]. More recently, an alignment of cell polarity toward the direction of the total interfacial force has also been implemented [69] (Section 2.3.2). 12 singlecells(seeRef.[21]forareview)andmorerecentlytothedescriptionofafewmigratingcells[22]andthestudyofconfluentepithelia[23 -- 26].WestartbydescribinganepitheliumconsistingofNcellsusingadifferentphasefieldϕiforeachindividualcell.Valuesofϕi¼1andϕi¼0denotetheinteriorandtheexteriorofacell,respectively,andthecellboundaryisdefinedtolieatthemidpointϕi¼1=2.Wedefinethedynamicsofthefieldsϕias∂tϕiþvi∇ϕi¼−δFδϕi;i¼1;…;N;ð1ÞwhereFisafreeenergyandviisthetotalvelocityofcelli.ThisvelocityentersEq.(1)onlythroughtheadvectionterm,effectivelypushingeachcelluniformlyasawholewithoutinducinganydeformationofitsinterface.ThefreeenergyFdefinesthedynamicsoftheindividualinterfacesandiswrittenasF¼FCHþFareaþFrep,whereFCH¼XiγλZdxf4ϕ2ið1−ϕiÞ2þλ2ð∇ϕiÞ2g;Farea¼Xiμ1−1πR2Zdxϕ2i2;Frep¼XiXj≠iκλZdxϕ2iϕ2j:ð2ÞTheCahn-HilliardfreeenergyFCHstabilizesthecellinterface.Ourformulationisguidedbysimplicitybutcouldbereplacedbymorerealisticmodelsofthecellularboundary[27 -- 29].ThecontributionFareaprovidesasoftconstraintfortheareaoftheindividualcellsaroundthevalueπR2,whereRisthecellradius,suchthatthecellsarecompressible[30].Finally,therepulsiontermFreppenal-izesregionswheretwocellsoverlap.Normalizationhasbeenchosensuchthatthewidthoftheinterfacesatequilibriumisλandsuchthatthepropertiesofthecellsareroughlypreservedwhenλisrescaled(seeSupplementalMaterial[31]).Theparametersγ,μ,andκsettherelaxationtimescaleofshapedeformations,areachanges,andrepul-siveforces,respectively(seeSupplementalMaterial[31]fortheparametervalues).Thisformulationallowsthecellularinterfacestoberesolved,andintracellularforcestobedefinedatthemicroscopiclevel[Fig.1(b)].Therearemanyphysicalforcesofimportanceatthecellularlevel[32,33],andforcetransmissionbetweencellshasbeenshowntocontributetocollectivephenomenasuchascollectivedurotaxis[34]orcoordinationduringmorphogenesis[35,36]andwoundhealing[2,37].Weconcentratehereonasimplifieddescriptionandconsideronlysubstratefrictionandforcesgeneratedatthecellularinterfaces,leadingtotheforcebalanceequation:ξvi¼Finti;ð3ÞwhereξisasubstratefrictioncoefficientandFintiisthetotalforceactingontheinterfaceofcelli.Byanalogywithcontinuumtheories,wedefinethesemicroscopicinterfaceforcesintermsofamacroscopictissuestresstensorσtissueasFinti¼Zdxϕi∇·σtissue¼−Zdxσtissue·∇ϕi:ð4ÞThefirstexpressionistheintegralofthelocalforce∇·σtissueweightedbythephasefieldϕi,whilethesecondistheintegraloftheforceexertedbythestresstensoronthevector−∇ϕinormaltotheinterfaceandpointingoutwards.Equation(4)effectivelybridgesscalesbetweenlocalforcesattheleveloftheindividualcellsandpropertiesofthewholetissue.Equations(1) -- (4)defineagenericmodeloftwo-dimensionalepithelialmonolayersthatonlyrequiresanappropriatedefinitionofthestresstensorasinput.Followingouranalogywithcontinuumtheories,weintro-ducetheusualseparationintopassiveandactivestressesσtissue¼−PI−ζQ;ð5ÞwherethefieldsPandQarethetissuepressureandtissuenematictensortobedefinedbelow.AspointedoutinRef.[25],thereisinfactmuchfreedomindefiningPfromthetotalfreeenergyF.HereweusethethermodynamicallyconsistentdefinitionP¼XiδFrepδϕi−δFCHδϕi−δFareaδϕi;ð6Þwhichincludescontributionsfromcompressionandsurfacetensionterms(seeSupplementalMaterial[31]fordetails).KeytoourresultsisthedefinitionofthetissuenematictensorinEq.(5),whichisbasedonthefollowingcelldeformation:(a)(b)FIG.1.(a)AmicroscopeimageofaMDCKmonolayer(adaptedfromRef.[11])and(b)asimulationsnapshot.InterfacesbetweencellsaredefinedastheoverlapofthephasefieldsPi≠jϕiϕj(left).Contoursϕi¼1=2withanillustrationoftheprincipalaxisofthedeformationtensortogetherwiththeresultingdipolarforcedensity(right).Thedirectorofthedipolarforceisparalleltothemaindeformationaxis.PHYSICALREVIEWLETTERS122,048004(2019)048004-2Collectivemigration.Forlowcelldensities,thecollectivemigrationisdominatedbybinaryinteractions.Fullyinelasticcollisions,seeFig.1a),canleadtothealignmentofthemigrationdirectionsandtheonsetofcollectiveunidirectionalmotion.Westudiedtwogenericsituations.First,asystemwithperiodicboundaryconditions,correspondingtoacellpopulationthatisfarfromallboundaries.AsimilarsituationhasbeenrealizedexperimentallyinRef.7.Second,acircularconfineddomainwherethecellscanadhere,surroundedbyaregionwhereadhesiontothesubstrateissuppressed.ThisgeometrywasstudiedexperimentallyinRefs.5,6,wherecirculardomainswerepreparedbymicro-contactprintingofadhesiveligands.Figure3a) -- c)showstheemergenceoftranslationalcollectivemigrationintheperiodicdomain.WedefinedanorderparameterforthetranslationalcollectivemotionviawTtðÞ~1NXNi~1^vitðÞ,ð1Þwhere^viistheunitvelocityvectorofthei-thcell.Forlargecellnumbers,theorderparameterwillvanishifthevelocitiesarerandom.Itwilltendto1ifallcellsarealigned.TheredcurveinFigure3d)illustratestheemergenceofcollectiveunidirectionalmotionfromanensembleofcellswithinitiallyrandomdirections:afteratransientofaboutt53000,theorderparameterwTapproachesavalueofone.Figure3e) -- g)showscollectivemotionintheconfinedcirculardomain.Afteratransient,all24cellsperformacounter-clockwiserotation.Againthiscanbequantifiedbyintroducinganorderpara-meterforrotationalcollectivemotionviawRtðÞ~1NXNi~1^ehitðÞ:^vitðÞ,ð2Þwhere^ehiistheunitvectorinangulardirectionofthei-thcell.TheredcurveinFig.3h)showsitsevolutionforthescenariodisplayedine) -- f),thefinalvaluecloseto1correspondingtocounter-clockwiserotation.Inbothcases,thecollectivemotionwasestablishedaftersometransient,typicallywhenthecellshavemigratedadistanceoftheorderof50 -- 100timestheirownsize(roughly3000 -- 4000dimension-lesstimeunits).Asacounterexample,wesimulatedcellswithelasticcollisions[asinFigure1b)],whichdidnotexhibitanycollectivemigrationontheconsideredtimescales(upto8000timeunits).Hencethesimplepictureofinelasticcollisionsinducingthetrans-ition,deducedfromthebinaryinteractions,prevailsuptomoderatecelldensities(volumefractionisabout0.4 -- 0.5).Inthecirculardomain,theinteractionsofcellswiththeboundary(dependingonparametersandincidenceangle,theyarereflectedortrappedbytheboundary,cf.45,47)forcesatransitionfromtranslationaltorotationalcollectivemotion,similartothatobservedinRef.5.Letusnowstudytheeffectofcell-celladhesion,whichsofarwasabsent(k50).Increasingthecell-celladhesionparametertomod-eratevalues(k56)leadstoabreak-downofthecollectivelyrotatingstate,seeFig.3i) -- k),asanticipatedfromthereductionofthecollisionFigure3Collectivemigrationofcells.(a) -- (c)Emergenceofatranslationalcollectivemigrationof20cellsinaperiodicdomainwithoutcell-celladhesion.(d)TheorderparameterwT(t)forcellswithout(red,k50)andwith(blue,k56)cell-celladhesion.Here,thealignmentduetoinelasticcollisions,cf.Fig.1,leadstoacollectiveunidirectionalmotion,seeSupplementaryMovie5.(e) -- (g)Emergenceofarotationalcollectivemotioninacircularconfineddomain.Intheredregion,theadhesivebondformationtothesubstrate(parametera0,seeSupplementaryinformation)isreducedbyafactorof9.SeeSupplementaryMovie6.(h)TheorderparameterwR(t)forcellswithout(red,k50)andwith(blue,k56)cell-celladhesion.(i) -- (k)Adhesion(k56)suppressescollectiverotationalmotion,althoughlargefluctuationsoftheorderparameterwR(t),asshowninh),indicatetransientcollectivebehavior,seeSupplementaryMovie7.(l)AveragevelocitynormalizedbythetotalnumberofcellsNforcellsmovinginaperiodicdomainfordifferentvalueofthecell-celladhesion(k).Initialradiusofcells:r0510,domainsize:L5100.www.nature.com/scientificreportsSCIENTIFICREPORTS5:9172DOI:10.1038/srep091724velocitiesasshownFig.2h)correspondto0.1 -- 0.2micronspersec-ond,wellintherangeforkeratocytespeeds52.Theobservedbehaviorshighlighttwokeyphenomena.Firstly,thebistabilityofthecellpolarizationdynamicsleadstoacoexistenceofstationaryandmotilestates,asfoundinexperiments.Secondly,thattransitionsbetweenthesestatescanbetriggeredbytheenvir-onmentalconditions,interactions,aswellasbyeffectiveparameterslikecelldensityandcollisionprobability.Figure1Binaryinteractions(collisions)ofmotilecells.(a)Stronglyinelasticcollisionoftwocanoe-shapedcells(c50.5,s51.3),leadingtoaneffectivealignmentofthedirectionsofmotion.(b)Analmostelasticcollisionoftwobell-shapedcells(c50.7,s50.6).In(a)and(b),contoursofthecellsaregiveninwhite,theabsolutevalueoftheactinorientationinblueandregionswithhighadhesioningreen.Thevelocitiesareindicatedasyellowarrows.Thelateralsizeis200dimensionlessunits.(c)Effectoftheincidenceangleonthecells'centerofmasstrajectories.Theredcurvecorrespondstothesnapshotsshownin(a).Thedirectionofmotionofcellsisindicatedbythearrows.(d)Effectofadhesionstrengthkonthecells'centerofmasstrajectories:increasingadhesionreducestheeffectivealignmentofcells.Initialradiusofcells:r0515,domainsize:L5200,periodicboundaryconditions.SeeSupplementaryMovies1&2.Figure2Transitionstriggeredbycelldensityandenvironmentalconditions.(a) -- (c)Sequenceofsnapshotsillustratinghowinitiallymovingcellscometorestandspreadonthesubstrate.(e) -- (g)Afewmotilecellsexcitethemotionofallcells.Thevaluesoftheparametersarethesameasin(a) -- (c)exceptforanincreasednumberofcellsandincreasedvalueoftheparameteranlfrom1to1.1(cf.themodelingoftheadhesivebondformationintheSupplementaryInformation).Panels(d)and(h)showtheaveragevelocityAEV(t)aeforthetwoscenariosabove,respectively.SeealsoSupplementaryMovies3&4.www.nature.com/scientificreportsSCIENTIFICREPORTS5:9172DOI:10.1038/srep091723Shape-based anisotropic active stressVelocity alignmentCollective motionTimeTimewww.nature.com/scientificreports/3Scientific RepoRts 5:11745 DOi: 10.1038/srep11745(cid:30)φφδδφ∂∂+⋅∇=−,()tv121nnnnwhere (cid:30) is the monolayer free energy, vn is the translational velocity of cell number n and δ denotes a functional derivative. Note that this equation is written in terms of dimensionless units, which will be used throughout the paper. The relationships between dimensionless and real units are detailed in the Supplementary Information, briefly reviewed at the end of the Methods section. The right-hand side of Eq. (1) describes cell shape dynamics, which are determined by free energy changes. Details of the model free energy are given in the Methods section and in particular, by Eqs (7) and (10). The free energy contains several parameters and its minimum determines the prefered state of the system. Briefly, one parameter controls the elastic response of each cell to shape deformation (γn in Eq. (7)). Another controls the preferred radius of the cells (R in see Eq. (7)), which tend to be circular when they are not perturbed by other cells. Also, there is a parameter that controls the energy penalty for overlapping cells (κ in Eq. (10), which is chosen to be large). Note that the interactions between cells are strictly repulsive.The velocity of each cell is the sum of two distinct contributions as described in Eq. (11): 1) The inactive part, vI, is due to the interaction force exerted by the other cells. Like the cell shape dynamics, the inactive part of the velocity is determined by free energy changes. 2) The active, or self-propulsion, part of the velocity, vA, is due to the cell engine. The relative strength of the two contributions to the velocity is determined by another parameter (ξ in Eq. (12)) which also controls the maximum cell shape deformation. The active part of the velocity is chosen so that the motion of isolated cells on the substrate is described by a persistent random walk34,35 where the characteristic cell speed and the reorientation statistics match the experimental observations9. Further, we assume that there is a large separation of time scales between the cell shape relaxation (fast) and the cell translation dynamics (slow). This approx-imation is physical since 1 µm deep indentations on cells typically relax within seconds36 and the motile cells we model translate by 1 µm within minutes9.To highlight the role of cell elasticity mismatch, we considered 2 types of cell monolayers assemblies: a single cancer cell in a layer of normal cells (hereafter referred as the "soft-in-normal" case) and normal cells only (hereafter referred as the "all-normal" case). We performed simulations at two packing frac-tions, ρ = 0.85 and 0.9, describing nearly confluent monolayers, where,ρπ≡,()NRA2cellB2where AB is the area of the simulation box. Overall, we will present a total of 4 simulations, each of which contains Ncell = 72 cells. All model parameters are listed in Table 1 and explained in details in the Methods section and in the Supplementary Information. Our aim is to isolate the effect of the mechani-cal properties of motile cells on their migratory behavior. Hence, all parameters will be identical for both types of cells with the exception of the parameter which controls the cell stiffness (we set γcancer/γnormal = 0.35, Figure 1. An example of a model monolayer comprising one soft cancer cell and normal cells. Each cell is described by a field, φ(x, y), that is defined to be 0 outside the cell and 1 inside. The field rapidly varies in the region of the cell boundary. A. The field of a single cancer cell. B. The monolayer is reconstructed by showing the boundary of all cells (Blue curves). A tagged cell, the cancer cell, is filled in Green with a Black boundary.Phase field3.2.4 Collective phenomena Phase-field models have primarily addressed the emergence of collective motion from cell-cell interactions. Whereas some works focused on explicit orientational interac- tions [98], other studies showed that, when cell polarity is coupled to cell shape asym- metry (Section 2.3.4), collisions between deformable cells lead to cell-cell velocity alignment and collective motion [66, 67] (Figs. 4c and 4d). Recently, the phase-field model has been employed to explain the emergence of extensile nematic behavior [100] and to recapitulate collective velocity oscillations [69] in epithelial monolayers. 3.2.5 Discussion The phase-field formalism provides a detailed description of cell shape while tackling some of the issues of the CPM. Foremost, it introduces a force balance Eq. (5) that provides a physical dynamics, thus going beyond the energy minimization process of the CPM, which imposes static mechanical equilibrium at each step. Moreover, the phase-field model is currently better connected to tissue mechanics (Eq. (6)), and it can explicitly account for cell-cell and cell-substrate friction as well as for active stresses (Eqs. (5) to (7)) [69, 100]. 3.3 Active network models With precedents in the physics of foams [102], network models describe epithelial tissues as networks of polygonal cells [103]. Thus, albeit in less detail than lattice and phase-field models, these models still describe sub-cellular features of cell shape. They encompass two subtypes of models: vertex and Voronoi models. 3.3.1 Vertex and Voronoi models In vertex models, the degrees of freedom are the vertices of the polygons. Alterna- tively, the network can be described by the cell centers, which reduces the number of degrees of freedom. These descriptions are known as Voronoi models because, given the positions of the cell centers, the cell-cell boundaries are delineated by the Voronoi tessellation (Fig. 5a). The difference in the number of degrees of freedom has impor- tant consequences for the mechanical properties of the network, which may thus differ between vertex and Voronoi models [104]. Moreover, cell motion, as well as cell divi- sion and cell death, may entail topological rearrangements of the network of cell-cell interfaces. In Voronoi models, the network is dynamic, evolving with each recomputa- tion of the tessellation. In vertex models, in contrast, network rearrangements entail the appearance and disappearance of vertices, which requires implementing specific rules. 3.3.2 Energy function In both descriptions, as in the previous approaches, cellular properties and interactions are encoded in an energy function, usually parametrized in terms of the areas Aa and perimeters Pa of cells a = 1, . . . , N: F = N(cid:88)a=1(cid:20) κ 2 (Aa − A0)2 + ΛPa + Γ 2 P 2 a(cid:21) . (8) 13 Here, κ is the modulus of cell area around its preferred value A0 (Section 2.2.3). Re- spectively, Λ = γc − w/2 is the line tension of the cell-cell interfaces that connect the vertices, which results from the coaction of the cortical tension along cell-cell con- tacts, γc, and the cell-cell adhesion energy w (Sections 2.2.1 and 2.2.4) [105, 106]. When cell-cell adhesion dominates, the line tension Λ becomes negative and the cell- cell interface tends to expand. This expansion is eventually saturated by other cellular processes. This saturation is encoded in the third term of Eq. (8), which gives rise to a perimeter-dependent line tension. This term is a key difference between models of tissues and foams; for the latter, Λ is always positive and the quadratic perimeter term is absent [102]. The two perimeter contributions in Eq. (8) can be recasted as an energetic penalty for departures from a preferred perimeter P0 = −Λ/Γ: (Pa − P0)2(cid:21) . (9) F = (Aa − A0)2 + Γ 2 N(cid:88)a=1(cid:20) κ 2 3.3.3 Cell migration and force balance Cell motility can then be implemented by applying active polar forces either on the ver- tices or on the cell centers, giving rise to Active Vertex Models (AVM) [77, 110 -- 113] and Self-Propelled Voronoi models (SPV) [107, 114], respectively. Thus, the corre- sponding degrees of freedom i = 1, . . . , n move according to Eq. (5), albeit with (cid:126)F int i = −(cid:126)∇(cid:126)riF (Fig. 5b). In addition, to account for interfacial effects at tissue bound- aries, Salm and Pismen added a 'wetting force' at the tissue edge [110], whereas Barton et al. included surface tension and bending forces [68]. 3.3.4 Polarity dynamics The most popular orientational interaction in SPV models has been polarity-velocity alignment [68, 108, 114 -- 116] (Fig. 5c and Section 2.4.1). However, polarity-shape alignment [68, 111] (Section 2.3.4), polarity alignment [68] (Section 2.3.1), CIL [77, 111] and force-induced polarization [110] (Section 2.3.2), as well as couplings to self- secreted chemoattractants [110] have also been considered. Coburn et al. have pro- posed a hybrid model that accounts for CIL and polarity-shape alignment [117]. 3.3.5 Collective phenomena Using the SPV model, Bi et al. studied how cell motility modifies the solid-fluid tran- sition displayed by passive vertex models, showing that both self-propulsion speed and persistence favor the fluid phase [107] (Fig. 5d). Other studies have focused on the onset of collective motion, showing that cell-autonomous polarity-velocity alignment (Fig. 5c and Section 2.4.1) gives rise to emergent cell-cell alignment, which leads to coherent rotations [114] and flocking [68, 108, 115]. Altogether, SPV models predict four distinct phases: solid, liquid, solid flock, and liquid flock (Fig. 5e). The solid phase supports elastic collective oscillations excited by self-propulsion [68, 116]. 3.3.6 Discussion By construction, most network models describe confluent tissues, in which cells are packed without free space between them. Therefore, these models are restricted to col- lective migration of epithelial cell groups. This limitation has been addressed in recent 14 Figure 5 Active Network Models. a, Cell centers at positions {(cid:126)r} are connected by the Delaunay triangulation (black). Its dual is the Voronoi tessellation (red) that defines cell boundaries and vertices at positions {(cid:126)h}. Reprinted from [107]. b, Cells are parametrized by an area Ai and a perimeter Pi, and experience a self-propulsion force Ta(cid:126)pi (orange) and an interaction force (cid:126)F int i = −(cid:126)∇(cid:126)riF (dashed black), which give the resultant force (black). Adapted from [108] with permission from The Royal Society of Chemistry. c, Polarity-velocity alignment with a time-scale τ. Adapted from [108] with permission from The Royal Society of Chemistry. d, Schematic phase diagram of the fluid-solid transition in the SPV model in terms of the shape index p0 = P0/√A0, and the self-propulsion speed v0 = Ta/ξ and persistence r . Adapted from [107]. e, Schematic phase diagram of the SPV model with D−1 polarity-velocity alignment at rate J = τ−1 (see c). Reprinted from [109] by permission from Springer Nature. work that generalizes the Voronoi model to non-confluent tissues [118]. In general, network models are particularly suited to study the role of cell geometry and topo- logical rearrangements on cell motion. Similar to the CPM, a current limitation of network models is that they account for neither internal dissipation nor anisotropic ac- tive stresses in the tissue. Recent efforts to include cell-cell friction [119] and to relate network geometry to the tissue stress tensor [120] offer possible ways to address these limitations. 3.4 Particle models Particle models are rooted in the physics of particulate media such as granular mate- rials. Compared to previous descriptions, particle models resolve even less details of cell shape by treating each cell as one or two circular particles. Using two particles 15 africtionalforcebetweencellsproportionaltothelengthoftheedgesharedbetweentwocells,andweknowfrompreviousworkonparticulateglassesthattheselocalizedfrictionscanchangethelocationofjammingorglasstransitionandthenatureofspatialcorrelationsinaglass[61,62].Itisalsotemptingtospeculateabouttherelationshipbetweentheunjammingtransitioncapturedbyourmodelandtheepithelial-mesenchimaltransitionthatprecedescellescapefromasolidtumormass.TheEMTinvolvessignificantchangesincell-celladhesionandcytoskeletalcomposition,withassociatedchangesincellshapeandmotility.Thissuggeststhatescapefromthetumormassiscontrollednotjustbythechemicalbreakdownofthebasementmembrane,butalsobyspecificchangesinmechanicalpropertiesofbothindividualcellsandthesurroundingtissue[63].Onecouldthenhypothesizethatthecollectiveunjammingwedescribeheremayprovidethefirstnecessarysteptowardsthemechanicalchangesneededforcellescapefromprimarytumors.Inparticular,recentworksuggeststhatcancertumorsaremechanicallyheterogeneous,withmixturesofstiffandsoftcellsthathavevaryingdegreesofactivecontractility[38].Ourjammingphasediagramsuggeststhatthesoftcells,whichoftenexhibitmesenchymalmarkersandpresumablycorrespondtohighervaluesofp0,mightunjamandmovetowardstheboundaryofaprimarytumormoreeasilythantheirstiffcounterparts.Examiningtheeffectsoftissueheterogeneityontissuerigidityandpatternsofcellmotilityis,therefore,averypromisingavenuefordevel-opingpredictivetheoriesfortumorinvasivenessandmetastasis.ACKNOWLEDGMENTSM.L.M.acknowledgessupportfromtheAlfredP.SloanFoundation.M.L.M.andD.B.acknowledgesupportfromNSF-BMMB-1334611,NSF-DMR-1352184,andNIHR01GM117598-02,theGordonandBettyMooreFoundation,andtheResearchCorporationforScientificAdvancement.M.C.M.acknowledgessupportfromtheSimonsFoundation.M.C.M.andX.Y.acknowledgesupportfromNSF-DMR-1305184andNSF-DGE-1068780.TheauthorsalsoacknowledgetheSyracuseUniversityHTCCampusGrid,NSFAwardNo.ACI-1341006,andtheSoftMatterProgramatSyracuseUniversity.D.B.acknowledgessupportfromtheCenterforStudiesinPhysicsandBiologyatRockefellerUniversity.APPENDIXA:SIMULATIONALGORITHMFORTHESPVMODELTocreateaninitialconfigurationforthesimulation,wefirstgenerateaseedpointpatternusingrandomsequentialaddition[64]andannealitbyintegratingEq.(2)withv0¼0for100MDsteps.Theresultingstructurethenservesasaninitialstateforallsimulationsruns.Theuseofrandomsequentialadditionservesonlytospeeduptheinitialseedgeneration,asusingaPoissonrandompointpatterndoesnotchangetheresultspresentedinthispaper.Ateachtimestepofthesimulation,aVoronoitessella-tioniscreatedbasedonthecellcenters.TheintercellularforcesarethencalculatedbasedonshapesandtopologiesoftheVoronoicells(seediscussionbelow).WeemployEuler'smethodtocarryoutthenumericalintegrationofEq.(2);i.e.,ateachtimestepofthesimulationtheintercellularforceiscalculatedbasedonthecellcenterpositionsintheprevioustimestep.InaDelaunaytriangulation,atrioofneighboringVoronoicentersdefineavertexofaVoronoipolygon.Forexample,inFig.6,(~ri,~rj,~rk)definethevertex~h3,whichisgivenby~h3¼α~riþβ~rjþγ~rk;ðA1Þwherethecoefficientsaregivenbyα¼∥~rj−~rk∥2ð~ri−~rjÞ·ð~ri−~rkÞ=D;β¼∥~ri−~rk∥2ð~rj−~riÞ·ð~rj−~rkÞ=D;γ¼∥~ri−~rj∥2ð~rk−~riÞ·ð~rk−~rjÞ=D;D¼2∥ð~ri−~rjÞ×ð~rj−~rkÞ∥2:ðA2ÞInthevertexmodel,thetotalmechanicalenergyofatissuedependsonlyontheareasandperimetersofcells:E¼XNi¼1½KPðAi−A0Þ2þKPðPi−P0Þ2:ðA3ÞInaVoronoitessellation,theareaandperimeterofacellicanbecalculatedintermsofthevertexpositions:FIG.6.Cellcenterpositionsarespecifiedbyvectorsf~rg.TheyformaDelaunaytriangulation(blacklines).ItsdualistheVoronoitessellation(redlines),withverticesgivenbyf~hg.MOTILITY-DRIVENGLASSANDJAMMINGTRANSITIONS…PHYS.REV.X6,021011(2016)021011-9INSIGHT REVIEW ARTICLENATURE PHYSICSmotors, collective gradient sensing is often impaired. Another striking emergent phenomenon in cell monolayers is their abil-ity to propagate mechanical waves. In response to sudden uncon-finement, the first row of cells at the monolayer edge spreads and migrates towards the freely available substrate, whereas the cells behind them remain static66,75. With time, every cell row becomes progressively engaged in collective motion following a wave of deformation and force generation. Propagation of mechanical waves has now been observed in confined clusters76,77, in expanding colonies66,75 and in colliding monolayers78 (Fig. 3). Propagation of mechanical waves in inertial matter is a trivial physical phenom-enon that can be simply explained by an exchange of potential and kinetic energy. In cell monolayers, which lack inertia, alterna-tive mechanisms must be at play to introduce second derivatives with respect to time and provide an effective inertia76. This could be achieved through the interplay between cell mechanics and molecular circuits involving mechanotransduction, such as those involving the mechanosensing protein merlin, which links forces at cell -- cell junctions with the regulation of the protrusive and force-generating action of RHO-GTPases6,79.A general feature of cell monolayers -- and possibly one that is intuitive to any cell biologist who has ever performed cell culture -- is that cells growing on a dish slow down their motion as the cell density increases80. From a physical perspective, this behaviour is reminiscent of that of granular materials such as sand or coffee beans close to a jamming transition; as the system density increases, each constitutive element becomes trapped by its neighbours, the energy required for structural rearrangements rises, and the sys-tem transitions from a fluid to a disordered solid81. Careful analysis of cell velocity fields in proliferating cell monolayers showed that the analogy between the behaviour of cell monolayers and granu-lar materials close to a jamming transition is deeper than expected. Like particles in granular materials, cells in dense monolayers move in large groups whose length scale grows with cell density5. Moreover, cells in a monolayer exhibit dynamic heterogeneities in cell migration, a non-Arrhenius dependence of relaxation times on cell density, peaks in the vibrational density of states, and a shift in the position of the four-point susceptibility function80.A perhaps less intuitive observation is that fluid-to-solid tran-sitions in cell monolayers can also occur at constant density. This striking result was first predicted by theoretical analysis of vertex models82 (Box 3). These models aim at capturing the dynamics of tissues in terms of the motion of vertices representing junctions between three or more neighbouring cells83. A key ingredient behind the success of vertex models is that they readily enable neighbour exchanges. In the absence of mitosis and extrusion, neighbour exchanges in monolayers occur through T1 transitions, whereby one edge between two cells shrinks until vanishing and a new edge is created at the same point between two different cells. Bi et al.82 identified that the energy distribution of T1 transitions in a ver-tex model is a function of a dimensionless geometric factor called the target shape index =∕pPA000, where P0 and A0 are preferred perimeter and area, respectively. For regular polygons, p0 increases with the number of sides, with p0 = 3.72 for a regular hexagon and p0 = 3.81 for a regular pentagon. More strikingly, they found that below a critical value p0 = 3.81, energy barriers are finite and the system behaves like a jammed solid. By contrast, above p0 = 3.81, energy barriers become vanishingly small and the monolayer behaves like an unjammed liquid. This result, which has now been extended to 3D bulk tissues84, suggested that cell monolayers exhibit a jamming transition controlled by a geometric parameter indepen-dent of tissue density. This prediction was successfully tested using bronchial epithelial cells maturing in an air/liquid interface85. As the maturation time increased, p0 decreased without a noticeable change in the cell density, and the monolayer dynamics slowed down. Remarkably, bronchial epithelial cells from asthmatic donors exhibited higher values of p0 and remained unjammed for a longer time than cells obtained from healthy donors.The ability of early 2D vertex models to precisely capture the jam-ming transition in cell monolayers is remarkable taking into account that they did not include the ability of cells to exert traction forces on their substrate or to self-propel their body. One approach to include self-propulsion in the description of 2D monolayers was the devel-opment of SPV models (Box 3)86 -- 88. These models predict four rel-evant states in monolayers defined in terms of the shape index p0 and the characteristic time of alignment between neighbours J (Fig. 3). These states are solid, liquid, solid flock and liquid flock86,87. The solid (or jammed) phase emerges at low p0 and low J, and it is char-acterized by the absence of cell rearrangements and of global tissue movement. If p0 is increased at low values of J, the monolayer is pre-dicted to enter a liquid phase in which the spatial correlation length remains low but local rearrangements become frequent. These two phases are analogous to those predicted by 2D vertex models in the absence of self-propulsion or with self-propulsion but in the absence of alignment86 -- 88. At higher values of J, the SPV model predicts the emergence of flocking (that is, an increase in the spatial correlation Shape index p0 Alignment interaction JSolid flockSolidLiquid flockLiquidSolid flockLiquid flockTimeTimeabcFig. 3 State diagrams of cell monolayers. Dynamic states of a cell monolayer. a, State diagram depicting cell states as a function of the shape index and the alignment rate. b, Time evolution of a cell cluster in a flocking solid state. Cells comprising the initial cluster move cohesively without neighbour exchanges. c, Time evolution of a cell cluster in a flocking liquid state. Cells comprising the initial cluster drift with frequent neighbour exchanges. Credit: adapted from ref. 86, RSC, and ref.!87, Macmillan Publishers Ltd (a); and courtesy of R. Cerbino and F. Giavazzi (b,c).INSIGHT REVIEW ARTICLEhttps://doi.org/10.1038/s41567-018-0194-9INSIGHT REVIEW ARTICLENATURE PHYSICSNATURE PHYSICS www.nature.com/naturephysics3472SoftMatter,2018,14,3471--3477Thisjournalis©TheRoyalSocietyofChemistry2018essentiallyalwaysunity.Recenttheoreticalworkhascombinedthewell-establishedVertexModel,thatdescribesaconfluentepithelialcellsheetasadisorderedpolygonaltilingoftheplaneandhasbeenusedsuccessfullytomodelthedevelopmentofthefruitflyembryo,19withideasfromactivematterphysicstodevelopaSelf-PropelledVoronoi(SPV)modelofmotiletissue.TheSPVmodelexhibitsaJUTtunedbycellmotilityandcellularshape,whichinturnembodiesthecompetitionofcontractilityandcell -- celladhesion.7,20 -- 24Noneofthesemodels,however,accountsforastrikingsetofexperimentalobservations,11inwhichtheelevationofRAB5A,amasterregulatorofendocytosis,induceslarge-scaledirectedmigratorypatterns,whichresembletheonsetofflockinginotherlivingsystems.25Additionally,experimentsshowthatcellsaltertheirpolarizationanddirectionofmigrationduetointeractionswithsurroundingcells.26 -- 28Thismechanicalfeedbackwasincorporatedinref.11byextendingtheSPVmodeltoincludealocalinteractionthattendstodirectcellpolarization.Whilethissimplemodificationyieldsflockingphasessimilartothoseobservedintheexperiments,11suggestingthattheobservedreawakeningofmotilityrequiresasimultaneousincreaseofcell -- celladhesionandcoordinationofcellpolarization,thepropertiesofsuchflockingphasesandthecomparisonwithflockingtransitionsinparticle-basedmodelshavenotyetbeenexplored.HerewecharacterizethefullphasediagramoftheSPVmodelwithalignmentinteractionsandexaminethestructuralanddynamicalpropertiesofthevariousphases.Weobtainanumberofsignificantresults.First,theflockingtransition,knowntobefirstorderinparticlemodelswheretheinteractionrangeisdefinedthroughametriccriterion,whereeachagentinteractsonlywithagentswithinaprescribedrange,29appearstobecontinuoushereinboththeliquidandthesolid.ThisisconsistentwithresultsfromVicsekmodelswhichdisplayacontinuoustransitionwhenagentsalignwiththeirtopologicalneighborsdefinedasthosebelongingtothefirstshellofaVoronoitessellation,insteadofmetricones.30GiveninteractionsintheSPVarecontrolledbytopologyandnotbymetricdistance,thisfindingsuggeststhepossibilitythatacontinuoustransitionmaybeagenericpropertyofsystemswithmetric-freeinteractions.Secondly,weexaminetheinter-playofalignmentandstructuralpropertiesinbothsolidandliquidstatesandshowthatalignmentpromotessolidificationbysuppressingfluctuationstransversetothedirectionofmeanmotion.Flockingglassyandcrystallinestateshavebeenreportedbeforeinparticlemodels31,32andincontinuumtheories,33butalignmentwassuggestedtodrivefluidificationinparticulatesystems.32Extendingtoolsfromstatisticalphysicstothesefar-from-equilibriumsystems,wealsodevelopageneralizedexpressionfortheeffectivetemperatureandcagingtimescaletoself-propelledsystemsthathelpsusunderstandhowalignmentinteractionsdrivesolidificationinconfluentmodels.Finally,theflockingstatesarecharacterizedbystronglyanisotropicfluctuations.Inthefluidsuchanisotropyisevidentinbothstructuralanddynamicalpropertiesthatrevealabehaviorsimilartothatof2Dsmectics,asrecentlysuggestedforincompressibleparticulateflockingfluids.34Collectiverearrangementsobservedasthesolidisapproachedfromthefluidsidearestronglyanisotropicandtaketheformofcorrelatedcellstreaming.Thismorphologyprovidesadistinctsignatureofthecollectivedynamicsofflockingliquidsthatcouldbeprobedinfutureexperiments.2ModelTheSPVmodeldescribesaconfluentmonolayerasanetworkofpolygonscoveringtheplane.20 -- 22EachcellischaracterizedbyitspositionriandcellshapeasdeterminedbytheVoronoitessellationofallcellpositions(Fig.1).Asinthevertexmodel,35cell -- cellinteractionsaredeterminedbyaneffectivetissueenergy,19,35 -- 39EXiKAAiA0ðÞ2þKPPiP0ðÞ2hi;(1)withAiandPithecross-sectionalareaandtheperimeterofthei-thcell,andKAandKPareaandperimeterstiffnesses.Thefirstterm,quadraticinthefluctuationsofthecellareaaroundthetargetvalueA0,arisesfromtheconstraintofincompressibilityinthreedimensionsandencodesbulkelasticity.Thesecondterm,quadraticinthedeviationofcellperimeterfromthetargetvalueP0,representsthecompetitionbetweenactivecontractilityintheactomyosincortexandcell -- celladhesion,resultinginaneffectiveboundarytensionproportionaltoP0.WeconsiderNcellsinasquareboxofareaL2withperiodicboundaryconditions.Inthefollowing,wesetboththeaveragecellarea%A=L2/NandthetargetareaA0equaltoone,%A=A0=1,thoughchangingA0inaperiodicsystemhasnoeffectonthecelldynamics.40ThesystemisinitializedwithrandominitialpositionsfortheNcells.Theconfigurationalenergyineqn(1)hasbeenextensivelyusedinthepasttomodelbiologicaltissues,butonlyrecentlyithasbeenshownthatthissimplemodelexhibitsarigiditytransitionthattakesplaceatconstantdensityanditiscontrolledbyasinglenon-dimensionalpara-meter,thetargetshapeindexp0P0ffiffiffiffiffiffiA0p.21Inourmodel,weFig.1Schematicrepresentationofthemodel.(a)EachcellisapolygonobtainedbytheVoronoitessellationofinitiallyrandomcellpositionsri,characterizedbytheareaAiandtheperimeterPiofthepolygon.ThecellexperiencesaforceFi=rriEduetoitsneighborsandaninternalpropulsiveforcefisalongthedirectionniofitspolarization(eqn(2)).(b)Anactiveorientationmechanismreorientseachcell'spropulsiveforcetowardsitsmigrationvelocityoveracharacteristicresponsetimet=J1(eqn(3)).PaperSoftMatterPublished on 12 April 2018. Downloaded by Princeton University on 10/05/2018 17:39:33. View Article Onlineaverageinthesolidphaseandanisotropicinthefluidphase.Thishighlightsthefactthatqcanbeusedasastructuralorderparameterfortheglasstransitionatallcellmotilities,providingapowerfulnewtoolforanalyzingtissuemechanics.IV.THREE-DIMENSIONALJAMMINGPHASEDIAGRAMFORTISSUESHavingstudiedtheglasstransitionasafunctionofv0andp0atalargevalueofDr,weinvestigatethefullthree-dimensionalphasediagrambycharacterizingtheeffectofDrontissuemechanicsandstructure.Drcontrolsthepersistencetimeτ¼1=DrandpersistencelengthorPécletnumberPe∼v0=Drofcelltrajectories;smallervaluesofDrcorrespondtomorepersistentmotion.InFig.3(a),weshowseveral2Dslicesofthethree-dimensionaljammingboundary.Solidlinesillustratethephasetransitionlineidentifiedbythestructuralorderparameterq¼3.813asfunctionofv0andp0foralargerangeofDrvalues(from10−2to103).(InAppendixB2,wedemonstratethatthestructuraltransitionlineq¼3.813matchesthedynamicaltransitionlineforallstudiedvaluesofDr.)Incontrasttoresultsforparticulatematter[22],thisfigureillustratesthattheglasstransitionlinesmeetatasinglepoint(p0¼3.81)inthelimitofvanishingcellmotility,regardlessofpersistence.Figure3(b)showsanorthogonalsetofslicesofthejammingdiagram,illustratinghowthephaseboundaryshiftsasafunctionofp0andDratvariousvaluesofv0.ThishighlightstheinterestingresultthatasolidlikematerialathighvalueofDrcanbemadetoflowsimplybyloweringitsvalueofDr.Thecrossoverinbehavioratlargev0occurswhenthepersistencetime1=Drisapproximatelyequaltotheviscousrelaxationtime1=ðμKAA0Þ¼1.Theseslicescanbecombinedtogenerateathree-dimensionaljammingphasediagramforconfluentbiologi-caltissues,showninFig.3(c).Thisdiagramprovidesaconcrete,quantifiablepredictionforhowmacroscopictissuemechanicsdependsonsingle-cellpropertiessuchasmotileforce,persistence,andtheinterfacialtensiongeneratedbyadhesionandcorticaltension.WenotethatFig.3(c)issignificantlydifferentfromthejammingphasediagramconjecturedbySadatietal.[12],whichwasinformedbyresultsfromadhesiveparticulatematter[14].Forexample,inparticulatematteradhesionenhancessolidification,whileinconfluentmodelsadhe-sionincreasescellperimetersorsurfaceareaandenhancesfluidization.Inaddition,weidentify"persistence"asanewaxiswithapotentiallysignificantimpactoncellmigrationratesindensetissues.Tobetterunderstandwhypersistenceissoimportantindensetissues,wefirsthavetocharacterizethetransitionsbetweendifferentcellularstructures.Inthelimitofzerocellmotility,thesystemcanbedescribedbyapotentialenergylandscapewhereeachallowablearrangementofcellneighborscorrespondstoametastableminimuminthelandscape.Therearemanypossiblepathwaysoutofeachmetastablestate:someofthemcorrespondtolocalizedcellrearrangements,whileotherscorrespondtolarge-scalecollectivemodes.Themaximumenergyrequiredtotran-sitionoutofametastablestatealongeachpathwayiscalledanenergybarrier[27].WeobservethattissuefluiditycanincreasedrasticallywithdecreasingDratfinitecellspeeds.Thissuggeststhatdifferentpathways(withlowerenergybarriers)mustbecomedynamicallyaccessibleatlowervaluesofDr.Onehintaboutthesepathwayscomesfromtheinstanta-neouscelldisplacements,shownfordifferentvaluesofDrinFig.4.AthighvaluesofDr(p0¼3.78,v0¼0.1),theinstantaneousdisplacementfieldisessentiallyrandomand(a)(b)(c)FIG.3.(a)Theglasstransitioninv0−p0phasespaceshiftsasthepersistencetimechanges.Linesrepresenttheglasstransitionidentifiedbythestructuralorderparameterq¼3.81.Thephaseboundarycollapsetoasinglepointatp0¼3.81,regardlessofDr,inthelimitv0→0.(b)Theglasstransitioninp0−Drphasespaceshiftsasafunctionofv0(fromtoptobottom:v0¼0.02,0.08,0.14,0.2,0.26).Forlargev0thereisacrossoverinthebehavioratDr∼μKAA0¼1,asdiscussedinthemaintext.(c)Thephaseboundarybetweensolidandfluidasafunctionofmotilityv0,persistence1=Dr,andp0,whichistunedbycell-celladhesion,canbeorganizedintoaschematic3Dphasediagram.Redlinesonthesurfacecorrespondtoiso-v0contoursandbluelinescorrespondtoiso-Drcontours.MOTILITY-DRIVENGLASSANDJAMMINGTRANSITIONS…PHYS.REV.X6,021011(2016)021011-5Fluid-likePropulsion speedPersistenceShape indexFigure 6 Particle Models. a, Schematic representation of forces in particle models. In different situations, cells experience various amounts of cell-cell repulsion (blue) and adhesion (orange), and cell-substrate active traction (red) and passive friction (green). Adapted from [46]. b, In a simple model, CIL rotates cell polarity (cid:126)pi towards the direction (cid:126)p CIL pointing away from cell-cell contacts. With these interactions, a cell cluster can spontaneously polarize and undergo collective motion, as indicated by the center-of-mass velocity (cid:126)v. The color code indicates the cell polarity angle θi. Adapted from [123]. c, In growing tissues, CIL gives rise to tension profiles similar to experimental measurements. Reprinted from [76]. d, Increasing the packing fraction of confined self-propelled particles leads to jamming. The jammed phase supports undamped low-frequency modes. Red arrows indicate cell velocity. Adapted with permission from [124]. Copyright (2011) by the American Physical Society. i still allows capturing cell shape anisotropy [121], and even a head-tail asymmetry if the particles have different size [122]. Otherwise, details of cell shape are entirely overlooked. 3.4.1 Cell-cell interaction potential Positional cell-cell interactions are implemented via a central interparticle potential V ((cid:126)ri − (cid:126)rj). As for the other kinds of energy functions, no general principle pre- dicts the exact form of the potential. Rather, simple forms are often proposed on phenomenological grounds. Typically, the potential features a short-range repulsion, which usually includes a hard core to prevent cell overlaps (Section 2.2.3). However, to capture cell extrusion, a recent model proposed a soft-core repulsion with a finite energy plateau [123]. In addition to the repulsive part, the potential often features a mid-range attraction to account for cell-cell adhesion (Fig. 6a and Section 2.2.1). Un- less modeling biochemical signaling or substrate-mediated elastic interactions, long- range non-contact interactions are not included and, hence, the potential is cut off at the maximal cell radius. 16 outward,andfillingtheavailablespace.Wecalculatedsubstratetractionstressesandintercellularstresseswithinthemodeltissue,andtheresultsareshowninFig.3.Inagreementwithexperimentaldataandincontrasttopreviousmodelversions(27,30),stressesinthetissuearealmostexclusivelytensilebecauseoftheintracellularcontractionforceandCIL(Fig.S2).Moreover,tensionincreasesfromthetissueedgetothecenter(Fig.3C).Tractionstressesexertedbythetissueareheterogeneous(Fig.3B);however,thereisabiastowardtheedges,whichleadstothecharacteristictensionprofileonintegrationalongthexdirection(Fig.3F).Thetractionforcepattern(Fig.3B)andbell-shapedtensioncurve(Fig.3F)arebothinexcellentagreementwithexperimentalobservationsinaspreadingMDCKcellcolony(13).Wealsoseeinoursimulationthatthecelldensityinthetissuecentersteadilyincreasesasthecolonyexpands(Fig.3D).Thesameeffecthasbeenobservedintheexperiment(13,41).Tensioninthetissuefollowstheexactsametrend,givingrisetoalinearrelationshipbetweencelldensityandintercellulartension(Fig.3I),againinexcellentagreementwiththeexperimentaldata(13).Asthecelldensityincreasestowardthetissuecenter,thespeedofoutwardmigrationofthecellsdecreasesandeventually,dropstoalmostzeroasthecelldensitycrossesathresholdvalue(Fig.3H).Cellsinthetissuecenterexperiencekineticarrest(41).Anin-creaseincellspeedtowardthetissueedgehasbeenmeasuredexperimentallyinrefs.44and45.Therelationshipbetweencellspeedandcelldensity(Fig.3H)agreeswellwiththeexperimentaldatainref.46.Weobserve,consistentwiththeexperimentsinref.41,thatCILprecedescontactinhibitionofproliferation(i.e.,thedensitycontinuestoincreasewhilethemotionofcellshasalreadystalled).Whatisthefeedbackmechanismleadingtoaclosecouplingbetweencelldensityandintercellulartension?Contactinhibitionalignsthepropulsionforcesofcellsatthetissueedgeawayfromthebulk,suchthatthosecells"escape."Consequently,theyreachadistancetotheirneighborsclosetothemaximuminteractiondis-tance(abovethedistanceformaximaladhesion)andexperienceweakintercellularforces.Atthesametime,theymakeroomfortheirfollowers,whichalsoexertlargepropulsionforces,becausethenumberofneighborsislowandCILisweak.Theabilitytoexertlargepropulsionforces,however,increasesthecelllength,leadingtoalowercelldensity.Asthenumberofneighborsin-creases,CILdecreasesforcesonthesubstrate.Therefore,cellscannotescapeandgetclosertotheirneighbors,andthecell -- celladhesionforcetendstoassumeitsmaximumvalue,makingitevenlesslikelyforcellstomoveawayfromtheirneighborsanddrivingthemintoajammedstate.Moreover,lowpropulsionforcesentailsmallcells,andthedensityofadhesionforceincreaseswithcelldensity,leadingtoahigherintercellulartension.Wehaveverifiedthismechanismbychangingsomeofourmodelparameters(Fig.S3).WeranasimulationwithoutCIL,suchthatallcellsalwaysexertpropulsionforcesofgivenmagni-tudealongthecellaxislikeindividualcells,andcell -- cellalignmentisthedominantcauseoftissuespreading(Fig.S3A).Inthatcase,pressureinsteadoftensionstartsbuildingupinthetissuecenterasthecelldensityincreases.Thisbehaviorseemsplausible,becausecellscontinuetoexertpropulsionforcesevenathighcelldensitiesandconsequently,pushagainsttheirneighbors.WecanalsovarytheextentofCIL(SIText).WhenloweringthelevelofCIL,stressinthetissueistensilebutdoesnotincreasetowardthecenter(Fig.S3B).Hence,strongCILisrequiredforthetensionbuildup.Whendecreasingthemaximumcell -- celladhesionforceinouroriginalsimulationtoaverylowvalue,thetissueisunderuniformlowtension,whereasthedensityslightlyincreasesovertime,showingthatincreasingvaluesofthecell -- celladhesionforcede-terminethehightensioninthetissuecenter(Fig.S3D).Moreover,thetissuespreadsfasterbecauseofthelowercelldensityandweakerCIL.Whendecreasingtheintracellularcontractionpa-rameter,thetissueisunabletosupportincreasingtensionathighcelldensity(Fig.S3E).Whenrunningasimulationwithoutcell -- cellalignment,thespreadingvelocityslightlydecreases,butthetensionbuildupisundisturbed(Fig.S3F).Wenextexamineddistributionsofparticletractionforcesandvelocities.Weperformedananalysissimilartothatinref.13,wheredistributionsoftractionstressesweremeasured,andwecalculatedthedistributionofthexcomponentofinstantaneousparticletractionforcesinastripinthecenterofthetissue(Fig.3G).Atlatestagesoftissuespreadingwithhighcelldensity,ourtractionforcedistributionagreeswellwiththemeasureddata.ItisbroaderthanGaussianandwell-fittedbyanexponentialfunction.Attheearlystagesoftissuegrowth,however,the-0.6-0.4-0.2 0 0.2 0.4 0.6-0.8-0.4 0 0.4 0.80.000.020.040.060.080.10 0 100 200 300Average speed vmx 0 0.02 0.04 0.06 0.08 0.1 0 0.2 0.4 0.6 0.8Average speedCell density 0 0.2 0.4 0.6 0 0.2 0.4 0.6 0.8Average intercellular tensionCell density0.00.20.40.60.8 0 100 200 300Average cell densityx10-410-310-2-1.5-1.0-0.50.00.51.01.5ProbabilityTx0100510310103151032010325103 0 500 1000 1500Number of cellsTime0.00.20.40.6 0 100 200 300Average intercellular tension -(cid:86)xxxAEGIxyBx−component of traction stressAverage normal intercellular stressCDHFFig.3.Simulationofaspreadingcellcolony;N=500cellswereseededinthecenterofthecomputationaldomainatt=0andallowedtodivideandmigrateoutward.(A)Totalnumberofcellsinthetissueatdifferenttimepoints.TheredcircleindicatesthatthissimulationtimeframeisshowninBandC.(B)Snapshotofthexcomponentofthetractionstressatt=1,400andRtrac=2.0.(C)Averagenormalstressðσmax+σminÞ=2calculatedwiththeHardymethod(30)att=1,400andRh=2.0.(D)Averagecelldensityasafunctionofthexcoordinate.(E)Averagespeedjvmjasafunctionofthexcoordinate.Thevelocityisaveragedovertrelax=50.(F)Averagetensioninthexdirection−σxxcalculatedbyintegratingaveragetractionstressTx.InD -- F,averagesaretakenalongtheyaxiswithintheregionindicatedbydottedlinesinB.(G)DistributionofthexcomponentofparticletractionforceswithintheregionindicatedinBfort=400,t=800,andt=1,400(colors).Exponentialfitofthedatafort=1,400(solidblackline)andGaussianfitofthedatafort=400(dashedblackline).(H)Averagespeedasafunctionofcelldensity.(I)Averagetensionasafunctionofcelldensity.Colorsindicatedatafromdifferenttimeframes(A).ParametersarethesameasinTableS1.Allunitsaresimulationunits(Fig.S1andMovieS2).Zimmermannetal.PNASMarch8,2016vol.113no.102663CELLBIOLOGYPHYSICSwww.nature.com/scientificreports/2SCIENTIFIC REPORTS(cid:3)(cid:513)(cid:3)(cid:893)(cid:483) 9720 (cid:3)(cid:513)(cid:3)(cid:7)(cid:18)(cid:12)(cid:483)(cid:887)(cid:886)(cid:484)(cid:887)(cid:886)(cid:889)(cid:894)(cid:512)(cid:149)(cid:890)(cid:887)(cid:891)(cid:895)(cid:894)(cid:486)(cid:886)(cid:887)(cid:893)(cid:486)(cid:887)(cid:886)(cid:886)(cid:892)(cid:895)(cid:486)(cid:894)phenomenological descriptions have provided important insights into the generic behaviors of collective cellular movements at length scales much larger than cell size24, 25. Discrete SPP models inspired by flocking or school-ing behavior of animal groups can reproduce coherent collective cell behavior through local velocity alignment rules24, 29. These models have been shown to successfully reproduce important features of large scale collective cell behavior, but do not explain important features of the dynamics of small groups of cells in which the specific characteristics of cellular interactions, including behaviors such as CIL or FIR, may play an important role. In general, SPP models can be used to describe the dynamics of small groups of cells and study the effects of impor-tant cell behaviors and parameters. Indeed, models of SPP have started to explore the role of CIL in the collective dynamics of cells in 2D, but either focus on large 2D monolayers or do not account for FIR30 -- 32. It remains unclear how cell behaviors such as CIL and FIR contribute to collective cell migration, especially for small groups of cells, such as those observed in developing embryos or during cancer metastasis.We introduce a theoretical description that successfully describes the motion of groups of cells of arbitrary numbers, from single cell motion to the collective migration of small groups of cells and large scale sheet migra-tions. The collective dynamics is obtained by balancing the forces in the system and specifying the dynamics of traction forces (or cell polarization) for individual cells, accounting for both CIL and FIR. We show that small groups of cells (3 or more cells) display coherent collective behavior, with persistence times that depend on the group size, despite their effective repulsion during the collision of cells pairs. We find an optimal size for small groups of cells that depends on cellular adhesion and traction strengths and maximizes the persistence of their coherent motion. Beyond small groups of cells, our description reproduces the diffusive behavior of individual cells in the absence of external cues, the observed behaviors upon pairwise cell collisions, as well as the traction force profiles reported in large scale cell migrations. Finally, we show that groups of identical cells can display coherent collective behavior or dispersal behavior by changing their confinement.(cid:23)(cid:138)(cid:135)(cid:145)(cid:148)(cid:135)(cid:150)(cid:139)(cid:133)(cid:131)(cid:142)(cid:3)(cid:7)(cid:135)(cid:149)(cid:133)(cid:148)(cid:139)(cid:146)(cid:150)(cid:139)(cid:145)(cid:144)We seek a minimal theoretical description accounting for key phenomenological observations regarding cell-cell interactions, namely CIL and FIR. To this end, we describe cells as particles and consider the pairwise physical interactions between them when moving along a 1D strip (Fig. 1A). While minimal, the 1D geometry has proven very useful to study collective cell migration at the experimental level13, 14, 33, as it simplifies the system consider-ably while preserving the essential features of collective cell migration.(cid:19)(cid:131)(cid:148)(cid:150)(cid:139)(cid:133)(cid:142)(cid:135)(cid:486)(cid:132)(cid:131)(cid:149)(cid:135)(cid:134)(cid:3)(cid:134)(cid:135)(cid:149)(cid:133)(cid:148)(cid:139)(cid:146)(cid:150)(cid:139)(cid:145)(cid:144)(cid:3)(cid:145)(cid:136)(cid:3)(cid:149)(cid:139)(cid:144)(cid:137)(cid:142)(cid:135)(cid:3)(cid:133)(cid:135)(cid:142)(cid:142)(cid:3)(cid:143)(cid:145)(cid:152)(cid:135)(cid:143)(cid:135)(cid:144)(cid:150)(cid:149)(cid:484)(cid:3)In order to control their movements, cells reg-ulate the forces they apply on their surroundings. A given cell generates a traction force !"T that causes its move-ment. Both dissipative processes inside the cell and friction with the substrate lead to a friction force opposing the cell movement which, in its most basic form, reads ξ−(cid:71)v, with ξ being an effective friction coefficient and (cid:71)v the cell velocity. For the specific case of a single cell, it is instructive to consider also the effect of an external force !"Fext, as previously done experimentally by applying a controlled force with optical or magnetic tweezers15, 16. Neglecting inertial terms, force balance on the cell readsξ=+!"!"!vTF,(1)extFigure 1. Description of the system, interaction forces and phenomenological cell behaviors. (A) Schematic representation of cells moving along a 1D strip (top) and particle-based representation of the system (bottom). Cells can be subject to adhesion forces (orange), excluded volume repulsion forces (blue) and friction forces (green), as well as generate traction forces (red). (B) Schematic representation of lamellipodial ruffluing (right) and a stable lamellipodium (left). (C) Pairwise interaction forces fij between cells as a function of their relative distance. Schematic representation of CIL (D) and FIR (E), leading to an effective repulsion between cells. (F) Schematic representation of neighbor-enabled repolarization (NER). (G) Schematic representation of cellular configurations during collisions and the associated values of the contact matrix Cij for each configuration and cell.Theresultsmaybeinterpretedinbiologicaltermsbyassociatingeachstatetocommonphenotypes,namelygrid-likedistributionsofmesenchymalcells,collectivelymigratingepithelialmonolay-ers,andcellularspheroids.Boththesoftcharacterofthepoten-tialandtheCILinteractionsarekeyinproducingstructuresandcollectivebehaviorsobservedincellcolonies.Inparticular,theformerenablestheformationof3Dtissuesviacellextrusion.Inturn,CILgivesrisetoself-organizedcollectivemotionincontin-uouscellmonolayers.Inlinewithref.17,wefindthatthiseffec-tiverepulsioninducestensilestressesincellmonolayers.ModelWemodela2Dcolonyofcellsasasuspensionofoverdampedself-propelleddisks.Neglectingtranslationalnoise,theequa-tionofmotionofcelliwithpositionxiandpolaritypi=(cos✓i,sin✓i)readsFmpi=sxi+nnXj⇥Fccijnij+(xixj)⇤,[1]forcontactingnearest-neighborcellsj,withnij=(xjxi)/dijanddij=xjxi.Here,Fmisthemagnitudeofthecellself-propulsionforce,andsandarecell -- substrateandcell -- cellfrictionconstants,respectively.ThecentralforceFccijincludesasoftrepulsionFrijassoci-atedtothereductionofthecell -- substrateadhesionareawhentwocellsarecloserthantheirspreadsize2R,andanattrac-tiveforceFaijthataccountsforactivecontractilitytransmittedthroughcell -- celladhesions.Frijisassumedtoincreaselin-earlywithdecreasingintercellulardistancedijuptodij=R.Hence,Frij=2Ws/R2(2Rdij),withWs=R2RRFrijddijthecell -- substrateadhesionenergy(grayinFig.1A).Nofurtherreductionofthecell -- substratecontactareaisallowedfordij<R.Asaresult,cellscanapproachatsmallerdistancesundercompression.Inthisregimecellsdonotexertanyforceonthesubstrateandareconsideredtobeextrudedfromthemonolayer(Fig.1AandB).Cellextrusionsmayleadto3Dtissues,whosestructureanddynamicsarenotdescribedbyour2Dmodel.Faijisassumedtoincreaselinearlywithdistanceuptodij=2R.Hence,Faij=2Wc/R2(dijR),withWc=R2RRFaijddijthecell -- celladhesionenergy(redinFig.1A).Accordingly,thetotalinterac-tionforce(blackinFig.1A)readsFccij(dij)=⇢2R[WsWs+WcR(dijR)],ifRdij2R0,else.[2]Inturn,CILtendstoorientthecellpolaritypiinthedirectionpfipointingawayfromtheweightedaveragepositionofthecon-ABCFig.1.Amodelofself-propelledparticleswithcell-likeinteractions.(A)Centralcell -- cellforceFccij(black),includingasoftrepulsionduetoreduc-tionofcell -- substrateadhesionarea(gray)andattractionduetoactivecon-tractilitythroughcell -- celladhesions(red).(B)Cellextrusionforintercellulardistancesdij<R,resultinginvanishingcell -- cellforcesintheplane.(C)Cellu-larself-propulsionforceFminthedirectionofthecellpolaritypi.CILrotatesthepolaritytowardthedirectionpfipointingawayfromcell -- cellcontacts.Fig.2.Phasebehaviorofcellcoloniesasafunctionofcell -- celladhesionWcandcellrepolarizationrate associatedtoCIL.Colorsindicatethepredictedregionsfornoncohesive(green),cohesive(blue),andoverlapped(red)orga-nizations.Inadditiontocapturingthesestructuraltransitions,simulationsallowustoidentifydynamicallydistinctstatessuchasanactivegas,aclus-tercrystal,agel-likepercolatednetwork,dynamicclusters,andanactivepolarliquid,asillustratedinsnapshots.tactingcells(Fig.1CandSIAppendix).Similarlytoref.18,wemodelthisinteractionviaaharmonicpotentialforthepolariza-tionangle✓ithat,inadditiontorotationalnoise,yields✓i=fcil(✓i✓fi)+p2Dr⇠.[3]Here,fcilisthecellularrepolarizationrateuponcell -- cellcon-tact,whereas⇠(t)isatypifiedGaussianwhitenoise,andDristherotationaldiffusioncoefficientofcellmotion.Theparametersofthemodelmaybereducedtofivedimen-sionlessquantities:thepackingfractionofcells,cell -- cellandcell -- substrateadhesionenergiesWc:=Wc/(2RFm)andWs:=Ws/(2RFm),cell -- cellfriction:=/s,andaparam-eter :=fcil/(2Dr)thatcomparesthetimescaleofcytoskele-talrepolarizationassociatedtoCILtotherotationaldiffusion.Hereafter,weset=0.85,Ws=1,and=0andfocusontheeffectsofintercellularadhesionandCILontheorganizationofcellcolonies.TheresultsaresummarizedinthephasediagraminFig.2.Includingcell -- cellfrictionleadstojammedconfigurationsofcohesivetissues(SIAppendix),inlinewithref.16.Inturn,celldensitydoesnotaffectthephasetransitionsbutmodifiesthedynamicalbehaviorofthecellcolony(SIAppendix).Thus,cellproliferationmaydrivethecolonythroughdifferentdynamicalstates(SIAppendix).ResultsNoncohesivePhase.Wefirststudythetransitionbetweenacohe-sivephaseinwhichcellsremainincontact,dij<2R,andanoncohesivephaseinwhichtheylosecontact.Lossofcell14622www.pnas.org/cgi/doi/10.1073/pnas.1521151113Smeetsetal.CIL repolarizationCollective motionUponphaseseparation,thecolonyformsacontinuouscellmonolayerthatexhibitsself-organizedcollectivemotion(MovieS5).ThisisreflectedintheMSDexponentthatevolvesfromdif-fusive(↵=1)towardalmostballistic(↵=2)aboveWc⇡0.4(Fig.4A).CILinducesacouplingbetweencellpolarityandden-sityfluctuationsinthefluidphasethatgivesrisetoamacroscopicpolarizationviaaspontaneoussymmetrybreaking.Theoutwardmotionofcellsattheboundaryofthemonolayercreatesfreespacebehindthem,whichpolarizesneighboringcellsbeforetheleadingcellcanreorientback.Throughthismechanism,self-organizedcollectivecellmotionemergesfromCIL,leadingtoanactivepolarliquidstate.Thepolarorderisstableiftheconfinementimposedbyneigh-borsrestoresthepositionandorientationofacellbeforeitspolar-ityturnstowardanewfreedirection.Therepolarizationoccurswithinatimescale1/fcil,andthecharacteristictimeofpositionrelaxationinadenseenvironmentis⇠s/k,withk=4(Ws+Wc)/R2thestiffnessofatwo-neighborconfinement.Thus,anapproximatestabilitycriterionreadss/k.f1cil,whichissatisfiedforthewholeparameterrangeinFig.2(SIAppendix).AsillustratedinFig.5A,isolatedfluidmonolayersmayacquireaglobalpolarityandconsequentlyperformpersistentrandomwalkswithapersistencemuchlargerthanthatofsinglecells(MovieS6).Forrandomlyorientedcells,theaveragepolarityofNcellsscalesasPN=PNi=1pi/N⇠N1/2.Ifcellpolaritiesalign,theaveragepolarityofasmallregionofcellsdecreasesslowerwithitssize,sothatpNPN>1.Thelargertherepolar-izationrate is,thefastertheincreaseofpolaritywithN(SIAppendix).Atsufficientlylargesizes,multiplemisalignedpolar-itydomainsappearthatrestoretherandomscaling(Fig.5B).Hence,wedefinetheonsetofmacroscopicpolarization(cir-clesinFig.2)bytheconditionthatpNPNhasamaximumatN=75,namelythatconnectedclustersconsistingofupto75cellsmayformasinglepolaritydomain.TheappropriatechoiceofNdependsonsystemsize.However,forthesizesexplored,thetransitionline(circlesinFig.2)ishardlysensitivetoval-uesaroundN=75(SIAppendix).Inconclusion,byensuringacompletephaseseparationwhilestillallowingforcellrearrange-ments,sufficientlystrongcell -- celladhesionandCILarerequiredtoformapolar,collectivelymovingcellmonolayer.Finally,theeffectivepotentialenergyEpofcell -- cellinterac-tionsgivesinformationonthemechanicsofthecolony.Positive(negative)potentialenergiescorrespondtotensile(compres-sive)intercellularstresses.Noncohesivecoloniesatlowcell -- cellABCDFig.5.Collectivemotion,mechanics,anddewettingofcellmonolayers.(A)Snapshotofagloballypolarized,collectivelymigratingcellmonolayer.(B)RescaledaveragepolaritypNPNofamonolayerofNcellsfordifferentCILrepolarizationrates atacell -- celladhesionWc=0.7.pNPN=1cor-respondstorandomlyorientedcells.CILinducesaglobalpolarity(pNPN>1)thatgivesrisetocollectivemotion.Theappearanceofseveralpolaritydomainsreducestheaveragepolarityoflargecellgroups.Thetransitiontotheactivepolarliquidstate(circlesinFig.2)isdefinedbytheconditionthatthemaximumofpNPNisatN=75.(C)Averagecell -- cellpotentialenergyEp=Ep/(2RFm)asafunctionofcell -- celladhesionWcandCILrepolarizationrate .CIL-associatedrepulsioninducestensilestresses(Ep>0)incellmonolayers.(D)Averagedistancebetweencontactingcellshdijic=⌦dij↵c/(2R)asafunctionofWcand .Thetransitionbetweencellmonolayersand3Daggregatesispredictedtooccuratavanishingaveragecell -- cellforce(dashedline)andisidentifiedbythecondition⌦dij↵c=3R/2(solidline,crossesinFig.2).adhesionfeatureaverageattractiveinteractionsleadingtotheformationofclusters.Inturn,bypolarizingbordercellsout-ward,CILinducestensilestressesincellmonolayers(Fig.5C),inagreementwithref.17.OverlappedPhase.Wefinallyfocusonthetransitionto3Dtis-sues.Whentheaveragetotalcell -- cellforceisattractive,cellseventuallyovercometheenergybarrierassociatedtothesoftrepulsivepotential(Fig.1A),whichcorrespondstocellextrusionevents.ExtrudedcellsareconfinedatdistancessmallerthanR,wheretheyexertneithercell -- cellnortractionforces.Thus,ourmodelcanpredicttheonsetofthetransitionto3Dcellarrange-ments.Assumingahomogeneousdistributionofcells,andusingEq.2,theaverageinteractionforcereads⌦Fccij↵=R2RR2⇡dijFccijddijR2RR2⇡dijddij=29R(4Ws5Wc).[6]ThisforceaddstotheeffectiverepulsionFpijassociatedtoantialignedself-propulsionforces(Eq.4),sothatthetransitionbetweenmonolayers(blueinFig.2)and3Dcellarrangements(redinFig.2)ispredictedbythecondition⌦Fccij↵+Fpij=0.Thissetsacriticalcell -- celladhesionenergyW3Dc=154Ws+94exp✓14 ◆,[7]abovewhichcellsareexpectedtofullyoverlapor,alternatively,acriticalCILrepolarizationrateabovewhichcellextrusionisprevented.Therefore,byopposingcellextrusion,CILhindersthecollapseofcellmonolayersinto3Daggregates.Indeed,asufficientlyfastrepolarizationofcellmotilitymaystabilizecellmonolayersevenwhencell -- celladhesionisstrongerthancell -- substrateadhesion,Wc>Ws=1(Fig.2).Insimulations,wecharacterizethedegreeofcelloverlapintermsoftheaveragedistancebetweencontactingcellshdijic(Fig.5D).Wethenidentifythetransitionwhenhalfofthecontact-ingcellsareatthecriticaldistanceforextrusion,dij=R,whiletheotherhalfarefullyspread,dij=2R.Hence,thetransitionisdefinedbyhdijic=12R+122R=3R/2(crossesinFig.2),inqual-itativeagreementwiththemean-fieldanalyticalprediction.Monolayerinstabilityoccursthroughadewettingprocesswherebyholesappearinthecellmonolayer,whichrapidlyevolvesintoanetworkstructure,asobservedinref.34.Sub-sequently,differentregionsofthenetworkslowlycollapseinto14624www.pnas.org/cgi/doi/10.1073/pnas.1521151113Smeetsetal.Uponphaseseparation,thecolonyformsacontinuouscellmonolayerthatexhibitsself-organizedcollectivemotion(MovieS5).ThisisreflectedintheMSDexponentthatevolvesfromdif-fusive(↵=1)towardalmostballistic(↵=2)aboveWc⇡0.4(Fig.4A).CILinducesacouplingbetweencellpolarityandden-sityfluctuationsinthefluidphasethatgivesrisetoamacroscopicpolarizationviaaspontaneoussymmetrybreaking.Theoutwardmotionofcellsattheboundaryofthemonolayercreatesfreespacebehindthem,whichpolarizesneighboringcellsbeforetheleadingcellcanreorientback.Throughthismechanism,self-organizedcollectivecellmotionemergesfromCIL,leadingtoanactivepolarliquidstate.Thepolarorderisstableiftheconfinementimposedbyneigh-borsrestoresthepositionandorientationofacellbeforeitspolar-ityturnstowardanewfreedirection.Therepolarizationoccurswithinatimescale1/fcil,andthecharacteristictimeofpositionrelaxationinadenseenvironmentis⇠s/k,withk=4(Ws+Wc)/R2thestiffnessofatwo-neighborconfinement.Thus,anapproximatestabilitycriterionreadss/k.f1cil,whichissatisfiedforthewholeparameterrangeinFig.2(SIAppendix).AsillustratedinFig.5A,isolatedfluidmonolayersmayacquireaglobalpolarityandconsequentlyperformpersistentrandomwalkswithapersistencemuchlargerthanthatofsinglecells(MovieS6).Forrandomlyorientedcells,theaveragepolarityofNcellsscalesasPN=PNi=1pi/N⇠N1/2.Ifcellpolaritiesalign,theaveragepolarityofasmallregionofcellsdecreasesslowerwithitssize,sothatpNPN>1.Thelargertherepolar-izationrate is,thefastertheincreaseofpolaritywithN(SIAppendix).Atsufficientlylargesizes,multiplemisalignedpolar-itydomainsappearthatrestoretherandomscaling(Fig.5B).Hence,wedefinetheonsetofmacroscopicpolarization(cir-clesinFig.2)bytheconditionthatpNPNhasamaximumatN=75,namelythatconnectedclustersconsistingofupto75cellsmayformasinglepolaritydomain.TheappropriatechoiceofNdependsonsystemsize.However,forthesizesexplored,thetransitionline(circlesinFig.2)ishardlysensitivetoval-uesaroundN=75(SIAppendix).Inconclusion,byensuringacompletephaseseparationwhilestillallowingforcellrearrange-ments,sufficientlystrongcell -- celladhesionandCILarerequiredtoformapolar,collectivelymovingcellmonolayer.Finally,theeffectivepotentialenergyEpofcell -- cellinterac-tionsgivesinformationonthemechanicsofthecolony.Positive(negative)potentialenergiescorrespondtotensile(compres-sive)intercellularstresses.Noncohesivecoloniesatlowcell -- cellABCDFig.5.Collectivemotion,mechanics,anddewettingofcellmonolayers.(A)Snapshotofagloballypolarized,collectivelymigratingcellmonolayer.(B)RescaledaveragepolaritypNPNofamonolayerofNcellsfordifferentCILrepolarizationrates atacell -- celladhesionWc=0.7.pNPN=1cor-respondstorandomlyorientedcells.CILinducesaglobalpolarity(pNPN>1)thatgivesrisetocollectivemotion.Theappearanceofseveralpolaritydomainsreducestheaveragepolarityoflargecellgroups.Thetransitiontotheactivepolarliquidstate(circlesinFig.2)isdefinedbytheconditionthatthemaximumofpNPNisatN=75.(C)Averagecell -- cellpotentialenergyEp=Ep/(2RFm)asafunctionofcell -- celladhesionWcandCILrepolarizationrate .CIL-associatedrepulsioninducestensilestresses(Ep>0)incellmonolayers.(D)Averagedistancebetweencontactingcellshdijic=⌦dij↵c/(2R)asafunctionofWcand .Thetransitionbetweencellmonolayersand3Daggregatesispredictedtooccuratavanishingaveragecell -- cellforce(dashedline)andisidentifiedbythecondition⌦dij↵c=3R/2(solidline,crossesinFig.2).adhesionfeatureaverageattractiveinteractionsleadingtotheformationofclusters.Inturn,bypolarizingbordercellsout-ward,CILinducestensilestressesincellmonolayers(Fig.5C),inagreementwithref.17.OverlappedPhase.Wefinallyfocusonthetransitionto3Dtis-sues.Whentheaveragetotalcell -- cellforceisattractive,cellseventuallyovercometheenergybarrierassociatedtothesoftrepulsivepotential(Fig.1A),whichcorrespondstocellextrusionevents.ExtrudedcellsareconfinedatdistancessmallerthanR,wheretheyexertneithercell -- cellnortractionforces.Thus,ourmodelcanpredicttheonsetofthetransitionto3Dcellarrange-ments.Assumingahomogeneousdistributionofcells,andusingEq.2,theaverageinteractionforcereads⌦Fccij↵=R2RR2⇡dijFccijddijR2RR2⇡dijddij=29R(4Ws5Wc).[6]ThisforceaddstotheeffectiverepulsionFpijassociatedtoantialignedself-propulsionforces(Eq.4),sothatthetransitionbetweenmonolayers(blueinFig.2)and3Dcellarrangements(redinFig.2)ispredictedbythecondition⌦Fccij↵+Fpij=0.Thissetsacriticalcell -- celladhesionenergyW3Dc=154Ws+94exp✓14 ◆,[7]abovewhichcellsareexpectedtofullyoverlapor,alternatively,acriticalCILrepolarizationrateabovewhichcellextrusionisprevented.Therefore,byopposingcellextrusion,CILhindersthecollapseofcellmonolayersinto3Daggregates.Indeed,asufficientlyfastrepolarizationofcellmotilitymaystabilizecellmonolayersevenwhencell -- celladhesionisstrongerthancell -- substrateadhesion,Wc>Ws=1(Fig.2).Insimulations,wecharacterizethedegreeofcelloverlapintermsoftheaveragedistancebetweencontactingcellshdijic(Fig.5D).Wethenidentifythetransitionwhenhalfofthecontact-ingcellsareatthecriticaldistanceforextrusion,dij=R,whiletheotherhalfarefullyspread,dij=2R.Hence,thetransitionisdefinedbyhdijic=12R+122R=3R/2(crossesinFig.2),inqual-itativeagreementwiththemean-fieldanalyticalprediction.Monolayerinstabilityoccursthroughadewettingprocesswherebyholesappearinthecellmonolayer,whichrapidlyevolvesintoanetworkstructure,asobservedinref.34.Sub-sequently,differentregionsofthenetworkslowlycollapseinto14624www.pnas.org/cgi/doi/10.1073/pnas.1521151113Smeetsetal.Uponphaseseparation,thecolonyformsacontinuouscellmonolayerthatexhibitsself-organizedcollectivemotion(MovieS5).ThisisreflectedintheMSDexponentthatevolvesfromdif-fusive(↵=1)towardalmostballistic(↵=2)aboveWc⇡0.4(Fig.4A).CILinducesacouplingbetweencellpolarityandden-sityfluctuationsinthefluidphasethatgivesrisetoamacroscopicpolarizationviaaspontaneoussymmetrybreaking.Theoutwardmotionofcellsattheboundaryofthemonolayercreatesfreespacebehindthem,whichpolarizesneighboringcellsbeforetheleadingcellcanreorientback.Throughthismechanism,self-organizedcollectivecellmotionemergesfromCIL,leadingtoanactivepolarliquidstate.Thepolarorderisstableiftheconfinementimposedbyneigh-borsrestoresthepositionandorientationofacellbeforeitspolar-ityturnstowardanewfreedirection.Therepolarizationoccurswithinatimescale1/fcil,andthecharacteristictimeofpositionrelaxationinadenseenvironmentis⇠s/k,withk=4(Ws+Wc)/R2thestiffnessofatwo-neighborconfinement.Thus,anapproximatestabilitycriterionreadss/k.f1cil,whichissatisfiedforthewholeparameterrangeinFig.2(SIAppendix).AsillustratedinFig.5A,isolatedfluidmonolayersmayacquireaglobalpolarityandconsequentlyperformpersistentrandomwalkswithapersistencemuchlargerthanthatofsinglecells(MovieS6).Forrandomlyorientedcells,theaveragepolarityofNcellsscalesasPN=PNi=1pi/N⇠N1/2.Ifcellpolaritiesalign,theaveragepolarityofasmallregionofcellsdecreasesslowerwithitssize,sothatpNPN>1.Thelargertherepolar-izationrate is,thefastertheincreaseofpolaritywithN(SIAppendix).Atsufficientlylargesizes,multiplemisalignedpolar-itydomainsappearthatrestoretherandomscaling(Fig.5B).Hence,wedefinetheonsetofmacroscopicpolarization(cir-clesinFig.2)bytheconditionthatpNPNhasamaximumatN=75,namelythatconnectedclustersconsistingofupto75cellsmayformasinglepolaritydomain.TheappropriatechoiceofNdependsonsystemsize.However,forthesizesexplored,thetransitionline(circlesinFig.2)ishardlysensitivetoval-uesaroundN=75(SIAppendix).Inconclusion,byensuringacompletephaseseparationwhilestillallowingforcellrearrange-ments,sufficientlystrongcell -- celladhesionandCILarerequiredtoformapolar,collectivelymovingcellmonolayer.Finally,theeffectivepotentialenergyEpofcell -- cellinterac-tionsgivesinformationonthemechanicsofthecolony.Positive(negative)potentialenergiescorrespondtotensile(compres-sive)intercellularstresses.Noncohesivecoloniesatlowcell -- cellABCDFig.5.Collectivemotion,mechanics,anddewettingofcellmonolayers.(A)Snapshotofagloballypolarized,collectivelymigratingcellmonolayer.(B)RescaledaveragepolaritypNPNofamonolayerofNcellsfordifferentCILrepolarizationrates atacell -- celladhesionWc=0.7.pNPN=1cor-respondstorandomlyorientedcells.CILinducesaglobalpolarity(pNPN>1)thatgivesrisetocollectivemotion.Theappearanceofseveralpolaritydomainsreducestheaveragepolarityoflargecellgroups.Thetransitiontotheactivepolarliquidstate(circlesinFig.2)isdefinedbytheconditionthatthemaximumofpNPNisatN=75.(C)Averagecell -- cellpotentialenergyEp=Ep/(2RFm)asafunctionofcell -- celladhesionWcandCILrepolarizationrate .CIL-associatedrepulsioninducestensilestresses(Ep>0)incellmonolayers.(D)Averagedistancebetweencontactingcellshdijic=⌦dij↵c/(2R)asafunctionofWcand .Thetransitionbetweencellmonolayersand3Daggregatesispredictedtooccuratavanishingaveragecell -- cellforce(dashedline)andisidentifiedbythecondition⌦dij↵c=3R/2(solidline,crossesinFig.2).adhesionfeatureaverageattractiveinteractionsleadingtotheformationofclusters.Inturn,bypolarizingbordercellsout-ward,CILinducestensilestressesincellmonolayers(Fig.5C),inagreementwithref.17.OverlappedPhase.Wefinallyfocusonthetransitionto3Dtis-sues.Whentheaveragetotalcell -- cellforceisattractive,cellseventuallyovercometheenergybarrierassociatedtothesoftrepulsivepotential(Fig.1A),whichcorrespondstocellextrusionevents.ExtrudedcellsareconfinedatdistancessmallerthanR,wheretheyexertneithercell -- cellnortractionforces.Thus,ourmodelcanpredicttheonsetofthetransitionto3Dcellarrange-ments.Assumingahomogeneousdistributionofcells,andusingEq.2,theaverageinteractionforcereads⌦Fccij↵=R2RR2⇡dijFccijddijR2RR2⇡dijddij=29R(4Ws5Wc).[6]ThisforceaddstotheeffectiverepulsionFpijassociatedtoantialignedself-propulsionforces(Eq.4),sothatthetransitionbetweenmonolayers(blueinFig.2)and3Dcellarrangements(redinFig.2)ispredictedbythecondition⌦Fccij↵+Fpij=0.Thissetsacriticalcell -- celladhesionenergyW3Dc=154Ws+94exp✓14 ◆,[7]abovewhichcellsareexpectedtofullyoverlapor,alternatively,acriticalCILrepolarizationrateabovewhichcellextrusionisprevented.Therefore,byopposingcellextrusion,CILhindersthecollapseofcellmonolayersinto3Daggregates.Indeed,asufficientlyfastrepolarizationofcellmotilitymaystabilizecellmonolayersevenwhencell -- celladhesionisstrongerthancell -- substrateadhesion,Wc>Ws=1(Fig.2).Insimulations,wecharacterizethedegreeofcelloverlapintermsoftheaveragedistancebetweencontactingcellshdijic(Fig.5D).Wethenidentifythetransitionwhenhalfofthecontact-ingcellsareatthecriticaldistanceforextrusion,dij=R,whiletheotherhalfarefullyspread,dij=2R.Hence,thetransitionisdefinedbyhdijic=12R+122R=3R/2(crossesinFig.2),inqual-itativeagreementwiththemean-fieldanalyticalprediction.Monolayerinstabilityoccursthroughadewettingprocesswherebyholesappearinthecellmonolayer,whichrapidlyevolvesintoanetworkstructure,asobservedinref.34.Sub-sequently,differentregionsofthenetworkslowlycollapseinto14624www.pnas.org/cgi/doi/10.1073/pnas.1521151113Smeetsetal.RAPIDCOMMUNICATIONSSILKEHENKES,YAOUENFILY,ANDM.CRISTINAMARCHETTIPHYSICALREVIEWE84,040301(R)(2011)FIG.1.(Coloronline)Samplesnapshotsofthesystemintheliquidphase(φ=0.6,left)andinthejammedphase(φ=0.95,right)forv0=0.025.Thegluedboundaryisshownindarkgray.Thered(darkgray)arrowsrepresenttheinstantaneousvelocityfield,withv=v0correspondingtoanarrowoflength1inunitsoftheparticlediameter.PleaseseeSupplementalMaterial[23]formoviesofthesetworuns.dynamics[24].Suchpolarizationisnot,however,permanent,butratheritisactivelyregulatedbybothbiochemicalpro-cessesinsidethecellandfeedbackfromneighboringcells.Finally,wedefinedimensionlessquantitiesbyscalingalllengthswiththeaverageradiusaofthespheresandalltimeswiththelagtimeτ.Additionally,wefixµk=10andσ=10−1.Usingthismodel,weperformmoleculardynamicssim-ulationswithNt=64to10000particles.Unlessotherwisespecified,weshowresultsforNt=1000particles.Toelimi-natetheglobaltranslationalmodeobtainedathighdensityinanopensystem,weconfinetheparticlestoacircularboxofradiusRwithsoftrepulsiveboundaryconditions.Theseareimplementedby"gluing"arowofsoftspherestothebox'sboundary,asshowninFig.1.Weexplorethephasediagrambyvaryingtheself-propulsionspeedv0andthepackingfractionφ=!ia2i/R2.Wefirstcharacterizethestateofthesystembystudyingthemean-squaredisplacement(MSD)ofindividualparticlesasafunctionoftime,shownontheleftsideofFig.2.Atlowpackingfractionorhighvelocity,theMSDgrowsmonotonicallywellbeyonda,correspondingtoaflowingFIG.2.(Coloronline)Left:Mean-squaredisplacementvstimeasafunctionofdensityatv0=0.025,showingatransitionfromrotationaldiffusionatlowφ,topolaralignmentforφ<0.8andtothejammedstatearoundφ=0.842.Right:Phasediagramintheφ-v0plane,showingthetransitionfromtheliquidstate(blueorlightgray)tothesolidstate(redordarkgray).Thedotsaresimulated(φ,v0)pairsshadedwhitetoblackproportionaltothefractionofjammedruns.system.Conversely,athighφorlowv0,theMSDisboundedandsmallerthana,i.e.,theparticlesaretrappedinthecageformedbytheirneighbors.TypicalsnapshotsofthesystemineachphaseareshowninFig.1.Atv0=0,theangulardegreeoffreedomψibecomesirrelevantandtheproblemisequivalenttotheathermaljammingofsoftspheres.Thetransitionbetweenaflowingphaseandatrappedoneatverylowv0isconsistentwiththislimit;inparticular,thecriticalpackingfractioncoincideswiththeexpectedvalueφc≈0.842.Byextension,wecallthetwoactivephases"liquid"and"jammed,"respectively.Theshapeofthe(φ,v0)phasediagramasinferredfromtheMSDisshownonFig.2(right).TheliquidphasecanbefurtherdividedbynotingthatthebehavioroftheMSDisnotuniform.Atverylowdensity,interactionsarenegligibleandeachparticleinde-pendentlyperformsapersistentrandomwalk,with⟨[r(t)−r(0)]2⟩=(4v20/σ2)[t+(2/σ2)(e−σ2t/2−1)]andacrossoverfromballisticbehavior⟨[r(t)−r(0)]2⟩∼v20t2fort≪σ−2todiffusivebehavior⟨[r(t)−r(0)]2⟩∼(4v20/σ2)tfort≫σ−2.Here,σ−2=102andballisticbehaviorisobservedatallbutthelongesttimes(butshorterthanthelimitimposedbytheboxsize,notshownonFig.2),asexpectedforindividualself-propelledparticles[25].Atintermediatedensity,however,clustersofalignedparticlesstarttoformandtheMSDremainsballisticatallobservedtimes.Thisbehaviorisreminiscentofthoseobservedinotheractivesystems[10,11].Anothersignatureofthesymmetrybreakingintroducedbytheactivevelocityintheliquidphaseistheexistenceof"giantnumberfluctuations"[6,7,12,26].ThescalingofthestandarddeviationNofthenumberofparticleswiththeaveragenumberofparticlesNinsubsystemsofvarioussizesisshowninFig.3.WeseeatransitionfromN∼N1/2,asexpectedinanidealgasorinapassivethermalliquid,toN∼Nαwithα>1/2atpackingfractionφ∼0.5,consistentwiththechangeofbehaviorobservedintheMSDandwithpreviousobservationsonself-propelledsystems[6,7,12,26].Inthejammedphase,weobserveregularoscillationsoftheparticledisplacementsaroundtheirmeanpositions,resultingFIG.3.(Coloronline)ScalednumberfluctuationsforNt=10000andacutthroughthephasediagramatv0=0.025.Weobservethreeregimes:gaslikefluctuationsatlowdensity(green,topthreecurves),giantnumberfluctuationsatintermediatedensity(red,middlefivecurves),andstronglysuppressedfluctuationsinthejammedphase(blue,bottomcurves).ThedashedlinecorrespondstoN/N1/2=1.040301-2Active polar fluidJammedCell-cell distanceForce3.4.2 Cell migration and force balance As in other approaches, cell motility is often accounted for by an active polar force Ta(cid:126)pi on the particles. In addition to central and self-propulsion forces, cell-substrate viscous friction −ξ(cid:126)vi and cell-cell friction with coefficient ξc can also be added to the force balance (Fig. 6a). Thus, the equation of motion of cell i = 1, . . . , N reads ξ(cid:126)vi = Ta(cid:126)pi +(cid:88)(cid:104)i,j(cid:105)(cid:104)−(cid:126)∇(cid:126)riV ((cid:126)ri − (cid:126)rj) + ξc[(cid:126)vi − (cid:126)vj](cid:105) . (10) Here, the sum is restricted to contacting cells. In addition, to account for interfacial phenomena, Tarle et al. added surface tension along the tissue edge as well as a cou- pling of the motility force to the edge curvature [125]. 3.4.3 Polarity dynamics A wide range of polarity interactions have been studied in particle models. Interest- ingly, as in phase-field and SPV models, the combination of short-range forces with autonomous polarity-velocity alignment (Section 2.4.1) can lead to cell-cell velocity alignment and flocking [121, 126]. Nevertheless, other studies have explicitly imple- mented Vicsek-like velocity alignment rules [127]. Other polarity interactions that have been modeled include CIL [45,46,76,122,123,128 -- 130], CFL [45], and force-induced repolarization [46] (Section 2.3.2). 3.4.4 Collective phenomena Particle models have unveiled that not only polarity-velocity alignment but also CRL interactions can give rise to collective motion [45, 46, 123, 129]. In particular, because it tends to anti-align the polarities of cell pairs, CIL would not be expected to lead to a state with net polarity. However, in cell clusters, CIL induces a coupling between the polarity and density fields that gives rise to a spontaneous symmetry breaking to- wards collective motion [123] (Fig. 6b). Moreover, adjusting CIL strength to the local concentration of a chemoattractant enables intrinsically collective modes of chemo- taxis [131]. Particle models have also shown that CIL can stabilize cell monolayers against dewetting [123] and ensure tensile intercellular stresses during tissue spread- ing [76] (Fig. 6c). Finally, jamming due to increasing cell density and friction has also been studied using particle models [32, 124]. As in the SPV model, jammed packings of self-propelled particles exhibit collective oscillations [124] (Fig. 6d). 3.4.5 Discussion On the one hand, particle models miss details related to cell shape and its coupling to polarity, which are relevant for some aspects of epithelial dynamics. On the other hand, the particle description is suited to study collective migration not only of epithe- lial but also of mesenchymal cells, which is coordinated by weak and transient cell-cell contacts [132]. Moreover, the description of cell-cell interactions in terms of a poten- tial allows to compute the tissue stress tensor, thus enabling to study how and which interactions determine the tensile mechanical state of a tissue [46, 76, 123]. Finally, in- cluding active cell-cell forces in the force balance Eq. (10) is possible and it may lead to important insights in the future. 17 3.5 Continuum models Continuum models do not describe individual cells but set the coarse-graining level at the multicellular scale [133]. In this approach, the cell colony is described by fields such as velocity (cid:126)v((cid:126)r, t), polarity (cid:126)p((cid:126)r, t), and cell density ρ((cid:126)r, t) that locally average these variables over many cells. Then, generic dynamical equations for the fields can be written based on the principles of hydrodynamic descriptions, observing symmetries and conservation laws. Here, we review how the dynamics of compressible polar media may be applied to model collective cell migration in epithelial monolayers. Because they contain many terms, the general equations are usually simplified to include only a few effects that are deemed most important for a particular phenomenon. We discuss the most common simplifications. 3.5.1 Free energy of compressible polar media The starting point is the free energy of quiescent compressible polar media [134 -- 136], which reads D K 2 (δρ)2 + a 2(cid:126)p2 + F =(cid:90)A(cid:20) κ 2 (cid:126)∇(cid:126)p : (cid:126)∇(cid:126)p + b 4(cid:126)p4 + w δρ (cid:126)∇ · (cid:126)p + 2 (cid:126)∇ρ2(cid:21) d2(cid:126)r. (11) The first term penalizes density variations δρ((cid:126)r, t) = ρ((cid:126)r, t) − ρ0 around ρ0 with a bulk modulus κ (Section 2.2.3). The second and third terms correspond to a Landau expansion on the polarity field. The non-polarized and polarized state are stable for a > 0 and a < 0, respectively, with b > 0 for stability. The fourth term couples the density and the polarity fields. In equilibrium, this term contributes a polarity (cid:126)p ∝ w (cid:126)∇ρ, pointing towards increasing or decreasing density for w > 0 and w < 0, respectively. Thus, with w < 0, this term may model interactions like CIL [137] (Section 2.3.2), which favor cell motility away from dense regions. Finally, the last two terms penalize spatial variations of the density and polarity fields, thus endowing them with a finite correlation length. In fact, the fifth term corresponds to the orientational Frank elasticity of liquid crystals in the so-called one constant approximation, which assumes that bend and splay deformations have a common modulus K [138]. This term captures polarity alignment interactions between cells (Section 2.3.1). 3.5.2 Density and polarity dynamics Then, one writes down dynamical equations. First, cell number balance is imposed by means of a continuity equation for the density field: ∂tρ + (cid:126)∇ · (ρ(cid:126)v) = k(ρ)ρ, (12) where k(ρ) is the net cell proliferation rate combining cell divisions and deaths [139]. Second, the long-wavelength dynamics of the polarity field is given by the theory of polar media, (cid:126)v, (13) D(cid:126)p Dt = 1 γ (cid:126)h − ¯ν 2 ((cid:126)∇ · (cid:126)v) (cid:126)p − ν v · (cid:126)p + νs γ where we have neglected higher-order active terms [6 -- 8]. Here, the corotational deriva- tive of a vector (cid:126)A reads D (cid:126)A/Dt = (∂t +(cid:126)v· (cid:126)∇) (cid:126)A + ω· (cid:126)A, where ω = ((cid:126)∇(cid:126)v− ((cid:126)∇(cid:126)v)T )/2 is the vorticity tensor. This derivative accounts for the advective and co-rotational transport of the polarity field. In the first term on the right-hand side of Eq. (13), the 18 so-called molecular field (cid:126)h = −δF/δ(cid:126)p is the generalized force (torque) acting on the polarity field to minimize the free energy F. The ensuing polarity changes are damped by the rotational friction γ, which may capture dissipation due to both cell-substrate friction (Section 2.1.2) and cytoskeleton reorganizations. The following two terms ex- press the couplings of the polarity to bulk and shear flows, with coefficients ¯ν and ν, respectively. Bulk flows are described by the velocity divergence (cid:126)∇ · (cid:126)v, whereas shear flows are described by the symmetric and traceless part of the strain rate tensor, v = ((cid:126)∇(cid:126)v + ((cid:126)∇(cid:126)v)T − (cid:126)∇·(cid:126)v I)/2. These terms might capture the tendency of the polarity to align with normal stresses [64] (plithotaxis, Section 2.3.3). Finally, the last term couples the polarity to uniform flows, with coefficient νs. This coupling might capture polarity-velocity alignment interactions (Section 2.4.1), but it has not been considered yet in continuum models of collective cell migration. 3.5.3 Force balance In addition to the dynamical equations Eqs. (12) and (13), force balance is established between the internal forces in the tissue, given in terms of its stress tensor σ((cid:126)r, t), and cell-substrate (traction) forces (cid:126)T ((cid:126)r, t), (cid:126)∇ · σ = (cid:126)T ; σ = −P I + σs + σa. (14) It is convenient to separate the stress tensor into the pressure P and the deviatoric stress with symmetric and antisymmetric parts σs and σa = 1/2 ((cid:126)p (cid:126)h − (cid:126)h (cid:126)p). The pressure can be computed via the Gibbs-Duhem thermodynamic relation: P = µρ − f, where µ = δF/δρ is the chemical potential and f is the free energy density, namely the integrand of Eq. (11). Then, the key modeling step is to specify constitutive equa- tions that relate the deviatoric stress tensor σs and the traction forces (cid:126)T to the velocity, polarity, and density fields, thus phenomenologically capturing cellular interactions at a coarse-grained level. Given the tissue rheology, the theory of active polar media provides generic constitutive equations [6 -- 8], which we review in the following two subsections. Different models in the literature have described migrating tissues as either elastic or fluid media, and both descriptions have successfully reproduced experimental ob- servations. Elastic models have been recently reviewed [133]. Here, we present the basis and general formulation of viscoelastic fluid models. Nevertheless, most of the formalism can be readily adapted to elastic descriptions by taking the limit of a long viscoelastic relaxation time τ → ∞. We discuss key differences between how elastic and fluid models describe tissue spreading in Section 3.5.8. 3.5.4 Constitutive equation for the deviatoric stress Cell aggregates largely devoid of extracellular matrix behave as viscoelastic fluids, exhibiting an elastic response at high frequencies and a viscous response at low fre- quencies [25, 140]. Thus, the simplest rheological choice is the Maxwell model, for which stress relaxes with a time-scale τ. In this case, the constitutive equation reads (cid:18)1 + τ D Dt(cid:19)(cid:104)σs − ν 2 ((cid:126)p (cid:126)h + (cid:126)h (cid:126)p − ((cid:126)p · (cid:126)h) I) − ¯ν ((cid:126)p · (cid:126)h) I + ζ Q + (¯ζ + ζ(cid:48)(cid:126)p2) I(cid:105) = = 2η v + ¯η (cid:126)∇ · (cid:126)v. (15) 19 The corotational derivative of a second rank tensor A reads DA/Dt = (∂t +(cid:126)v· (cid:126)∇)A+ ω · A − A · ω, where ω = ((cid:126)∇(cid:126)v − ((cid:126)∇(cid:126)v)T )/2 is the vorticity tensor. The relaxation time τ is set by the processes that dominate energy dissipation. These processes may be intracellular, such as cytoskeleton reorganizations, or intercellular, such as cell-cell sliding. They are thought to release stress at time-scales of protein turnover in the cytoskeleton and in cell-cell junctions, which are of the order of tens of minutes at most [25, 141]. In addition, other processes such as cell division, death, and extrusion [142,143], as well as cell shape fluctuations [144 -- 146] and topological rearrangements [145,147] also fluidize the tissue over different time-scales. In general, cell migration is a really slow process, imposing strain rates of ∼ h−1, which are slower than the fastest relaxation times. Hence, migrating cell monolayers are generally expected to behave as liquids and, indeed, they exhibit liquid-like phenomena like wetting transitions [140, 148, 149] and fingering instabilities [10, 150] (further discussion in Section 3.5.8). Besides the rheological model, the constitutive equation includes different types of stresses. Shear and bulk viscous stresses, proportional to the respective viscosi- ties η and ¯η, account for cell-cell friction (Section 2.2.2) and for the dissipation as- sociated to density changes (Section 2.2.3). For elastic deformations, η/τ and ¯η/τ are, respectively, the shear (Section 2.2.1) and bulk (Section 2.2.3) moduli of the cell monolayer. In turn, anisotropic active stresses are proportional to the nematic tensor Q = (cid:126)p (cid:126)p−1/2(cid:126)p2 I, with coefficient ζ, whereas isotropic active stresses have the coef- ficients ¯ζ, ζ(cid:48) (Section 2.2.4). Finally, the terms with ν and ¯ν are the stresses associated to the flow-polarity coupling discussed above (Sections 2.3.3 and 3.5.2). 3.5.5 Constitutive equation for the traction forces The next step is to specify the constitutive equation for the traction forces. Its expres- sion is less conventional than that for the internal stress, but it was recently shown to read [72] (cid:126)T = ξ(cid:126)v − Ta(cid:126)p + νs (16) in the long-time limit. Here, the first term is a cell-substrate viscous friction (Sec- tion 2.1.2), the second term is the active polar force that drives cell migration (Sec- tion 2.1.1 and Fig. 7a), and the third term accounts for the cell-substrate forces associ- ated to polarity-velocity alignment (Section 2.4.1 and Eq. (13)). (cid:126)p 3.5.6 Boundary conditions In addition to the equations, boundary conditions must be specified. First, many mod- els impose vanishing density and stress at the tissue edge. Alternatively, to capture interfacial effects, some models include a line tension [150 -- 152]. Second, the ten- dency of several cell types to polarize towards free space (Section 2.3.2) can be cap- tured by imposing perpendicular (homeotropic) anchoring of the polarity at the tissue edge [149, 150]. In this case, for an unpolarized tissue (a > 0 in Eq. (11)), the po- larity decays from the boundary with the characteristic length Lc = (cid:112)K/a, which defines the width of the polarized boundary layer observed in experiments [28, 149] (Fig. 7a). If cells align rather tangential the boundary, planar or tilted anchoring may be imposed [65] 20 3.5.7 Common simplifications Different models have simplified the equations in different ways. Most models that focus on tissue flow do not describe the density field altogether. Moreover, many of these models neglect flow-polarity interactions [28, 149, 150, 153] and even consider the polarity dynamics to be quasi-static (∂t(cid:126)p ≈ 0) [28,149,150]. Other models work in the limit of small correlation length of the polarity field, Lc =(cid:112)K/a → 0, whereby the polarity field drops from the description [139, 152, 154, 155]. In this case, active tractions are strictly localized at the tissue edge, amounting to a non-zero boundary stress. Another common simplification is to focus on a single dissipation source, thus only keeping either internal viscosity [38,149,156] or cell-substrate friction [154,157]. Fric- tion dominates over hydrodynamic interactions above the screening length λ =(cid:112)η/ξ. However, some studies suggest that this length may sometimes be comparable to tissue size, and hence both friction and internal viscosity must be kept [65, 150, 158]. On a related note, some continuum models approximate two-dimensional tissue flows as incompressible [154]. Even though it may be a valid approximation in some situations, monolayer area is not conserved (Section 2.2.3), and hence 2D incompressibility is not a general feature of epithelial monolayers. Finally, different models have included only isotropic [75, 159] or anisotropic [36, 64, 65] active stresses, or a combination of them [149, 153]. 3.5.8 Collective phenomena Continuum models have focused on the spreading of epithelial monolayers, in partic- ular addressing the formation of multicellular fingers [150, 153, 160, 161]. Consistent with experiments [53] (Fig. 7b), a recent model has shown that, even in the absence of motility regulation at the monolayer edge, there is an active instability that leads to a fingering pattern with an intrinsic wavelength [150] (Fig. 7c). Using the same model, recent work demonstrated a wetting transition between tissue spreading and re- traction as a result of the competition between active tractions and contractile stresses. The balance between these active forces depends on monolayer size and, hence, only monolayers larger than a critical size can spread [149] (Fig. 7d). This model has been extended to account for the role of substrate stiffness on tissue spreading, hence making predictions about tissue durotaxis [158]. A long-standing discussion in the field has been whether elastic or fluid models are more appropriate to describe tissue spreading [133]. Whereas spreading monolayers are expected and observed to behave like liquids (see Section 3.5.4), they also exhibit effective elastic responses even at long times [11,163], probably due to mechanotrans- duction processes. To capture this behavior, some models directly assume an elastic rheology of the monolayer [75, 159, 160, 164], whereas others explain it based on a viscous model with time-dependent parameters [28]. This controversy has also led to different explanations of the elastic-like mechanical waves observed during spread- ing [11]. Elastic models posit mechanochemical feedbacks, whereby either active stresses are coupled to the concentration of a strain-regulated protein [75,159,160,164] (Fig. 7e), or the polarity is coupled to strain [157]. In contrast, fluid models rely on the combination of active forces and either flow-polarity [64] or density-polarity [162] couplings to obtain effectively elastic waves (Fig. 7f). 21 Figure 7 Continuum Models. a, Scheme of a continuum model of a spreading monolayer. The red shade indicates the polarized boundary layer of width Lc. The stress σ accounts for intercellular and intracellular tension. Adapted from [149]. b, The monolayer edge develops multicellular fingers (red arrows) with a typical spacing. Scale bar, 100 µm. Reprinted from [53]. c, Positive growth rates of tissue shape perturbations of wavenumber q indicate a fingering instability. The maximal growth rate identifies the wavelength of the fingering pattern (dashed line). Adapted from [150]. Copyright (2019) by the American Physical Society. d, Phase diagram of tissue spreading. For a given contractility, monolayers larger than a critical radius spread (wetting) whereas smaller monolayers retract (dewetting). Adapted from [149]. e, Stability diagram of mechanochemical waves in an elastic model. The active stress is −¯ζ = β ln(c/c0), where c is the concentration of a strain-regulated protein. Adapted from [159]. Copyright (2015) by the American Physical Society. f, Stability diagram of mechanical waves in a fluid model. δa adds an active contribution to the polarity-density coupling w in Eq. (11), and γa adds a density-dependent term to the contractility: −¯ζ + γaρ. Adapted from [162] by permission of The Royal Society of Chemistry. 3.5.9 Discussion A different type of continuum models are not based on liquid crystal physics but on the Toner-Tu equations for flocking [165 -- 167]. Even though they also describe com- pressible polar fluids, the Toner-Tu equations do not include hydrodynamic interactions ("dry fluids" [6]) and do not distinguish between the polarity and the velocity fields. The facts that traction forces are observed even in static cell monolayers [75,149,168], and that traction and velocity fields are sometimes misaligned (kenotaxis) [74, 75], calls for a separation of polarity and velocity. Hence, even though they may correctly capture some phenomena, flocking-type continuum models do not seem generically appropriate to describe collective cell migration. As network models, continuum models mostly restrict to describing the migration of cohesive cell groups. The main strength of this approach is the analytical tractability of the field equations, which often allows getting analytical predictions, hence yielding insights without having to explore parameter space in simulations. However, the high degree of coarse-graining is double-edged. On the one hand, the generic equations are versatile, physically well-grounded, and can be written even without knowledge of microscopic details. On the other hand, this means that cell-cell interactions are not implemented at the cellular level but rather encoded in phenomenological couplings 22 Contractility − ³ (kPa)01020300102030Time t (h)Nematic length L (µm)0102030010203040Time t (h)a01020304050Contractility−³(kPa)-15-10-50510152025-³LLWettingDewetting-³L =(2´)T0L =´Spreadingvelocity V (µm/h)Monolayerradius R (µm)0100800700600500400300200bcdegfMaximal traction T 0 (kPa)01020300.00.20.40.60.81.0Time t (h)Figure 3Side viewL-T0 p´ (@ v + @ v ) - ³p p»vLTop viewLR®Focal adhesionCell -- cell junctionTractionTensionFrictionPolaritycc®¯¯®¯20406080100-0.20.00.20.40.60.81.0Radial coordinate r (µm)Traction T (kPa)T0Lr®®ccc´cccWettingDewettingV < 02 T0 �� RMonolayerradius R (µm)V > 001501005020000.050.10.150.20.250.300.010.020.030.040.05Growthrate!(h1)Wavenumberq(µm1)incomparisontothecontrolcase.Remarkably,ineachcase,theforcecorrelationlengthmatchedthecorrespondingleader-to-leaderdistance(Fig.3a -- c,SupplementaryFig.8,Supplementarymovie9).HaCaTcellsalsoshowedsimilartrendsofchangesinleader-to-leaderdistance(SupplementaryFig.7).Complementingthechemicalperturbation,wealsoalteredtheforcecorrelationlengthbyphysicalmeans,bychangingthestiffnessofsubstrateoverwhichthecellsmigrated(Fig.3c,rightpanel).Inthiscase,bothforcecorrelationlengthandleader-to-leaderdistanceincreasedwithincreasingstiffnessofthesubstrate(Fig.3a -- d).Together,theseresultsconfirmedsystematicmechano-biologicalregulationofleadercellgenerationduringcollectivemigrationof75 µmCellsPDMST 0T 120T 120T 120T 0ab100200Unbiased75 µm300 µmDistance (µm)T 120NScedDistance (µm)100200100200dLLFCLDistance (µm)dLLFCLMDCKMDCKHaCaTHaCaTNSNS300 µm T 120 T 0 -- 51005Avg. normal stress (Pa)NATURECOMMUNICATIONSDOI:10.1038/s41467-018-05927-6ARTICLENATURECOMMUNICATIONS(cid:0)(2018)(cid:0)9:3469(cid:0)DOI:10.1038/s41467-018-05927-6www.nature.com/naturecommunications5Contractility − ³ (kPa)01020300102030Time t (h)Nematic length L (µm)0102030010203040Time t (h)a01020304050Contractility−³(kPa)-15-10-50510152025-³LLWettingDewetting-³L =(2´)T0L =´Spreadingvelocity V (µm/h)Monolayerradius R (µm)0100800700600500400300200bcdegfMaximal traction T 0 (kPa)01020300.00.20.40.60.81.0Time t (h)Figure 3Side viewL-T0 p´ (@ v + @ v ) - ³p p»vLTop viewLR®Focal adhesionCell -- cell junctionTractionTensionFrictionPolaritycc®¯¯®¯20406080100-0.20.00.20.40.60.81.0Radial coordinate r (µm)Traction T (kPa)T0Lr®®ccc´cccWettingDewettingV < 02 T0 �� RMonolayerradius R (µm)V > 0015010050200Stresspropagatingwaves,thestressinitiallyshowsafewlocalmaxima[Fig.1(b)],whichevolvetowardsasinglemaxi-mumatthecenterofthemonolayer,asobservedinexperiments[8,22].Theconcentrationofcontractileele-mentsalsooscillatesandbuildsupatthecenterofthemonolayer[Fig.1(c)].Thestresswavespropagatenearlyinphasewiththestrainfield,whereasthestrainratefluctuatesnearlyoutofphasewiththestress[Fig.1(d)].Thus,theresponseofthematerialisdominatedbyelasticrelaxationwithdissipationinducedbyturnoversincontractilityonatimescaleτ.Thewavesspantheentirelengthofthemonolayerandconsistofastrainratewavefrontthatpropagatesinwardsfromtheedge,andthentravelsbacktotheedge,resemblinganXpattern,asobservedexperi-mentally[8].Withthegivenparametervaluesournumeri-calsimulationscapturethemechanicalwavesasevidentinthekymographsofstress,strainrate,andconcentrationofcontractileunits[Figs.2(b) -- (d)].Tounderstandtheoriginofwavepropagationandestimatethewavefrequency,itisusefultoexaminethelinearfluctuationsinthestrainfieldδεandtheconcen-trationfieldδc,aboutthequiescenthomogeneousstateu¼0,c¼c0,andnospreadingforce.UsingEqs.(1)and(2),onecantheneliminateδcfromsuchlinearizedequationstoobtainthelinearizeddynamicsofstrainfluctuations,τΓ∂2tδεþΓ∂tδε¼h0ðBeffþηeff∂t−τBD∂2xÞ∂2xδε;ð3ÞTheaboveequationshowsthatthecouplingofstraintoconcentrationfieldyieldsaneffectivemassdensity(iner-tia),τΓ,andviscoelasticitycharacterizedbyaneffectiveelasticmodulus,Beff¼Bþαβτ=c0,andaneffectiveviscosityηeff¼ðB−βþDΓ=h0Þτ.ThedynamicsofstrainfluctuationsresemblesadampedKelvin-Voigtoscil-latorwithacharacteristicfrequencyofoscillations,ω0¼qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffih0ðBeffþτq2BDÞ=ðτΓÞp,withqthewavevector.Theestimateforthetimeperiod2π=ω0agreeswellwiththetimeperioddeterminedfromnumericsforq≃4π=L0[seeFig.3(a)]andwiththevaluemeasuredinrecentexperi-ments[8].Finally,wenotethatiftheconcentrationcisconserved(τ→∞;α¼0),stablepropagatingwavesarespontaneouslygeneratedfor0<B−βþDΓ=h0<2ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiDBΓ=h0p.Ifdiffusionisslowcomparedtoelasticrelaxation,DΓ=Bh0≪1,stablepropagatingwavesarenotobserved[21].Intheoppositelimitofinfinitelyfastturnoversincontractility(τ→0),strainfluctuationsdecaydiffusivelyatarate≃Bh0=ΓL2.Meanfieldmodel. -- Themeanfieldlimitofthecon-tinuummodelisobtainedbyneglectingspatialvariationsincandεanditisformulatedintermsofthelength(L),height(h),andtheaverageconcentrationofcontractileelements,¯cðtÞ¼ð1=LÞRL0dxcðx;tÞ,withγdLdt¼F0−AðtÞσðtÞ;ð4aÞd¯cdtþ¯cLdLdt¼−1τð¯c−c0Þþαε;ð4bÞwithF0thepropulsionforce,γthefriction,AðtÞ¼dhðtÞthecross-sectionalarea,εðtÞ¼LðtÞ=L0−1thestrain,andσðtÞtheinternalstressgivenbyσðtÞ¼BεðtÞþβ½¯cðtÞ=c0−1.Theheightisdeterminedusingtheincompress-ibilitycondition,withthesizeintheydirection,d,fixed.ThesteadystatesolutionisL∞¼L0=ð1−ΛÞ,h∞¼h0ð1−ΛÞandc∞¼c0þατΛ=ð1−ΛÞ,withΛ¼c0F0=dh0ðBc0þαβτÞthenetcompressivestraininthezdirection.ForagivenvalueofelasticmodulusB,themean-fieldmodelpredictsoscillatorysolutionsforβ>βc,whereβcðBÞdefinesthephaseboundaryin(B;β)planeseparatingtheregionsofpropagatingwavesanddiffusivespreading[dashedlineinFig.2(a)].Forβ<βcthemonolayerdiffusivelyapproachesthesteadystate(c∞;L∞).Thissimplemean-fieldapproachallowsustostudythematerialresponseofthemonolayercharacterizedbyaneffectiveelasticmodulus,BMF¼dσ=dε.Theoscillatoryregime(β>βc)exhibitssustainedoscillationsinthematerialrigidity,BMF,withaslowperiodofstiffeningfollowedbyasharpturnover[seeFig.3(b)].Forβ<βc,thematerialgraduallystiffenswithBMFasymptoticallyapproachingthevalueBeff.Theseoscillationsreflectself-sustainedturn-oversinthecytoskeletonwithperiodicreinforcementandfluidizationondifferenttimescales,whichwasinvokedtoFIG.2(coloronline).(a)Phasediagramofthespreadinggel.TheverticalaxisrepresentsthecontractileactivityβandthehorizontalaxisisthecompressionalmodulusB.Threebehaviorsareobserved:stablediffusive,stablepropagatingwaves,andoscillatoryinstability.Theredsquaresareobtainedfromthenumericalsolutionsofthefullnonlinearmodel,theblacksolidlinesaretheresultsofthelinearstabilityanalysis(LSA)oftheequilibriumstate(atq¼13.5=L0)[21],andthedashedgreenlinesrefertotheLSAofthemean-fieldmodelgiveninEqs.(4).Kymographsof(b)themonolayerstressfield,(c)strainrate∂tεðx;tÞ,and(d)cðx;tÞ=c0.TheparametervaluesaretakentobethesameasinFig.1.PRL114,228101(2015)PHYSICALREVIEWLETTERSweekending5JUNE2015228101-3Contractility (Pa)Elastic modulus B (Pa)Fingering patternThisjournalis©TheRoyalSocietyofChemistry2017SoftMatter,2017,13,7046--70527049anoiseofsmallamplitude.Weusedafinite-differencemethodandintegratedeqn(14) -- (16)onadiscretized1DspacewithmeshsizeDx=0.01andtimestepDt=0.0001asfollows:giventhedensityfieldr(x,t)andthepolarityfieldp(x,t)atatimestept,wesolveeqn(15)forthevelocityfieldv(x,t).Next,usingeqn(14)andeqn(16),wedeterminethedensityfieldr(x,t+Dt)andthepolarityfieldp(x,t+Dt)atthenexttimestept+DtwiththeexplicitEulermethod.InFig.2,wesetallparameterstoone(K=re=a2=Z=l=Gp=ga=ta=aa=da=1)exceptforx=p0=0.5andK4=0.0005.Aspredictedbythelinearstabilityanalysisintheprevioussection,weconfirmnumericallythattheinstabilityoccursandthepropagationvelocity(cB1.07)isslightlylargerthanthemeantissuevelocityv0=tap0/x=1.Thetransienttimescaleisoftheorderof102.InFig.3,weshowforthesameparametervaluesthespatialprofilesofthecelldensity,stressandstrainrate_exx=qxv(x,t)att=1000.Thestressandstrainrateareout-of-phase,whilethestressandthecelldensityareinantiphase.Ifweconsiderthatthecellareameasuredexperimentallycorrespondstotheinverseofthecelldensity,bothphaserelationsagreewithexperimentalobservations,whichhavebeeninterpretedasevidenceforanelasticrheology.8However,ourconstitutiveeqn(7)isthatofacompressible,viscousmaterial,alsoendowedwithactiveandpolarproperties.Numericalsimulationsallowtomimicinhibitionassaysbytuningthevalue(s)ofparameter(s)thatwouldbechangedduetotheapplicationofadrug.Whenwelowerthevalueoftheactiveparameterstoda=0.5,ga=0.5att=200,weobservethatthetravelingwaverapidlydisappears(seeFig.4a),asobservedexperimentallywheninhibitingcontractility.Inasimilarman-ner,whenwechangethereferencepolarityvaluefromafinitep0top0=0(bysettinga2=500att=200),thetravelingwavesalsodisappearsquickly(seeFig.4b),asobservedexperimen-tallywheninhibitingArp2/3.Finally,Fig.5showsanexampleofatravelingwaveobservedfartherfromthreshold,withaa=2,ga=da=3,andotherparametersasinFig.2.ThepropagationvelocityisaccordinglylargercB1.3.Fig.1Linearstabilityanalysis.(a)Plotoftherealpartofthelargestgrowthratesr+(q)asafunctionofwavenumberq.Allparametersaresetequalto1(includinggaandda),exceptp0=0.5,x=0.5.Wefindqmax=argmaxq(sr+(q))=0.56,withawavevelocityc=si+(qmax)/qmax=1.07greaterthanthemeantissuevelocityv0=tap0/x=1.Wecheckedthatsr(q)o0,8q.(b)Bifurcationdiagraminthe(ga,da)plane.Theyellow(resp.blue)domaincorrespondstoastable(resp.unstable)homogeneousstate.Fig.2Atravelingwave:allparametersaresetequaltooneK=re=a2=Z=l=Gp=ta=aa=da=ga=1,exceptforx=p0=0.5andK4=0.0005.Thecelldensity(a)andvelocity(b)kymographsarepresentedwhentZ1000,aftertransientshavediedout.Thepolarityandstressfieldsalsoexhibitstabletravelingwavesolutionsforthesameparametervalues(notshown).Fig.3Spatialprofilesofthedensity,stressandstrainratefieldsatt=1000.ParametervaluesareasinFig.2.PaperSoftMatterPublished on 29 August 2017. Downloaded by Universitat de Barcelona on 12/10/2017 16:43:08. View Article OnlineStress-density couplingPolarity-density couplingStableElastic-like Waves01020304050050100150200WettingDewetting2TaContractility⇣(kPa)MonolayerradiusR(µm)whose relationship to cellular processes may be unclear. Finally, at the vicinity of the jamming transition, cell shape fluctuations become a slow field that must be incorpo- rated into hydrodynamic descriptions. To this end, Czajkowski et al. have recently proposed a model that couples a cell shape anisotropy field to the polarity field [169]. 4 Conclusions and outlook Concurrent advances in experiments and theory are quickly shaping a solid under- standing of collective cell migration. Following the footsteps of more mature areas of physics, the field can now aim towards making quantitative predictions and testing specific models and assumptions in theory-inspired experiments. Besides progress in this direction, fundamental challenges remain. For example, here we have restricted our attention to 2D migrating cell sheets. However, as mentioned in the introduction, cells also migrate within 3D environ- ments, which are remodeled by and mediate mechanical interactions between cells, and even provide single cells with migration modes that are unavailable in 2D [170, 171]. Whether and how cells integrate these modes and interactions to move collectively is unknown. Moreover, techniques to probe mechanical forces in 3D remain limited in accuracy and throughput [5]. Generalizing 2D theories and techniques to 3D environ- ments is thus one of the current challenges of cell migration biophysics. Another central challenge is to bridge descriptions at different scales. For example, deriving continuum models from cell-scale models would map cellular interactions to tissue-scale mechanical properties and phenomena. Another pending task is to link the molecular mechanisms of actin polymerization and cell-substrate adhesion with the ac- tual self-propulsion and friction forces included in models of collective cell migration. In particular, the most appropriate type of cell-substrate friction is still unclear [172], largely due to the lack of experimental evidence. A third limitation is the lack of a unifying picture that captures the mechanical consequences of cell-cell contact interactions. Upon collision, cells can adhere, repel, or ignore each other. Within each of these three behaviors there exist many nuances as well (see Section 2). A better understanding of the mechanisms that define the me- chanics and migration of two cells upon contact is crucial to advance our understanding of collective cell migration. A closer collaboration between physicists and life scien- tists is needed to incorporate the broad diversity of biological mechanisms involved in collective cell migration into a unifying physical picture. Acknowledgments We apologize to the many colleagues whose work could not be cited owing to space constraints. We thank Jaume Casademunt, Carles Blanch-Mercader, and Carlos Pérez- González for a stimulating collaboration and many discussions. We thank Anna Laber- nadie for assistance with Fig. 2, and Amin Doostmohammadi, José Muñoz, Vito Conte, and Matej Krajnc for a critical reading of the manuscript. R.A. acknowledges support from the Human Frontiers of Science Program (LT-000475/2018-C). X.T. is supported by MINECO/FEDER (BFU2015-65074-P), the Generalitat de Catalunya (2014-SGR- 927 and the CERCA Programme), the European Research Council (CoG-616480), the European Commission (Grant Agreeement SEP-210342844), and Obra Social "La Caixa". IBEC is recipient of a Severo Ochoa Award of Excellence from the MINECO. 23 References [1] Born G. 1897. Arch. für Entwickelungsmechanik der Org. 4:517 -- 623 [2] Holmes SJ. 1914. J. Exp. Zool. 17:281 -- 295 [3] Herrick EH. 1932. Biol. Bull. 63:271 -- 286 [4] Vaughan RB, Trinkaus JP. 1966. J. Cell Sci. 1:407 -- 413 [5] Roca-Cusachs P, Conte V, Trepat X. 2017. Nat. Cell Biol. 19:742 -- 751 [6] Marchetti MC, Joanny JF, Ramaswamy S, Liverpool TB, Prost J, et al. 2013. Rev. Mod. Phys. 85:1143 -- 1189 [7] Prost J, Jülicher F, Joanny JF. 2015. Nat. Phys. 11:111 -- 117 [8] Jülicher F, Grill SW, Salbreux G. 2018. Reports Prog. Phys. 81:076601 [9] Good M, Trepat X. 2018. Nature 563:188 -- 189 [10] Poujade M, Grasland-Mongrain E, Hertzog A, Jouanneau J, Chavrier P, et al. 2007. Proc. Natl. Acad. Sci. U. S. A. 104:15988 -- 15993 [11] Serra-Picamal X, Conte V, Vincent R, Añón E, Tambe DT, et al. 2012. Nat. Phys. 8:628 -- 634 [12] du Roure O, Saez A, Buguin A, Austin RH, Chavrier P, et al. 2005. Proc. Natl. Acad. Sci. U. S. A. 102:2390 -- 5 [13] Trepat X, Wasserman MR, Angelini TE, Millet E, Weitz DA, et al. 2009. Nat. Phys. 5:426 -- 430 [14] Friedl P, Noble PB, Walton PA, Laird DW, Chauvin PJ, et al. 1995. Cancer Res. 55:4557 -- 60 [15] Brugués A, Anon E, Conte V, Veldhuis JH, Gupta M, et al. 2014. Nat. Phys. 10:683 -- 690 [16] Friedl P, Gilmour D. 2009. Nat. Rev. Mol. Cell Biol. 10:445 -- 457 [17] Cai D, Chen SC, Prasad M, He L, Wang X, et al. 2014. Cell 157:1146 -- 1159 [18] Clark AG, Vignjevic DM. 2015. Curr. Opin. Cell Biol. 36:13 -- 22 [19] Campàs O. 2016. Semin. Cell Dev. Biol. 55:119 -- 130 [20] Hakim V, Silberzan P. 2017. Reports Prog. Phys. 80:076601 [21] Ladoux B, Mège RM. 2017. Nat. Rev. Mol. Cell Biol. 18:743 -- 757 [22] Xi W, Saw TB, Delacour D, Lim CT, Ladoux B. 2019. Nat. Rev. Mater. 4:23 -- 44 [23] Schwarz US, Safran SA. 2013. Rev. Mod. Phys. 85:1327 -- 1381 [24] Elosegui-Artola A, Trepat X, Roca-Cusachs P. 2018. Trends Cell Biol. 28:356 -- 367 24 [25] Khalilgharibi N, Fouchard J, Recho P, Charras G, Kabla A. 2016. Curr. Opin. Cell Biol. 42:113 -- 120 [26] Bi D, Lopez JH, Schwarz JM, Manning ML. 2015. Nat. Phys. 11:1074 -- 1079 [27] Park JA, Kim JH, Bi D, Mitchel JA, Qazvini NT, et al. 2015. Nat. Mater. 14:1040 -- 8 [28] Blanch-Mercader C, Vincent R, Bazellières E, Serra-Picamal X, Trepat X, Casademunt J. 2017. Soft Matter 13:1235 -- 1243 [29] Iyer KV, Piscitello-Gómez R, Paijmans J, Jülicher F, Eaton S. 2019. Curr. Biol. 29:578 -- 591.e5 [30] Czirók A, Varga K, Méhes E, Szabó A. 2013. New J. Phys. 15:075006 [31] Peglion F, Llense F, Etienne-Manneville S. 2014. Nat. Cell Biol. 16:639 -- 651 [32] Garcia S, Hannezo E, Elgeti J, Joanny JF, Silberzan P, Gov NS. 2015. Proc. Natl. Acad. Sci. U. S. A. 112:15314 -- 15319 [33] Zehnder SM, Suaris M, Bellaire MM, Angelini TE. 2015a. Biophys. J. 108:247 -- 250 [34] Zehnder SM, Wiatt MK, Uruena JM, Dunn AC, Sawyer WG, Angelini TE. 2015b. Phys. Rev. E 92:032729 [35] Kocgozlu L, Saw TB, Le AP, Yow I, Shagirov M, et al. 2016. Curr. Biol. 26:2942 -- 2950 [36] Saw TB, Doostmohammadi A, Nier V, Kocgozlu L, Thampi S, et al. 2017. Na- ture 544:212 -- 216 [37] Chen T, Saw TB, Mège RM, Ladoux B. 2018. J. Cell Sci. 131:jcs218156 [38] Beaune G, Stirbat TV, Khalifat N, Cochet-Escartin O, Garcia S, et al. 2014. Proc. Natl. Acad. Sci. U. S. A. 111:8055 -- 8060 [39] Maruthamuthu V, Sabass B, Schwarz US, Gardel ML. 2011. Proc. Natl. Acad. Sci. U. S. A. 108:4708 -- 13 [40] Stramer B, Mayor R. 2017. Nat. Rev. Mol. Cell Biol. 18:43 -- 55 [41] Mayor R, Etienne-Manneville S. 2016. Nat. Rev. Mol. Cell Biol. 17:97 -- 109 [42] Li D, Wang YL. 2018. Proc. Natl. Acad. Sci. U. S. A. 115:10678 -- 10683 [43] Fujimori T, Nakajima A, Shimada N, Sawai S. 2019. Proc. Natl. Acad. Sci. U. S. A. 116:4291 -- 4296 [44] d'Alessandro J, Solon AP, Hayakawa Y, Anjard C, Detcheverry F, et al. 2017. Nat. Phys. 13:999 -- 1005 [45] Desai RA, Gopal SB, Chen S, Chen CS. 2013. J. R. Soc. Interface 10:20130717 [46] George M, Bullo F, Campàs O. 2017. Sci. Rep. 7:9720 25 [47] Desai RA, Gao L, Raghavan S, Liu WF, Chen CS. 2009. J. Cell Sci. 122:905 -- 911 [48] Weber GF, Bjerke MA, DeSimone DW. 2012. Dev. Cell 22:104 -- 115 [49] Ng MR, Besser A, Danuser G, Brugge JS. 2012. J. Cell Biol. 199:545 -- 563 [50] Roca-Cusachs P, Sunyer R, Trepat X. 2013. Curr. Opin. Cell Biol. 25:543 -- 549 [51] Das T, Safferling K, Rausch S, Grabe N, Boehm H, Spatz JP. 2015. Nat. Cell Biol. 17:276 -- 287 [52] Ladoux B, Mège RM, Trepat X. 2016. Trends Cell Biol. 26:420 -- 433 [53] Vishwakarma M, Di Russo J, Probst D, Schwarz US, Das T, Spatz JP. 2018. Nat. Commun. 9:3469 [54] Davis JR, Luchici A, Mosis F, Thackery J, Salazar JA, et al. 2015. Cell 161:361 -- 73 [55] Scarpa E, Szabó A, Bibonne A, Theveneau E, Parsons M, Mayor R. 2015. Dev. Cell 34:421 -- 434 [56] Roycroft A, Mayor R. 2016. Cell. Mol. Life Sci. 73:1119 -- 1130 [57] Roycroft A, Szabó A, Bahm I, Daly L, Charras G, et al. 2018. Dev. Cell 45:565 -- 579.e3 [58] Aigouy B, Farhadifar R, Staple DB, Sagner A, Röper JC, et al. 2010. Cell 142:773 -- 86 [59] Marel AK, Podewitz N, Zorn M, Rädler JO, Elgeti J. 2014. New J. Phys. 16:115005 [60] Tambe DT, Corey Hardin C, Angelini TE, Rajendran K, Park CY, et al. 2011. Nat. Mater. 10:469 -- 475 [61] Trepat X, Fredberg JJ. 2011. Trends Cell Biol. 21:638 -- 46 [62] He S, Liu C, Li X, Ma S, Huo B, Ji B. 2015. Biophys. J. 109:489 -- 500 [63] Zaritsky A, Welf ES, Tseng YY, Angeles Rabadán M, Serra-Picamal X, et al. 2015. Biophys. J. 109:2492 -- 2500 [64] Blanch-Mercader C, Casademunt J. 2017. Soft Matter 13:6913 -- 6928 [65] Duclos G, Blanch-Mercader C, Yashunsky V, Salbreux G, Joanny JF, et al. 2018. Nat. Phys. 14:728 -- 732 [66] Coburn L, Cerone L, Torney C, Couzin ID, Neufeld Z. 2013. Phys. Biol. 10:046002 [67] Löber J, Ziebert F, Aranson IS. 2015. Sci. Rep. 5:9172 [68] Barton DL, Henkes S, Weijer CJ, Sknepnek R. 2017. PLOS Comput. Biol. 13:e1005569 26 [69] Peyret G, Mueller R, d'Alessandro J, Begnaud S, Marcq P, et al. 2018. bioRxiv :492082 [70] Brotto T, Caussin JB, Lauga E, Bartolo D. 2013. Phys. Rev. Lett. 110:038101 [71] Kumar N, Soni H, Ramaswamy S, Sood AK. 2014. Nat. Commun. 5:4688 [72] Oriola D, Alert R, Casademunt J. 2017. Phys. Rev. Lett. 118:088002 [73] Maitra A, Srivastava P, Marchetti MC, Ramaswamy S, Lenz M. 2019. arXiv:1901.01069v1 [74] Kim JH, Serra-Picamal X, Tambe DT, Zhou EH, Park CY, et al. 2013. Nat. Mater. 12:856 -- 63 [75] Notbohm J, Banerjee S, Utuje KJ, Gweon B, Jang H, et al. 2016. Biophys. J. 110:2729 -- 2738 [76] Zimmermann J, Camley BA, Rappel WJ, Levine H. 2016. Proc. Natl. Acad. Sci. U. S. A. 113:2660 -- 2665 [77] Lin SZ, Ye S, Xu GK, Li B, Feng XQ. 2018. Biophys. J. 115:1826 -- 1835 [78] Bun P, Liu J, Turlier H, Liu Z, Uriot K, et al. 2014. Biophys. J. 107:324 -- 335 [79] Breckenridge MT, Desai RA, Yang MT, Fu J, Chen CS. 2014. Cell. Mol. Bioeng. 7:26 -- 34 [80] Marzban B, Yi X, Yuan H. 2018. Biomech. Model. Mechanobiol. 17:915 -- 922 [81] Saez A, Ghibaudo M, Buguin A, Silberzan P, Ladoux B. 2007. Proc. Natl. Acad. Sci. U. S. A. 104:8281 -- 8286 [82] Gupta M, Sarangi BR, Deschamps J, Nematbakhsh Y, Callan-Jones A, et al. 2015. Nat. Commun. 6:7525 [83] Gupta M, Doss B, Lim CT, Voituriez R, Ladoux B. 2016. Cell Adh. Migr. 10:554 -- 567 [84] Gupta M, Doss BL, Kocgozlu L, Pan M, Mège RM, et al. 2019. Phys. Rev. E 99:012412 [85] Prager-Khoutorsky M, Lichtenstein A, Krishnan R, Rajendran K, Mayo A, et al. 2011. Nat. Cell Biol. 13:1457 -- 65 [86] Camley B, Rappel WJ. 2017. J. Phys. D. Appl. Phys. 50:113002 [87] Spatarelu CP, Zhang H, Nguyen DT, Han X, Liu R, et al. 2019. arXiv:1903.04921 [88] Graner F, Glazier JA. 1992. Phys. Rev. Lett. 69:2013 -- 2016 [89] Chiang M, Marenduzzo D. 2016. EPL (Europhysics Lett. 116:28009 [90] Rappel WJ, Nicol A, Sarkissian A, Levine H, Loomis WF. 1999. Phys. Rev. Lett. 83:1247 -- 1250 27 [91] Kabla AJ. 2012. J. R. Soc. Interface 9:3268 -- 78 [92] Szabó A, Ünnep R, Méhes E, Twal WO, Argraves WS, et al. 2010. Phys. Biol. 7:046007 [93] Ouaknin GY, Bar-Yoseph PZ. 2009. Biophys. J. 97:1811 -- 21 [94] Coburn L, Lopez H, Schouwenaar IM, Yap AS, Lobaskin V, Gomez GA. 2018. Phys. Biol. 15:024001 [95] Segerer FJ, Thüroff F, Piera Alberola A, Frey E, Rädler JO. 2015. Phys. Rev. Lett. 114:228102 [96] Thueroff F, Goychuk A, Reiter M, Frey E. 2019. bioRxiv :548677 [97] Gonzalez-Cinca R, Folch R, Benitez R, Ramirez-Piscina L, Casademunt J, Hernandez-Machado A. 2004. Phase-field models in interfacial pattern forma- tion out of equilibrium. In Advances in Condensed Matter and Statistical Me- chanics, eds. E Korutcheva, R Cuerno, chap. 9. Nova Science Publishers, 203 -- 236 [98] Camley BA, Zhang Y, Zhao Y, Li B, Ben-Jacob E, et al. 2014. Proc. Natl. Acad. Sci. U. S. A. 111:14770 -- 14775 [99] Palmieri B, Bresler Y, Wirtz D, Grant M. 2015. Sci. Rep. 5:11745 [100] Mueller R, Yeomans JM, Doostmohammadi A. 2019. Phys. Rev. Lett. 122:048004 [101] Najem S, Grant M. 2016. Phys. Rev. E 93:052405 [102] Weaire D, Hutzler S. 1999. The Physics of Foams. Oxford University Press [103] Alt S, Ganguly P, Salbreux G. 2017. Philos. Trans. R. Soc. London B Biol. Sci. 372:20150520 [104] Sussman DM, Merkel M. 2018. Soft Matter 14:3397 -- 3403 [105] Manning ML, Foty RA, Steinberg MS, Schoetz EM. 2010. Proc. Natl. Acad. Sci. U. S. A. 107:12517 -- 22 [106] Winklbauer R. 2015. J. Cell Sci. 128:3687 -- 93 [107] Bi D, Yang X, Marchetti MC, Manning ML. 2016. Phys. Rev. X 6:021011 [108] Giavazzi F, Paoluzzi M, Macchi M, Bi D, Scita G, et al. 2018. Soft Matter 14:3471 -- 3477 [109] Trepat X, Sahai E. 2018. Nat. Phys. 14:671 -- 682 [110] Salm M, Pismen LM. 2012. Phys. Biol. 9:026009 [111] Mathur J, Sarker B, Pathak A. 2018. Biophys. J. 115:2474 -- 2485 [112] Schaumann EN, Staddon MF, Gardel ML, Banerjee S. 2018. Mol. Biol. Cell 29:2835 -- 2847 28 [113] Staddon MF, Bi D, Tabatabai AP, Ajeti V, Murrell MP, Banerjee S. 2018. PLoS Comput. Biol. 14:e1006502 [114] Li B, Sun SX. 2014. Biophys. J. 107:1532 -- 1541 [115] Malinverno C, Corallino S, Giavazzi F, Bergert M, Li Q, et al. 2017. Nat. Mater. 16:587 -- 596 [116] Petrolli V, Le Goff M, Tadrous M, Martens K, Allier C, et al. 2019. Phys. Rev. Lett. 122:168101 [117] Coburn L, Lopez H, Caldwell BJ, Moussa E, Yap C, et al. 2016. Mol. Biol. Cell 27:3436 -- 3448 [118] Teomy E, Kessler DA, Levine H. 2018. Phys. Rev. E 98:042418 [119] Koride S, Loza AJ, Sun SX. 2018. APL Bioeng. 2:031906 [120] Yang X, Bi D, Czajkowski M, Merkel M, Manning ML, Marchetti MC. 2017. Proc. Natl. Acad. Sci. U. S. A. 114:12663 -- 12668 [121] Basan M, Elgeti J, Hannezo E, Rappel WJ, Levine H. 2013. Proc. Natl. Acad. Sci. U. S. A. 110:2452 -- 2459 [122] Schnyder SK, Molina JJ, Tanaka Y, Yamamoto R. 2017. Sci. Rep. 7:5163 [123] Smeets B, Alert R, Pešek J, Pagonabarraga I, Ramon H, Vincent R. 2016. Proc. Natl. Acad. Sci. U. S. A. 113:14621 -- 14626 [124] Henkes S, Fily Y, Marchetti MC. 2011. Phys. Rev. E 84:040301 [125] Tarle V, Ravasio A, Hakim V, Gov N. 2015. Integr. Biol. 7:1218 -- 1227 [126] Szabó B, Szöllösi G, Gönci B, Jurányi Z, Selmeczi D, Vicsek T. 2006. Phys. Rev. E 74:061908 [127] Sepúlveda N, Petitjean L, Cochet O, Grasland-Mongrain E, Silberzan P, Hakim V. 2013. PLoS Comput. Biol. 9:e1002944 [128] Bindschadler M, McGrath JL. 2007. J. Cell Sci. 120:876 -- 884 [129] Woods ML, Carmona-Fontaine C, Barnes CP, Couzin ID, Mayor R, Page KM. 2014. PLoS One 9:e104969 [130] Copenhagen K, Malet-Engra G, Yu W, Scita G, Gov N, Gopinathan A. 2018. Sci. Adv. 4:eaar8483 [131] Camley BA, Zimmermann J, Levine H, Rappel WJ. 2016. Phys. Rev. Lett. 116:098101 [132] Theveneau E, Mayor R. 2013. Cell. Mol. Life Sci. 70:3481 -- 3492 [133] Banerjee S, Marchetti MC. 2018. arXiv:1805.06531v1 [134] Kung W, Marchetti MC, Saunders K. 2006. Phys. Rev. E 73:031708 [135] Voituriez R, Joanny J, Prost J. 2006. Phys. Rev. Lett. 96:028102 29 [136] Cates ME, Tjhung E. 2018. J. Fluid Mech. 836:P1 [137] Marcq P. 2014. Eur. Phys. J. E 37:29 [138] de Gennes PG, Prost J. 1993. The Physics of Liquid Crystals. Oxford University Press, 2nd ed. [139] Recho P, Ranft J, Marcq P. 2016. Soft Matter 12:2381 -- 2391 [140] Gonzalez-Rodriguez D, Guevorkian K, Douezan S, Brochard-Wyart F. 2012. Science 338:910 -- 917 [141] Wyatt T, Baum B, Charras G. 2016. Curr. Opin. Cell Biol. 38:68 -- 73 [142] Ranft J, Basan M, Elgeti J, Joanny JF, Prost J, Jülicher F. 2010. Proc. Natl. Acad. Sci. U. S. A. 107:20863 -- 20868 [143] Matoz-Fernandez DA, Agoritsas E, Barrat JL, Bertin E, Martens K. 2017. Phys. Rev. Lett. 118:158105 [144] Marmottant P, Mgharbel A, Käfer J, Audren B, Rieu JP, et al. 2009. Proc. Natl. Acad. Sci. U. S. A. 106:17271 -- 17275 [145] Etournay R, Popovi´c M, Merkel M, Nandi A, Blasse C, et al. 2015. Elife 4:e07090 [146] Tlili S, Gay C, Ladoux B, Graner F, Delanoë-Ayari H. 2018a. arXiv:1811.05001v1 [147] Krajnc M, Dasgupta S, Ziherl P, Prost J. 2018. Phys. Rev. E 98:022409 [148] Douezan S, Guevorkian K, Naouar R, Dufour S, Cuvelier D, Brochard-Wyart F. 2011. Proc. Natl. Acad. Sci. U. S. A. 108:7315 -- 7320 [149] Pérez-González C, Alert R, Blanch-Mercader C, Gómez-González M, Kolodziej T, et al. 2019. Nat. Phys. 15:79 -- 88 [150] Alert R, Blanch-Mercader C, Casademunt J. 2019. Phys. Rev. Lett. 122:088104 [151] Ravasio A, Cheddadi I, Chen T, Pereira T, Ong HT, et al. 2015. Nat. Commun. 6:7683 [152] Williamson JJ, Salbreux G. 2018. Phys. Rev. Lett. 121:238102 [153] Lee P, Wolgemuth CW. 2011. PLoS Comput. Biol. 7:e1002007 [154] Cochet-Escartin O, Ranft J, Silberzan P, Marcq P. 2014. Biophys. J. 106:65 -- 73 [155] Yabunaka S, Marcq P. 2017a. Phys. Rev. E 96:022406 [156] Douezan S, Brochard-Wyart F. 2012. Eur. Phys. J. E 35:116 [157] Tlili S, Gauquelin E, Li B, Cardoso O, Ladoux B, et al. 2018b. R. Soc. Open Sci. 5:172421 [158] Alert R, Casademunt J. 2018. Langmuir :acs.langmuir.8b02037 [159] Banerjee S, Utuje KJC, Marchetti MC. 2015. Phys. Rev. Lett. 114:228101 30 [160] Köpf MH, Pismen LM. 2013. Soft Matter 9:3727 [161] Ben Amar M, Bianca C. 2016. Sci. Rep. 6:33849 [162] Yabunaka S, Marcq P. 2017b. Soft Matter 13:7046 -- 7052 [163] Vincent R, Bazellières E, Pérez-González C, Uroz M, Serra-Picamal X, Trepat X. 2015. Phys. Rev. Lett. 115:248103 [164] Köpf MH. 2015. Phys. Rev. E 91:012712 [165] Zimmermann J, Basan M, Levine H. 2014. Eur. Phys. J. Spec. Top. 223:1259 -- 1264 [166] Nesbitt D, Pruessner G, Lee CF. 2017. Phys. Rev. E 96:062615 [167] Bogdan M, Savin T. 2018. R. Soc. Open Sci. 5:181579 [168] Mertz AF, Banerjee S, Che Y, German GK, Xu Y, et al. 2012. Phys. Rev. Lett. 108:198101 [169] Czajkowski M, Bi D, Manning ML, Marchetti MC. 2018. Soft Matter 14:5628 -- 5642 [170] Friedl P, Alexander S. 2011. Cell 147:992 -- 1009 [171] Bergert M, Erzberger A, Desai RA, Aspalter IM, Oates AC, et al. 2015. Nat. Cell Biol. 17:524 -- 529 [172] Christensen A, West AKV, Wullkopf L, Terra Erler J, Oddershede LB, Math- iesen J. 2018. Phys. Biol. 15:066004 31
1811.09536	1	1811	2018-11-19T22:22:26	Protein Folding and Machine Learning: Fundamentals	[ "physics.bio-ph", "q-bio.BM" ]	In spite of decades of research, much remains to be discovered about folding: the detailed structure of the initial (unfolded) state, vestigial folding instructions remaining only in the unfolded state, the interaction of the molecule with the solvent, instantaneous power at each point within the molecule during folding, the fact that the process is stable in spite of myriad possible disturbances, potential stabilization of trajectory by chaos, and, of course, the exact physical mechanism (code or instructions) by which the folding process is specified in the amino acid sequence. Simulations based upon microscopic physics have had some spectacular successes and continue to improve, particularly as super-computer capabilities increase. The simulations, exciting as they are, are still too slow and expensive to deal with the enormous number of molecules of interest. In this paper, we introduce an approximate model based upon physics, empirics, and information science which is proposed for use in machine learning applications in which very large numbers of sub-simulations must be made. In particular, we focus upon machine learning applications in the learning phase and argue that our model is sufficiently close to the physics that, in spite of its approximate nature, can facilitate stepping through machine learning solutions to explore the mechanics of folding mentioned above. We particularly emphasize the exploration of energy flow (power) within the molecule during folding, the possibility of energy scale invariance (above a threshold), vestigial information in the unfolded state as attractive targets for such machine language analysis, and statistical analysis of an ensemble of folding micro-steps.	physics.bio-ph	physics	Protein Folding and Machine Learning: Fundamentals Walter A. Simmons October 2018 Department of Physics and Astronomy University of Hawaii at Manoa 1 Abstract In spite of decades of research, much remains to be discovered about folding: the detailed structure of the initial (unfolded) state, vestigial folding instructions remaining only in the unfolded state, the interaction of the molecule with the solvent, instantaneous power at each point within the molecule during folding, the fact that the process is stable in spite of myriad possible disturbances, potential stabilization of trajectory by chaos, and, of course, the exact physical mechanism (code or instructions) by which the folding process is specified in the amino acid sequence. Simulations based upon microscopic physics have had some spectacular successes and continue to improve, particularly as super-computer capabilities increase. The simulations, exciting as they are, are still too slow and expensive to deal with the enormous number of molecules of interest. In this paper, we introduce an approximate model based upon physics, empirics, and information science which is proposed for use in machine learning applications in which very large numbers of sub-simulations must be made. In particular, we focus upon machine learning applications in the learning phase and argue that our model is sufficiently close to the physics that, in spite of its approximate nature, can facilitate stepping through machine learning solutions to explore the mechanics of folding mentioned above. We particularly emphasize the exploration of energy flow (power) within the molecule during folding, the possibility of energy scale invariance (above a threshold), vestigial information in the unfolded state as attractive targets for such machine language 2 analysis, and statistical analysis of an ensemble of folding micro- steps. 3 INTRODUCTION In spite of more than half a century of intensive research, the fundamental mechanisms of protein folding are still uncertain; the 'code' that carries the folding instructions in the amino acid sequence is still not known; in fact, even the most elementary physical structures that carry the instructions are not entirely agreed upon. Since protein folding is a self-organizing process, the common conceptualization of a control function and a state space is complicated. Furthermore, with ten degrees of freedom per residue, identification of a simple subset, such as dihedral angles only, is questionable. Finally, many folding processes occur near the limit of thermodynamic instability. In short, the simulation of many proteins has to manage against the accumulation of errors while maneuvering a low energy signal, over a long time frame, through a noisy background based upon uncertain parameters (1), (2). However, proteins are not disrupted by small changes in initial conformation, in temperature, or in chemical potential with the solution, and fold quickly to unique final states. Evidently, natural molecules have some aspects that make small changes in ambient conditions unimportant. (3), (4), (5), (6). Simulation is the most common approach to determining the ground state structure from the sequence. In this paper, we discuss the foundations of an alternate approach using machine learning (ML). 4 There have been many successes with simulations, but a major breakthrough will be needed to increase the speed and lower the cost, so that the structure of the tens of thousands (or more) medically important molecules can be obtained by computation alone. One major limitation of simulation, by which we mean computer simulations built upon physics and chemistry based models of the molecules, is the fact that simulations in 1010 or more time femtosecond time intervals might require intervals. The models use changing physical parameters obtained from models or from chemistry experiments with finite precision. Also, the natural instruction set has some (unknown) finite precision. The undefined parts of the instructions, whatever that might mean, enter in a non-linear way. Therefore, these limitations taken together or separately, imply that accumulative errors may develop and invalidate the simulation. This is typically treated by intensive computation, (such as minimizing the length of the time step), which is a substantial time and cost burden. The acid test of a simulation is the final state calculation; however, we propose that much can be learned using ML even if the detailed ground state cannot be calculated precisely. A final remark on simulations. The unfolded, or initial state, is only partially known. It may be the instructions essential to folding are partially encoded in conformations of this state, over and beyond the sequence (7). If so, a model that is tested against randomly constructed conformations may fail in some cases even if the instructions inherent in the sequence are properly accounted for. The dramatic success of AlphaGo-Zero in solving Go (8) has undoubtedly set the imaginations of protein scientists aflame. ML has been applied in many ways; some interesting applications: (9), (10), (11), (12), (13) {High energy physicists have also found 5 Artificial Intelligence (AI) to be a set of powerful tools (14), (15), (16), (17), (18).} This paper explores the folding fundamentals upon which machine-learning models might be built. An obvious advantage of ML over simulation is that a constant looping on a known structure (in the machine learning phase) can, in principle, reduce or eliminate accumulative calculational errors. In contemplating the application of machine learning, a statistical matrix is often adjusted in a loop to achieve pattern recognition. The fact that many valid pathways may pass near thermodynamic instabilities means that many pathways derived in the machine learning would not be valid in nature. Therefore, eliminating accumulated errors inherent in simulations, eliminating unphysical trajectories, and accommodating regions of energy scale invariance are desirable features of computer based folding. From information theory, (19), (20), (21), we note that if the symbols used in communication are energy-symmetric (same energy to make and use a zero as a one), then the message need not be modulated in energy. We discuss, below, how that might appear in folding as, 'thermodynamic perfection'. Also from information theory, we have the concept that the distribution function of the set of all possible messages with a given format and a common reservoir of symbols with fixed probabilities, is a Gaussian. Furthermore, the number of messages with the highest probabilities are those beneath the top of the Gaussian (i.e. within the variance). This useful approach has been applied to molecular biology and we discuss how this might be useful in folding research in particular. We summarize our objectives in this paper as follows. 6 Develop a model of folding that is simple but which respects the physics and especially the geometry of the molecules. Base the model upon a probabilistic process (e.g. directed Brownian motion); the process will define an ensemble of microscopic folding steps based upon informational 'words'. Propose ML in the learning phase to eliminate accumulation errors and to accommodate the weak ambient energy dependence of the final state structure. Consider an ensemble of molecules and show how the 'words' close to the center of the Gaussian can be expected to dominate; this enables immediate application of the maximum entropy method of pattern resolution. We shall not enter into the specifics of algorithms (22), which can be written in many ways. We are attempting to lay a groundwork based on empirical knowledge and physics, which can be used in variously in algorithms. Parameterization As mentioned, there is an open question of which physical proprieties of the sequence carry the instructions and exactly how that works. It is useful to consider the following generic description. The folding consists of two major processes: a random Brownian component and a directed part driven by the (limited) energy initially resident in the unfolded molecule (23). The directed force takes the specific form of driven torsion waves. The wave motion is enormously complicated since the boundary conditions change continuously during folding due to flexing of the molecule and change discontinuously due to contact formation. 7 Rotations in general do not commute and the fact the folding goes to a unique native structure, strongly suggests that there are few alternative directed paths. It is worth remarking that it is well-known in physics (23), (24), (25), that stochastic forces can smooth out perturbations in addition to driving a system down-hill. Whether that occurs in folding is not known. The use of a standard parameterization used in simulations is not particularly attractive for the ML applications envisioned here. Instead, we propose an informational parameterization based upon microscopic changes in angles consistent with molecular geometry. Detailed models can be exploited later. In this paper, we attack the first step in ML learning about folding. We parameterize fundamental units of instructions that can change on the computer without taking the model into unphysical territories. We begin by recognizing that changes in shape across some range of residues is fundamental; moreover, rotations are the most important changes, so the chain must encode directions as well as magnitudes. We collect together pieces of various models, all well known. In the earliest days of folding simulations, attention turned to triplets of residues; that is, one residue and two nearest neighbors. It is now understood that coding instructions apply over domains, if not over larger structures, and not just locally around such a triplet. In spite of these limitations, we suggest the triplet, a change in which we shall call a 'word' of instruction, be used in ML designs. The major reasons for this choice are as follows. 8 a.) With three dihedral angles per residue, a triplet of adjacent residues has six degrees of freedom. That is just as required to describe the relative positions (three) and orientations (three) of the ends of the triplet. b.) The instructions in the sequence change during folding. The angular change and energy change can be accommodated in a triplet. c.) The changes in the triplet are, in part, associated with the deterministic gross motion and, in part, by random Brownian effects. The random part is geometrical and probabilistic and can be treated without reference to energy; i.e. non energy directed motion must be included in the computer calculation. d.) Except for steric hindrance, the triplet can take the shape of any structural element. It is also flexible enough to traverse the passage through the exit of the ribosome. e.) Because of the short length, inevitable computation errors can be controlled over short distances. f.) The short word length is convenient for analysis of wave motion and is especially convenient for studying the change in wave structure resulting from contact formation. g.) Power: success in seeing folding take place step by step would create opportunities to study instantaneous power during folding. Note that some misfolds, or alternate viable conformations, are blocked by local power limitations even while the average power may be sufficient to overcome 9 some local thresholds. The importance of such detailed understanding was emphasized by Dobson in 2003 (7). See (26) for recent and detailed discussion. The unfolded state must also receive some attention. As mentioned, some unrecognized aspects of this collection of states may define part of the folding process. To carry that one step further, we suggest that the various microscopic features follow probability distribution functions. In that case, as we discussed in another context, the Central Limit Theorem suggests that these instructional features follow a Gaussian and only a microscopic number of possible such features are actually in use. In other words, the instructions in the amino acid sequence also specify the unfolded state and a Gaussian form suggests that the number of unfolded states is likely very small. This can be conceptualized by visualizing energy flowing through a limited number of Levinthal pathways; the power may originate in a highly non-uniform distribution of initial energy in the unfolded sates. In initial development the energy dependence of the word can be simplified because the word is essential geometric and intended to have sufficient flexibility to fit nearly any chain shape. When the machine learns enough to require an energy model that can be inserted at that phase. Model Construction and Goals As mentioned, we follow information theory to develop an analysis that is broader than statistical mechanics in that changes the do 10 not depend upon the energy stored in the initial state are included. The basic 'word' of code instruction embodied in the sequence is, according to the ideas just introduced, is a six-dimensional differential, (or difference), change in the six dihedral angles of the triplet. Contact formation must be modeled by computing the position of the chain, detecting contact, and making an evaluation of binding probability. Connecting the words of the sequence introduces a problem common to all computer simulations: the steps are carried out individually and have to be smoothed out. The ML approach is much less sensitive to this problem than strictly model based simulations. There are various statistical methods that follow along the lines just described. A newly created one can be found at (14), which is a physics application. Since we have referred to the common approach of treating folding as a deterministic process driven down slope by random impulse, we now analyze the proposed model with that in mind. To begin, we note that if the machine learns to fold specific structures, then one can test the effect of a change in temperature VS a change in chemical potential. That is, the relative energy scale invariance of Brownian VS driven forces. We consider an ensemble of ML calculations based on a single set of known structures and a wide range of computer generated initial states. We propose that the number of occurrences of any given six-dimensional word VS the specific words. The result will consist of two parts. The random Brownian motion will yield a 11 Gaussian. The structure arising from the directed part will be different and we can only speculate; if it is also probabilistic (over an ensemble of folds with varying initial conditions), then it too will be Gaussian with the variance representing the most common words, as in information theory. A related analysis is to vary the energy of the initial state. If the information theoretic model described here works, then the final structure will be independent of small changes in the initial energy (i.e. energy scale invariance). The distribution of words involved may be similar if there are a few fundamental modes of folding, or different if there are many folding pathways. If the conjecture about a non-uniform distribution of energy in the unfolded state, (mentioned above), is correct, then random initial configurations will not work. In detail, the calculation process has a Gaussian distribution about the directed folding path. If this compares well with experiment, then it defines the folding path rather tightly. A major point here is that the number of words appearing in a large number of folds is likely to be miniscule compared to the possible words. Said differently, certain sets of angular changes in the triplet of residues dominate over all possible sets of angles. The words that occur most frequently can be different from the words with the largest instantaneous change in energy (steepest slope). Thus, this model, in spite of being obviously simplified, accommodates energy-directed and probabilistic changes. The model can statistically reveal the relative roles of directed and random motions as well as the presence of a few, (or of many), folding pathways. 12 Conclusions: We emphasized the advantages of ML over simulations. The two most important advantages are: (i) that uncertainties in parameters and in calculations can be better suppressed in ML than in simulations and, (ii) (if only the learning phase is used, and if a sufficiently accurate model is used), then stepping through the calculation may very efficiently reveal new phenomena in folding. We argued that correct predication of the ground state, which is the acid test of simulations, is not necessary to learn about how proteins fold. A particularly interesting short term goal is understanding instantaneous power at each point in the molecule during folding (power landscape). Some folding pathways may be allowed by available energy but forbidden by the instantaneous power requirement. More generally, researchers frequently encounter situations in which general principles of folding might be very useful. Examples would be instantaneous power distribution, changes in the power landscape with contact formation, (proposed) energy scale invariance, underlying principles of process stability, other informational symmetry features, (e.g. such as rules for misfolding). Dynamically, the model suggested is based closely upon the directed Brownian model and upon the idea that torsional wave motion predominates folding. 13 Following information theory, we introduce a 'word' describing a small change in a three-residue sub segment of the chain. We suggest that ensembles of ML tests be performed; each test has a unique initial state but all tests use the same final structure. The statistics of any probabilistic processes yield Gaussians in word frequency. The number of words found for any given molecule is expected to be minuscule compared to the number possible. We also propose searching the data for energy scale invariance. The specific 'word' proposed is an approximation but should be very useful because of its properties listed earlier. Finally, we remark on the extension of the application of information science to the motions involved in folding (27). If all probabilities were the same (uniform probability case) then the number of 'possible' protein sequences for a 100 residue molecule would be, as is often stated, is to be viewed from here as probabilistic and the number of protein sequences that will actually occur is a microscopic fraction of the above number (20), (27). We have proposed that the same logic be applied to some defined individual angular changes during folding; the number of folding pathways is microscopic compared to the number of 'possible' pathways in the uniform probability case. [We remark that if cooperativity can be similarly limited then the number of possible folds is reduced by another exponential factor (Gaussian).] The often quoted timescale for years, is also reduced uniform probability folding, which is exponentially by a cooperativity Gaussian and possibly another Gaussian factor due to the distribution function of energy in the unfolded state. We believe that these calculations can be carried out using ML as described in this paper. . However, evolution (20) 100 T ≥ 1010 14 References: [1] Bottaro, S.2018, & K., Lindorff-Larsen, "Biophysical experiments and biomolecular simulations: A perfect match?", Science 361, 355. [2] Bereau, T., 2018, J.F. Rudzinski, "Circumventing additivity in molecular mechanics with conformationally-dependent surface hopping",arxiv.org/abs/1808.05644v1 [3] Levinthal, C., 1968, "Are there pathways in protein folding?" Journal de chimie physique. [4] Dill, K.A., 1990, "Dominant Forces in Protein Folding", Biochemistry, 29(31), 7133-7155. [5] Simmons, W. & Weiner, J.L., 2015, "Topology, Geometry, and Stability: Protein Folding and Evolution", arxiv 1505.07153. [6] Simmons, W., 2017, "Protein Folding Problem: Scientific Basics]. Arxiv 1709.07953. [7] Dobson, C.M., 2003, "Protein folding and misfolding", Nature 426, 884. [8] Silver, D., 2017, J. Schrittwieser, K. Simonyan, I., Antonoglou, A.,Huang, A., Guez, , T., Hubert, L., Baker, M., Lai, A., Bolton, Y., Chen, T., Lillicrap, F., Hui, L., Sifre, G., van den Driessche, T., Graepel, D., Hassabis, "Mastering the game of Go without human knowledge", Nature 550, 354-359. [9] Bengo, Y., 2016. Goodfellow, I.J., & Courville, A., "Deep Learning", MIT. [10] Jurtz, V. I., 2017, A.R., Johansen, M., Nielsen, J.J.A. , Armenteros, H., Nielsen, C.K. ,Sonderby, O., Winther, S.K., Sonderby, "An introduction to deep learning on biological 15 sequence data: examples and solutions" Bioinformatics 33, 3685- 3690. [11] de Oliveira, L., Michaela Paganini, & Benjamine Nachman, "Learning Particle Physics by Example: Location-Aware Generative Adversarial Networks for Physics Synthesis" [12] Stahl, K., Schneider, M., Brock, O., 2017,, "EPSLON-CP: using deep learning to combine information from multiple sources for protein contact prediction." BMC Bioinformatics 18, 303. [13] Zhao, Z.Q., Zu-Gou,Y., Vo, A., Jing-Yang, W., Guo-Sheng, H. 2016, "Protein folding kinetic order prediction from amino acid sequence based on horizontal visibility network." Bioinformatics 11, 173-185. [14] Aartsen, M.G., Ice-Cube-Gen2 Collaboration, 2018, "Computational Techniques for the Analysis of Small Signals in High-Statistics Neutrino Oscillation Experiments", arXiv: 1803. 05390; pages 1-32. [15] Baldi, P., 2014, Sadowski, P., & D. Whiteson, "Searching for exotic particles in high-energy physics with deep learning." Nature Communications volume 5, Article number: 4308. [16] Lloyd, S., 2018 & C. Weedbrook, "Quantum Generative Adversarial Learning" PHYSICAL REVIEW LETTERS 121, 040502-1, 040502-5. [17] Dunjko, V., 2016, J.M., Taylor, and H.J. Briegel,"Quantum Enhanced Machine Learning" PHYSICAL REVIEW LETTERS 117, 130501-1. 13050-6. [18] Larrañaga, P., 2006, B. Calvo, R. Santana, C. Bielza, J. Galdiano, I. Inza, J.A. Lozano, R. Armananzas, G. Santafe, A. Perez, and V. Robles, "Machine learning in bioinformatics", 16 Briefings in Bioinformatics, Volume 7, Issue 1, , Pages 86 -- 112, https://doi.org/10.1093/bib/bbk007 [19] Ben-Naim, A., 2017, "Information Theory", World Scientific. [20] Yockey, H.P., 1992, "Information theory and molecular biology", Cambridge University Press. [21] Yockey, H.P., 2006, "Information Theory and the Origin of Life", Cambridge University Press, . [22] Witten, I.H., 2017, Frank, E., Hall, M.A., C.J Pal, Data Mining, Morgan Kaufmann Cambridge, Ma . [23] Huang, K., "Statistical Physics and Protein Folding", World Scientific (2005). [24] Yang, L. "Fighting Chaos with Chaos in Lasers", Science 361, 1201 (2018) [25] Cao, H., 1999, Zhao, Y.G., Ho, S.T., Seelig, E.W., Wang, Q.H., and Chang, R.P.H., "Random Laser Action in Semiconductor Powder", Physical Review Letters 82, 2278- 2281. [26] Chung, H.S., 2018, & Eaton, W.A. ,"Protein folding transition path times from single molecule FRETG", Current Opinion in Structural Biology, 48, 30-39. [27] Presse, S., 2013, Ghosh, K., Lee, J. and Dill, K., "Principles of maximum entropy and maximum caliber in statistical physics", Reviews of Modern Physics, 85, 1115-1141. 17 18
1110.5997	4	1110	2011-11-16T21:23:04	How the DNA sequence affects the Hill curve of transcriptional response	[ "physics.bio-ph", "q-bio.SC" ]	The Hill coefficient is often used as a direct measure of the cooperativity of binding processes. It is an essential tool for probing properties of reactions in many biochemical systems. Here we analyze existing experimental data and demonstrate that the Hill coefficient characterizing the binding of transcription factors to their cognate sites can in fact be larger than one -- the standard indication of cooperativity -- even in the absence of any standard cooperative binding mechanism. By studying the problem analytically, we demonstrate that this effect occurs due to the disordered binding energy of the transcription factor to the DNA molecule and the steric interactions between the different copies of the transcription factor. We show that the enhanced Hill coefficient implies a significant reduction in the number of copies of the transcription factors which is needed to occupy a cognate site and, in many cases, can explain existing estimates for numbers of the transcription factors in cells. The mechanism is general and should be applicable to other biological recognition processes.	physics.bio-ph	physics	How the DNA sequence affects the Hill curve of transcriptional response Department of Physics and Astronomy, Vrije Universiteit, Amsterdam, The Netherlands M. Sheinman Y. Kafri Department of Physics, Technion, Haifa 32000, Israel (Dated: July 11, 2018) The Hill coefficient is often used as a direct measure of the cooperativity of binding processes. It is an essential tool for probing properties of reactions in many biochemical systems. Here we analyze existing experimental data and demonstrate that the Hill coefficient characterizing the binding of transcription factors to their cognate sites can in fact be larger than one -- the standard indication of cooperativity -- even in the absence of any standard cooperative binding mechanism. By studying the problem analytically, we demonstrate that this effect occurs due to the disordered binding energy of the transcription factor to the DNA molecule and the steric interactions between the different copies of the transcription factor. We show that the enhanced Hill coefficient implies a significant reduction in the number of copies of the transcription factors which is needed to occupy a cognate site and, in many cases, can explain existing estimates for numbers of the transcription factors in cells. The mechanism is general and should be applicable to other biological recognition processes. PACS numbers: Molecular recognition plays an important role in bi- ological systems ranging from antigen-antibody identifi- cation to protein-protein binding [1]. In many cases the recognition process is driven by free-energy differences between a desired reaction and many competing unde- sired reactions [2 -- 5]. One example, of particular impor- tance, is that of protein-DNA interactions. Its role in un- derstanding regulation in cells has led to large experimen- tal effort to which aims at mapping the binding energy between transcription factors (TFs) in their specific state and different subsequences on the DNA [6]. It is known that to a good approximation the energy can be written as a sum of energies representing the binding energy of a nucleotide on the DNA to the region on the protein with which it is aligned [7, 8]. Specifically, the binding energy of a nucleotide s = A, C, G, T to position j = 1, 2, ..., L (where L is the length of the protein's DNA binding do- main in units of basepairs) is usually described by a 4× L position weight matrix (PWM), ǫs,j. By now, the PWM is known for many proteins and, together with a knowl- edge of the genomic sequence, it specifies the binding energy landscape of TFs to the DNA. Irrespective of the energy landscape properties the ac- tivation of a cognate site -- a specific location on the DNA -- by a TF is usually described by a Hill curve [9]. Namely, if we consider a DNA molecular inside a con- tainer representing, say, a prokaryotic cell the activation probability of an operator by a TF is given by, PT = (1) 1 m (cid:1)n . 1 +(cid:0) m1/2 Here m is the number of TFs in the cell and at m = m1/2 the occupation probability is one half (the conversion to concentrations is trivial). The Hill coefficient (HC), n, governs the steepness of the curve and is widely used to extract qualitative information about the regulation of genes from experimental data [10 -- 13]. In the simplest cases, when there is no cooperative binding involved one expects n = 1. In the presence of cooperative interac- tions n is different than one. For example, in the case of activation by dimers one expects n = 2 if m is the number of monomers. In this article we demonstrate that this simple intuitive picture for the Hill curve can fail. This is a direct conse- quence of a non-trivial combination of variations in the binding energy of TFs to different sites along the DNA and the steric repulsion between them. This leads to (a) a disorder enhanced Hill coefficient which is larger than one even in the absence of any cooperative binding to the operator, and (b) a dramatic increase in the occupation probability of the cognate site as compared to a system with no steric interactions between the TFs or a constant non-cognate binding energy. Importantly, we show that the results are essential for explaining the number of TFs found in cells. I. RESULTS The Hill curve, Eq. (1), is directly related to a for- mulation of the problem using statistical mechanics and the knowledge of the experimentally measured binding energy landscape of the TF to the DNA. To illustrate this we first focus on a simple case where: (i) There is no cooperativity associated with the structure of the TF or its binding properties, such that one would naively expects n = 1. (ii) The probability of the TF to be off the DNA or in a non-specific conformation on the DNA is negligible. Note that by a non-specific conformation we mean one where the TF is on the DNA but does not interact with the bases. This conformation, which typically occurs due to electrostatic interactions, exists 106 (b) m1/2 103 106 (a) 104 m1/2 102 100 108 106 m1/2 104 102 −60 −50 −40 −30 −20 (d) −30 −25 −20 (c) mc 100 2 1.8 1.6 n 1.4 1.2 1 −15 −10 −5 ET −30 −25 −20 ET FIG. 1: The value of m1/2 is presented as a function of the energy of different cognate sequences, ET for (a) LexA, (b) RpoN and (c) Lrp. The dotted lines are based on a numeri- cal calculation using the E.coli genome. The circles represent energies calculated from all known cognate sites of the TF. The black solid lines are based on the freezing regime approx- imation, Eq. (13), while the red dashed lines are based on the non-steric approximation, Eq. (8). Filled, grey horizon- tal areas show the typical range of the TF's copy number in E.coli. The symbol mc in (c) marks the crossover value of the Lrp TF between the uncrowded and crowded regimes, pre- dicted by Eqs. (27) and (30). (d) The HC, obtained by a fit of the numerical data to Eq. 1, is shown as a function of the cognate site energy for the LexA TF. The solid line is the analytical prediction, Eq. (14), while the circles represent nu- merical data based on real DNA and cognate sites sequence. The filled circle represents the HC of a hypothetical cognate site with a perfect consensus sequence. The dashed line rep- resents the result of the non-steric approximation, Eq. (5), which gives n = 1. on any location along the DNA, including the cognate site. The effects of both simplifications are discussed in the Supplementary Information (SI). Using standard statistical mechanics it is straightforward to calculate m1/2 and PT (m) numerically (see Methods). We use the PRODORIC database for PWMs of E.coli TFs [16], their cognate sequences and the genomic sequence of E.coli of In the next section we length N = 2 × 4686077 [17]. discuss the results of this approach and show that it can lead to rather counterintuitive Hill curves. We then give a simple theory which accounts for the results analyti- cally and gives precise conditions for these effects to be important. A. E.coli TFs Hill curves First, we evaluate numerically m1/2 for all known cog- nate sites of three TFs. The results are shown in Fig. 1 (for the moment focus on the data shown as circles in 2 panels a, b and c). As can be seen (and intuitively clear), the weaker the binding energy of the cognate site (larger ET ) the larger m1/2. The values of m1/2 span a large range for different TFs and for different cognate sites of the same TF. Next, in Figs. 2 and 3 (b and c panels) we present the occupation probability of typical cognate sites for two representative TFs, one cognate site of LexA and two of RpoN. In addition we fit the results to a Hill curve. With the value of m1/2 given the only fit param- eter is the HC, n. Surprisingly, for the three cases (and others not shown) we obtain n > 1 (in Figs. 2 and 3 we obtain n = 1.7, 8.3, 2.03), despite the absence of coop- erativity in the model between the TF copies apart from steric interactions. As shown below this occurs for many TFs. Remarkably, performing the same procedure but ig- noring the steric repulsion of the TFs everywhere apart from the cognate site one obtains qualitatively and quan- titatively distinct results (see Figs. 1, 2 and 3). This is despite the fact that in most cases the ratio of the num- ber of TFs in the cell to the length of the DNA is very small. In fact, ignoring steric repulsion leads to a signifi- cant increasing in the value of m1/2 (in some cases much above the measured estimate of number of TFs in the cell). Moreover, this simplification always yields a HC of n = 1. Similarly, it is clear that without disorder in the binding energy of the TF to different sites along the DNA one would obtain n = 1. In sum, to account properly for the Hill curve of TFs binding to DNA one must (a) account for the disordered binding energy of the TF to different sites along thee DNA and (b) account for steric interactions between dif- ferent copies of the TFs. Without these the HC would be lower than the actual one (later we show that this is true even for cases where naively one expects a HC which is larger than one) and m1/2 would be much larger than the real one. Those are the main results of this paper. In what follows we explain the obtained results using an analytical approach. We show that the effect is generic and illustrate its importance for many TFs. B. Analytical solution As stated above, the binding energy to any sequence on the DNA is a sum of L = 10 − 30 independent vari- ables. Therefore, assuming a pseudorandom sequence on the non-cognate DNA, the probability distribution of the binding energy, Pr (Ei), is close to normal (see [4, 5] and Figs. 2(a) and 3(a)): Pr (Ei) ∝ exp(cid:18)− E2 i 2σ2(cid:19) . (2) Here we define the centre of the normal distribution as zero energy. The value of σ characterizes the width of the disordered binding energy profile along the DNA and is usually in the range 2−8, where throughout the paper we measure energy in units of kBT with kB the Boltzmann (a) 104 102 100 100 (b) 10−2 PT 10−4 mc −30 −20 −10 0 E 10 20 30 100 102 m 104 3 103 m 104 10 1 (c) 0.8 0.6 PT 0.4 0.2 0 102 FIG. 2: (a) A histogram of the binding energies of the LexA TF to the E. coli genome. The red points represent numerical data while the line is based on a Gaussian fit with σ = 5.76. The big filled circle marks the binding energy to the recQ operon sequence with ET = −20.6. (b) The occupation probability of the LexA protein to the recQ operon as a function of m. The circles represent numerical data based on the PMW of the protein and the E. coli DNA sequence. The dashed line is based on the non-steric approximation, Eq. (3), while the solid line on Eq. (12). The solid arrow shows the crossover value (nonphysical in this case) of the protein copy number, mc = 0.65, predicted by Eqs. (27), and (30). (c) The same data as shown in panel (b) with a Hill function, Eq. (1), fit (thin line) to the numerical data (circles) with n = 1.7 and m1/2 = 3000. The gray areas in (b) and (c) show the typical range of the LexA copy number in E.coli (200 − 4000) [14]. 106 105 104 103 102 101 100 1 2 100 10−2 PT 10−4 10−6 10−8 (a) 2 1 1 1 1 0.8 0.6 PT 0.4 0.2 2 2 1 −40 −20 0 20 40 E 100 102 m 104 (b) 0 100 (c) 102 m 104 FIG. 3: (a) A histogram of the binding energy of the RpoN TF to the E. coli genome. The red points represent the numerical data while the line is based on a Gaussian fit with σ = 7.8. The circle shows the binding energy to the argT operon with ET = −41.8. The square shows the binding energy to the fixABCX operon with ET = −28.9. (b) The occupation probability of the RpoN protein for the operons argT (circles and 1 symbol) and fixABCX (squares and 2 symbol) as a function of m. The circles and squares represent numerical data based on the PWMs of the TFs and the E. coli DNA sequence. The dashed (dotted) line is based on the non-steric approximation, Eq. (3), for the argT (fixABCX) operon, while the solid lines are based on Eq. (12). (c) The same numerical data as in panel (b) with a fit of a Hill function, Eq. (1), (thin, solid, black lines) to the numerical data with n = 8.3 and m1/2 = 3.3 for the argT operon and with n = 2.07 and m1/2 = 1300 for the fixABCX operon. Vertical areas in panels (b) and (c) show the typical number of the RpoN proteins in E.coli (∼ 110) [15]. constant and T the temperature. The validity of this approximation is illustration in Figs. 2(a) and 3(a) where we plot the histograms of the binding energies of two typical TFs on the E.coli DNA. the cognate site is given by P non−steric T = 1 1 + N m eET + σ2 2 , (3) Before presenting a full analysis of the problem it is interesting to consider a case where the different copies of the TFs have steric interactions only on the cognate site. We refer to this as the non-steric approximation. As shown in the SI, here the occupation probability of where ET is the energy of the operator of interest. Thus, by comparing with Eq. (1) one has m1/2 = N eET + σ2 2 , n = 1 . (4) (5) In particular, this implies that without steric interactions (a) (b) 4 We note that the disorder width, σ, has different values for different TFs with typical values in the range of 2− 8. Thus, the crossover value of the protein's copy number varies from a non-physical 10−7 (where the non-steric approximation surely fails) to 105 where the non-steric approximation is expected to hold unless the number of TFs is extremely high. The above derivation implies that the non-steric ap- proximation (valid in the uncrowded regime) will give a good estimation of the HC and the number of particles at half occupation, σ σ FIG. 4: An illustration of the crowding effect in an energy landscapes with high disorder. Typical disordered energy pro- files, with small and high disorder strength, are plotted in the (a) and (b) panels respectively. The circles represent TFs who's number is much smaller than the possible binding sites on the DNA. Therefore, in the small disorder case, (a), they spread evenly across the DNA. When the disorder strength is high (panel (b)) the TFs compete on a limited number of traps. Then steric interaction between those competing on these small number of traps become important. As the traps fill with TFs the occupation probability of the cognate site increases dramatically. This leads to a reduction in m1/2 and an enhanced HC. there is no enhanced HC and one obtains the naive re- sults. From the numerical experiment presented before, clearly, the non-steric approximation can fail. In addi- tion to giving a wrong value for n it gives, in some cases, unreasonable values of m1/2. Indeed, in Figs. 2 and 3 we show numerical results for LexA compared with Eq. (3). The values of m1/2 differ by more than two orders of mag- nitude and as stated above the value of n is larger than one. Moreover, the number of LexA TFs in the cell is estimated to be between 200 to 4000, a value much lower than the m1/2 = 105 found in the non-steric approxi- mation. A similar disagreement between the numerical results and the non-steric approximation occurs for RpoN as well as for many other TFs. To analyze the problem more carefully, taking into account steric interactions, we calculate the occupation probability of the cognate site in two limiting cases (see Methods and SI): a. Uncrowded regime: In this regime, illustrated in Fig. 4(a), the number of TFs is smaller than a crossover value defined by the DNA length and the disorder width: m ≪ mc = N e− σ2 2 . (6) The occupation probability in this regime is given by the non-steric approximation, Eq. (3), so that: P uncrowded T = 1 1 + N m eET + σ2 2 = P non−steric T . (7) m1/2 = N eET + σ2 2 n = 1, (8) (9) only when m1/2 ≪ mc. Furthermore, to be in the self averaging regime near m ≃ m1/2 the cognate site of in- terest, with energy ET , has to satisfy the condition ET < −σ2 . (10) b. Crowded regime: In this regime, illustrated in Fig. 4(b), m ≫ mc = N e− σ2 2 . and the occupation probability is given by (11) −1 .   P crowded T 1 + =  N exp(cid:20)ET + σr2 ln(cid:16) σr2 ln(cid:16) (cid:20) 1 n√1+2σ2(cid:17)(cid:21)−1 n√1+2σ2(cid:17)(cid:21) − 1 N (12) If close to saturation of the cognate site the system is in the crowded regime, m1/2 ≫ mc. Then comparing Eqs. (12) and (1) one can see that m1/2 = n = W 2(e−ET σ2) 2σ2 N e− √1 + 2σ2 1 + W (cid:0)e−ET σ2(cid:1) W 2 (e−ET σ2) σ2, (13) (14) where W is the Lambert W -function [19]. The function is well approximated by W (X) ≃ ln(X) for large X. Note that to be in the crowded regime close to saturation of the cognate site the condition ET > −σ2 (15) has to be satisfied. To check our analytical results we plot both the non- steric approximation and the results from the crowded regime in Figs. 2 and 3. The non-steric approximation, (3), clearly fails. In contrast Eq. (12) agrees well with the numerical data (strictly the uncrowded regime result In gives a good approximation below mc = N e− particular, the HC of the numerical data in the analyzed σ2 2 ). r o t c a f n o i t p i r c s n a r T 20 15 10 5 0 −20 0 20 40 5 F adR σ = 4.6 m = 295 mc = 207 T yrR σ = 5.0 m = 145 mc = 33 N arL σ = 3.7 m = 230 mc = 104 M odE σ = 5.6 m = 290 mc = 1 GntR σ = 4.7 m = 200 mc = 57 F ur σ = 5.3 m = 570 mc = 10 M etJ σ = 5.6 m = 455 mc = 2 Lrp σ = 3.4 m = 5000 mc = 104 F N R σ = 4.3 m = 450 mc = 1000 DnaA σ = 4.0 m = 110 mc = 3000 RpoN σ = 7.8 m = 110 mc = 10−6 F is σ = 2.7 m = 100 − 50000 mc = 105 C RP σ = 4.4 m = 1300 mc = 600 LacI σ = 4.7 m = 10 mc = 57 OxyR σ = 5.8 m = 160 mc = 6 argR σ = 5.0 m = 330 − 510 mc = 40 LexA σ = 5.8 m = 200 − 4000 mc = 0.65 −10 −8 −6 −4 −2 0 2 4 6 10 12 14 16 18 20 22 24 26 28 30 8 ET − ¡−σ2¢ FIG. 5: A classification of cognate sites of different TFs. For each TF (y-axis) ET − (cid:0)−σ2(cid:1) is plotted for all its known cognate sites (taken from [16]). Empty rectangles represent a hypothetical cognate site's energy that is half-occupied if the number of TFs is as specified in the database [18] (see legend). For TFs with m < mc filled red areas represent the same hypothetical energy, as predicted by Eqs. (8). When the typical number of the TFs in the cell is larger than mc the hypothetical cognate sites' energies calculated from the non-steric approximation (4) are presented by the red filled area while the results of applying Eq. (13) are presented by the blue filled areas. cases is clearly above one and depends on the disorder width and the cognate site's energy, as demonstrated by Eq. (12). It is interesting to note that close to saturation the Hill curve (determined by the value of m1/2 and the steepness of PT ) is well described by Eqs. (8),(9),(13) and (14), as presented in Fig. 1. However, in some cases there is a disagreement between the theory and the numerical results for small values of the protein copy number (see the small m values in Fig. 3(b) and (c) and the low ET values in Fig. 1(b) and (d)). This disagreement is due to the deviation of the binding energy probability density from a normal distribution at the low energy tail (see Figs. 2(a) and 3(a)). This effect, which is easy to calculate numerically, increases both the value of m1/2 and n relative to the analytical predictions. As shown in Fig. 3(a) and (b) in some cases this effect may be significant and lead to a very large HC (n = 8.3 for the argT operon of RpoN). In all the presented cases the non-steric approximation clearly fails by several orders of magnitude. To examine the validity and relevance of these results to other TFs we use a database of 17 known PWMs of TFs of E. coli, chosen randomly, and their known binding sites' sequences [16]. Many of the TFs are present in large copy numbers, larger than their mc value (see the legend in Fig. 5). Therefore, the non-steric approximation of the occupation probability, (3), does not approximate well the occupation probability of many cognate sequences in the biologically relevant regime. In addition, as one can see in Fig. 5, the value of ET + σ2 for many cognate sites of many proteins is positive. As suggested by Eq. (15), to describe the occupation probability of such cognate sites Eqs. (7), and (12) have to be used, while the non-steric approximation, Eq. (3), fails. It is instructive to calculate the chemical potential or the value of a hypothetical cognate site energy, denoted by µ, which is half-occupied when the number of the TF, m, is as it is measured in experiments. The results are shown in Fig. 5. One can clearly see that the value of µ is smaller (larger) than −σ2 if m is smaller (larger) than mc, as suggested by conditions (10), and (15). Moreover, the non-steric approximation predicts well the value of µ for m < mc but fails for m > mc predicting much lower values and, therefore predicting very small occupa- tion probability for all the cognate sites with an energy above µ. In contrast, Eqs. (28), and (31) predict well the location of the hypothetical, half-occupied energy for both weak and strong cognate sequences. II. DISCUSSION The considerations discussed in this paper suggest the existence of a disorder enhanced HC. They provide an estimate of the TF's copy number needed to significantly occupy its cognate sites. This estimate is shown in many cases to be much smaller and more consistent with the existing data than a naive estimate, based on the non- steric approximation. By analyzing the disordered statis- tical mechanics problem analytically, we show that two regimes are possible. In the uncrowded regime the num- ber of TFs, m is much smaller than the crossover value mc and steric interactions can be ignored. In contrast, in the crowded regime, m ≫ mc, the steric interactions play an important role and change dramatically the satu- ration curve, both quantitatively and qualitatively. The crossover value, mc = N e−σ2/2, can be much smaller than the DNA length and in many cases is much smaller than the measured number of the TF in the cell. In addition, for weak cognate sites with a binding energy, ET > −σ2, the HC is always greater than one, even in the absence of cooperative binding, and in principle un- bounded from above. To understand intuitively this effect we note that in the high disorder regime (where the non-steric approximation fails) the partition function of the system is dominated by a small number of sites with a low energy [20]. Therefore, any TF added to the system will immediately bind to one of these low energy sites. Their number can be much smaller than the total number of sites and comparable to the number of TFs. With this in mind it is clear that steric interactions play an important role in this regime and change quantitatively and qualitatively the saturation curve, including a disorder induced HC (see Fig. 4 for illustration). In the SI we show that this effect persists as long as the proteins spend most of their time in a specific state on the DNA. When this is not the case the value of m1/2 grows dramatically making this a rather costly option for many TFs [21]. In fact, as we show for many of the TFs we consider, data on TF numbers in E. coli [14] suggests a value of m1/2 which does not agree with the partition function which is not dominated by specific conforma- tions of the TF on the DNA. Finally, we show that when naive considerations give a HC of n the disorder will lead to a HC which is given by n · n, with n the disorder en- hanced HC obtained when there is no cooperative binding (see SI). The above results are summarized in Fig. 5 where we plot for several TFs and all their known cognate sites the value of ET + σ2. The regime ET + σ2 smaller/greater than zero corresponds to a without/with disorder en- hanced HC. In addition we calculate using data on TF numbers the value of ET +σ2 which would yield PT = 1/2 with the typical estimated number of TFs in E. coli. As can be seen the numbers indicate that a disorder induced HC is likely for a significant fraction of the TFs and their cognate sites. This happens since, as shown in the leg- end of Fig. 5, many TFs are present in numbers much larger than mc and, therefore, are located in the crowded regime. This study was performed using the measured PWMs of several TFs and can easily be extended to oth- ers. Acknowledgments 6 We acknowledge useful comments from A. Horovitz, R. Voituriez, O. Benichou, L. Mirny, E. Braun, A. Sharma and M. Depken. M.S. thanks FOM/NWO for financial support. III. METHODS For a cognate site with energy ET the occupation prob- ability is given by PT = 1 1 + eET −µ . (16) The chemical potential, µ, is set by the solution of the equation e−Fns+µ + 1 1 + eEi−µ = m . N Xi=1 (17) Here Ei is the binding energy of the TF at location i = 1, 2, . . . N , with N the number of accessible DNA binding sites, on the DNA. We assume that different copies of the TF exhibit steric interactions, resulting in Fermi-Dirac statistics in Eq. (17). Fns is the free energy associated with a TF in the solution or in nonspecific conformations on the DNA [4, 5, 22, 23] and, for now, under our assumption (ii) is negligible. In what follows it will be useful to bear in mind that a non-negligible con- tribution can only reduce the value of PT and enhance m1/2. We discuss its possible contribution in SI. To ob- tain the occupation probability of the cognate sites of the analyzed TFs we solve numerically Eqs. (16) and (17). IV. SUPPLEMENTARY INFORMATION The analysis in the article relied on (i) no inherent co- operativity in the TF, and (ii) a negligible probability to be in non-specific states either on or off the DNA. We now turn to discuss the influence relieving these assumptions. A. Effects of cooperativity: In our study, so far, we ignored the possibility of coop- erative effects between distinct molecules of the TF. How- ever, many TFs are active only in their ¯n-meric form. In this case the HC is usually expected to be close to ¯n since the number of active molecules of the TF is proportional to the ¯n'th power of the concentration of monomers [9]. One can reformulate the above derivation with m acting as the number of active, ¯n-meric copies of the TF. In this case the HC according to the arguments given above is ¯n· n. Thus, a combination of cooperativity and quenched disorder can naturally lead to a high values of the HC. B. The effect of nonspecific states: In our study, so far, we ignored nonspecific states of the protein. These states exist and correspond either to the TF being in the solution or in a nonspecific conformation bound to the DNA [4, 5, 22, 23]. Namely, the nonspecific free energy is given by Fns = ln(cid:0)e−E3D + N e−E2(cid:1) , (18) where E3D is the free energy of an unbound TF (mod- eled say by the free-energy of a solution of TFs) and E2 is the energy of the bound protein in the nonspecific con- formation which for simplicity we assume to have the same energy for all sites along the DNA (the results are unchanged, being proportional to N , in the presence of small disorder in this binding energy with a slight rein- terpretation of E2). Clearly, these nonspecific states can only reduce the occupation probabilities of the cognate sites calculated in Eqs. (7) and (12). However, as sug- gested previously the existence of these nonspecific states can be an important component for the dynamics of the search and recognition process carried out by the TF [4, 5, 22, 24]. From Eq. (17) one can see that the nonspecific energy is negligible when Fns ≫ µ + ln m. (19) Otherwise a significant fraction of the TFs are in the nonspecific state so that the chemical potential is well approximated by Fns. Thus, the occupation probability of the cognate site may be approximated by PT = min 1 eET −µ + 1 1 eET −Fns , m + 1! , (20) where µ is given by Eqs. (7), and (12). In Fig. 6 we show that indeed this results agree well with the numerical calculations for different values of the nonspecific energy. Thus, condition (19) implies that a half occupation of the cognate site occurs in the freezing regime (so that n > 1) if Fns ≫ ET + ln m1/2. (21) Since we are not aware of direct measurements of the nonspecific binding it is hard to estimate if this condition holds. We do stress, however, that the experimentally measured numbers of TFs in the cell suggest that it does not play an important role for many TFs. In sum, the nonspecific states may be ignored if most of the TFs are mostly bound to the DNA in a specific conformation. Otherwise the chemical potential takes a value which is closer to the nonspecific free energy. This causes the occupation probability to be well approx- imated by the non-steric approximation which results in a Hill curve with n = 1 and m1/2 = eET −Fns. However, in many cases the free energy of the nonspecific states 7 100 PT Theory Fns > −25 Fns = −30 Fns = −32 100 Fns = −34 Fns = −36 m 100 10−2 PT 10−4 10−6 100 102 m 104 FIG. 6: The occupation probability of the LexA TF to one of its cognate sequences, the recQ operon, is presented for dif- ferent values of the nonspecific energy. Different symbols rep- resent the numerical data (see legend) while the solid curves are based on Eq. (20). which is needed to bring the system to the self averaging regime is very low. Then the cognate site is significantly occupied only if the number of TFs is much larger than its typical value in the cell. In Fig. 7 the parameters of the Hill curve are shown for different values of Fns. One can see that to decrease the HC to one, the value of m1/2 has to be much larger than the typical copy number of the protein. Note, that even in this case the occupation probability might scale as mn with n > 1 far from the saturation regime, mc ≪ m ≪ exp (Fns − µ) (see Fig. 6, for an example). In fact the existence of the nonspecific states can totally eliminate the disorder induced HC for all values of m only if the condition Fns ≪ µ + ln mc is satisfied. C. Analytical solution Assuming a pseudorandom DNA sequence implies that Eq. (17) is well approximated, for N ≫ 1, by m N Pr (E) 1 + eE−µ dE = ∞ (cid:28) 1 1 + eE−µ(cid:29) = −∞ , (22) where the binding energy probability density is given by Eq. (2) and the angular brackets are defined by the in- tegral. First we discuss the non-steric approximation. In this case the Boltzmann statistics takes place of the Fermi- Dirac one everywhere except from the cognate site. Since occupation of the target site is much smaller than m it can be neglected and the constraint on the chemical po- tential is given by (cid:10)e−E+µ(cid:11) = e σ2 2 +µ = m N (23) where E ∗ may be solved self-consistently in two limiting cases: is the value of E at the saddle point. Eq. (25) c. Uncrowded regime: In this regime the saddle 8 point occurs at E ∗ ≃ −σ2, so that Eq. (25) implies = −σ2(cid:16)1 − eσ2/2(cid:17) . m N E ∗ Self-consistency in this regime requires m ≪ mc = N e− σ2 2 . The chemical potential (24) in this limit is given by µ = − σ2 m(cid:19) 2 − ln(cid:18) N (26) (27) (28) and the occupation probability is given by Eq. (7). d. Crowded regime: In this regime the saddle point satisfies E ∗ ≪ σ2 and Eq. (25) implies E ∗ = −σs2 ln(cid:18) N m√1 + 2σ2(cid:19) (29) (30) 105 104 m1/2 103 102 −32 −30 −28 −26 −24 −22 Fns n 1.8 1.7 1.6 1.5 1.4 1.3 1.2 1.1 1 FIG. 7: The parameters of the fitted Hill curve, Eq. (1), are presented for the occupation probability of the LexA TF to one of its cognate sequences, the recQ operon as a function of the nonspecific free energy. The circles represent m1/2 while the squares represent n. The filled gray horizontal area shows the typical range of the LexA copy number in E.coli (200 − 4000) [14]. Substituting the solution for µ in Eq. (16) one gets the non-steric approximation (3). As is shown in the main text the non-steric approxi- mation clearly fails. We turn now evaluate the integral in Eq. (22) using a saddle point approximation. The resulting set of equations are Self-consistency requires m ≫ mc = N e− σ2 2 . The chemical potential in this limit is then given by µ = E ∗ + ln(cid:18)− σ2 E ∗ − 1(cid:19) and ∗ E2 2σ2 e− σ2 + 1 E∗+σ2 + = m N , 1 (E∗+σ2)2 σ2q 1 (24) µ = −σs2 ln(cid:18) N n√1 + 2σ2(cid:19)+ln  σ r2 ln(cid:16) N n√1+2σ2(cid:17) − 1 (31) (25) and the occupation probability by Eq. (12). [1] B. Alberts, D. Bray, J. Lewis, M. Raff, K. Roberts, and J. D. Watson, The molecular biology of the cell (Garland, New York, 1994), 4th ed. [10] N. E. Buchler, U. Gerland, and T. Hwa, Proc. Nat Acad. Sci. USA 102, 9559 (2005). [11] T. Kuhlman, Z. Zhang, M. H. Saier, and T. Hwa, Proc. [2] A. Perelson and G. Weisbuch, Rev. Mod. Phys. 69, 1219 Natl Acad. Sci. USA 104, 6043 (2007). (1997). [12] H. D. Kim, T. Shay, E. K. O'Shea, and A. Regev, Science [3] J. Zhang, S. Maslov, and E. I. Shakhnovich, Mol. Sys. 325, 429 (2009). Biol. 4, 1 (2008). [13] H. G. Garcia, A. Sanchez, T. Kuhlman, J. Kondev, and [4] U. Gerland, J. D. Moroz, and T. Hwa, Proc. Natl. Acad. R. Phillips, Trends in Cell Biology 20, 7233 (2010). Sci. USA 99, 12015 (2002). [14] A.-M. Dri and P. L. Moreau, Mol. Microbiol. 12, 621 [5] M. Slutsky and L. A. Mirny, Biophys. J. 87, 4021 (2004). [6] J. Wang, J. Lu, G. Gu, and Y. Liu, J. Endocrinology 210, 15 (2011). (1994). [15] M. Jishage, A. Iwata, S. Ueda, and A. Ishihama, J. Bac- teriol. 178, 5447 (1996). [7] O. G. Berg and P. H. von Hippel, J. Mol. Biol. 193, 723 [16] R. Munch, K. Hiller, H. Barg, D. Heldt, S. Linz, E. Win- (1987). [8] G. Stormo and D. Field, Trends Biochem. Sci. 23 (1998). [9] R. Phillips, J. Kondev, and J. Theriot, Physical biology of the cell (Taylor and Francis, New York, 2009). gender, and D. Jahn, Nucl. Acid. Res. 31, 266 (2003). [17] F. Blattner, G. I. Plunkett, C. Bloch, N. Perna, V. Bur- land, M. Riley, J. Collado-Vides, J. Glasner, C. Rode, G. Mayhew, et al., Science 277, 1453 (1997). [18] Y. Ishihama, T. Schmidt, J. Rappsilber, M. Mann, F. U. Hartl, M. Kerner, and D. Frishman, BMC Genomics 9, 102 (2008). [19] M. Abramowitz and I. Stegun, Handbook of mathematical functions with formulas, graphs, and mathematical tables (Dover Publications, Incorporated, 1970). [20] B. Derrida, Phys. Rev. Lett. 45, 79 (1980). [21] G.-W. Li, O. G. Berg, and J. Elf, Nature Phys. 5, 294 (2009). 9 [22] O. G. Berg, R. B. Winter, and P. H. von Hippel, Biochem. 20, 6929 (1981). [23] C. G. Kalodimos, N. Biris, A. M. J. J. Bonvin, M. M. Levandoski, M. Guennuegues, R. Boelens, and R. Kaptein, Science 305, 386 (2004). [24] M. Sheinman, O. Bénichou, Y. Kafri, and R. Voituriez, Rep. Prog. Phys. (2011), to appear.
1811.09501	1	1811	2018-11-23T14:55:36	Comparisons of wave dynamics in Hodgkin-Huxley and Markov-state formalisms for the Sodium (Na) channel in some mathematical models for human cardiac tissue	[ "physics.bio-ph" ]	We compare and contrast spiral- and scroll-wave dynamics in five different mathematical models for cardiac tissue. The first is the TP06 model, due to ten Tusscher and Panfilov, which is based on the Hodgkin-Huxley formalism; the remaining four are Markov-state models, MM1 WT and MM2 WT, for the wild-type (WT) Na channel, and MM1 MUT and MM2 MUT, for the mutant Na channel. Our results are based on extensive direct numerical simulations of waves of electrical activation in these models, in two- and three-dimensional (2D and 3D) homogeneous simulation domains and also in domains with localised heterogeneities, either obstacles with randomly distributed inexcitable regions or mutant cells in a wild-type background. Our study brings out the sensitive dependence of spiral- and scroll-wave dynamics on these five models and the parameters that define them. We also explore the control of spiral-wave turbulence in these models.	physics.bio-ph	physics	Comparisons of wave dynamics in Hodgkin-Huxley and Markov-state formalisms for the Sodium (Na) channel in some mathematical models for human cardiac tissue Mahesh Kumar Mulimani 1, Alok Ranjan Nayak 2, and Rahul Pandit,1,∗ 1Centre for Condensed Matter Theory, Department of Physics, Indian Institute of 2International Institute of Information and Technology, Bhubaneshwar, Orissa, India. Science, Bangalore 560012, India. Abstract We compare and contrast spiral- and scroll-wave dynamics in five different mathematical models for cardiac tissue. The first is the TP06 model, due to ten Tusscher and Panfilov [1], which is based on the Hodgkin-Huxley formalism; the remaining four are Markov-state models, MM1 WT and MM2 WT, for the wild-type (WT) Na channel, and MM1 MUT and MM2 MUT, for the mutant Na channel [2, 3]. Our results are based on extensive direct numerical simulations of waves of electrical activation in these models, in two- and three-dimensional (2D and 3D) homogeneous simulation domains and also in domains with localised heterogeneities, either obstacles with randomly distributed inexcitable regions or mutant cells in a wild-type background. Our study brings out the sensitive dependence of spiral- and scroll-wave dynamics on these five models and the parameters that define them. We also explore the control of spiral-wave turbulence in these models. Keywords: Mathematical Models for Cardiac Tissue, Hodgkin-Huxley Models, Markov-state Models, Wild-Type Markov Models, Mutant Markov Models. 1 Introduction The development of an understanding of the dynamics of waves of electrical activation in cardiac tissue is a problem of central importance in research on life-threatening cardiac arrhythmias, because sudden cardiac death is responsible for roughly half of the deaths because of cardiovascular disease, i.e., 15% of all deaths globally [4]. Approximately 80% of sudden cardiac deaths arise from ventricular arrhythmias [4]. Such arrhythmias are often associated with the formation of spiral or scroll waves of electrical activation; unbroken spirals or scrolls lead to ventricular tachycardia (VT), whereas broken waves, with spiral- or scroll-wave turbulence [5, 6, 7, 8, 9, 10], are responsible for ventricular fibrillation (VF); VT and VF lead to the malfunctioning of the pumping mechanism of the heart, so, in the absence of medical intervention, VF leads to sudden cardiac death. It is very important, therefore, to study VT and VF by using all means possible, namely, in vivo, in vitro, and in silico investigations, ∗Also at Jawaharlal Nehru Centre For Advanced Scientific Research, Jakkur, Bangalore, India; [email protected] 1 which play complementary roles. In silico investigations require mathematical models for cardiac cells (cardiomyocytes or, simply, myocytes) and for cardiac tissue. Studies of the electrical behavior of myocytes require models for the dynamics of the ion channels [5, 7, 8, 11]. The first, successful ion-channel model, due to Hodgkin and Huxley [12], considers the opening and closing of the channel to be governed by the gates, which depend, in turn, on the myocyte transmembrane potential Vm; these Hodgkin-Huxley-Model (HHM) gates are defined by deterministic, first-order, ordinary differential Equations (ODEs); and each gating variable is independent of other gating variables. However, ion channels are proteins that can have many conformational states and, therefore, channels open and close stochastically; hence, discrete-state Markov models (MMs) have been developed to model ion channels; in some cases, these Markov models can be reduced to HHMs [13]. Clearly, these Markov models are more general than HHMs; in particular, the discrete states in an MM depend on each other; and MMs have more parameters than HHMs. Markov-state models are especially useful when there are ion-channel mutations in which the functionality of an ion-channel subunit is disturbed. For example, mutations in the HERG subunit in the rapid, delayed, rectifier potassium (Kr) channel lead to the prolongation of the myocyte action- potential duration (APD); this is referred to as the LQT2 syndrome [14]; and mutations in the α-subunit in the Na channel result in the LQT3 syndrome [15], which can lead to sudden cardiac death. The MM formalism has been used to study the effects of mutations in a variety of ion channels [16] and especially on the LQT syndrome because of mutations in the Kr channel [14] and in the Na channel [3, 15]. In particular, such studies have elucidated the effects of different mutations on the myocyte action potential (AP) [15] and the interaction between drugs and the discrete states in an MM [3, 17, 18]. Challenges in MM studies include the difficulties in estimating the large number of parameters in these models [16] and the higher computational cost relative to HHM investigations. Recently it has been shown that, at the level of a single cardiomyocyte, the dynamics of wild-type (WT) and mutant (MUT) ion-channels can be modeled well by the HHM formalism, if it is obtained from the Markov-state Model (MM) [19]. In particular, HHM action potentials, their morphological properties, the action-potential-duration restitution (APDR), and the conduction-velocity restitution (CVR) are comparable to their MM counterparts [19]. The authors of Ref. [19] have considered both WT and MUT cases for Kr and Na channels in the MM and their HHM counterparts; their results are encouraging, insofar as they suggest that we can use simple, effective HHM models, whose parameters are obtained from comparisons with their complicated MM counterparts, to obtain the properties of action potentials and their dependence on mutations. A careful comparison of these Markov models and the Hodgkin-Huxley model for an ion channel, at the cellular level, brings out the differences in the action potential and its morphology. To compare the characteristic properties of excitation waves in these models, it behooves us to carry out studies of spiral- and scroll-wave dynamics in homogeneous and heterogeneous tissue in two- and three-dimensional (2D and 3D, respectively) simulation domains; we embark on such a study here. In particular, we focus on the Na channels in these MM and HHM models, as the Na channel is important in controling the upstroke-velocity, at the cellular level, and CV, at the tissue level. In our Hodgkin-Huxley model (HHM) Na-channel formalism, we use the human-ventricular-tissue TP06 model, due to ten Tusscher and Panfilov [1]. We compare spiral- and scroll-wave states in this model with their counterparts in two different Markov-state models, which we call MM1 [2] and MM2 [3]. In these models, we study both wild-type (WT) and mutant (MUT) Na channels, by replacing the Na-channel formalism in the TP06 model by their MM1 and MM2 versions; and we use the TP06 formulation for all other ion channels (see the section on Methods). Therefore, we examine three models for the WT Na channel (these are variants of the TP06 model): the original HHM (TP06) and two MMs (MM1 WT and MM2 WT); and we use two models for the mutant channels (again variants of the TP06 model), specifically, the MM1 MUT and MM2 MUT models. Note that the TP06 HHM is not obtained from the Markov models as in Ref. [19]. We first compare activation and inactivation properties of the Na channels in all the five models mentioned above. We then contrast the effects of these changes on the action potentials and their morphologies in these models, at the single-cell level. We show that, for the wild-type (WT) Na-channel, 2 the probability of opening of this channel is different for the TP06, MM1 WT, and MM2 WT models. The peak value of this probability and the time duration of this opening are also dissimilar in MM1 WT and MM2 WT models. These differences alter the action potential (AP) and its morphology. We show that, for the mutant (MUT) Na channels, the failure of inactivation leads to early afterdepolarizations (EADs), in the APs in MM1 MUT and MM2 MUT models [20]. The differences in the WT Na peak amplitude lead to disparate upstroke velocities in these models, which manifest themselves in dissimilar CVRs. Furthermore, the conduction velocities (CVs) in MM1 WT and MM2 WT models turn out to be outside (lower than) the accepted range for CV in the human myocardium; we show that we can obtain CVs in this range if we increase the diffusion constant D in both MM1 WT and MM2 WT models. The differences in our single-cell and cable-level results motivate our study of wave dynamics in mathematical models for cardiac tissue, which use these different models. We carry out a variety of simulations in 2D homogeneous domains to show that spiral-wave dynamics, in TP06, MM1 WT, MM2 WT, MM1 MUT, and MM2 MUT models, depends sensitively on these models. For example, we demonstrate that, in the MM1 WT (MM2 WT) model, the spiral wave is stable (unstable, meandering spiral). The formation of certain EADs can lead to backward propagation of the wave, and rapid spiral breakup, in the MM2 MUT model; by contrast, in the MM1 MUT model, EADs are somewhat different, so we do not find such backward propagation, the mother rotor is unaffected, and there is only far-field breakup of the spiral. Furthermore, the spatiotemporal evolution of a spiral wave in the MM2 WT model depends sensitively on the time τS2 between the application of the S1 and S2 impulses, which we use to initiate spiral waves. In the case of mutant models, because of the different kinds of EADs that we find in MM1 MUT and MM2 MUT, these models display qualitatively different electrical-wave dynamics. Furthermore, in the spirit of the studies of Refs. [6, 7, 21, 22], we investigate the effects of two types of inhomogenieties on spiral-wave dynamics in these models: (a) Two-dimensional (2D), circular or three-dimensional (3D), cylindrical obstacles, with a random distribution of inexcitable regions, to model fibrotic patches in Markov-state WT models; Pf , the percentage of inexcitable obstacles, and the radius of the obstacle are important control parameters. (b) A circular patch of mutant cells in an otherwise homogeneous, 2D WT domain; we find that a spiral wave is formed in the MM2 MUT model, but not in the MM1 MUT model, if we pace the tissue at a high frequency. The elimination of spiral- and scroll-wave turbulence is of central importance in developing low- amplitude defibrillation schemes for the elimination of VT and VF. In the Supplementary Material, we describe one such defibrillation scheme (control of spiral waves) for the models of we study. We carry out a few illustrative studies of scroll waves in 3D TP06, MM1 WT, and MM2 WT models. We show, in a homogeneous domain, that scroll waves are stable in TP06 and MM1 WT models, but not in the MM2 model. We also investigate when scroll waves are anchored or broken up by cylindrical obstacles, of the type described above. Finally, we perform a parameter-sensitivity analysis for TP06 and MM1 WT and MM2 WT models, in which we consider three important dependent variables, namely, the APD, Vmax, and Vrest, at the cellular level and CV and APD at the cable level (see Supplementary Material). The remaining part of this paper is organized as follows. Section 2 is devoted to Methods and Simulations. In Section 3 we report our Results for single-cell studies and tissue-level simulations in 2D square and 3D slab domains for WT models; we also present, for MUT models, single-cell and 2D-simulation results. In Section 4 , Discussion and Conclusions, we end with concluding remarks. 3 2 Methods and Simulations 2.1 Model The electrical behavior of a single cardiac myocyte is governed by the following ordinary differential Equation (ODE) for the transmembrane potential Vm: dVm dt Iion = (cid:88) = − Iion Cm Ii; ; (1) (2) i here, Iion is the sum of all the ionic currents, Ii is the current because of the ith ion-channel, and Cm is the normalized, transmembrane capacitance. In the parent TP06 model, Iion is the sum of the following 12 ionic currents (Table 1): Iion = IN a + ICaL + Ito + IKs + IKr + IK1 + IN aCa + IN aK + IpCa + IpK + IbN a + IbCa. (3) The spatiotemporal evolution of Vm, at the tissue level, is governed by the following reaction-diffusion partial differential Equation (PDE): ∂Vm ∂t = D ∇2V − Iion Cm , (4) where D is the diffusion constant; for simplicity we consider the case in which D is a scalar. Hodgkin-Huxley Model The TP06 model uses the Hodgkin-Huxley formalism for the WT Na channel. The macroscopic current through this channel is governed by the three gating variables m, h, and j [1]; the first of these is an activation gate and the latter two are inactivation gates; the gating dynamics and the Na current are given by a∞ − an dan dt IN a = GN a m3hj(Vm − EN a); τn = ; here an can be m, h, or j; (5) (6) GN a is the maximal sodium-channel conductance, a∞ is the steady-state value of an, τn the time constant of this gating variable, and EN a is the sodium-channel Nernst potential. Markov-state models We consider four Markov-state models (MMs): two of these are for the wild-type (WT) and the other two for the mutant (MUT) Na channels. We use the Markov-state formalisms of Ref. [2] for the first WT and MUT Na channels; we refer to these as MM1 WT and MM1 MUT, respectively. We use the Markov models of Ref. [3] for the second WT and MUT Na channels, which we label MM2 WT and MM2 MUT, respectively. We then replace the Na current in the TP06 model by these two different WT and two different MUT models. Finally, we have three different WT models, i.e., the original TP06, MM1 WT, and MM2 WT; and we have two different MUT models, namely, MM1 MUT and MM2 MUT. All the other currents in the original TP06 model are unaltered in our studies below. Schematic diagrams of MM1 WT and MM2 WT models are shown in the top panel of Figure 1. The MM1 WT model has nine states: the open state (O), the three closed states (C1, C2, C3), and the five inactivation states (IF, IM1, IM2, IC2, IC3). The MM2 WT model has eight states: the open state (O), the three closed states (C1, C2, C3), and the four inactivation states (IF, IS, IC2, IC3). 4 The orange, double-headed arrows indicate transitions between such Markov states; transition rates for the rightward (leftward) transition are given above (below) these arrows, e.g., a111 (b111) for the IC3 → IC2 (IC2 → IC3) transition in MM1 WT. Similar schematic diagrams for the MM1 MUT and MM2 MUT models are given in the bottom panel of Figure 1. The MM1 MUT model has the same number of states as the MM1 WT model, but the transition rates between the Markov states are different. The MM2 MUT model has 12 states: 8 of these are as in the MM2 WT model; in addition, there are 4 bursting states, namely, BO, BC1, BC2, BC3. The dynamics of transitions between the states of these Markov models and the Na-channel current IN a are given, respectively, by Equations 7 and 8 below: (cid:88) αlPl − (cid:88) l,l→k l,k→l dPk dt = βlPk, (7) where l, k label the states O, C1, C2, C3, IC2, IC3, IM1, IM2, IF, IS, BO, BC1, BC2, BC3 (Fig- ure 1); αl and βl are generic labels for forward and backward transition rates, respectively; IN a = GN a(PO)(Vm − EN a), (8) where PO is the probability of the opening of the Na channel. We use the values of GN a that are employed either in the original TP06 model [1] or in Ref. [2]; specifically, we use GN a = (cid:40) 14.838 nS/pF, TP06, MM2 WT, and MM2 MUT; 16 nS/pF, MM1 WT and MM1 MUT. The major differences between the two MM (WT and MUT) models are as follows; • The number of states and the connections between them (see Figure 1 • The MM2 WT model has a Na current with a late component; this is absent in the MM1 WT model. We show below that these differences can have significant effects on single-cell and spiral-wave properties in these models. 2.2 Numerical Simulations For our single-cell simulations, we use the Rush-Larsen method to solve Equation 5 for the HHM; for the MM Equation 7 we use the implicit trapezoidal method of Ref. [3] and the forward-Euler method for Equation 1. In our 2D tissue simulations for Equation 4, we use a square domain with N × N grid points, a fixed space step of size ∆x = 0.025 cm, and a time step ∆t = 0.02 ms; in 3D we use a slab domain (see below). The accuracy of the numerical scheme is tested and is reported in Supplementary Material.For the homogeneous tissue we consider, D is a scalar; we use D = 0.00154 cm2/ms, as in Ref. [1]; this yields a maximum plane-wave conduction velocity 70cm/s, which is in the biophysically reasonable range for human ventricular tissue [23]. For the Laplacian in Equation 4, we use five-point and seven-point stencils, respectively, in 2D and 3D. We impose no-flux boundary conditions. When we study the effects of heterogeneities, we introduce, in a localized region of our simulation domain, a patch of inexcitable cells, which are decoupled from adjoining cells (effectively, D = 0 in this patch). To obtain the inactivation and activation properties of MM1 WT, MM2 WT, MM1 MUT, and MM2 MUT Na channels, we use the voltage-clamp-simulation protocol of Ref. [2]. In the activation protocol, we clamp the cell with a voltage Vc, which ranges from the hyper-polarized regime (below the resting membrane potential of an action potential) to the depolarized regime ((cid:39) 50 mV); we do this in steps of 5 mV; the clamping is maintained for a clamping time tc = 1s. We then record the peak 5 current IpeakN a(Vc) and divide it by the driving force (V − EN a) to obtain the conductance G(Vc), which we normalize to obtain the activation variable A as follows: Vc = −100 mV to 50 mV , t = 1s; IpeakN a(Vc) = max (PO) (Vc − EN a); G(Vc) = A ≡ IpeakN a (V − EN a) G(Vc) ; G(Vc = 50 mV ) . (9) Similarly, for the inactivation protocol, we use a holding potential (Vh), ranging between hyper- polarized and the depolarized values, and apply it for (cid:39) 250 ms; we then apply a test potential Vt = 0 mV , record the peak Na current, and then define the inactivation variable I as follows: (cid:40) Vh = −130 mV to − 10 mV , t < 250 ms; Vt = 0 mV, t = 250 ms; Vc = IpeakN a(Vc) = max (PO) (Vc − Ena), t > 250 ms; I ≡ Ipeak N a(Vc) Ipeak N a(Vc = −130 mV ) . (10) We record the single-cell AP and its morphology, after we have paced the cell with n pulses, each with a constant pacing cycle length (PCL); we use n = 500 pulses. We obtain the static action potential duration restitution (APDR) (s1-s2) as follows: (a) we apply several pulses ((cid:39) 17) with a fixed PCL (s1); then, once the system reaches the steady state, we change the diastolic interval (DI) (s2) by recording the time at which the cell is 90% repolarized; this is the action potential duration (APD90), or simply the APD. We next obtain the dynamic conduction velocity restitution (CVR) (s1-s1) as follows: We consider a cable of cells (of dimension 832 × 10) and pace it by applying a current stimulus (s1) at one of its ends; we obtain CV from the time that an iso-potential line takes to move between two cells, which are separated by a fixed distance. The CVR is the plot of CV versus DI. In our 2D studies we use two representative square domains, namely, one with 1024 × 1024 grid points, for our spiral-wave studies, and another with 512 × 512 grid points, when we pace of the simulation domain along an edge. In our studies of scroll-wave dynamics we use a 3D slab domain with 1024 × 1024 × 40 grid points ( 25.6cm × 25.6cm × 1cm). We initiate spiral and scroll waves in such domains by using the following S1-S2 cross-field protocol, with stimuli amplitudes of 150 pA/pF and durations of 3 ms: We allow a plane wave (S1) to propagate in the domain along a particular direction; as it propagates, we start another plane wave (S2), in a direction perpendicular to the S1 wave; this results in a conduction block and, eventually, the formation of a spiral wave (2D) or scroll wave (3D). We vary the time interval τS2 between the S1 and S2 impulses to study the sensitive dependence of the spiral-wave dynamics on τS2. We have carried out the following two sets of simulations of electrical-wave dynamics with certain localised inhomogeneities in an otherwise homogeneous simulation domain. (a) In the first set of simulations we introduce circular (2D) and cylindrical (3D) regions with a random distribution of inexcitable obstacles to mimic localised fibrotic patches in Markov-state WT models; an important control parameter here is Pf , the percentage of inexcitable obstacles in the circular or cylindrical regions. (b) In the second set of simulations, we examine electrical-wave dynamics in the presence of a circular patch of mutant cells in an otherwise homogeneous, 2D WT domain. 6 3 Results 3.1 Single-cell results Activation and Inactivation We begin with a comparison of the activation A and the inactivation I in the models MM1 (WT and MUT) and MM2 (WT and MUT) with their TP06 counterparts. Figures 2A and 2B show, for different values of Vc, plots of IN a versus time t. These plots illustrate, respectively, the activation and inactivation protocols (see 2.1), whence we obtain Figures 2C and 2D, which depict, respectively, the dependences of A and I on Vm for the MM1 WT, MM2 WT, MM1 MUT, MM2 MUT, and TP06 models; for the TP06 model A = m3∞ and I = j∞ × h∞. If we contrast the plots of A in Figure 2C, we see that the curves for both MM1 WT and MM1 MUT models lie to the right of, and are less steep than, their TP06, MM2 WT, and MM2 MUT counterparts; hence, activation occurs most slowly (with respect to Vm) in MM1 WT and MM1 MUT models. In particular, the Na channels in the TP06, MM2 WT, and MM2 MUT models reach near-complete activation at V (cid:39) 2mV , V (cid:39) −14mV , V (cid:39) −8mV , respectively, whereas the MM1 WT and MM1 MUT models do so only for V (cid:39) 60mV and V (cid:39) 60mV , respectively. The potential at which A = 0.5 is VA=0.5,TP06 (cid:39) −36mV , VA=0.5, MM2 WT (cid:39) −32 mV , VA=0.5, MM2 MUT (cid:39) −24 mV , VA=0.5,MM1 WT (cid:39) −20 mV , and VA=0.5, MM1 MUT (cid:39) −34 mV . From Figure 2D, we find that the plots of I, for the MM1 WT, MM1 MUT, MM2 WT, and MM2 MUT models, are shifted to the right (the depolarized-potential side) compared to I for the TP06 model. Note also that the MM1 MUT model shows inactivation earlier than the MM1 WT model. In the MM2 MUT case, the inactivation occurs earlier than in the MM2 WT model in the range −72mV ≤ Vm ≤ −32mV . The potential at which I = 0.5 is VI=0.5,TP06 (cid:39) −84 mV , VI=0.5,MM2 WT (cid:39) −69/mV , VI=0.5,MM2 MUT (cid:39) −70/mV , VI=0.5,MM1 WT (cid:39) −61 mV and VI=0.5,MM1 MUT (cid:39) −66 mV . Our results for the two Markov-state models are in agreement with those shown in Figure 2 of Refs. [2, 19]. Probabilities of the Markov states Let us examine now the temporal evolution of the probabilities of different Markov states during the course of an action potential. As we have mentioned in Section 2.1, there are three main classes of Markov states for the Na channel, namely, the open states (O, BO), the inactivation states (IF, IS, IM1, IM2, IC2, IC3), and the closed states (C1, C2, C3, BC1, BC2, BC3). The prob- abilities of these three classes of states are as follows: PO is the open state probability; PI is the sum of probabilities of all the inactivation states; and PC is the sum of probabilities of all the closed states. In the case of MM1 WT and MM2 WT models, they are as follows:  P robability = PO (MM1 WT and MM2 WT); PI = PIF + PIM 1 + PIM 2 + PIC2 + PIC3 (MM1 WT); PI = PIF + PIS + PIC2 + PIC3 (MM2 WT); PC = PC1 + PC2 + PC3 (MM1 WT and MM2 WT). The probabilities of these three main classes of states, in MM1 MUT and MM2 MUT models, are as follows: 7  P robability = PO (MM1 MUT); PI = PIF + PIM 1 + PIM 2 + PIC2 + PIC3 (MM1 MUT); PC = PC1 + PC2 + PC3 (MM1 MUT); PO ≡ PO + PBO (MM2 MUT); PI = PIF + PIS + PIC2 + PIC3 (MM2 MUT); PC = PBC1 + PBC2 + PBC3 + PC1 + PC2 + PC3 (MM2 MUT). Figure 3 shows plots of PO, PI , and PC versus time t for the Na channel in the course of an action potential for MM1 (MM2) models in the top (bottom) panel; the blue and red curves are for WT and MUT models, respectively. We obtain these plots by pacing a single cell with P CL = 3000 ms. By comparing the blue curves in Figures 3 (A), (B) , (D) and (E)) we find that the duration for which the Na channel is in the inactivation or closed states, i.e., the time interval during which PI = 1 and PC = 0 (inactivation state) or PI = 0 and PC = 1 (closed state), is approximately the same in MM1 WT and MM2 WT models. In contrast, the duration for which PO is significantly greater than 0 differs in MM1 WT and MM2 WT models (compare the blue curves in Figures 3 (C) and (F)); this duration, measured by the full-width-at-half-maximum (FWHM) of PO, is (cid:39) 0.37 ms and PO,max (cid:39) 0.17, for MM1 WT, and (cid:39) 0.13 ms and PO,max (cid:39) 0.28, for MM2 WT. The blue curves in the insets of Figures 3 (C) and (F) show that, in the MM1 WT model, there is no late-Na current because PO = 0 in the repolarisation phase of the action potential (AP); in contrast, the MM2 WT model yields PO (cid:39) 0.0003 at t (cid:39) 350 ms, which demonstrates that the Na channel opens in the repolarisation regime of the AP. In the MUT cases, the time duration for which the Na channel is in the inactivation or closed states is prolonged compared to that in the WT cases (see Figures 3 (A), (B), (D), and (E)); from the insets of these figures we see that PI decreases slightly below 1 [there are corresponding increases in PO and PC (see Figures 3 (C) and (F))], for 360ms (cid:46) t (cid:46) 1110ms in MM1 MUT and 360ms (cid:46) t (cid:46) 1700ms and MM2 MUT. The duration for which the Na channel is in the inactivation state PI = 1 and PC = 0, for MM2 MUT, is much longer than that in MM1 MUT. We find the following FWHMs: for PI FWHM (cid:39) 1729.2ms (MM2 MUT) and (cid:39) 1200.25ms (MM1 MUT); for PC FWHM (cid:39) 1721.2ms (MM2 MUT) and (cid:39) 1176.2ms (MM1 MUT). PO,max is markedly different in both MUT models: (cid:39) 0.4796 (MM2 MUT) and (cid:39) 0.0861 (MM1 MUT); and the FWHM of PO is (cid:39) 0.14ms (MM2 MUT) and (cid:39) 1.03ms (MM1 MUT). Action Potential, APDR, and CVR We pace a single cell with the following three different values of PCL: high frequency (PCL=300 ms), intermediate frequency (PCL=650 ms), and low frequency (PCL=1000 ms). We present the steady-state AP and the Na current IN a at the top panel of Figure 4(A),(B), and (C) (for PCL = 1000 ms), for the three WT models; and we compare the morphological properties of the APs of these models in Table 2. PCL = 1000ms Given the differences in the activation profiles in Figures 2(C),(D) and the plots of PO for the MM models in Figures 3(C).(F) , we observe that (a) the times at which the Na channels open are different in all the three WT models; and (b) the amplitude of IN a is comparable in MM2 WT (−312.74 pA/pF ) and TP06 (−300.43 pA/pF ) models, but it is significantly lower in the MM1 WT model (−144 pA/pF ) as we show in Figure 4(B) . These differences in the amplitude of IN a affect the maximum voltage and the upstroke-velocity of the AP (Table 2). The upstroke velocities for TP06, MM1 WT, and MM2 WT models are markedly different (Table 2). Also, there is the late component of the Na current IN a,L (Figure 4(C)) in the case of MM2 WT; this component is clearly absent in TP06 and MM1 WT models. PO becomes significant at (cid:39) 350ms in the MM2 WT model (Figure 3(F)), so the APD for this model is larger than its counterparts in the TP06 and MM1 WT models (see Figure 4(C) and Table 2). 8 PCL = 300ms As we decrease PCL, say to 300ms, we find that both IN amax (the maximal value of −IN a) and the upstroke velocity in the MM2 WT model increase relative to their counterparts in the TP06 model as we show in Table 2 (contrast this with our results for PCL = 1000ms). These increases occur principally because PO,max is higher in the MM2 WT model than in the TP06 model. PCL = 3000ms, MUT Na channel In Figures 4(D),(E) and (F), we show, respectively, plots of Vm, IN af , and IN aL versus time t; we use dashed curves for the MM1 MUT (blue) and MM2 MUT (red) models and the illustrative value PCL = 3000ms. Clearly, the MM1 MUT and MM2 MUT APs in Figure 4(D) show early afterdepolarizations (EADs) [20, 24], insofar as their APs are prolonged considerably relative to the the APs for MM1 WT and MM2 WT models, because of the failure of inactivation near the repolarisation region (insets in Figure 3). APDR and CVR (WT) For the TP06, MM1 WT, and MM2 WT models, we present plots of the single-cell static APDR (Figure 5(A)) and the dynamic CVR (Figure 5(B)), for a one-dimensional cable of cells. The APDR profiles for TP06 and MM1 WT lie close to each other, but the MM2 WT curve lies above these, because of the late current component IN a,L (see above). The slopes of the APDR and CVR profiles are given, respectively, in Figures 5(C) and (D). Note that, in all these three models, the maximal slope of the APDR profile > 1 (it is highest in the MM1 WT model). The Na channel determines the upstroke velocity at the cellular level; therefore, this channel plays an important role in determining CV, in cardiac tissue, and also CVR plots (Figure 5(B)). From these plots we find that, for TP06, MM1 WT, and MM2 WT models, CV is nearly independent of DI, for large DI; the ranges spanned by CV are 60.51 − 70.55 cm/s (TP06), 35.5 − 40.4 cm/s (MM1 WT), and 51.43 − 54.89 cm/s (MM2 WT), for DI in the interval 90 − 900 ms; and the saturation values of CV are (cid:39) 70.55 cm/s (TP06), (cid:39) 40.41 cm/s (MM1 WT), and 54.89 cm/s (MM2 WT). In the human myocardium, CV is (cid:39) 60 − 75 cm/s [23, 1]. To obtain CV in this physiological range, we must increase the diffusion constant D in both MM1 WT and MM2 WT models; we find that, if we multiply D by 2.915 (MM1 WT) and 1.299 (MM2 WT), then the saturated value of CV is (cid:39) 64.65 cm/s (MM1 WT) (cid:39) 71.75 cm/s (MM2 WT); these multiplicative scale factors can be obtained by noting that CV ∝ √ D [5, 10] and by using the saturated CV value in the TP06 model. With these changes in D, CV can be brought to a physiologically realistic value; but its variation is small: 60.69 − 64.65 cm/s (MM1 WT) and 62.34 − 71.75 cm/s (MM2 WT) over the DI range of 90 − 900ms. 3.2 2D results We have explored differences between the TP06, MM1, and MM2 models at the single-cell and the cable levels. We now compare spiral- and scroll-wave dynamics in these models by carrying out detailed numerical simulations in 2D (Section 3.2) and 3D (Section 3.3) domains. Wild-type Na channel We contrast, in the top panel of Figure 6, spiral waves in these three models, with D = 0.00154 cm2/ms. We find that spiral waves in TP06 and MM1 WT are stable and they rotate with frequencies ω (cid:39) 4.75 Hz and ω (cid:39) 4.25 Hz, respectively; in particular, the low value of CV (40.41 cm/s), in the MM1 WT model with D = 0.00154 cm2/ms, does not alter the spiral-wave dynamics qualitatively. By contrast, in 9 the MM2 WT model, the spiral wave is unstable and exhibits transient breakup; it is not possible to isolate a single cause for this break up, but the late Na current IN a,L(Figure 4(C)) plays an important role in this instability; we have checked that, by increasing β12, we can reduce the magnitude of this late current and thus suppress spiral-wave turbulence (the spiral meanders but does not break up into multiple spirals as we show in the Movie (M0) in the Supplementary Material 4). We have carried out another set of studies in 2D simulation domains, with the values of D scaled up to D ∗ 2.915 (MM1 WT) and D ∗ 1.299 (MM2 WT), to bring the values of CV close to the range of values in human ventricular tissue [1]. These scaled values of D do not change our qualitative results about spiral-wave stability (TP06 and MM1 WT) or their breakup (MM2 WT). However, the spiral-arm width increases when we scale up the value of D (Figure (S1) and Movie (M1) in the Suppelmental Material 4); furthermore, because CV increases when we scale up D, the spiral-wave rotation frequency ω also increases with D. Henceforth, in our 2D and 3D simulations we use the same fixed value D = 0.00154 cm2/ms for all three models (TP06, MM1 WT, and MM2 WT). We employ the S1-S2 protocol to initiate spiral waves in all these models (Section 2). The pseudocolor plots of Vm in Figure 7 show that the spiral-wave activity in the TP06 and MM1 WT models is independent of the time τS2, at which the S2 pulse is applied after the S1 pulse (we use 560ms ≤ τS2 ≤ 620ms). By contrast, in the MM2 WT model, we observe spiral-wave breakup for τS2 = 560ms and 580ms until the end of our simulation, i.e., 10s; but spiral-wave activity vanishes for τS2 = 600ms at (cid:39) 5.5s and for τS2 = 620ms at (cid:39) 6.8s (Figure 7 and Movie (M2) in the Supplementary Material 4). Spiral-wave dynamics in the MM2 WT model depends on the time τS2 at which we initiate the S2 pulse. It behooves us, therefore, to examine whether obstacles (or conduction inhomogeneities) affect spiral-wave activity in the MM1 WT and MM2 WT models, for it has been shown, for HH-type models for cardiac tissue, that spiral-wave dynamics depends sensitively on the position, size, and shape of such obstacles [6, 7, 21, 22]. Our obstacles consist of inexcitable points that are distributed randomly within a circular region of radius R; Pf is the percentage of the area of the circle that has inexcitable obstacles. Given our experience with studies of spiral-wave dynamics with such obstacles in HH-type models, we expect that, as Pf increases, such an obstacle should anchor a spiral wave [25, 26]. Therefore, we investigate the dependence of spiral-wave dynamics on Pf and R in the MM1 WT and MM2 WT models and compare this with its counterpart in the TP06 model, for different values of τS2. Illustrative plots from our simulations are shown in Figure 8. We find that, for the TP06 and MM1 WT models, the anchoring of the spiral wave depends on R and on Pf , but not on τS2. The time period T of the anchored spiral increases with R and Pf as we show in Figure 9(A); but T decreases for lower percentages (e.g., Pf = 30%) in TP06 and MM1 WT models at large values of R (Figure 9(A)). The interaction of the tip of the spiral wave with the obstacle is complicated. In particular, this depends on how much of the region, inside the circular patch, is excitable. For low values of Pf , this excitable region forms a tortuous but spanning cluster (in the sense of percolation theory [27]), so the tip of the spiral propagates inside the obstacle, the wave of activation is slightly deformed there, but then it re-emerges into the homogeneous part of the simulation domain. If Pf is large, the excitable region can still be tortuous, but it does not form a spanning cluster, so the tip of the spiral rotates around the obstacle, and is anchored to it, but does not propagates inside it. To quantify the effect of our obstacle on the spiral wave we calculate δT ≡ (T − T0), where T is the time period (or inverse of the rotation frequency ω), at a given set of values of Pf and R, T0 is the time period (or inverse of the corresponding frequency ω0) with Pf = 100% for the same value of R. Clearly T must depend on Pf and R. The plots in Figures 9(B)and (D) show, for TP06 and MM1 WT models, the dependence of δT on R for different values of Pf . Given these plots, we identify three regions, namely, (i) δT < 0, i.e., ω > ω0, (ii) δT > 0, i.e., ω < ω0 and (iii) δT = 0 i.e., ω = ω0. If δT > 0, then the frequency ω ∼ T −1, for a given pair (R, Pf ), is less than ω0 ∼ T −1 (for R, Pf = 100%); this may occur because the spiral core penetrates the obstacle because of a spanning cluster of excitable regions inside the obstacle. In Figures 9(C) and (E) we show different colored regions in the (R, Pf ) plane for TP06 and MM1 WT models, respectively: light blue indicates an increase in ω relative to ω0 0 10 (caused by penetration of the spiral core); light green is for a decrease in ω relative to ω0 (accompanied by penetration of the spiral core); yellow indicates no penetration of the spiral core into the obstacle; dark blue depicts regions in which there is no change in ω relative to ω0 even though the spiral core penetrates into the obstacle. For the MM2 WT model, the minimum size Rmin for spiral anchoring is large, compared to that in TP06 and MM1 WT model, and is Rmin = 1.875 cm (Movie (M5) in the Supplementary Material 4). The threshold percentage in the MM2 WT case is Pf,min (cid:39) 50%. Once we reach the values Rmin and Pf,min required for anchoring, the spiral activity is independent of τS2, as we show in the fourth row of Figure 8. The dependence of the spiral rotation time period T on R, for different values of Pf , is shown in Figure 9(A). Stability diagrams for the spiral-wave activity, in the presence of localized, inexcitable obstacles distributed within a circular region of radius R, are shown in the (R, τS2) plane, for different values of Pf in the MM2 WT model, in Figure 10; brown, green, and blue denote regions with an anchored spiral, spiral breakup, and no activity, respectively. Mutant Na channel The mutant Na channel fails to inactivate completely in the MM1 MUT and MM2 MUT models; this leads to prolonged EADs, as we have shown in Sec 3.1 and Figure 4. We find that two of the types of EADs that have been discussed in Ref. [20] occur in both these MUT models: there is a single EAD (of type 2 in the nomenclature of Ref. [20]), in the MM1 MUT model, and an oscillatory EAD (roughly of type 3 in the nomenclature of Ref. [20]), in the MM2 MUT model. These two types of EADs affect the wave dynamics differently, as we demonstrate explicitly by simulating plane-wave propagation in our 2D domain, but with all mutant myocytes. We observe backward propagation of the plane wave in the MM2 MUT model because of the oscillatory EADs; by contrast, there is no such backward propagation in the MM1 MUT model. If we initiate a spiral wave in both these models, then, (a) in the MM1 MUT model, we get almost-instantaneous far-field breakup away from the core, but the mother rotor is unaffected, and (b) in the MM2 MUT model, we obtain almost-instantaneous spiral break-up (bottom panels of Figures 6(D) and (E)). The complete spatiotemporal evolution of such spiral-wave dynamics, for both these cases, is shown in the Movie (M3) in the Supplementary Material 4. Although there are several studies of the effects of different types of inhomogeneities on spiral-wave dynamics in mathematical models for cardiac tissue (see, e.g., Refs. [6, 7, 21, 22] and references therein), to the best of our knowledge there has been no study, based on Markov-state models, of an inhomogeneity comprising mutant myocytes in a background of wild-type myocytes. Therefore, we present a representative study of spiral-wave dynamics in the presence of a clump of only mutant cells, of radius R = 1.125 cm, embedded in a background of wild-type cells. We then explore the possibility of spiral-wave formation via high-frequency stimulation by pacing the simulation domain from the left boundary (pacing frequency 3.7Hz). We find that, in the MM1 MUT model, no spiral wave forms (Figure 11(A)); by contrast, in the MM2 MUT model a spiral wave forms (Figure 11(B)). [The spatiotemporal evolution of these waves is shown in the Movie(M10) in the Supplementary Material 4.] This qualitative difference arises because of the different types of EADs (discussed above) in MM1 MUT and MM2 MUT models. 3.3 3D results We end with an illustrative study of scroll waves in 3D TP06, MM1 WT, and MM2 WT models. These waves are shown via color isosurface plots of the transmembrane potential Vm in Figures 12 A, B, and C, respectively, for both a homogeneous domain (top panel) and with localized obstacles [Pf = 10% (middle panel) and Pf = 50% (bottom panel)]. In a homogeneous domain, scroll waves are stable in TP06 and MM1 WT models, but not in the MM2 model. An inexcitable obstacle, with Pf = 10%, has no significant impact on scroll waves in TP06 and MM1 WT models. An increase in Pf , say to Pf = 50%, leads to an anchoring of the scroll waves at the obstacle (as in the study of inexcitable obstacles in Ref. [9]). For the MM2 model, with Pf = 10%, scroll-wave break-up is enhanced; but for 11 Pf = 50%, the scroll wave gets anchored to the obstacle. The spatiotemporal evolution of these scroll waves is shown in the Movie(M10) in the Supplementary Material 4. 4 Discussion and Conclusions Earlier studies of Markov models for cardiac myocytes have focused on the effects of mutations in subunits of N a+, and K + channels in the context of the Brugada and LQT syndromes [28, 15, 14, 29]; these studies have elucidated the effects of changes in the kinetic properties of these ion-channel, and their consequences, such as the prolongation of the APD, which leads, in turn, to EADs. Also, Markov models have been used to asses the importance of a particular functionality of an ion channel, e.g., the role of IKs on AP repolarisation [30]. Furthermore, Markov models have been used to investigate the activation and inactivation properties of N a+, K + and Ca+2 ion channels as, e.g., in [31, 32, 33, 34]. In addition, some studies have focused on theraupetics and drug-channel interactions [18, 3, 17], from cellular to the anatomically realistic tissue levels. We have investigated, from cellular to tissue levels, the differences in kinetic properties of Na channels in TP06, MM1 (WT and MUT), and MM2 (WT and MUT) models. We have shown that Na channels in TP06 and MM2 (WT and MUT) models are activated faster, with respect to Vm, than their counterparts in the MM1 (WT and MUT) models; also the inactivation of these channels is faster in the TP06 model than in MM1 (WT and MUT) and MM2 (WT and MUT) models. These differences leads to different times of openings of the Na channel and the amplitudes of IN a,f are completeley determined by the amplitude of PO in the MM models. These changes in the amplitudes of IN a,f , IN a,L, and the activation-inactivation dynamics lead to disparate CVR and maximal CVs in cable simulations. To the best of our knowledge, our study is the first to compare spiral-wave dynamics in different Markov models for the Na (WT and MUT) ion channels in realistic mathematical models for cardiac tissue. We have carried out in silico studies, in both homogeneous simulation domains and domains with inhomogeneities, to compare and contrast spiral- and scroll-wave dynamics in five different models for cardiac tissue (Hodgkin-Huxley type, TP06 model [1], and Markov-state models such as MM1 WT and MM2 WT, for the WT Na channel, and MM1 MUT and MM2 MUT, for the mutant Na channel [2, 3]). Our study explores the sensitive dependence of spiral- and scroll-wave dynamics on these five models and the parameters that define them. We also examine the control of spiral-wave turbulence in these models. To the best of our knowledge, such a comparative study of wave dynamics in HHM and Markov-state models has not been carried out hitherto. In our opinion, such a comparison is even more valuable than the comparison of single-cell properties of models for cardiac myocytes. We hope our study will lead to more comparisons of wave dynamics in different mathematical models for cardiac tissue and in in vitro experiments. Furthermore, we have carried out a detailed parameter-sensitivity study, principally for the WT models by using multivariable linear regression (see the Supplementary material). We mention some of the limitations of our study. We have studied the differences in Na ion-channel modeling, which is important in the context of the LQT syndrome; but we have not carried out such a study for Kr-channel modeling, as mutations in the Kr channel also leads to the LQT syndrome. We have considered a few, illustrative Markov-state models for Na channels, given our computational resources; many more such Markov-state models have been developed for the Na channel [16, 35]; a comprehensive comparison of all these models lies beyond the scope of this paper. Also, we have used a single base model, i.e., TP06 model upon which we build the Markov-state models (MM1 WT, MM1 MUT, MM2 WT and MM2 MUT); other base models, e.g., the O'Hara Rudy model [36], can be used; a comprehensive comparison of all these models lies beyond the scope of this paper. We do not use an anatomically realistic simulation domain [37], with information about the orientation of muscle fibers [9, 38]; and we use a monodomain description for cardiac tissue. These considerations lie beyond the scope of this paper. We note, though, that the study of Ref. [39, 40] has compared results from monodomain and bidomain models and has shown that the differences between them are small. 12 Acknowledgments We thank the Department of Science and Technology (DST), India, for the funding and the Supercom- puter Education and Research Centre (SERC, IISc) for computational resources. Author Contributions Conceived and designed the study: MKM ARN RP. Performed the simulations: MKM. Analyzed the data: MKM ARN RP. Contributed analysis: MKM ARN RP. Wrote the paper: MKM ARN RP. Supplementary Data Figure.S1: Spiral waves in the MM1 WT and MM2 WT models, with increased values of D (see text). Apart from an increase in the conduction-velocity CV and the arm-length, the qualitative features of spiral-wave dynamics here remain the same as those with the unaltered value of D. See the Movie(M1). Figure.S2: Pseudocolor plots of the transmembrane potential Vm illustrating the elimination of spiral waves by electrical stimulation on a square mesh. The domain is divided into square cells of dimension 128× 128 grid points (for a domain size 1024× 1024) or 64× 64 (for a domain size 512× 512); and then a stimulus, of amplitude 50pA/pF , is applied for 100ms along the edges of the square cells as in Ref. [41]; this leads to the elimination of spiral-wave activity in all the five models. Movie(M0): This movie shows the quasi-stable behavior of spiral waves in the MM2 WT model when we increase one of the rate constant, namely, β12 → β12 ∗ 25 . Movie(M1): This movie (5 frames per second (fps)) shows pseudocolor plots of Vm that illustrate spiral waves in the MM1 WT and MM2 WT models, with increased values of D (see text): (a) MM1 WT(D), (b) MM1 WT (D2.915), (c) MM2 WT (D), and (d) MM2 WT (D1.299). Apart from an increase in the conduction-velocity CV and the arm-length, the qualitative features of spiral-wave dynamics here remain the same as those with the unaltered value of D. Movie(M2): This movie (5 fps) shows pseudocolor plots of Vm that illustrate the sensitive depen- dence of wave activity on τS2 (see text) in the MM2 WT model: (a) τS2 = 560ms, (b) τS2 = 580ms, (c) τS2 = 600ms, and (d) τS2 = 620ms. Movie(M3): This movie (5 fps) shows pseudocolor plots of Vm that illustrate the spatiotemporal evolution of the spiral waves of MM1 MUT and MM2 MUT models. Movie(M4,M5): This movie (5 fps) shows pseudocolor plots of Vm that illustrate the depen- dence on τS2 of spiral-wave initiation, in the MM2 WT model in the presence of inhomogeneities (see text) with Pf = 30, 50, 70, and 100, within circular regions of radii R = 1.875cm and R = 1.375cm . Movie(M6-M8): These movies (5 fps) show isosurface plots of Vm that illustrate scroll-waves in homogeneous media in TP06, MM1 WT, and MM2 WT models. Movie(M10): These movies (5 fps) show isosurface plots of Vm that illustrate the pacing of the tissue in MM1 and MM2 models, with a circular heterogeneities of Na mutant cells (radius of R = 1.125cm) surrounded by the Na wild-type cells. Movie(M11): These movies (5 fps) show isosurface plots of Vm that illustrate the successful elimination of the spiral waves and spiral-wave turbulence (by using the method described in the 13 Supplementary material) for the MM1 and MM2 (WT and MUT) models. References [1] K.H.W.J. Ten Tusscher and A.V. Panfilov. Alternans and spiral breakup in a human ventricular tissue model. American Journal of Physiology-Heart and Circulatory Physiology, 291(3):H1088 -- H1100, 2006. [2] S. Vecchietti, I. Rivolta, S. Severi, C. Napolitano, S.G. Priori, and S. Cavalcanti. Computer simulation of wild-type and mutant human cardiac na+ current. Medical and Biological Engineering and Computing, 44(1-2):35 -- 44, 2006. [3] J.D. Moreno, Z.I. Zhu, P.-C. Yang, J.R. Bankston, M.-T. Jeng, C. Kang, L. Wang, J.D. Bayer, D.J. Christini, N.A. Trayanova, et al. A computational model to predict the effects of class i anti-arrhythmic drugs on ventricular rhythms. Science translational medicine, 3(98):98ra83 -- 98ra83, 2011. [4] R. Mehra. Global public health problem of sudden cardiac death. Journal of electrocardiology, 40(6):S118 -- S122, 2007. [5] A.G. Kl´eber and Y. Rudy. Basic mechanisms of cardiac impulse propagation and associated arrhythmias. Physiological reviews, 84(2):431 -- 488, 2004. [6] T.K. Shajahan, S. Sinha, and R. Pandit. Spiral-wave dynamics depend sensitively on inhomo- geneities in mathematical models of ventricular tissue. Phys. Rev. E, 75(1):011929, 2007. [7] T.K. Shajahan, A.R. Nayak, and R. Pandit. Spiral-wave turbulence and its control in the presence of inhomogeneities in four mathematical models of cardiac tissue. PLoS One, 4(3):e4738, 2009. [8] R.H. Clayton, O. Bernus, E.M. Cherry, H. Dierckx, F.H. Fenton, L. Mirabella, A.V. Panfilov, F.B. Sachse, G. Seemann, and H. Zhang. Models of cardiac tissue electrophysiology: progress, challenges and open questions. Progress in biophysics and molecular biology, 104(1):22 -- 48, 2011. [9] R. Majumder, A.R. Nayak, and R. Pandit. Scroll-wave dynamics in human cardiac tissue: lessons from a mathematical model with inhomogeneities and fiber architecture. PLoS One, 6(4):e18052, 2011. [10] R. Majumder, A.R. Nayak, and R. Pandit. An overview of spiral-and scroll-wave dynamics in mathematical models for cardiac tissue. In Heart Rate and Rhythm, Eds. Tripathi, O., Ravens, U., Sanguinetti, M.C., pages 269 -- 282. Springer, 2011. [11] M.P.. Clerx. Multi-scale modeling and variability in cardiac cellular electrophysiology. PhD thesis, 2017. [12] A.L. Hodgkin and A.F. Huxley. A quantitative description of membrane current and its application to conduction and excitation in nerve. The Journal of physiology, 117(4):500 -- 544, 1952. [13] J.P. Keener. Invariant manifold reductions for markovian ion channel dynamics. Journal of Mathematical Biology, 58(3):447 -- 457, 2009. [14] C.E. Clancy and Y. Rudy. Cellular consequences of herg mutations in the long qt syndrome: precursors to sudden cardiac death. Cardiovascular research, 50(2):301 -- 313, 2001. [15] C.E. Clancy and Y. Rudy. Linking a genetic defect to its cellular phenotype in a cardiac arrhythmia. Nature, 400(6744):566 -- 569, 1999. 14 [16] M. Fink and D. Noble. Markov models for ion channels: versatility versus identifiability and speed. Philosophical Transactions of the Royal Society of London A: Mathematical, Physical and Engineering Sciences, 367(1896):2161 -- 2179, 2009. [17] J.D. Moreno, P.-C. Yang, J.R. Bankston, E. Grandi, D.M. Bers, R.S. Kass, and C.E. Clancy. in silico pharmacologic Ranolazine for congenital and acquired late ina linked arrhythmias: screening. Circulation research, pages CIRCRESAHA -- 113, 2013. [18] C.E. Clancy, Z.I. Zhu, and Y. Rudy. Pharmacogenetics and anti-arrhythmic drug therapy: a theoretical investigation. American Journal of Physiology-Heart and Circulatory Physiology, 292(1):H66 -- H75, 2007. [19] B. Carbonell-Pascual, E. Godoy, A. Ferrer, L. Romero, and J.M. Ferrero. Comparison between hodgkin -- huxley and markov formulations of cardiac ion channels. Journal of theoretical biology, 399:92 -- 102, 2016. [20] S. Zimik, N. Vandersickel, A.R. Nayak, A.V. Panfilov, and R. Pandit. A comparative study of early afterdepolarization-mediated fibrillation in two mathematical models for human ventricular cells. PloS one, 10(6):e0130632, 2015. [21] R. Majumder, R. Pandit, and A.V. Panfilov. Turbulent electrical activity at sharp-edged inexcitable obstacles in a model for human cardiac tissue. American Journal of Physiology-Heart and Circulatory Physiology, 307(7):H1024 -- H1035, 2014. [22] S. Zimik and R. Pandit. Reentry via high-frequency pacing in a mathematical model for human- ventricular cardiac tissue with a localized fibrotic region. Sci. Rep., 7(1):15350, 2017. [23] KHWJ Ten Tusscher, Denis Noble, Peter-John Noble, and Alexander V Panfilov. A model for human ventricular tissue. American Journal of Physiology-Heart and Circulatory Physiology, 286(4):H1573 -- H1589, 2004. [24] N. Vandersickel, I.V. Kazbanov, A. Nuitermans, L.D. Weise, R. Pandit, and A.V. Panfilov. A study of early afterdepolarizations in a model for human ventricular tissue. PloS one, 9(1):e84595, 2014. [25] Z.Y. Lim, B. Maskara, F. Aguel, R. Emokpae, and L. Tung. Spiral wave attachment to millimeter- sized obstacles. Circulation, 114(20):2113 -- 2121, 2006. [26] Takanori Ikeda, Masaaki Yashima, Takumi Uchida, Dustan Hough, Michael C Fishbein, William J Mandel, Peng-Sheng Chen, and Hrayr S Karagueuzian. Attachment of meandering reentrant wave fronts to anatomic obstacles in the atrium: role of the obstacle size. Circulation research, 81(5):753 -- 764, 1997. [27] D. Stauffer and A. Aharony. Introduction to Percolation Theory. CRC press, 1994. [28] Lisa A Irvine, M Saleet Jafri, and Raimond L Winslow. Cardiac sodium channel markov model with temperature dependence and recovery from inactivation. Biophysical journal, 76(4):1868 -- 1885, 1999. [29] Colleen E Clancy and Yoram Rudy. Na+ channel mutation that causes both brugada and long-qt syndrome phenotypes: a simulation study of mechanism. Circulation, 105(10):1208 -- 1213, 2002. [30] Jonathan Silva and Yoram Rudy. Subunit interaction determines iks participation in cardiac repolarization and repolarization reserve. Circulation, 112(10):1384 -- 1391, 2005. [31] Shimin Wang, Shuguang Liu, Michael J Morales, Harold C Strauss, and Randall L Rasmusson. A quantitative analysis of the activation and inactivation kinetics of herg expressed in xenopus oocytes. The Journal of Physiology, 502(1):45 -- 60, 1997. 15 [32] Vladimir E Bondarenko, Glenna CL Bett, and Randall L Rasmusson. A model of graded calcium release and l-type ca2+ channel inactivation in cardiac muscle. American Journal of Physiology- Heart and Circulatory Physiology, 286(3):H1154 -- H1169, 2004. [33] Shimin Wang, Vladimir E Bondarenko, Yu-jie Qu, Glenna CL Bett, Michael J Morales, Randall L Rasmusson, and Harold C Strauss. Time-and voltage-dependent components of kv4. 3 inactivation. Biophysical journal, 89(5):3026 -- 3041, 2005. [34] Vladimir E Bondarenko, Gyula P Szigeti, Glenna CL Bett, Song-Jung Kim, and Randall L Rasmusson. Computer model of action potential of mouse ventricular myocytes. American Journal of Physiology-Heart and Circulatory Physiology, 287(3):H1378 -- H1403, 2004. [35] Pietro Balbi, Paolo Massobrio, and Jeanette Hellgren Kotaleski. A single markov-type kinetic model accounting for the macroscopic currents of all human voltage-gated sodium channel isoforms. PLoS computational biology, 13(9):e1005737, 2017. [36] Thomas O'Hara, L´aszl´o Vir´ag, Andr´as Varr´o, and Yoram Rudy. Simulation of the undiseased human cardiac ventricular action potential: model formulation and experimental validation. PLoS computational biology, 7(5):e1002061, 2011. [37] Natalia A Trayanova and Brock M Tice. Integrative computational models of cardiac arrhythmias -- simulating the structurally realistic heart. Drug Discovery Today: Disease Models, 6(3):85 -- 91, 2009. [38] Rupamanjari Majumder, Alok Ranjan Nayak, and Rahul Pandit. Nonequilibrium arrhythmic states and transitions in a mathematical model for diffuse fibrosis in human cardiac tissue. PLoS one, 7(10):e45040, 2012. [39] M. Potse, B. Dub´e, J. Richer, A. Vinet, and R.M. Gulrajani. A comparison of monodomain and bidomain reaction-diffusion models for action potential propagation in the human heart. IEEE Transactions on Biomedical Engineering, 53(12):2425 -- 2435, 2006. [40] Yves Bourgault and Charles Pierre. Comparing the bidomain and monodomain models in electro- cardiology through convergence analysis. 2010. [41] S. Sinha, A. Pande, and R. Pandit. Defibrillation via the elimination of spiral turbulence in a model for ventricular fibrillation. Phys. Rev. Lett., 86(16):3678, 2001. 4.1 Figures 4.2 Tables 16 Figure 1: Schematic diagrams for Markov Models for Wild-Type (WT) and Mutant (MUT) cases: Top panel: MM1 WT and MM2 WT models; the MM1 WT model has 9 states, namely, the open state (O), the closed states (C1, C2, C3), and the inactivation states (IF, IM1, IM2, IC2, IC3); the MM2 WT model has 8 states, namely, the open state (O), the closed states (C1, C2, C3), and the inactivation states (IF, IS, IC2, IC3). The orange, double-headed arrows indicate transitions between such Markov states; transition rates for the rightward (leftward) transition are given above (below) these arrows, e.g., a111 (b111) for the IC3 → IC2 (IC2 → IC3) transition in MM1 WT. Bottom panel: schematic diagrams for the MM1 MUT and MM2 MUT modes; the MM1 MUT model has the same number of states as the MM1 WT model, but the transition rates between the Markov states are different; the MM2 MUT model has 12 states: 8 of these are as in the MM2 WT model; in addition there are 4 bursting states, namely, BO, BC1, BC2, BC3. fast inward N a+ current L-type inward Ca++ current Transient outward current Slow delayed rectifier outward K + current Rapid delayed rectifier outward K + current Inward rectifier outward K + current IN a ICaL Ito IKs IKr IK1 IN aCa N a+/Ca++ exchanger current IN aK IpCa IpK IbN a IbCa N a+/K + pump current plateau Ca++ current plateau K + current background inward N a+ current background inward Ca+ current Table 1: The various ionic currents in the TP06 model Ref. [1]; the symbols used for the currents follow Ref. [1]. 17 Figure 2: Na-channel activation A and the inactivation I, in the models MM1 (WT and MUT) and MM2 (WT and MUT), compared with their TP06 counterparts. (A) Activation- and (B) inactivation-protocol plots, for different values of Vc, of IN a versus time t (see 2.1), whence we obtain the plots of (C) and (D), which depict, respectively, the dependences of A and I on Vm, for the MM1 WT, MM2 WT, MM1 MUT, MM2 MUT, and TP06 models; for the TP06 model A = m3∞ and I = j∞ × h∞. 18 Figure 3: Plots of the probabilities PI , PC, and PO (see text) versus time t for the Na channel in the course of an action potential. Plots for the MM1 (MM2) model are in the top (bottom) panel; the blue and red curves are for WT and MUT models, respectively. We obtain these plots by pacing a single cell with a pacing-cycle length PCL = 3000 ms in MM1 and MM2 models for both WT and MUT cases (plots for the n = 501 stimulation). The plots in the insets show the sudden opening of the Na channel in the mutant cases because of delayed inactivation and closing. PCL (ms) 300 650 1000 Model TP06 MM1 WT MM2 WT TP06 MM1 WT MM2 WT TP06 MM1 WT MM2 WT INamax (pA/pF) −177.19 −82.23 −252.13 −298.19 −127.6 −280.98 −312.74 −144.46 −300.43 Vmax (mV) dV dt max (mV/ms) 23.42 21.17 18.56 37.17 32.78 23.74 39.82 36.58 25.11 227.97 81.76 256.8 349.94 128.8 280.79 373.65 146.18 300.49 APD (ms) 218.76 219.66 226.76 291.56 292.36 304.6 302.12 302.64 314.84 Table 2: Characteristic properties of the action potentials in TP06, MM1 WT, and MM2 WT models. These data are for three representative cases: low-frequency (PCL= 1000ms), intermediate-frequency (PCL= 650ms), and high-frequency (PCL= 300ms) pacing. 19 Figure 4: Plots of action potentials, the fast Na current IN a,f , and the late Na current IN a,L. Top panel: cell paced for PCL = 1000 ms for n = 501 stimulations for TP06, MM1 WT, and MM2 WT models (plots for the n = 501 stimulation). Bottom panel: cell paced for PCL = 3000 ms for MUT models and comparing MM1 WT with MM1 MUT and MM2 WT with MM2 MUT (plots for the n = 501 stimulation). Note that the late opening of the mutant Na channel in the repolarization regime causes a release of IN a,L that leads, in turn, to early afterdepolarizations (EADs) in the AP for mutant models. 20 Figure 5: Restitution plots for the TP06, MM1 WT and and MM2 WT models. (A) Single- cell static APDR (profiles for TP06 and MM1 WT models lie close to each other, but the MM2 WT curve lies above these) and (B) the dynamic CVR, for a one-dimensional cable of cells (640 × 10). The slopes of the profiles are given in (C) for the APDR and in (D) for the CVR; in all these three models, the maximal slope of the APDR profile > 1; the CVR profiles in MM1 WT and MM2 WT models do not depend sensitively on DI over the range of values in these plots. The solid lines are calculated for the same diffusion constant D; for the MM1 WT and MM2 WT models this yields steady-state CVs of 40.41 cm/s and 54.89 cm/s, respectively, which are not in the normal range (for the myocardium) (cid:39) 60 − 75 cm/s; if we increase D, for the MM1 WT and MM2 WT models (see text) CV can be brought to this normal range, as we show by the dashed-line plots. 21 Figure 6: Pseudocolor plots of the transmembrane potential Vm illustrating the spatiotem- poral evolution of spiral waves. First row: WT models (see text) (A) TP06, (B) MM1 WT, and (C) MM2 WT; in the MM1 WT (MM2 WT) model the spiral wave is stable (unstable). Second row: MUT models (see text) (D) MM1 MUT and (E) MM2 MUT. The formation of type-3 EADs (see text) leads to backward propagation in the MM2 MUT; by contrast, the type-2 EAD in the MM1 MUT model does not lead to such backward propagation. However, these EADs create, at the cellular level, dynamical heterogeneities, far from the stable spiral core, as shown in (D); this leads to spiral break up in a homogeneous MM1 MUT simulation domain. The complete spatiotemporal evolution of the spiral waves of MM1 MUT and MM2 MUT is shown in the Movie(M3) (5 frames per second (fps)) in the Supplementary Material 4. Figure 7: Pseudocolor plots of the transmembrane potential Vm illustrating the spatiotem- poral evolution of spiral waves for different values of τS2, the time interval between the S1 and S2 impulses (see text). (A) TP06 and (B) MM1 WT models at 5.9s after spiral-wave initiation; and (C) the MM2 WT model at 6.9s after such initiation. Clearly, the spatiotemporal evolution of the spiral waves in the TP06 and MM1 WT models is independent of τS2, in the range of values investigated here, but not so for the MM2 WT model. The spatiotemporal evolution of the spiral waves for different τS2 of MM2 WT is shown in the Movie(M2) (5 fps) in the Supplementary Material 4. 22 Figure 8: Pseudocolor plots of the transmembrane potential Vm illustrating the spatiotem- poral evolution of spiral waves in the presence of in-excitable obstacles for different values of Pf and different τS2 (see text) in TP06, MM1 WT and MM2 WT models. The spiral waves in TP06 (first row) and MM1 WT (second row) models anchor to the obstacle (for the values of Pf and τS2 used here) and so are independent of τS2. By contrast, the waves in the MM2 WT model (last two rows) depend on R, Pf , and on τS2. The spatiotemporal evolution of the spiral waves in the presence of inexcitable obstacles for two representative radius (R = 1.375 and R = 1.875cm) for the MM2 MUT model is shown in the Movie(M5,M4) (5 fps) in the Supplementary Material 4. 23 Figure 9: The dependence of the time period T of the anchored spiral wave on the radius R and percentage of fibrosis Pf . (A) Plots of T versus R for different values of Pf for TP06, MM1 WT and MM2 WT models; note spiral anchoring starts around R (cid:39) 1.875cm for the MM2 WT model. (B) and (D): Plots of the change in time period ∆T versus, (obtained from five recording points in the domain of which one grid point is in the region with heterogeneity) for TP06 and MM1 WT models (T0 is the time period for a completely inexcitable obstacle (Pf = 100%), for different Pf . If ∆T > 0, then the frequency ω ∼ T −1, for a given pair (R, Pf ), is less than ω0 ∼ T −1 (for R, Pf = 100%); this may occur because the spiral core penetrates the obstacle because of a spanning cluster of excitable regions inside the obstacle. In (C) and (E) we show different regions in the (R, Pf ) plane for TP06 and MM1 WT models, respectively, with the following the color code: Light Blue: an increase in ω relative to ω0 (caused by penetration of the spiral core). Light Green: decrease in ω relative to ω0 (accompanied by penetration of the spiral core); (R, Pf ). Yellow: No penetration of the spiral core into the obstacle. Dark Blue: No change in ω relative to ω0 even though the spiral core penetrates into the obstacle. 0 Figure 10: Stability diagrams for spiral-wave activity, in the presence of localized, in- excitable obstacles distributed within a circular region of radius R, for different values of Pf , in the MM2 WT model. Color code: Brown, Green, and Blue show regions with an anchored spiral, spiral breakup, and no activity, respectively. 24 Figure 11: Pseudocolor plots of the transmembrane potential Vm illustrating the spatiotem- poral evolution of electrical-activation waves when there is a circular clump of mutant cells. The clump radius R = 1.125 cm (shown via a black circle); this clump is surrounded by wild-type (WT) cells, and the simulation domain is paced from the left boundary (pacing frequency 3.7Hz). (A) MM1 MUT model (no spiral wave forms); and (B) MM2 MUT model (a spiral wave forms). For the complete spatiotemporal evolution movie see the Movie(M10) (5 fps) in the Supplementary Material 4. The qualitative difference between (A) and (B) arises because of the different types of EADs in MM1 MUT and MM2 MUT models. 25 Figure 12: Color isosurface plots of the transmembrane potential Vm illustrating scroll waves in (A) TP06, (B) MM1 WT, and (C) MM2 WT models. The isosurfaces lie between −10mV and 30mV in both the homogeneous domain (top panel) and with localized obstacles [Pf = 10% (middle panel) and Pf = 50% (bottom panel)]; these illustrative plots are at t = 2s. In a homogeneous domain scroll waves are stable in TP06 and MM1 WT models, but not in the MM2 model. Note that the scroll wave is anchored to the obstacle with Pf = 50% in the MM2 model. For the spatiotemporal evolution of these scroll waves see Movies(M6-M8) (5 fps) in the Supplementary Material 4. 26
1712.05784	1	1712	2017-12-15T18:39:34	Measuring behavior across scales	[ "physics.bio-ph", "q-bio.NC", "q-bio.QM" ]	The need for high-throughput, precise, and meaningful methods for measuring behavior has been amplified by our recent successes in measuring and manipulating neural circuitry. The largest challenges associated with moving in this direction, however, are not technical but are instead conceptual: what numbers should one put on the movements an animal is performing (or not performing)? In this review, I will describe how theoretical and data analytical ideas are interfacing with recently-developed computational and experimental methodologies to answer these questions across a variety of contexts, length scales, and time scales. I will attempt to highlight commonalities between approaches and areas where further advances are necessary to place behavior on the same quantitative footing as other scientific fields.	physics.bio-ph	physics	Measuring behavior across scales Gordon J. Berman1 1Department of Biology, Emory University∗ The need for high-throughput, precise, and meaningful methods for measuring behavior has been amplified by our recent successes in measuring and manipulating neural circuitry. The largest challenges associated with moving in this direction, however, are not technical but are instead conceptual: what numbers should one put on the movements an animal is performing (or not performing)? In this review, I will describe how theoretical and data analytical ideas are interfacing with recently-developed computational and experimental methodologies to answer these questions across a variety of contexts, length scales, and time scales. I will attempt to highlight commonalities between approaches and areas where further advances are necessary to place behavior on the same quantitative footing as other scientific fields. As modern techniques for recording and manipulat- ing neural circuits have expanded our toolbox for de- constructing the molecular and cellular components of animals' nervous systems, an accompanying realization has gradually developed: to more fully comprehend the function of neural circuits and the computations under- lying them, we must understand their output in an ac- cordingly precise manner [1, 2]. Specifically, we need to measure behavior. More careful measurements of the ac- tions animals perform is key not just for advancing our basic understanding of nervous system function, but also in our assessment and categorization of psychiatric dis- orders and the development of brain-machine interfaces [3, 4]. But what type of behavior do we want to measure, and once we decide on this, how do we measure it? Answering these questions has proven difficult, but this is largely due to conceptual limitations rather than technical ones. If watching an animal behave, what are some precise, yet manageable, numbers we should use to describe its movements? Is it the center of mass motion of the whole animal? The position and velocity of the organism's body and limbs? The dynamics of individual myosin motors within muscle tissue? A more coarse- grained measure related to the animal's "intended" ac- tion? A collective variable describing the combined dy- namics of many animals? And how do we connect these scales to make inferences from the cellular and the molec- ular up to the movement of a limb, a wing, a finger, or an eyebrow? This is the dilemma that those of us who attempt to measure behavior commonly face. While selecting the proper representation for one's measurements is hardly a problem exclusive to the study of animal behavior (e.g. "more is different" and not wanting to "model bulldozers with quarks" [5, 6]), it is felt acutely by researchers in this field due to the multi- scale and distributed dynamics inherent to almost any behavioral process. Cognition or sensation acts to drive muscles that drive joints that drive limbs that drive lo- comotion or other motions, which then send feedback signals in the reverse direction, and the cycle continues. Where in this loop do we define behavior? Or is it the ∗ [email protected] whole loop? And what numbers should we use to de- scribe the observed dynamics? These are the questions that I will focus on here, asking how to best represent be- havioral data in a manner that bridges length and time scales, highlighting particularly fruitful approaches. Before progressing, though, it should be noted that there has been a recent proliferation of review arti- cles discussing behavior, detailing concepts ranging from computational techniques for measuring behavior [7–10] to finding simplicity in "big behavioral data" [11, 12] to the advent of computational psychiatry and measuring emotional states [3, 13–17] to the need for measuring be- havior in the first place [1, 2] to the reproducability and robustness of said behavioral measures [18, 19]. While there will inevitably be a great deal of overlap between this review and those that have come before it, here I will focus less on the practical aspects of behavioral quan- tification and more on the consequences of the repre- sentational choices one makes, highlighting areas where further progress is required. MEASURING BEHAVIOR ON THE ORGANISMAL SCALE Beyond being a mere technical inconvenience, the rela- tive lack of a quantitative language for measuring behav- ior has shaped the types of questions we have been able to ask. In a laboratory setting, behavioral experiments have usually been designed to observe a restricted set of actions within the scope of a restricted environment [20, 21]. To wit, the behavior measured in most of these experiments is typically performed within a "paradigm" – with the accompanying implication that we have tuned the animal to our quantification scheme rather than the other way around. Examples of this approach would be placing an animal in a maze where it can only turn left or right or head-fixing a rodent, where it is asked to perform a whisking-based detection task. While this reliance on non-naturalistic behavior sometimes emerges from a cul- ture of treating behavior as a read-out variable of the neural hardware in question, more commonly, it is driven by an understandable desire to have a repeatable mea- surement that can generate high-throughput data while recording activity from neurons. Nevertheless, the end result is to measure an over-constrained behavior that likely lies outside of an animal's typical repertoire of ac- tions. To move forward with the analysis of more natural behavior, we can try to imagine the best case scenario, ignoring all of the technical worries. If we have an ar- bitrarily large amount of high quality data from an an- imal behaving with minimal artificial constraints, how should we describe it quantitatively? To an extent, the answer here is the same as in most other scientific mea- surements: we desire consistency (repeatable results), fidelity (describing the system as accurately and com- pletely as possible), interpretability (ease of relating the found numbers to their biological underpinnings), and scalability (requiring minimal manual labor or scoring without impractically taxing computational or human resources). While consistency and scalability can be theoretically obtained independently of the other two, fidelity and in- terpretability are, by definition, in tension, with mea- surements typically being more understandable but less accurate as we remove details. Our goal for measur- ing behavior, then, is to find descriptive representations of these multi-scale processes that are as parsimonious as possible. This trade-off naturally suggests a contin- uum of solutions, and in the rest of this article, we will see how researchers have represented behavioral data in varying ways, tying-together the length and time scales of naturalistic behavior at many levels of abstraction. SELECTING A REPRESENTATION A good place to start investigating behavioral repre- sentations is to note the options available to researchers a decade ago if they wished to measure ethological behav- ior at the organismal scale. One option would have been the previously mentioned paradigmatic approach, where the quantification is ingrained into the experimental ap- paratus itself. Quantifying behavior in this manner has the advantages of high-throughput and consistent mea- surements, but it captures a very low-dimensional and potentially unnatural measurement [22]. Another approach would be to measure a coarse, yet non-paradigmatic, variable such as mean velocity or the fraction of time moving (including the laser-crossing ex- periments that are typical in circadian rhythm studies [23]). These measurements are more naturalistic than paradigmatic ones, while allowing for a similar level of throughput, however, they only capture dynamics at a single scale. This is a plausible approach when studying the effects of genetic manipulations on sleep-wake cycles, but it not be able to capture, say, the precise grooming patterns of an animal or other movements that are un- likely to be apparent by treating the animal as a point moving through space. Alternatively, if a researcher desired a richer descrip- tion of an animal's behavior, they could have developed a human-defined classification system for an animal's be- 2 havior that was then scored by a trained observer. While providing a great deal more description, this approach is extremely labor-intensive, often requiring significant effort to devise the scoring scheme, followed by poten- tially months of researcher-hours to apply it. Moreover, although the scheme one uses can be elaborated in de- tail, there will inevitably be user-specific variability in its application. More problematic, since behaviors are defined and delineated intuitively, it is difficult to quan- titatively argue that one individual's or group's repre- sentation of the behavior is more accurate or appropriate than another's, further limiting reproducibility. Lastly, this approach implicitly assumes that behavior can be described in terms of hopping between discrete states without showing, from the data, that such a model is indeed a reasonable representation in the first place. Skipping ahead to the present, all three of these op- tions are still frequently used, often generating novel insights into behavior and the mechanisms driving it. Automation has greatly increased the throughput of the first two described options, especially in small organisms like worms [24, 25], flies [26, 27], and zebrafish larvae [28–30]. Moreover, supervised machine learning tech- niques have greatly improved the repeatability and de- creased the manual effort required to analyze behavioral data with user-defined classification of behavior [31–35]. That being said, the fundamental difficulties with these approaches remain - the level of behavioral description is either coarse-grained or behaviors are intuitively de- fined, explicitly encoding a human observer's underly- ing assumptions about an animal's behavior. Thus, we come to a fundamental query: how can we leverage mod- ern data-collection techniques to extract complex behav- ioral representations in manner that is transparent and repeatable, with explicitly-stated and testable assump- tions that shed light onto particular biological questions? The answer to this requires thinking about the general principles. STEREOTYPY AS A GENERAL PRINCIPLE One area of organismal-scale research where the chal- lenges in measuring behavior have become predomi- nantly technical is biolocomotion, the study of how an- imals move through their environments [36–39]. Here, while many deep questions regarding the performance, control, and evolution of these behaviors remain to be answered, there is a generally-agreed-upon framework for measuring behavior: most researchers study dynamic trajectories of motion, typically center-of-mass, body bending, and/or limb trajectories. What is it about bi- olocomotion that has made it amenable to this type of agreed-upon representation? Part of the reason for this advantage is the clear etho- logical context of the actions studied – moving from one place to another quickly, efficiently, and robustly. Thus, there is a natural mathematical formalism to translate between scales, namely Newtonian mechanics, and the 3 Inspired by these studies, much of the recent progress in developing tools for data-driven and unsupervised (i.e. without the aid of human-labeled examples) analysis of animal behavior has resulted from this observation that a large fraction of animal movements are low-dimensional compared to the animal's total capacity for movement and are often repeated in a similar manner (Fig. 1) [11, 51–53]. However, in order to proceed, we must have a more precise mathematical description of stereotypy (i.e. defining what we mean by "low-dimensional," "sim- ilar," and "movement"). The goal here is to put the human at the beginning of the analysis process (defining stereotypy) as well as at the end (interpreting behavioral outputs of the analysis process), rather than in the mid- dle, as is the case for label-based, or supervised, methods for behavioral analysis. Several recent studies have developed tools for find- ing these stereotyped behaviors across a range of model organisms during (relatively) free behavior, from worms to rodents [53–59]. Although these researchers have all taken widely-differing technical approaches, there are key similarities that join their efforts together. The com- mon logic between these methods points toward a shared definition of stereotyped behavior and forces us to ask a pair of fundamental questions: what does it mean for two behaviors to be similar or different, and how do we place a number on this difference? One thing that is important to note, though, is that none of the ap- proaches described below are strictly unbiased, despite the term being often brandished when describing their advantages. The implication in calling these unsuper- vised approaches unbiased is typically that the analyzer is removing themselves fully from the loop. Each choice a researcher makes, though, has consequences, regard- less of how explicit those choices are, but the key to all of these approaches is that the consequences of these op- tions are readily apparent. FIG. 1. Approaches for identifying stereotyped move- ments. A. Representation of all of the movements an ani- mal could theoretically make. For instance, each line could be the dynamics of two joint angles, say, the bending of a knee and an ankle, or another set of postural variables over time. Although an animal could potentially move with any of these postural trajectories, many of the motions here would be only rarely performed. B. How we observe most animals to move. Specifically, they use a relatively small portion of their potential behavioral repertoire (stereotyped behaviors, colored lines) along with a few instances of less-robustly ob- served ones (non-stereotyped behaviors, black lines). C. One way to isolate stereotyped behaviors is to break-up the ob- served trajectories into clusters (denoted by dashed lines). D. An alternate means of identifying stereotyped behaviors is to transform the dynamics in such a way that, for in- stance, each time one of the trajectories in B is performed, a dot is placed using a low-dimensional embedding to a differ- ent space. Similar trajectories are mapped near each other (dots), and stereotyped behaviors could be identified as peaks in the density contours (lines) of this map. behaviors in question are clearly separable from other actions that the animal performs. Even in cases where the mathematics underlying this translation are unten- ably difficult to analyze directly, robots can serve as the physical equivalent of generative models to bridge this gap [40–42]. Moreover, concepts such as optimal control or energy-efficiency provide a theoretical basis for pro- viding meaning to the investigations [43–45]. Another factor is that these behaviors are highly stereotyped, with physical constraints typically allowing for only a small number of movement patterns or gaits [46]. Al- though animals are capable of moving their limbs in an extremely large manner of ways [47], during locomotion, their motion lies on a very low-dimensional set of pos- tures, with small perturbations either corrected for or used to actuate control [48–50]. FINDING STEREOTYPED MOVEMENTS IN BEHAVIORAL DATA Although superficially distinct, there are surprising similarities in the underlying bases of different ap- proaches for automatically identifying stereotyped be- havior from videos (we will ignore other modalities for the moment) of freely-behaving animals. The general framework has been to first extract a low-dimensional postural time series from a data set, followed by a trans- lation of these postures into a dynamical representation that is used to create a behavioral representation that iso- lates individual stereotyped actions. If desired, an ani- mal's dynamics within this behavioral representation can be observed over time, finding patterns and sequences of behavior (Figure 2). ABDCPosture 1Posture 2Posture 1Posture 2Posture 1Posture 2Behavior Dimension 1Behavior Dimension 2All Possible MovementsAll Observed MovementsBehaviors from ClusteringBehaviors from Embedding4 imal or large amounts of manual correction. Recent ad- vances in experimental design [34, 63, 64] and computa- tional algorithms [7, 65–67] provide hope for improving the state-of-the-art moving forward, but for large data sets containing up to billions of images, tracking individ- ual body parts is not currently practical, especially for 2-dimensional images. Instead of directly tracking, a common approach has been to think about postural decomposition as an image compression problem. After doing some image process- ing to isolate the animal from the background and align it translationally and rotationally to a template image, the tactic taken by work in flies [53] and mice [56] has been to perform a dimensionality reduction operation like Principal Components Analysis (PCA) on the raw image pixel data. This process allows for images of an animal with complex morphology to be repeatably and continuously mapped into a relatively small set of time series, much like direct tracking of joint angles would do, but with vastly fewer errors and no need for manual in- spection (Fig. 3C). This process has the disadvantage, however, of creating relatively uninterpretable time se- ries, a fact we need take into account when moving to- ward a dynamical representation. Building representations of dynamical behavior When defining stereotyped behavior, we typically think of movements, not postures. For example, we wouldn't describe walking as bending the right knee at 73.1o, the right ankle at 15.23o, and so on, but rather as a trajectory of these angles through time. As a result, to measure stereotyped behaviors, we need to create a dynamical representation that describes how the mea- sured postural time series are changing. Building such a representation can be achieved by either directly fitting a differential equation to the postural data or through attributing features that incorporate dynamics such as temporal motifs or time-frequency features to individual segments of the data. We will see examples of each of these approaches momentarily. From here, one would like to create a behavioral representation, which can be thought of as longer-time scale changes in the underlying postural movements that generate the observed postural motions. For instance, giving the relative velocities of each of an insect's six legs might be a dynamical repre- sentation, but saying that the animal is walking with an alternating tripod gait would be a behavioral representa- tion. Of course, we need to make this idea more precise, and we will see how several different studies have done this, each with associated strengths and challenges. The most straight-forward process for building a dy- namic representation is to eliminate the step of find- ing postural time series and instead create a manually- curated set of dynamical features that are later used as the input to either a supervised classifier or a cluster- ing/embedding algorithm [32, 68–72]. While relatively easy to implement, this approach risks missing elements FIG. 2. Archetypical data analysis pipeline for identi- fying stereotyped behaviors automatically from data. Extracting postural time series The first step in almost any of these analyses is to isolate the animal's posture from the raw video data. Here, by posture, I mean a measure that describes the configuration of an animals' body and limbs at a given point in time (describing how they move will come in the next section). Usually, we prefer to describe this configuration in a manner that is in the body frame of the animal so that behavior is measured independently from spatial position or orientation. It is from this snapshot that further analyses will be devised, and it is here where organism-specific practicalities are most apparent. This latter point can be readily seen in the difference between describing a nematode like C. elegans and a fly like D. melanogaster (Figure 3). While almost all of the dynamics of worm behavior could be described by the motion of its centerline, a fly's movement is the combi- nation of six legs (each with two joints), two wings (each capable of moving with three degrees of freedom), and other body movements such as abdomen bending. These body plans clearly require different representations, even if we expect both to be relatively low-dimensional over the course of typical activities the animals perform. A rodent, with a more flexible body, or a human, with its typical bipedal walking gait, would require different rep- resentations still. In all cases, though, the aim is to take a high-dimensional measurement – say, thousands to mil- lions of pixel values – and reduce it to a low-dimensional set of numbers describing the animal's posture. The traditional, and in some senses optimal due to its interpretability, manner to achieve a set of low- dimensional time series has been to track the positions of individual body parts such as joints, leg tips, the tail, or the head. Outside of animals with relatively simple mor- phologies like C. elegans, this is an extremely difficult computer vision problem that has been the subject of comprehensive discussions elsewhere [7–9]. Even in the case of worms, new image analysis methods have been necessary to account for events where the worm crosses itself [60–62]. For legged animals, most automated meth- ods typically require either attaching markers to the an- Raw ImageDataImageProcessingVideo of a Behaving AnimalAlignment &SegmentationPosturalTime SeriesTracking orLow-dimensionalEmbeddingSequences& PatternsBehavioralRepresentationDescriptive or Generative Model of a Behaving AnimalClustering, Embedding, or Fixed PointsDynamicalRepresentationDynamical system,Motifs, or Time-Frequency5 FIG. 3. Examples of postural representations. A. A schematic for how posture is typically represented by assigning body frame coordinates, here for a fruit fly. This assignment is usually created from manual tracking or machine vision techniques. B. Using variations in the tracked centerline of the nematode C. elegans (left) to find a set of postural modes (right). Here, principal components analysis is used to find a set of "eigenworms," where the original centerline can be largely reconstructed through a linear combination of these centerline variations (adapted from [54]). C. In cases where tracking is not feasible due to occlusions, high-dimensionality, and/or large data sets, an alternative approach has been to use image compression to find postural modes, such as those seen in the fly images here (adapted from [53]). Here, red and blue represent positive and negative eigenvector magnitudes, respectively, that are the result of concentrating as much of the data's variance in as few directions as possible. The original image can be reconstructed via a linear combination of all the modes plus an overall mean, and time series can be generated by observing sequential images' projections onto these postural modes. of behavioral dynamics not captured in the list, and each of the measurements potentially has different units (e.g. velocity, angular velocity, acceleration, distance from an- other animal), requiring additional conversion factors or assumptions about equal variance that could affect any analysis' outcome in subtle ways. Ideally, an appropriate dynamical representation would emerge naturally from postural dynamics. To date, the clearest example of using postural data to ex- plicitly generate a dynamical system that provides a nat- ural behavioral representation is the work on C. elegans from Stephens et al [52, 54]. Here, the authors found that the majority of a worm's motion can be described by the progression of a single phase variable that can be thought-of as the advance of a traveling wave moving up or down the animal's body. Fitting the observed dynam- ics of this variable to a model with a deterministic and a stochastic component (Fig. 4A-B), they find that the worm's behavior can be described as a set of dynamic at- tractors with switching times that are predictable from the statistics of the underlying noise. Although applying such methods directly to higher-dimensional data sets like those generated from legged animals can be challeng- ing, recent advances in finding dynamical models that best describe a continuous time series provide future av- enues for exploration [75–77]. Another approach to building a dynamical system rep- resentation is to fit a statistical model to the data. A prominent example of this can be seen in the work of Wiltschko et al [56], who took collected postural time series data of mice and fit an Autoregressive Hidden Markov Model (AR-HMM) to their data (Fig. 4E-G). One can think of this approach as fitting small segments (less than 1 s) to linear dynamical systems, and that the animal is switching between these systems with time scales that are significantly longer than those of the dy- namics within a given system. This method creates a dynamical representation (bottom row of Fig. 4E) at the same time as it creates a behavioral representation (top row of Fig. 4E). While the ability to simultaneously represent dynam- ics and behavior is a distinctive advantage of the AR- HMM approach, it is also a limitation since it requires a parameter that sets the overall time scale of staying in a particular behavioral state. One could imagine amend- ing this limitation by adding additional time scale pa- rameters when fitting the model, but this still requires a hand-tuning of the time scales available to the system, as well as a corollary assumption that the amount of time an animal spends in a particular behavior must follow an exponential distribution. The time spent performing a behavior, however, can range over orders of magnitude – from a reflex lasting tens of milliseconds to a night's sleep – and long time-scale dynamics are often observed in behavioral data [52, 74]. Moreover, if one wishes to directly measure the time scales evident in a particular data set, the fact that the approach used relies on this type of structure as an assumption can be confounding. A complementary approach is to create a multi-scale dynamical representation that forms the basis for a be- havioral representation. This can be achieved through finding motifs of varied lengths in a data set [55, 73, 78] or using a time-frequency analysis approach like a wavelet transform [53, 59, 79] to represent postural dy- namics across a variety of time scales. For the case of motif-finding (Fig. 4C-D), one finds postural patterns that commonly occur throughout a data set and looks for when the animal exhibits similar dynamics. The rela- tive frequency and patterns of use for these motifs can be used to create "behavioral fingerprints" for individuals or collections of animals differing in genotypes, neural ma- nipulations, or other conditions of interest. The resulting φ(x0,y0)(xT1,yT1)(xJ1,yJ1)(xT2,yT2)(xJ2,yJ2)ABCDirect PostureTrackingReducing the Dimension of Tracked PosturesPosture Tracking ThroughImage CompressionMode 1Mode 2Mode 3Mode 4Image StackPostural EigenvectorsTrackedPosture6 FIG. 4. Examples of dynamical and behavioral representations. A. For C. elegans, a histogram of projections onto the first two "eigenworms" (the left two curves in Fig. 3B) shows a low-dimensional structure that can be parameterized by a single phase variable, φ. B. Fitting the dynamics of this variable to a deterministic dynamical system yields this phase map, with forward and backward locomotion naturally emerging as traveling wave trajectories at the top and bottom, respectively, and two fixed points in the middle corresponding to two different pause states (A and B are adapted from [54]). C. An alternative approach to represent C. elegans behavior is via motif-finding. Here, time-series of projections onto the eigenworms are scoured for repeated patterns (e.g. the blue and red curves here). These patterns are then catalogued and used as the basis for a behavioral representation (adapted from [55]). D. Instead of using dynamical motifs directly, the worm's behavior can be captured as a sequence of postures, as seen in this example from [73]. E. The approach taken by [56] was to fit an autoregressive hidden markov model (AR-HMM) to postural data of mouse movements, generated in a similar, but not identical, manner to that seen in Fig. 3C. Here, each Pt is a vector of the animal's postural mode values at time t, and St is an underlying state that affects the dynamics of postural outputs. Here, arrows imply direct dependence (i.e. Pt is a stochastic function of St, Pt−1, and Pt−2, and so on). It is assumed that the time scale for changes in P is much faster than that for changes in S. This latter time scale, a parameter in the model, sets the distribution for the length of time that an animal stays within a particular behavioral state. F. Average behavioral usage frequencies using an AR-HMM for four different mouse genotypes: Wild type, C57/BL6, as well as homozygous (Mut) and heterozygous (Het) mutations in the aretinoid-related orphan receptor 1β (Ror1β) gene. G. Distinct walking gaits found in the Mut (top) and C57/BL6 (bottom) mice (E-G adapted from [56]). H. An example of a time-frequency analysis representation from freely-moving fruitflies, where each set of axes represents a mode, and the colormap values indicate the continuous wavelet transform amplitudes for at each point in time. This approach allows for multiple time scales to enter the dynamical representation. I. Probability density resulting from embedding points into 2-d such that two instances when a fruit fly is moving similar parts of its body at similar speeds are mapped nearby. Note the peaks and valleys. Here, the peaks represent stereotyped behaviors. J. Break-down of the behavioral representation in I, with names for the behaviors within each of these regions manually labelled. Black lines are proportional to the transition probability between moving from one coarse region to another, with right-handedness implying the direction of transmission. (H-J adapted from [53, 74]). behavioral representation is thus the set, frequency, and ordering of motifs that an animal performs. A difficulty of this approach, however, is that results may not always be robust to slight changes in postural dynamics such as changes in frequency or relative phasing between limbs. An alternative approach to capture behavioral dynam- ics across multiple time scales is to use time-frequency analysis (Fig. 4H). Here, one takes the set of postural time series, determines a wide range of frequencies that are present in each time series and measures the relative importance of each of these frequencies as a function of time. This importance is often quantified via a wavelet transform [80], which uses a trade-off between accurate temporal resolution and poor frequency resolution at high frequencies and poor temporal resolution and ac- curate frequency resolution at low frequencies to gener- ate a multi-scale representation of the animal's postural movements. The resulting dynamical representation for a single point in time is thus a set of wavelet amplitudes for a collection of frequencies from each of the observed postural time series. Despite the fact that the wavelet transform contains both amplitude and phase informa- tion, it is typical to only use the amplitude information, as this eliminates many of the robustness issues experi- enced in the motif-finding case. Behavioral representa- tions can then be obtained from either clustering [58, 79] or low-dimensional embedding (Fig. 4I-J) [53, 59] of the resulting vector of amplitudes. Typically, when embed- ding these feature vectors, an anisotropic density across this space emerges, with local peaks corresponding to particular stereotyped behaviors. Accordingly, one could treat the behavioral representation as either the density itself or the sequence of peaks that an animal visits. including space, other behavioral modalities, other indi- viduals, and neural dynamics. 7 Discrete vs. continuous behavioral representations Note how we now have seen that behavioral represen- tations can either be discrete (e.g. clusters or motifs) or continuous (e.g. densities or non-piecewise dynami- cal models) and that discrete representations can often be derived from continuous ones (e.g. fixed points or peaks). So which is better? Ideally, one is able to iden- tify a discrete representation through the fixed points of a dynamical model, but this is not currently practi- cable for animal morphologies more complicated than a worm's. For other systems, though, like most method- ological questions, the answer depends on the experi- mental exigencies at play, and performing both often provides additional context and information. On one hand, in favor of continuous representations, it is more intellectually satisfying to show that a discrete representation arises naturally out of a data set with- out imposing such a structure a priori. Even in the case where a discrete representation is appropriate, it may be that the interesting measurements to note are the subtle distinctions on the edge of the peaks. Additionally, al- though many of the movements an animal performs are stereotyped, not all of them need to be. An important aspect of continuous representations is that they allow for the ability to have portions of time where the animal is performing non-stereotyped dynamics (i.e. they do not remain stationary on the map in Figure 4J). Results from fruit flies show that the animals perform non-stereotyped behaviors approximately half of the time [53, 59], imply- ing that one must be careful when interpreting a repre- sentation that places all time points into a cluster. On the other hand, though, if the data indeed has clusters, one should perform clustering in the high- dimensional space that retains all of the information in the data and where partitioning algorithms are more likely to succeed and one does not have to worry about the specific form of the length-scale distortions that any nonlinear embedding necessarily creates [58]. However, while formalisms such as AR-HMM allow for the build- ing of a type of dynamical model, they also rely on underlying assumptions about a single time scale that could over- or under- partition the data. Accordingly, re- searchers need to think carefully about the consequences of these choices of representation and tailor their ap- proach to the questions at hand. FUTURE CHALLENGES Many of the next steps in building representations for measuring behavior involve building representations that link postural dynamics to dynamics of other variables, Joint representation of space and posture An interesting observation about almost all of the rep- resentations in the previous section is that the typical quantities measured in coarse behavioral assays, namely spatial position, orientation, and their derivatives, are the first aspects to be eliminated. This is performed to ensure that one measures motions in an animal's own frame, but there are numerous scenarios in neuroscience and social behavior where we would like to look at the in- teractions between location, movement, and behavioral patterns, ideally generating a joint representation. A natural question here is, why not simply add the postural dynamics as an extra time series to be thrown- into one's favorite behavioral mapper or classifier? The difficulty here is that the variables describing dynami- cal representation – derivatives or spectral transforms of joint angles or postural modes – all have the same units, and these units differ from those of the spatial variables. Thus, a unit conversion must occur, requiring at least one arbitrarily-chosen parameter. Current solutions have been to measure behavior con- ditioned on position or position conditioned on behavior [56, 59, 81] or to measure a response field averaged across individuals [82], but this does not provide a true joint representation. As an example, if one animal performs the exact same motion twice, but in slightly different lo- cations, are those two behaviors closer or further away than the animal performing two slightly different mo- tions but at the exact same position? Finding system- atic and precise quantifications to answer this question (and the answer might change depending on the precise scientific investigation at hand) will be key to building joint positional-postural representations. Collective and social behavior Similar to the difficulty of representing space and pos- ture simultaneously, we face a problem when attempting to describe the collective dynamics of many individuals moving together. This is often achieved through measur- ing an order parameter that is related to the proportion of individual velocities pointed in the same direction [83– 85]. Ideally, though, one would like to capture metrics that describe the collective dynamics of many individu- als in a manner as rich as the previously-described ap- proaches for single animals. Particularly fruitful ideas here borrow techniques from fluid dynamics, including the use of Lagrangian coherent structures [86] and dy- namic mode decomposition [87] to generate continuum- based models of many organisms moving collectively. An additional challenge arising in social behavior is that much of the research described previously focuses on the physical motion of an animal's limbs and body, but in the case of social interactions, capturing other aspects of behavior such as the production of audio and substrate- borne signals will be necessary to fully describe the ani- mals' dynamics. There have been many recent successes relating behavioral dynamics to, for example, audio dy- namics through asking what behavioral features predict the performance of a particular song or song type using methods such as general linear models (GLMs) [88–90], and improvements in automated methods have increased the throughput of audio data analysis [91–94]. Ideally, though, we would be able to create a joint representation of the alternative behavioral modalities and the postural movements occurring at the same time that more fully links the dynamics of these processes. Linking neurons to behavior As our ability to record neurons in freely-behaving animals increases, the need to represent neural activ- ity jointly with behavior is becoming increasingly ap- parent. As with multi-modal dynamics, most current approaches to neuro-behavioral analysis [57, 68, 95–100] take a correlative or decoding approach: given one knows something about neural dynamics, what can one predict about behavior, or vice versa? This could take the form of "given a neural stimulation what did the animal do?" or "is it possible to predict an animal's behavior from neural dynamics?" While these are necessary first steps toward building our understanding of how neural circuits drive behavior, to more fully comprehend the interplay between these circuits and how behavior feeds back onto neural responses, we need to devise methods to simulta- neously analyze the combined dynamics of posture and neural activity. One potential avenue for achieving this aim is to com- bine experimentally-tested computational models of neu- ral dynamics with high-resolution behavioral measure- ments and perturbations. Ideas toward this end have been put forward in the nascent field of computational psychiatry, where neural models, ranging in scale from small collections of neurons to individual brain nuclei to whole brain dynamics [3, 16, 101, 102] are manipulated or systematically controlled to see how system-wide out- puts are affected. Although in these studies, outputs are usually measured in terms of neural activity alone, a joint representation of behavioral outputs in model or- ganisms or human patients and model-specific control dynamics of the neural circuit present an intriguing path forward. This would also allow ideas from control the- ory to inform the discussion [43], building a framework where feedbacks between neural activity and behavior could be more thoroughly linked. TOWARD THEORIES OF BEHAVIOR The previous point about the use of theory in gen- erating behavioral representations brings us back to the 8 beginning of our discussion. Our fundamental challenges still remain at the conceptual rather than the technical level. Despite the significant advances in measuring be- havior over the past few years, the ultimate goal of these approaches – understanding how and why animals con- trol and generate particular sequences of physical move- ments – requires developing theories and models to serve as connective tissue, providing context and justification for the measurements we make and allowing us to make predictions that suggest future experiments. But what should these theories look like? Does this mean we should turn behavior into particle physics? What is the atom or proton or quark of behavior? Does it even make sense to discuss behavior as if there is a set of underlying first principles from which all actions are derived? Like most questions in biology, we can begin to make progress by looking to evolution. Specifically, we cannot forget that almost every behavior has a goal: to increase an animal's probability of passing its genes to subsequent generations. Thus, all movements are placed in the context of how they aid in the performance of one or many tasks. This viewpoint, shared by many of us who refer to our- selves as "computational ethologists" [1] (whether this is different than "ethologists with fancy computers" is a discussion for a different article), makes an argument to engage in a parallel endeavor to the mapping and ma- nipulation of neural circuits. We should search for what Richard Dawkins referred to as "software explanations of behavior" [103]. The most famous example of this type of analysis is Tinbergen's hypothesis that animals' behavioral drives can be explained via a hierarchically- organized set of competing impulses, based on both ob- servations and ideas about optimality and evolvability [104]. This idea, independently developed by Herbert Simon in the context of engineered systems [105, 106], provides testable consequences that have lead to further investigations and theories across a wide variety of sys- tems [74, 107–109]. Similarly, ideas about optimizing feedback control and energy-efficiency have shaped bi- olocomotion studies [110], and concepts from reinforce- ment learning have served as a starting point toward in- vestigations into the neural implementation of learning [13, 111]. In each of these examples, observations about behav- ior have been used to make inferences about the brain's functioning that do not explicitly rely on detailed models or knowledge of brain dynamics or morphology, poten- tially providing general principles that apply across sys- tems. When deciding what type of behavior to measure and how to measure it, we either intentionally or unin- tentionally rely on theories such as these when we choose a behavioral context, select length and time scales, or decide how to analyze the data. Only through consciously generating and interacting with broad theoretical concepts can we create a fuller understanding of how neural systems function to pro- duce movement and behavior. For example, the idea of using stereotyped movements as a scale for behav- ioral measurements builds upon observations about low- dimensionality in movements and the commonality of neural circuitry such as central pattern generators de- voted to the performance of periodic activities. Taking these assumptions directly into account has allowed the methods discussed in this review to be developed, and the identification of further concepts will be essential to their expansion, refinement, and application. At its core, "What type of behavior do we want to measure?" is a question that relies on theoretical insight for its answer, and future efforts toward quantitatively linking behav- ior to its physiological underpinnings will greatly benefit from approaching experimental design and analysis ac- cordingly. 9 COMPETING INTERESTS The author declares no competing interests. ACKNOWLEDGEMENTS The author would like to thank Avani Gadani, Alex Gomez-Marin, Itai Pinkoviezki, Jennifer Rieser, Carlos Rodriguez, and Kun Tian for comments and suggestions on the manuscript. This work was partially supported by NIMH 1R01MH115831-01. [1] Anderson DJ, Perona P. Toward a Science of Compu- cations. Nature Neuroscience. 2016;19(3):404–413. tational Ethology. Neuron. 2014;84(1):18–31. [2] Krakauer JW, Ghazanfar AA, Gomez-Marin A, Maciver MA, Poeppel D. Neuroscience Needs Be- havior: Correcting a Reductionist Bias. Neuron. 2017;93(3):480–490. [3] Wang XJ, Krystal JH. Computational psychiatry. Neu- ron. 2014;84(3):638–654. [4] Lebedev MA, Nicolelis MAL. Brain-Machine Interfaces: From Basic Science to Neuroprostheses and Neuroreha- bilitation. Physiological reviews. 2017 Apr;97(2):767– 837. [5] Anderson PW. More Is Different. Science. 1972;177(4047):393–396. [6] Goldenfeld N, Kadanoff LP. Simple Lessons from Com- plexity. Science. 1999;284(5411):87–89. [7] Dell AI, Bender JA, Branson K, Couzin ID, De Polavieja GG, Noldus LPJJ, et al. Automated image-based tracking and its application in ecology. Trends in Ecology & Evolution. 2014;29(7):417–428. [8] Egnor SER, Branson K. Computational Analysis of Behavior. Ann Rev Neuro. 2016;39:217–236. [9] Robie AA, Seagraves KM, Egnor SER, Branson K. Ma- chine vision methods for analyzing social interactions. J Exp Bio. 2017;220(1):25–34. [10] Calhoun AJ, Murthy M. Quantifying behavior to solve sensorimotor transformations: advances from worms and flies. Current Opinion in Neurobiology. 2017 Aug;46:90–98. [11] Stephens GJ, Osborne LC, Bialek W. Searching for simplicity in the analysis of neurons and behavior. Proc Nat Acad Sci. 2011;108(Supp 3):15565–15571. [12] Gomez-Marin A, Paton JJ, Kampff AR, Costa RM, Mainen ZF. Big behavioral data: psychology, ethology and the foundations of neuroscience. Nature Neuro- science. 2014;17(11):1455–1462. [13] Montague PR, Dolan RJ, Friston KJ, Dayan P. Com- putational psychiatry. Trends in Cognitive Sciences. 2012;16(1):72–80. [14] Anderson DJ, Adolphs R. A Framework for Studying Emotions across Species. Cell. 2014;157(1):187–200. [15] Anderson DJ. Circuit modules linking internal states and social behaviour in flies and mice. Nature Rev Neuro. 2016;17(11):692–704. [16] Huys QJM, Maia TV, Frank MJ. Computational psy- chiatry as a bridge from neuroscience to clinical appli- [17] Stephan KE, Mathys C. Computational approaches Current Opinion in Neurobiology. to psychiatry. 2014;25:85–92. [18] Leshner A, Pfaff DW. Quantification of behavior. Proc Nat Acad Sci. 2011;108(Supp 3):15537–15541. [19] Fonio E, Golani I, Benjamini Y. Measuring behavior of animal models: faults and remedies. Nature Methods. 2012;9(12):1167–1170. [20] Altmann J. Observational study of behavior: sampling methods. Behaviour. 1974;49(3):227–267. [21] Martin P, Bateson P. Measuring behaviour: an intro- ductory guide. Cambridge, UK: Cambridge University Press; 2007. [22] Gao P, Ganguli S. On simplicity and complexity in the brave new world of large-scale neuroscience. Current Opinion in Neurobiology. 2015;32:148–155. [23] Panda S, Hogenesch JB, Kay SA. Circadian rhythms from flies to human. Nature. 2002;417(6886):329–335. [24] Yemini E, Jucikas T, Grundy LJ, Brown AEX, Schafer WR. A database of Caenorhabditis elegans behavioral phenotypes. Nature methods. 2013;10(9):877–879. [25] Churgin MA, Jung SK, Yu CC, Chen X, Raizen DM, Fang-Yen C, et al. imaging of Caenorhabditis elegans in a microfabricated device re- veals variation in behavioral decline during aging. eLife. 2017;6:e26652. Longitudinal [26] Ayroles JF, Buchanan SM, O'Leary C, Skutt-Kakaria K, Grenier JK, Clark AG, et al. Behavioral idiosyncrasy reveals genetic control of phenotypic variability. Proc Nat Acad Sci. 2015;112(21):6706–6711. [27] Branson K, Robie AA, Bender JA, Perona P, Dickin- son MH. High-throughput ethomics in large groups of Drosophila. Nature Methods. 2009;6(6):451–457. [28] P´erez-Escudero A, Vicente-Page J, Hinz RC, Arganda S, De Polavieja GG. idTracker: tracking individuals in a group by automatic identification of unmarked animals. Nature Methods. 2014;11(7):743–748. [29] Naumann EA, Fitzgerald JE, Dunn TW, Rihel J, Som- polinsky H, Engert F. From Whole-Brain Data to Func- tional Circuit Models: The Zebrafish Optomotor Re- sponse. Cell. 2016;167(4):947–960.e20. [30] Orger MB, De Polavieja GG. Zebrafish Behavior: Op- portunities and Challenges. Ann Rev Neuro. 2017;. [31] Dankert H, Wang L, Hoopfer ED, Anderson DJ, Perona P. Automated monitoring and analysis of social behav- ior in Drosophila. Nature Methods. 2009;6(4):297–303. [32] Kabra M, Robie AA, Rivera-Alba M, Branson S, Bran- son K. JAABA: interactive machine learning for auto- matic annotation of animal behavior. Nature Methods. 2013;10(1):64–67. [33] de Chaumont F, Coura RDS, Serreau P, Cressant A, Chabout J, Granon S, et al. Computerized video anal- ysis of social interactions in mice. Nature Methods. 2012;9(4):410–417. [34] Kain J, Stokes C, Gaudry Q, Song X, Foley J, Wil- son R, et al. Leg-tracking and automated behavioural classification in Drosophila. Nature Communications. 2013;4:1910. [35] Hong W, Kennedy A, Burgos-Artizzu XP, Zelikowsky M, Navonne SG, Perona P, et al. Automated measure- ment of mouse social behaviors using depth sensing, video tracking, and machine learning. Proc Nat Acad Sci. 2015;112(38):E5351–E5360. [36] Dickinson MH, Farley CT, Full RJ, Koehl MAR, Kram R, Lehman S. How Animals Move: An Integrative View. Science. 2000;288(5):100–106. [37] Childress S, Hosoi A, Schultz WW, Wang ZJ, editors. Natural locomotion in fluids and on surfaces: swim- ming, flying, and sliding. New York, NY: Springer; 2012. [38] Holmes P, Full RJ, Koditschek D, Guckenheimer J. The Dynamics of Legged Locomotion: Models, Analyses, and Challenges. SIAM Review. 2006;48:207–304. [39] Miller LA, Goldman DI, Hedrick TL, Tytell ED, Wang ZJ, Yen J, et al. Using computational and mechanical models to study animal locomotion. Int Comp Bio. 2012;52(5):553–575. [40] McInroe B, Astley HC, Gong C, Kawano SM, Schiebel PE, Rieser JM, et al. Tail use improves soft substrate performance in models of early vertebrate land locomo- tors. Science. 2016;353(6295):154–158. [41] Aguilar J, Zhang T, Qian F, Kingsbury M, McInroe B, Mazouchova N, et al. A review on locomotion robo- physics: the study of movement at the intersection of robotics, soft matter and dynamical systems. Reports on Progress in Physics. 2016;79(11):110001. [42] Dickinson MH, Lehmann FO, Sane SP. Wing rotation and the aerodynamic basis of insect flight. Science. 1999;284(5422):1954–1960. [43] Cowan NJ, Ankarali MM, Dyhr JP, Madhav MS, Roth E, Sefati S, et al. Feedback control as a framework for understanding tradeoffs in biology. Integrative And Comparative Biology. 2014;54(2):223–237. [44] Roth E, Sponberg S, Cowan NJ. A comparative ap- proach to closed-loop computation. Current Opinion in Neurobiology. 2014;25:54–62. [45] Alexander RM. Optima for Animals. Princeton, NJ: Princeton University Press; 1996. [46] Full RJ, Koditschek DE. Templates and anchors: Neu- romechanical hypotheses of legged locomotion on land. J Exp Bio. 1999;202(23):3325–3332. [47] Monty Python's Flying Circus. "The Ministry of Silly Walks". BBC. 1970;2(1). [48] Ristroph L, Bergou AJ, Ristroph G, Coumes K, Berman GJ, Guckenheimer J, et al. Discovering the flight autostabilizer of fruit flies by inducing aerial stumbles. Proc Nat Acad Sci. 2010;107(11):4820–4824. [49] Jindrich DL, Full RJ. Dynamic stabilization of rapid hexapedal locomotion. J Exp Bio. 2002;205(18):2803– 2823. 10 [50] Revzen S, Guckenheimer JM. Finding the dimension of slow dynamics in a rhythmic system. J Royal Soc Interface. 2012;9(70):957–971. [51] Osborne LC, Lisberger SG, Bialek W. A sensory source for motor variation. Nature. 2005;437(7057):412–416. [52] Stephens GJ, de Mesquita MB, Ryu WS, Bialek W. Emergence of long timescales and stereotyped behav- iors in Caenorhabditis elegans. Proc Nat Acad Sci. 2011;108(18):7286–7289. [53] Berman GJ, Choi DM, Bialek W, Shaevitz JW. Map- ping the stereotyped behaviour of freely moving fruit flies. J Royal Soc Interface. 2014;11(99):20140672. [54] Stephens GJ, Johnson-Kerner B, Bialek W, Ryu WS. Dimensionality and dynamics in the behavior of C. el- egans. PLoS Comp Bio. 2008;4(4):e1000028. [55] Brown AEX, Yemini EI, Grundy LJ, Jucikas T, Schafer WR. A dictionary of behavioral motifs reveals clusters of genes affecting Caenorhabditis elegans locomotion. Proc Nat Acad Sci. 2013;110(2):791–796. [56] Wiltschko AB, Johnson MJ, Iurilli G, Peterson RE, Ka- ton JM, Pashkovski SL, et al. Mapping Sub-Second Structure in Mouse Behavior. Neuron. 2015;88(6):1121– 1135. [57] Kato S, Kaplan HS, Schrodel T, Skora S, Lindsay TH, Yemini E, et al. Global Brain Dynamics Embed the Motor Command Sequence of Caenorhabditis elegans. Cell. 2015;p. 1–15. [58] Todd JG, Kain JS, de Bivort BL. Systematic explo- ration of unsupervised methods for mapping behavior. Physical Biology. 2017;14(1):015002. [59] Klibaite U, Berman GJ, Cande J, Stern DL, Shae- vitz JW. An unsupervised method for quantifying the behavior of paired animals. Physical Biology. 2017;14(1):015006. [60] Nagy S, Goessling M, Amit Y, Biron D. A Generative Statistical Algorithm for Automatic Detection of Com- plex Postures. PLoS Comp Bio. 2015;11(10):e1004517. [61] Broekmans OD, Rodgers JB, Ryu WS, Stephens GJ. Resolving coiled shapes reveals new reorientation be- haviors in C. elegans. eLife. 2016;5:e17227. [62] Nguyen JP, Shipley FB, Linder AN, Plummer GS, Liu M, Setru SU, et al. Whole-brain calcium imaging with cellular resolution in freely behaving Caenorhabditis el- egans. Proc Nat Acad Sci. 2016;113(8):E1074–E1081. [63] Mendes CS, Bartos I, Akay T, M´arka S, Mann RS. Quantification of gait parameters in freely walking wild type and sensory deprived Drosophila melanogaster. eLife. 2013;2:e00231. [64] Machado AS, Darmohray DM, Fayad J, Marques HG, Carey MR. A quantitative framework for whole-body coordination reveals specific deficits in freely walking ataxic mice. eLife. 2015;4:e07892. [65] Ristroph L, Berman GJ, Bergou AJ, Wang ZJ, Co- hen I. Automated hull reconstruction motion tracking (HRMT) applied to sideways maneuvers of free-flying insects. J Exp Bio. 2009;212(9):1324–1335. [66] Fontaine EI, Zabala F, Dickinson MH, Burdick JW. Wing and body motion during flight initiation in Drosophila revealed by automated visual tracking. J Exp Bio. 2009;212(9):1307–1323. [67] Uhlmann V, Ramdya P, Delgado-Gonzalo R, Benton R, Unser M. FlyLimbTracker: An active contour based approach for leg segment tracking in unmarked, freely behaving Drosophila. PLoS ONE. 2017;12(4):e0173433. [68] Vogelstein JT, Park Y, Ohyama T, Kerr RA, Truman JW, Priebe CE, et al. Discovery of brainwide neural- behavioral maps via multiscale unsupervised structure learning. Science. 2014;344(6182):386–392. [69] Geng W, Cosman P, Baek JH, Berry CC, Schafer WR. Quantitative classification and natural cluster- ing of Caenorhabditis elegans behavioral phenotypes. Genetics. 2003;165(3):1117–1126. [70] Ghosh R, Mohammadi A, Kruglyak L, Ryu WS. Multi- parameter behavioral profiling reveals distinct thermal response regimes in Caenorhabditis elegans. BMC Bi- ology. 2012;10:85. [71] Jhuang H, Garrote E, Mutch J, Yu X, Khilnani V, Pog- gio T, et al. Automated home-cage behavioural pheno- typing of mice. Nature Communications. 2010;1:68. [72] Golani I, Kafkafi N, Drai D. Phenotyping stereo- typic behaviour: collective variables, range of variation and predictability. Applied Animal Behaviour Science. 1999;65(3):191–220. [73] Gomez-Marin A, Stephens GJ, Brown AEX. Hierarchi- cal compression of Caenorhabditis elegans locomotion reveals phenotypic differences in the organization of be- haviour. J Royal Soc Interface. 2016;13(121):20160466. [74] Berman GJ, Bialek W, Shaevitz JW. Predictability and hierarchy in Drosophila behavior. Proc Nat Acad Sci. 2016;113(42):11943–11948. [75] Bongard J, Lipson H. Automated reverse engineering of nonlinear dynamical systems. Proc Nat Acad Sci. 2007;104(24):9943–9948. [76] Natale JL, Hofmann D, Hern´andez DG, Nemenman I. Reverse-engineering biological networks from large data sets. arXiv. 2017;. [77] Daniels BC, Nemenman I. Automated adaptive infer- ence of phenomenological dynamical models. Nature Communications. 2015;6:8133. [78] Schwarz RF, Branicky R, Grundy LJ, Schafer WR, Brown AEX. Changes in Postural Syntax Char- acterize Sensory Modulation and Natural Varia- tion of C. elegans Locomotion. PLoS Comp Bio. 2015;11(8):e1004322. [79] Sakamoto KQ, Sato K, Ishizuka M, Watanuki Y, Taka- hashi A, Daunt F, et al. Can Ethograms Be Automat- ically Generated Using Body Acceleration Data from Free-Ranging Birds? PLoS ONE. 2009;4(4):e5379. [80] Debnath L, Shah FA. Wavelet transforms and their applications. 2nd ed. New York, NY: Springer; 2015. [81] Benjamini Y, Fonio E, Galili T, Havkin GZ, Golani I. Quantifying the buildup in extent and complex- ity of free exploration in mice. Proc Nat Acad Sci. 2011;108(Supp 3):15580–15587. [82] Katz Y, Tunstrom K, Ioannou CC, Huepe C, Couzin ID. Inferring the structure and dynamics of interactions in schooling fish. Proceedings of the National Academy of Sciences. 2011;108(46):18720–18725. [83] Vicsek T, Czir´ok A, Ben-Jacob E, Cohen I, Shochet O. Novel type of phase transition in a system of self-driven particles. Physical Review Letters. 1995;75(6):1226– 1229. [84] Toner J, Tu Y. Flocks, herds, and schools: A quantitative theory of flocking. Physical Review E. 1998;58(4):4828–4858. [85] Buhl J, Sumpter D, Couzin ID, Hale J, Despland E, Miller E, et al. From disorder to order in marching locusts. Science. 2006;312(5778):1402–1406. 11 [86] Haller G. Lagrangian Coherent Structures. Ann Rev Fluid Mech. 2015;47(1):137–162. [87] Brunton BW, Johnson LA, Ojemann JG, Kutz JN. Extracting spatial-temporal coherent patterns in large- scale neural recordings using dynamic mode decompo- sition. Journal of neuroscience methods. 2016;258:1–15. [88] Coen P, Clemens J, Weinstein AJ, Pacheco DA, Deng Y, Murthy M. Dynamic sensory cues shape song struc- ture in Drosophila. Nature. 2014;507(7491):233–237. [89] LaRue KM, Clemens J, Berman GJ, Murthy M. Acoustic duetting in Drosophila virilis relies on the integration of auditory and tactile signals. eLife. 2015;4:e07277. [90] Neunuebel JP, Taylor AL, Arthur BJ, Egnor SER. Fe- male mice ultrasonically interact with males during courtship displays. eLife. 2015;4:e06203. [91] Arthur BJ, Sunayama-Morita T, Coen P, Murthy M, Stern DL. Multi-channel acoustic recording and auto- mated analysis of Drosophila courtship songs. BMC Biology. 2013;11:11. [92] Tabler JM, Rigney MM, Berman GJ, Gopalakrishnan S, Heude E, Al-Lami HA, et al. Cilia-mediated Hedge- hog signaling controls form and function in the mam- malian larynx. eLife. 2017;6:e19153. [93] Mets DG, Brainard MS. An Automated Approach to the Quantitation of Vocalizations and Vocal Learning in the Songbird. bioRxiv. 2017;. [94] Van Segbroeck M, Knoll AT, Levitt P, Narayanan S. MUPET-Mouse Ultrasonic Profile ExTraction: A Sig- nal Processing Tool for Rapid and Unsupervised Analy- sis of Ultrasonic Vocalizations. Neuron. 2017;94(3):465– 485. [95] Robie AA, Hirokawa J, Edwards AW, Umayam LA, Lee A, Phillips ML, et al. Mapping the Neural Substrates of Behavior. Cell. 2017;170(2):393–406.e28. [96] Clemens J, Girardin CC, Coen P, Guan XJ, Dickson BJ, Murthy M. Connecting Neural Codes with Be- havior in the Auditory System of Drosophila. Neuron. 2015;87(6):1332–1343. [97] Wang NXR, Olson JD, Ojemann JG, Rao RPN, Brun- ton BW. Unsupervised Decoding of Long-Term, Nat- uralistic Human Neural Recordings with Automated Video and Audio Annotations. Frontiers in human neu- roscience. 2016;10(251):78–13. [98] Billings J, Medda A, Shakil S, Shen X, Kashyap A, Chen S, et al. Instantaneous brain dynamics mapped to a continuous state space. Neuroimage. 2017;162:344– 352. [99] Gepner R, Skanata MM, Mihovilovic Skanata M, Bernat NM, Kaplow M, Gershow M. Computations un- derlying Drosophila photo- taxis, odor-taxis, and multi- sensory integration. eLife. 2015;4:e06229. [100] Dunn TW, Mu Y, Narayan S, Randlett O, Naumann EA, Yang CT, et al. Brain-wide mapping of neural activity controlling zebrafish exploratory locomotion. eLife. 2016;5:471. [101] Brodersen KH, Deserno L, Schlagenhauf F, Lin Z, Penny WD, Buhmann JM, et al. Dissecting psychiatric spectrum disorders by generative embedding. NeuroIm- age: Clinical. 2014;4:98–111. [102] Muldoon SF, Pasqualetti F, Gu S, Cieslak M, Grafton ST, Vettel JM, et al. Stimulation-Based Control of Dy- namic Brain Networks. PLoS Computational Biology. 2016;12(9):e1005076. 12 [103] Dawkins R. Hierarchical organisation: A candidate principle for ethology. In: Bateson PPG, Hinde RA, editors. Growing Points in Ethology. Cambridge, UK: Cambridge U Press; 1976. p. 7–54. [104] Tinbergen N. The Study of Instinct. Oxford, UK: Ox- ford University Press; 1951. [105] Simon HA. The Architecture of Complexity. Pro- the American Philosophical Society. ceedings of 1962;106(6):467–482. [106] Simon HA. The organization of complex systems. In: Pattee HH, editor. Hierarchy Theory. New York, NY: Braziller; 1973. p. 3–27. [107] Dawkins R, Dawkins M. Hierarchical Organization and Postural Facilitation - Rules for Grooming in Flies. An- imal Behaviour. 1976;24(4):739–755. [108] Lefebvre L. Grooming in crickets: timing and hierarchi- cal organization. Animal Behavior. 1981;29(4):973–984. [109] Solway A, Diuk C, C´ordova N, Yee D, Barto AG, Niv Y, et al. Optimal behavioral hierarchy. PLoS Compu- tational Biology. 2014;10(8):e1003779. [110] Sponberg S. The Emergent Physics of Animal Locomo- tion. Physics Today. 2017;70(9):34. [111] Niv Y. Reinforcement learning in the brain. Journal of Mathematical Psychology. 2009;53(3):139–154.
1004.4666	1	1004	2010-04-26T22:00:56	Error-Control and Digitalization Concepts for Chemical and Biomolecular Information Processing Systems	[ "physics.bio-ph", "cond-mat.soft", "q-bio.MN" ]	We consider approaches for controlling the buildup of noise by design of gates for chemical and biomolecular computing, in order to realize stable, scalable networks for multi-step information processing. Solvable rate-equation models are introduced and used to illustrate several recently developed concepts and methodologies. We also outline future challenges and possible research directions.	physics.bio-ph	physics	Error-Control and Digitalization Concepts for Chemical and Biomolecular Information Processing Systems Vladimir Privman* Department of Physics, Clarkson University, Potsdam NY 13699, USA _________________________________________________ * Phone: +1-315-268-3891; E-mail: [email protected] ABSTRACT We consider approaches for controlling the buildup of noise by design of gates for chemical and biomolecular computing, in order to realize stable, scalable networks for multi-step information processing. Solvable rate-equation models are introduced and used to illustrate several recently developed concepts and methodologies. We also outline future challenges and possible research directions. KEYWORDS: computing, biocomputing, digital, chemical, biomolecular, sensor, biomedical. – 1 – Introduction 1. There have been significant advances in the development of chemical1-4 and biomolecular5-12 systems which are intended to process information by realizing Boolean gate functions, for example, AND, OR, etc., in (bio)chemical kinetics. The range of the information carrying entities has not been limited to simple molecules, but included1-13 supra-molecular and biomolecular structures (enzymes, DNA, etc.), as well as whole cells. Most expected uses of chemical or biochemical “information processing”, to be termed “computing” for brevity, have not aimed at replacing the conventional computers but rather at offering additional functionalities for multi-signal sensing14,15 and interfacing/actuation15-17 in situations when a direct wiring to computers and power sources is not practical, such as in many biomedical applications. One of the main challenges for chemical and biochemical computing has been design of gates and other processing elements, with capabilities to connect them as network components for fault-tolerant information processing of increasing complexity.18-20 First results in developing networks for (bio)chemical information processing1-4,18-20 have included diverse systems, such as those performing elements of basic arithmetic operations,21-22 multifunctional molecules,23-25 DNA-based gates and circuits,26,27 and enzyme-catalyzed reaction networks of several concatenated gates.19,20,28,29 In this article we survey selected topics in, as well as offer illustrative model examples of theoretical analyses of noise reduction and control for scalability in biochemical computing, recently developed by our group10,12,14,17,19,30 primarily in conjunction with experimental data for enzyme-reaction based logic gates and networks. Theoretical studies, which generally apply to a broad range of chemical and biomolecular information processing systems, presently suggest that typical networks up to 10 gates can operate with the acceptable level of noise,10,12,17 similar to recent findings in networking of neurons.31,32 For larger networks, additional non-Boolean network elements, as well as proper network design to utilized redundancy for digital error correction will be needed for fault-tolerant operation.10,12,17,30 There is plentiful experimental evidence that the level of noise in chemical1-4 and biomolecular5-12,14,17,19-20,31-34 computing systems is quite high as compared to the electronic computer counterparts. This includes both the input/output signals and the “gate machinery” chemical concentrations, which typically vary at least several percent on the scale normalized to – 2 – the digital 0 to 1 range. Avoiding noise amplification by gate and network design is therefore quite important even for rather small networks. While we consider aspects of error control, here we do not address the origin/sources of stochastic noise in (bio)chemical reactions: This would take us into topics in statistical mechanics which are outside the scope of this article. We also do not review the experimental findings, which are illustrated in other articles in this Special Issue. Instead, here we devise solvable chemical rate equation models and use the resulting expressions to illustrate recently developed concepts in (bio)chemical computing gate design for noise control and suppression. Let us further discuss the plethora of chemical and biomolecular information processing systems. In the remainder of this section and in the next section, most literature citations aim at providing examples and are not exhaustive. We first reference an extensive body of ongoing experimental work on chemical processes reformulated35,36 in the language of computing operations, involving changes1-4,36-48 in various structural, chemical, or physical properties upon application of physical,49-76 chemical,77-84 or more than one type85-87 of input signals. The final spectroscopically88-93 output signals have typically been read out or electrically/electrochemically.94-96 Chemical computing can be done in a bulk system, specifically, in solution,97-98 or at surfaces/interfaces,14-17,99-102 such as electrodes or Si-chips. Supra-molecular ensembles can also operate as switchable “molecular machines” performing logic operations.103 Much effort has been invested in realizing chemical-computing equivalents of standard Boolean gate functions, such as AND,104,105 OR,106 XOR,103,107 NOR,108-111 NAND,112,113 INHIB,114-117 XNOR.118-119 Issues of reversibility,120,121 reconfiguring122-125 and resetting126,127 logic gates have been explored. Somewhat more complicated systems128-134 have carried out digital logic functions of several-gate rudimentary device components (e.g., keypad lock) and memory units.135-145 Chemical-computing systems can encode computational steps at the single- molecule146 “nano-scale” level,147 as well as perform parallel computations by numerous molecules.148 In summary, chemical computing shows great promise.149-151 However, as most unconventional computing paradigms,152 it is not considered a foreseeable-future viable alternative to the speed and versatility of Si-chip computers. Rather, it offers new functionalities and novel approaches to applications, such as microrobotics, multi-input (bio)sensors/actuators, – 3 – and bioimplantable devices. As with other unconventional computing systems, the main challenge for chemical computing has been networking of basic gates for achieving scalable, fault-tolerant information processing similar to “ordinary” electronic computers.153 Advances have recently been reported with small networks performing basic operations,1-4,21,22 for example, adder/subtractor and their sub-units.154-159 Multi-signal response to chemical or physical inputs has been explored,23-25 and attempts at scaling up the complexity of chemical computing along the lines of biological principles have been made.74 Most variants of biomolecular or biochemical information processing, to be termed “biocomputing” for brevity, can be considered a branch of chemical computing. However, it has developed into an independent, active field of research. The reasons have been several-fold. Indeed, biomolecules offer natural specificity when used in complex “chemical soup” environments, as well as biocompatibility, the latter important for biomedical and biotechnology applications. Biomolecules are also likely more suitable for possible future developments of scalability paradigms borrowing ideas from Nature. Furthermore, the biocatalytic nature of many biomolecular processes, offers certain advantages for analog noise control within the Boolean- gate circuit design paradigm.17 Biomolecules such as proteins/enzymes5-12,14-17,19-20,161-165 and DNA/RNA/DNAzymes166-182 have been extensively used for biocomputing, including systems realizing small networks and computational units, and those motivated183 by applications. This article is organized as follows. In Section 2, we introduce general concepts for considering (bio)chemical gate functions. Gate design for decreasing the degree of noise amplification is addressed in Section 3. Section 4 presents optimization of AND gates, while Section 5 describes gate design as part of a small network. Section 6 offers a summary and also discusses future challenges. 2. Analog/Digital Paradigm Processing (Bio)chemical for Information In order to realize networks processing large quantities of information at high levels of complexity, the approach envisioned in the chemical and biochemical computing literature has – 4 – usually been that gates will be connected similar to Si-chip electronic devices, paralleling the “conventional” design184,185 of fault-tolerant systems that can avoid buildup of noise without the prohibitive use of resources. Another approach, particularly with biomolecules involved, could be the use of design concepts borrowed from processes in living organisms, which are being actively studied in Systems Biology.186 Ultimately, hybrid solutions can be expected, with bio- inspired elements supplementing the conventional design. One can also utilize massive parallelism,182 such as in variants of DNA computing.187 There are several good reasons for the enzyme-based biocomputing gate and network realizations and analyses reported by our group,10,12,14-17,19,20,30 and for much of the rest of the biomolecular computing literature, to follow the conventional information processing paradigm of modern electronics: digital approach based on analog gates and other elements operating in a network.184,185 Indeed, biomolecular computing is presently far from the complexity and richness of coupled biochemical reaction sets needed for mimicking processes in living organisms. Furthermore, most applications of the near-future moderate-complexity biocomputing systems are expected to be for novel sensor development,15 involving processing of several input signals and or “Sense/Respond” to corresponding outputs, digital yielding Yes/No “Sense/Diagnose/Treat” actions. Thus, either in the biochemical stages or during signal transduction to electrodes/electronic computers for the “action” step, the Yes/No digitalization will be imposed, for example, by filtering, as addressed later. Most importantly, the use of the digital information processing paradigm offers a well established approach for control of the level of noise buildup in networks of (bio)chemical information processing reactions. Chemical and biochemical systems are much more prone to noise than electronic computers. The inputs reactant concentrations, and the “gate machinery” chemical concentrations, such as those of catalysts, are expected to fluctuate within at least a couple of percent of the range of values between the “digital” 0 and 1. As a result, consideration of control of noise build-up is needed already for small, 2-3 gate networks.10,15,19,20 Digital information processing is actually carried out by network elements which are analog in nature. Figure 1 illustrates the simplest “gate”: the identify function. A possible analog response curve is also shown. The input and output signals are not limited to the range bounded by the selected “digital” 0 and 1 values appropriate for a specific application. They can also be considered for values beyond the “digital” range, if physically allowed (for example, chemical – 5 – concentrations can only be nonnegative), as illustrated by the broken-line sections in the figure. The “digital” 0 does not have to be at the physical zero. Let us consider a simple model for a chemical reaction, described within the rate- equation approximation, of two atoms of the species A, of initial concentration (0)A A= , 0 0 = : combining to yield the product, C, of concentration C(t) at time Here k denotes the rate constant for the reaction, assumed irreversible. The rate equation is easily solved: t ≥ , where initially, 0 k A A C + → . (1) C (0) dA dt = − 2 kA 2 , C t ( ) = A 0 A t ( ) − 2 = 2 kA t 0 kA t 1 2 + 0 . (2) (3) We will further assume that the information-processing application involves a certain value, of the input which is regarded as the digital 1, and, for simplicity, we also take the physical maxA zero as the digital 0 input. We will also assume that the product of the reaction provides the output signal at the “gate time” time gt . It transpires that the digital value for the output is then set by the gate/application itself and cannot be conveniently selected. Digital 0 and 1 will be at, 0C = and respectively, C max = t 2 kA max kA 1 2 + max g . t g (4) Next, we consider logic-range variables, in terms of which the input, )gC t ( , are normalized to the “digital” range of values: output, (0)A A= 0 , and the – 6 – , x A A / = max 0 z C t C ) /g ( = max . (5) (6) From these relations, we can get the gate-response function shape, see Figure 2, z x ( ) = 2 p x (1 2 ) + px 1 2 + , which depends on the combination of parameters: p = kA t max g . (7) (8) , is generally set by the It is interesting to note that, while the digital-1 of the input, maxA application, we can control the reaction rate constant, k, by varying the physical and chemical conditions of the system to the extent allowed by the application. We can also adjust the reaction duration, gt . Thus, there is a certain degree of control of the “response function shape” which could be used for gate design and optimization. To discuss this further, we have to address the issue of control of noise, which is the topic of the following sections. Here we point out that the considered chemical reaction generally can only yield concave shapes of the type shown in Figure 2. However, as seen in the figure, even with this limitation the shape of the gate response does not vary significantly. Large variations in the parameter values are needed to achieve qualitatively different response. This difficulty has been noted and discussed earlier10,12,19 and is shared by most biocatalytic information processing systems which realize specific gates studied for novel sensor applications.15 Finally, we note that the ranges of both variables in Figure 2 need not be limited to [0,1]; they can be considered for x and z larger than 1 as well. – 7 – 3. Noise Amplification and Filtering In order to discuss noise amplification and filtering, we will, in this section, continue to consider the simplest “identity” gate as a reference. Two-input/one-output gates, such as AND, will be addressed later. However, let us introduce a different reaction which offers a response more realistic of typical chemical kinetics. We consider the process with the rate constant K and initial conditions )gC t ( 0A as the input set by the application (the environment in However, here we prefer to regard A< B which the gate is used), whereas 0 will be assumed small (so that it will limit rather than ) 0 drive the output) and regarded as a controllable-supply “gate machinery” chemical. The rate equation is then A= B= 0 0 . Obviously, this can also be perceived as a two-input AND gate. signal again measured as 0 = , and the output K A B C + → , , C (0) , (0)B (0)A (9) ( yielding dA dt = − KAB = − KA A A ( − 0 + B 0 ) , (10) C t ( ) = − ( e A B [1 − 0 0 − B e A − 0 0 ( A B Kt ) − 0 0 A B Kt ) − 0 0 ] . (11) Equations (5)-(6) are then used to rescale the input and output in terms of the “logic” ranges, with the result z x ( ) = x (1 − e (1 − ax b − + e a b − + )( a b − + a be )( − ax b − + ax be − , ) ) (12) – 8 – which depends on the two combinations of parameters: a KA t = max g , b KB t = 0 g . (13) As before, these parameters can be controlled by changing the physical and chemical conditions (vary K), the “gate machinery” chemical supply, 0B , and the reaction time, gt , as illustrated in Figure 3. An important observation, which can be proven by tedious algebraic considerations not reproduced here, is that the function in Equation (12) always gives the monotonically increasing, convex response curve; see Figure 3. Typically, in catalytic biochemical reactions such convex response curves (and surfaces, for more than one input) are also found: The output (the product of the reaction) is controlled by and is typically proportional to the input-signal chemical concentration(s) for small inputs. For large inputs, the output is usually limited, for example, by the reactivity of the available biocatalyst, and the response signal reaches saturation. We point out that the “digital” 0 and 1 signal values need not be sharply defined. In some applications, input or output signals in certain ranges of values may constitute 0 or 1. For example, a certain range of “normal” physiological concentrations can be 0, whereas another, “abnormal” range, can be 1. These ranges need not even be bounded, for instance, if the “abnormal” concentrations correspond to those above a certain “flag” value. There are several sources of error in gate functioning. The most obvious source of noise is that in the inputs, which is natural and actually quite large in chemical and biochemical environments in which applications of (bio)chemical information processing have been envisioned. The gate function will transfer this noise (the distribution of the input values) into noise in output signal(s). The gate function itself can also be noisy and, furthermore, perhaps somewhat displaced away from the desired digital values/ranges. In our earlier examples, noise and fluctuations in concentrations and physical parameters of the system can lead to a distribution of the values of z x , for each x, rather than a sharply defined function such as in Equations (7), (12). ( ) Furthermore, the mean values of this distribution need not pass precisely through the expected – 9 – might logic values connected by smooth response surfaces. For example, the mean value of (1)z somewhat deviate from 1 for our “identity gate.” We will denote as “analog” the noise due to the spread of the output signal about the desired “digital” values (or possibly ranges of values). In order counteract buildup of noise, we have to pass our signals through “filters” of the type shown in Figure 4. In fact, ideally we would like to have the sigmoid property — small slopes/gradients at and around the digital points — in all or most of our gates. Filters can also be used as separate elements/steps. There is evidence that such solutions for suppressing analog noise buildup are utilized by Nature.188,189 However, filtering can push those values which are far away from the correct digital result (the values which are in the tail of the distribution about the desired correct outcome) to the wrong answer. Thus, the process of digitalization itself introduces also the “digital” type of noise. Such errors are not very probable and only become important to actively correct for larger networks. Standard techniques based on redundancy are available190,191 for digital error correction. For biocatalyst (typically, enzyme) based computing gates studied by our group, for the presently realized network sizes and levels of noise, it is the analog error correction that is important and has recently received significant attention.10,12,14,15,17,19,20,30 It has been estimated10,12 that up to order 10 such gates can be connected in a network before digital error correction is warranted. Experimental realizations of the sigmoid behavior (Figure 4) are an ongoing effort and will hopefully be soon accomplished, along the lines of theoretical suggestions10,189 based on the idea that an additional reactant which depletes the product, but can only consume (react with) a small quantity of it, will suppressed response at small inputs without voiding the saturation property at large inputs, thus yielding a sigmoid response. No simple solvable rate-equation models of the type analyzed earlier, can be offered here for sigmoid behavior: Our group’s numerical simulations (work in progress) of rate equations for typical biocatalytic reactions support the proposed mechanism. As pointed out earlier, the final output signal of several (bio)chemical information processing steps in near-term future sensor applications of the “decision-making” type,14,15 will likely be coupled to conventional electronics. There are active, well developed research areas (not reviewed here) of interfacing enzyme-based logic with “smart” signal-responsive16,192-205 – 10 – materials and with electrodes and bioelectronic devices.16,206-212 This interfacing, involving the transduction of (bio)chemical signals to electronic ones, can also involve a well-defined filtering “sigmoid” property, as has been recently experimentally demonstrated.16 4. The AND Gate Logic AND gates are the most studied in chemical and biochemical computing, and the only ones explored in detail for the shape of their response surface, the latter in the literature on enzyme-catalyzed biochemical reactions, to be referenced later in this section. Indeed, the truth table for the AND gate is that the output 1 is obtained only when both inputs are 1, which is the most natural outcome when measured as a product of a two-input chemical reaction. Presently, let us introduce a simple model for the AND gate chemical-computing function. For this, we now regard the reaction in Equation (9): A B C + → , as a two-input, one-output process. We introduce the second variable reduced to the “logic” range [0,1], to supplement the definitions in Equations (5)-(6), where maxB is the reference value of the logic-1 for the second input, set by the application. We now regard the quantity z defined in z x y , describing the AND-gate response surface Equation (6), as a two-variable function, ( , ) shape. The solution of the rate equations, given by Equation (11), is now recast in terms of the reduced variables to yield y B B / = max 0 , (14) z x y ( , ) = x β α α e e e xy ( )( α β − x α α β xe e e )( ( α − − y β e − y β ye β ) ) , (15) where we defined the parameters – 11 – α= KA t max g , β= KB t max g . (16) This is similar to the set of (dimensionless) parameters in Equation (13). However, here we have less control over their values, because their ratio is fixed by the application (the environment) of the gate which in many cases dictates the values of maxA and maxB . Thus, technically only the gKt product can be adjusted. The shapes of the resulting response surfaces are illustrated in Figures 5 and 6. As mentioned earlier, the noise in the input signals is passed on to the output, with the added noise effects due to the gate functioning itself, such as the imprecise (on average) and fluctuating z x y . In addition to designing gates with as precise and sharply defined z x y as values of ( , ) ( , ) possible, we can also minimize the propagation of analog noise, and hopefully avoid noise gKt ) values that yield gates which suppress spread amplification, by finding parameter (such as in the input signals by having small slopes near the logic points. Let us consider this concept in greater detail. The absolute value of the gradient vector, z x y ( , ) , calculated at the logic points, (cid:71) ∇ ) measures the noise spread amplification or suppression. However, this is only relevant provided z x y is smooth (does not vary much) in regions about the logic points which the gate function ( , ) are approximately the size of the spread of the noise distributions. Our model example offers z x y shapes; see Figure 5. We can try to identify parameter ( , ) illustration of relatively smooth (cid:71) (cid:71) ∇ ∇ z x y ( , z x y ( , values for which the largest of the four gradients, (cid:71) (cid:71) ∇ ∇ is as small as possible (note that z x y ( , z x y ( , is always zero for this ) (cid:71) z∇ 1, = , , , 0, 1 1, 1 00 = = = = particular model). For this calculation, we will assume that both α and β can be adjusted independently, which gives additional freedom (we commented earlier that in applications the ratio of these two parameters might be fixed and not controllable). What we are after is an estimate of how little can noise amplification be for AND gates modeled by this reaction scheme, A B C + → . By numerical calculation, we find that for moderate values of α and β, the (cid:71) (cid:71) (cid:71) = ∇ = ∇ ∇ minimum is obtained for , and is given by 0.4966 1.1796 α β= . ≈ ≈ z z z = 0 = 0 ) x = 0, y ) x x y x y y 10 01 11 – 12 – This result is interesting in several aspects. First, gate functions of this type amplify analog noise even under optimal conditions. The noise amplification in the best case scenario is about 18%. Studies10,12,19 of enzyme-based AND gates, which have utilized more realistic (and thus more complicated and not exactly solvable) rate-equation models appropriate for biocatalytic reactions, found similar estimates. Experimental data were fitted and results were numerically analyzed by using both the rate equation approach and more phenomenological shape-fitting forms for the gate response function surface, the latter described in the next section. It transpires that smooth, convex gates corresponding to (bio)chemical reactions can have very large noise amplification, typically 300–500%, if the gate is not optimized. Reaching the optimal conditions is not always straightforward primarily because the gate function shape depends only weakly on parameter values. Even under optimal conditions, at least about 20% noise amplification is to be expected. For fast reactions, the maximum of the four gradients can actually be smaller than ~1.18, and in fact can even be somewhat less than 1 (which would suggest noise suppression). However, as seen in Figure 6, under these conditions the gate function surface develops sharp features, and the gradients can no longer be used, because they remain close to the logic-point values only in tiny regions near these points, as compared to the typical noise spread of at least several percent, for (bio)chemical signals. For such gates, and generally when the spread of the noise is larger than the x and y scales over which the gate function varies significantly, one can assume a certain shape of the noise distribution, such as a product of approximately Gaussian (half-Gaussian, if the logic zero is exactly at the physical zero) distributions in x and y, for inputs at each of the logic points. When this distribution is properly integrated with the use of the gate response function,10 one can numerically calculate the output signal distribution for each of the input options, and thus estimate noise propagation.10,12,14 Interestingly, a “ridged” gate response function was encountered12 in a study of an enzymatic system, which has also a smooth-response counterpart when a different chemical is used as one of the inputs.12 While the reaction kinetics was much more complicated than the present model, the finding has been quite general for such gate functions: The optimal conditions are obtained with a symmetrically (diagonally) positioned ridge, as in Figure 6, and noise amplification percentage is then very small (estimated not from gradients, but by integrating over – 13 – distributions). For gates operating in this regime, with the amplification factor only slightly larger than 1, noise amplification is practically avoided (when compared to other possible noises of source, from the gate-function itself). However, they do not have the noise-suppression, “filter” property. Figure 7 offers a schematic of another AND gate-response shape which was recently explored and experimentally realized: sigmoid in only one of the two inputs. Many allosteric enzymes have such a “self-promoter” property with respect to one of their substrates (input chemical species that the enzyme binds as part of its biocatalytic reaction scheme). Details14 for this specific experiment and its modeling are not given here. A key finding14 was that the single- sided sigmoid shape can be tuned by parameter adjustment to also have the noise amplification percentage very small (noise amplification factor only slightly above 1), so that there is practically no noise amplification. However, a desirable two-sided sigmoid response, also illustrated in Figure 7, has not been to our knowledge realized at the level of a single AND gate, in chemical or biomolecular computing literature. Certain biochemical processes in Nature, which are much more complex than our synthetic AND-gate systems, do realize188 a two-sided sigmoid response. 5. Network of AND Gates Optimization of (bio)chemical gates one at a time is not straightforward for several reasons. Indeed, we have seen that in most cases a rather large variation of the controllable parameters is needed: physical and chemical conditions, reactant concentrations and in some cases choice of species, which may not be experimentally feasible. In fact, the actual detailed kinetic modeling of the reactions involved, especially for biomolecular systems, is in itself a challenging and numerically taxing task, not reviewed here.10,12,14,15,17 Furthermore, the kinetics of most biomolecular processes, specifically those used for AND gates, is not only complex but also not well studied. The quality of the experimental data for the gate-response function shape is limited due to the noise in the gate-function itself, short life-time for constant activity of the biocatalytic species, etc. As a result, multi-parameter complex reaction schemes, even if – 14 – proposed, are difficult to substantiate by data fitting in the gate-design context which requires models to work for a large range of parameters. It is therefore useful to explore optimization of the relative gate functioning as part of a network, whereby each gate is modeled within an approximate, phenomenological curve/surface- fitting approach. Such ideas have recently been tested19 for coupled enzymatic reactions which include steps common in sensor development213 for maltose and its sources. A modular network representation of the biocatalytic processes involved is possible in terms of three AND gates; see Figure 8. This convenient representation is actually approximate, because it obscures some of the complexity of the processes involved, which are not reviewed here.19 The approach taken, has been as follows. We first propose an approximate, phenomenological fitting function for the gate response surface in terms of as few parameters as possible, but enough to capture the expected global, qualitative features of the shape. Specifically, for a typical convex “identity” gate, the proposed fitting function is here conveniently written as z x ( ) ≈ sx x 1) . + 1 ( s − (17) This is just a simple, single-parameter, s, rational form that “looks” qualitatively appropriate, provided we assume that x z x = Indeed, the curve is then convex and has slope s at x z x = , and 1 / s at , ( ) 1,1 . , ( ) 0, 0 For a convex, smooth AND gate, we use the two-parameter, say, 1u > , product 1s > and function, 1s > . (18) z x y ( , ) ≈ sx uy ( )( u 1][( + ) − . 1) y + 1] [( s − 1) x (19) – 15 – The gradient values are (cid:71) z∇ = 0 , 00 (cid:71) ∇ z 10 = u , (cid:71) ∇ z 01 = s , and (cid:71) ∇ z = 11 2 − s + u − 2 . The 4 2 u= = s , which is also minimum of the largest of the last three values is obtained for 1.189 ≈ the value of the gradient, consistent with the earlier reported expectation that smooth convex AND gates can typically be optimized at best to yield noise amplification somewhat under 20%. Having introduced our approximate fitting functions, we now vary selective inputs in the network; see Figure 8. In the experiment,19 each of the three inputs 1,2,3 x was separately varied between 0 (corresponding to the logic 0) and the reference value pre-defined as the logic 1, while 3y ) where at their reference logic-1 values. In fact, when the all the other inputs (including parameterization of Equation (19) is applied to all three gates in our network, Figure 8, we get a complicated rational expression for z as a function of all the four inputs ( 1,2,3 and 3y ). Setting x all of them but a single x-input to 1, we get the parameterization for the measurement with that input varied. We only keep the varying arguments for simplicity: z x ( 1 ) = ( s 1 z x ( 2 ) = z x ( 3 ) = ( ( , 1 + s x 1 1 x 1) − 1 s u x 2 1 2 x s u 1) − 2 1 2 s u u x 3 1 2 3 x s u u 1) − 3 3 1 2 + , 1 . + 1 (20) (21) (22) An interesting conclusion is that each data set only depends on a single parameter ( 1s , or one of s u or 3 1 2 s u u ). the products 2 1 Thus, we only get partial information on the gate functioning. However, we can attempt to “tweak” the relative gate activities in the network to improve its stability. We note that if the proposed approximate description is accurate for a given gate, then the parameters s and u for that gate will be functions of adjustable quantities, such as the gate time, input concentrations of some of the chemicals, reaction rates (which can in turn be controlled by the physical and chemical conditions). In addition, s and u can depend on other quantities which are not controllable in a specific application. Without detailed rate-equation kinetic modeling and – 16 – verification of applicability of the phenomenological functional form selected, this parameter dependence is not known. s u , 3 1 2 However, examination of the fitted quantities ( 1s , 2 1 s u u ) still provides useful information on the relative effect that the gates have in their contribution to the gradients at 1 / 2 1 / 4 2 2 logic points, when compared the optimal values ( various to s = 1 , s u = 2 1 , 3 / 4 2 ). The initial sets of data19 were collected with the experimentally convenient but s u u == 3 1 2 otherwise initially randomly selected values for the adjustable “gate machinery” and other parameters. Examination of the results19 has lead to a semi-quantitative conclusion that the deviations form the optimal values could largely by attributed to the gate which is the closest to z x y ): it was too “active” as compared to the other two gates the output in Figure 8 ( = AND 1 1 (means, its biocatalytic reaction was too fast). A new experiment was then carried out19 with the concentration of the enzyme catalyzing this gate’s reactions reduced by an order of magnitude (actually, by a factor of approximately 11); recall that large parameter changes are needed to effect qualitative changes. The new data collected for the modified network yielded 1s , 2 1 s u , s u u values significantly closer to optimal.19 3 1 2 6. Conclusions and Future Challenges We addressed certain aspects of and approaches to gate optimization for control of the analog noise buildup, which is an important consideration in connecting gates in functional networks, though for larger networks digital error correction by redundancy will also have to be implemented, and special network elements will have to be devised, notably filters, but also elements for signal splitting, balancing and gate-to-gate connectivity, memory, and interfacing with external input, output and control mechanisms. Our goal here has been to develop simple rate-equation models which allow exact solvability, and then use them to illustrate and motivate the discussion. Thus, we avoided presenting experimental data and their analysis, which can be found in the cited articles, while – 17 – various chemical and biochemical gate examples are offered in other reviews in this Special Issue. The reader will notice that our presentation has been limited to AND gates and related systems. The reason for this has been that all the recent studies of noise control in (bio)chemical computing have thus far been for AND gates and, furthermore, only those with the logic 0 set at the physical zeros of chemical concentrations. While such a limitation is natural for chemical reactions per se, it is definitely not typical for applications envisaged, especially in multi-input biomedical sensing.15 We expect that, as new experiments on mapping out (bio)chemical gate functions and probing network functioning are reported (some presently ongoing in our group), new features in noise and error control will be explored. Specifically, noise in the gate function itself, including spread of its values and imprecise mean-value — not exactly at the reference 0 or 1, with deviations possibly also somewhat different for various inputs that should ideally yield the same logic output — will have to be considered and corrected, most likely by filtering. Indeed, we expect that while long-term network design and scaling up will be crucial, the grand challenge in (bio)chemical to design versatile and effective is information processing short-term (bio)chemical filter processes that can be concatenated with various types of single logic gates. The author gratefully acknowledges research funding by the NSF (grant CCF-0726698) and by the Semiconductor Research Corporation (award 2008-RJ-1839G). – 18 – REFERENCES 1. A. Credi, Angew. Chem. Int. Ed. 46, 5472-5475 (2007). 2. A. P. de Silva, S. Uchiyama, T. P. Vance and B. Wannalerse, Coord. Chem. Rev. 251, 1623-1632 (2007). 3. A. P. de Silva and S. Uchiyama, Nature Nanotechnology 2, 399-410 (2007). 4. K. Szacilowski, Chem. Rev. 108, 3481-3548 (2008). 5. X. G. Shao, H. Y. Jiang and W. S. Cai, Prog. Chem. 14, 37-46 (2002). 6. A. Saghatelian, N. H. Volcker, K. M. Guckian, V. S. Y. Lin and M. R. Ghadiri, J. Am. Chem. Soc., 125, 346-347 (2003). 7. G. Ashkenasy and M. R. Ghadiri, J. Am. Chem. Soc. 126, 11140-1114 (2004). 8. R. Baron, O. Lioubashevski, E. Katz, T. Niazov and I. Willner, J. Phys. Chem. A 110, 8548-8553 (2006). 9. R. Baron, O. Lioubashevski, E. Katz, T. Niazov and I. Willner, Angew. Chem. Int. Ed. 45, 1572-1576 (2006). 10. V. Privman, G. Strack, D. Solenov, M. Pita and E. Katz, J. Phys. Chem. B 112, 11777- 11784 (2008). 11. G. Strack, M. Pita, M. Ornatska and E. Katz, ChemBioChem 9, 1260-1266 (2008). 12. D. Melnikov, G. Strack, M. Pita, V. Privman and E. Katz, J. Phys. Chem. B 113, 10472- 10479 (2009). 13. A. H. Flood, R. J. A. Ramirez, W. Q. Deng, R. P. Muller, W. A. Goddard and J. F. Stoddart, Austr. J. Chem. 57, 301-322 (2004). 14. V. Privman, V. Pedrosa, D. Melnikov, M. Pita, A. Simonian and E. Katz, Biosens. Bioelectron. 25, 695-701 (2009). 15. E. Katz, V. Privman and J. Wang, in: Proc. Conf. ICQNM 2010, edited by V. Ovchinnikov and V. Privman, Pages 1-9 (IEEE Comp. Soc. Conf. Publ. Serv., Los Alamitos, California, 2010). 16. M. Privman, T. K. Tam, M. Pita and E. Katz, J. Am. Chem. Soc. 131, 1314-1321 (2009). Privman, Chem. Soc. Rev., 17. E. Katz and V. in print, online http://dx.doi.org/10.1039/b806038j (2010). 18. N. Wagner and G. Ashkenasy, Chem. Eur. J. 15, 1765-1775 (2009). at – 19 – 19. V. Privman, M. A. Arugula, J. Halamek, M. Pita and E. Katz, J. Phys. Chem. B 113, 5301- 5310 (2009). 20. M. A. Arugula, J. Halamek, E. Katz, D. Melnikov, M. Pita, V. Privman and G. Strack, in: Proc. Conf. CENICS 2009, edited by K. B. Kent, P. Dini, O. Franza, T. Palacios, C. Reig, J. E. Maranon, A. Rostami, D. Zammit-Mangion, W. C. Hasenplaugh, F. Toran and Y. Zafar, Pages 1-7 (IEEE Comp. Soc. Conf. Publ. Serv., Los Alamitos, California, 2009). 21. G. J. Brown, A. P. de Silva and S. Pagliari, Chem. Commun. 2002, 2461-2463 (2002). 22. U. Pischel, Angew. Chem. Int. Ed. 46, 4026-4040 (2007). 23. F. M. Raymo and S. Giordani, J. Am. Chem. Soc. 123, 4651-4652 (2001). 24. X. Guo, D. Zhang, G. Zhang and D. Zhu, J. Phys. Chem. B 108, 11942-11945 (2004). 25. Y. Liu, W. Jiang, H.-Y. Zhang and C.-J. Li, J. Phys. Chem. B 110, 14231-14235 (2006). 26. M. N. Stojanovic and D. Stefanovic, Nature Biotechnol. 21, 1069-1074 (2003). 27. J. Macdonald, Y. Li, M. Sutovic, H. Lederman, K. Pendri, W. H. Lu, B. L. Andrews, D. Stefanovic and M. N. Stojanovic, Nano Lett. 6, 2598-2603 (2006). 28. T. Niazov, R. Baron, E. Katz, O. Lioubashevski and I. Willner, Proc. Natl. Acad. USA 103, 17160-17163 (2006). 29. G. Strack, M. Ornatska, M. Pita and E. Katz, J. Am. Chem. Soc. 130, 4234-4235 (2008). 30. L. Fedichkin, E. Katz and V. Privman, J. Comput. Theor. Nanosci. 5, 36-43 (2008). 31. O. Feinerman and E. Moses, J. Neurosci. 26, 4526-4534 (2005). 32. O. Feinerman, A. Rotem and E. Moses, Nature Physics 4, 967-973 (2008). 33. Y. Benenson, T. Paz-Elizur, R. Adar, E. Keinan, Z. Livneh and E. Shapiro, Nature 414, 430-434 (2001). 34. P. Fu, Biotechnol. J. 2, 91-101 (2007). 35. P. Dittrich, Lect. Notes Computer Sci. 3566, 19-32 (2005). 36. A. N. Shipway, E. Katz and I. Willner, in: Molecular Machines and Motors, edited by J.-P. Sauvage, Structure and Bonding, Vol. 99, Pages 237-281 (Springer, Berlin, 2001). 37. D. A. Weinberger, T. B. Higgins, C. A. Mirkin, C. L. Stern, L. M. Liable-Sands and A. L. Rheingold, J. Am. Chem. Soc. 123, 2503-2516 (2001). 38. H. R. Tseng, S. A. Vignon, P. C. Celestre, J. Perkins, J. O. Jeppesen, A. Di Fabio, R. Ballardini, M. T. Gandolfi, M. Venturi, V. Balzani and J. F. Stoddart, Chem. Eur. J. 10, 155-172 (2004). – 20 – 39. Y. L. Zhao, W. R. Dichtel, A. Trabolsi, S. Saha, I. Aprahamian and J. F. Stoddart, J. Am. Chem. Soc. 130, 11294-11295 (2008). 40. M. Suresh, A. Ghosh and A. Das, Tetrahedron Lett. 48, 8205-8208 (2007). 41. J. D. Crowley, D. A. Leigh, P. J. Lusby, R. T. McBurney, L. E. Perret-Aebi, C. Petzold, A. M. Z. Slawin and M. D. Symes, J. Am. Chem. Soc. 129, 15085-15090 (2007). 42. S. Iwata and K. Tanaka, J. Chem. Soc., Chem. Commun. 1995, 1491-1492 (1995). 43. S. H. Lee, J. Y. Kim, S. K. Kim, J. H. Leed and J. S. Kim, Tetrahedron 60, 5171-5176 (2004). 44. D. C. Magri, G. J. Brown, G. D. McClean and A. P. de Silva, J. Am. Chem. Soc. 128, 4950- 4951 (2006). 45. K. K. Sadhu, B. Bag and P. K. Bharadwaj, J. Photochem. Photobiol. A 185, 231-238 (2007). 46. E. Katz, A. N. Shipway and I. Willner, in: Photoreactive Organic Thin Films, edited by S. Sekkat and W. Knoll, Chapter II-7, Pages 219-268 (Elsevier, San Diego, 2002). 47. V. Balzani, Photochem. Photobiol. Sci. 2, 459-476 (2003). 48. S. Giordani, M. A. Cejas and F. M. Raymo, Tetrahedron 60, 10973-10981 (2004). 49. M. Lion-Dagan, E. Katz and I. Willner, J. Am. Chem. Soc. 116, 7913-7914 (1994). I. Willner, M. Lion-Dagan, S. Marx-Tibbon and E. Katz, J. Am. Chem. Soc. 117, 6581- 50. 6592 (1995). 51. F. M. Raymo and S. Giordani, Proc. Natl. Acad. USA 99, 4941-4944 (2002). 52. Z. F. Liu, K. Hashimoto and A. Fujishima, Nature 347, 658-660 (1990). 53. A. Doron, M. Portnoy, M. Lion-Dagan, E. Katz and I. Willner, J. Am. Chem. Soc. 118, 8937-8944 (1996). 54. N. G. Liu, D. R. Dunphy, P. Atanassov, S. D. Bunge, Z. Chen, G. P. Lopez, T. J. Boyle and C. J. Brinker, Nano Lett. 4, 551-554 (2004). 55. P. R. Ashton, R. Ballardini, V. Balzani, A. Credi, K. R. Dress, E. Ishow, C. J. Kleverlaan, O. Kocian, J. A. Preece, N. Spencer, J. F. Stoddart, M. Venturi and S. Wenger, Chem. Eur. J. 6, 3558-3574 (2000). 56. E. Katz and A. N. Shipway, in: Bioelectronics: From Theory to Applications, edited by I. Willner and E. Katz, Chapter 11, Pages 309-338 (Wiley-VCH, Weinheim, 2005). 57. S. Bonnet and J. P. Collin, Chem. Soc. Rev. 37, 1207-1217 (2008). – 21 – 58. I. Thanopulos, P. Kral, M. Shapiro and E. Paspalakis, J. Modern Optics 56, 686-703 (2009). 59. E. Katz, L. Sheeney-Haj-Ichia, B. Basnar, I. Felner and I. Willner, Langmuir 20, 9714- 9719 (2004). 60. E. Katz, R. Baron and I. Willner, J. Am. Chem. Soc. 127, 4060-4070 (2005). I. M. Hsing, Y. Xu and W. T. Zhao, Electroanalysis 19, 755-768 (2007). 61. 62. R. Hirsch, E. Katz and I. Willner, J. Am. Chem. Soc. 122, 12053-12054 (2000). 63. E. Katz, L. Sheeney-Haj-Ichia and I. Willner, Chem. Eur. J. 8, 4138-4148 (2002). I. Willner and E. Katz, Angew. Chem. Int. Ed. 42, 4576-4588 (2003). 64. 65. J. Wang and A. N. Kawde, Electrochem. Commun. 4, 349-352 (2002). 66. R. Laocharoensuk, A. Bulbarello, S. Mannino and J. Wang, Chem. Commun. 2007, 3362- 3364 (2007). 67. J. Wang, Electroanalysis 20, 611-615 (2008). 68. J. Lee, D. Lee, E. Oh, J. Kim, Y. P. Kim, S. Jin, H. S. Kim, Y. Hwang, J. H. Kwak, J. G. Park, C. H. Shin, J. Kim and T. Hyeon, Angew. Chem. Int. Ed. 44, 7427-7432 (2005). 69. J. Wang, M. Scampicchio, R. Laocharoensuk, F. Valentini, O. Gonzalez-Garcia and J. Burdick, J. Am. Chem. Soc. 128, 4562-4563 (2006). 70. O. A. Loaiza, R. Laocharoensuk, J. Burdick, M. C. Rodriguez, J. M. Pingarron, M. Pedrero and J. Wang, Angew. Chem. Int. Ed. 46, 1508-1511 (2007). 71. L. Zheng and L. Xiong, Colloids Surf. A 289, 179-184 (2006). 72. M. Riskin, B. Basnar, E. Katz and I. Willner, Chem. Eur. J. 12, 8549-8557 (2006). 73. M. Riskin, B. Basnar, V. I. Chegel, E. Katz, I. Willner, F. Shi and X. Zhang, J. Am. Chem. Soc. 128, 1253-1260 (2006). 74. V. I. Chegel, O. A. Raitman, O. Lioubashevski, Y. Shirshov, E. Katz and I. Willner, Adv. Mater. 14, 1549-1553 (2002). I. Poleschak, J. M. Kern and J. P. Sauvage, Chem. Commun. 2004, 474-476 (2004). 75. 76. X. T. Le, P. Jégou, P. Viel and S. Palacin, Electrochem. Commun. 10, 699-703 (2008). 77. P. R. Ashton, R. Ballardini, V. Balzani, I. Baxter, A. Credi, M. C. T. Fyfe, M. T. Gandolfi, M. Gomez-Lopez, M. V. Martinez-Diaz, A. Piersanti, N. Spencer, J. F. Stoddart, M. Venturi, A. J. P. White and D. J. Williams, J. Am. Chem. Soc. 120, 11932-11942 (1998). – 22 – 78. C. J. Richmond, A. D. C. Parenty, Y. F. Song, G. Cooke and L. Cronin, J. Am. Chem. Soc. 130, 13059-13065 (2008). 79. Y. Shiraishi, Y. Tokitoh, G. Nishimura and T. Hirai, Org. Lett. 7, 2611-2614 (2005). 80. F. Coutrot, C. Romuald and E. Busseron, Org. Lett. 10, 3741-3744 (2008). 81. G. Nishimura, K. Ishizumi, Y. Shiraishi and T. Hirai, J. Phys. Chem. B 110, 21596-21602 (2006). 82. W. D. Zhou, J. B. Li, X. R. He, C. H. Li, J. Lv, Y. L. Li, S. Wang, H. B. Liu and D. B. Zhu, Chem. Eur. J. 14, 754-763 (2008). 83. R. P. Fahlman, M. Hsing, C. S. Sporer-Tuhten and D. Sen, Nano Lett. 3, 1073-1078 (2003). 84. Y. Shiraishi, Y. Tokitoh and T. Hirai, Chem. Commun. 2005, 5316-5318 (2005). 85. R. Baron, A. Onopriyenko, E. Katz, O. Lioubashevski, I. Willner, S. Wang and H. Tian, Chem. Commun. 2006, 2147-2149 (2006). 86. A. Doron, M. Portnoy, M. Lion-Dagan, E. Katz and I. Willner, J. Am. Chem. Soc. 118, 8937-8944 (1996). 87. M. Biancardo, C. Bignozzi, H. Doyle and G. Redmond, Chem. Commun. 2005, 3918-3920 (2005). 88. S. Giordani and F. M. Raymo, Org. Lett. 5, 3559-3562 (2003). 89. B. Raychaudhuri and S. Bhattacharyya, Appl. Phys. B 91, 545-550 (2008). 90. S. D. Straight, J. Andrasson, G. Kodis, S. Bandyopadhyay, R. H. Mitchell, T. A. Moore, A. L. Moore and D. Gust, J. Am. Chem. Soc. 127, 9403-9409 (2005). 91. Z. Wang, G. Zheng and P. Lu, Organic Lett. 7, 3669-3672 (2005). 92. C.-J. Fang, Z. Zhu, W. Sun, C.-H. Xu and C.-H. Yan, New J. Chem. 31, 580-586 (2007). 93. S. D. Straight, P. A. Liddell, Y. Terazono, T. A. Moore, A. L. Moore and D. Gust, Adv. Funct. Mater. 17, 777-785 (2007). 94. F. Li, M. Shi, C. Huang and L. Jin, J. Mater. Chem. 2005, 15, 3015-3020 (2005). 95. G. Wen, J. Yan, Y. Zhou, D. Zhang, L. Mao and D. Zhu, Chem. Commun. 2006, 3016- 3018 (2006). 96. K. Szaciłowski, W. Macyk and G. Stochel, J. Am. Chem. Soc. 128, 4550-4551 (2006). 97. J. H. Qian, Y. F. Xu, X. H. Qian and S. Y. Zhang, ChemPhysChem 9, 1891-1898 (2008). – 23 – 98. G. Fioravanti, N. Haraszkiewicz, E. R. Kay, S. M. Mendoza, C. Bruno, M. Marcaccio, P. G. Wiering, F. Paolucci, P. Rudolf, A. M. Brouwer and D. A. Leigh, J. Am. Chem. Soc. 130, 2593-2601 (2008). 99. S. Nitahara, N. Terasaki, T. Akiyama and S. Yamada, Thin Solid Films 499, 354-358 (2006). 100. A. P. de Silva, Nature 454, 417-418 (2008). 101. P. M. Mendes, Chem. Soc. Rev. 2008, 37, 2512-2529 (2008). 102. T. Gupta and M. E. van der Boom, Angew. Chem. Int. Ed. 47, 5322-5326 (2008). 103. A. Credi, V. Balzani, S. J. Langford and J. F. Stoddart, J. Am. Chem. Soc. 119, 2679-2681 (1997). 104. A. P. de Silva, H. Q. N. Gunaratne and C. P. McCoy, Nature 364, 42-44 (1993). 105. A. P. de Silva, H. Q. N. Gunaratne and C. P. McCoy, J. Am. Chem. Soc. 119, 7891-7892 (1997). 106. A. P. de Silva, H. Q. N. Gunaratne and G. E. M. Maguire, J. Chem. Soc., Chem. Commun. 1994, 1213-1214 (1994). 107. A. P. de Silva and N. D. McClenaghan, Chem. Eur. J. 8, 4935-4945 (2002). 108. A. P. de Silva, I. M. Dixon, H. Q. N. Gunaratne, T. Gunnlaugsson, P. R. S. Maxwell and T. E. Rice, J. Am. Chem. Soc. 121, 1393-1394 (1999). 109. B. Turfan and E. U. Akkaya, Org. Lett. 4, 2857-2859 (2002). 110. Z. Wang, G. Zheng and P. Lu, Org. Lett 7, 3669-3672 (2005). 111. S. D. Straight, P. A. Liddell, Y. Terazono, T. A. Moore, A. L. Moore and D. Gust, Adv. Funct. Mater. 17, 777-785 (2007). 112. H. T. Baytekin and E. U. Akkaya, Org. Lett. 2, 1725-1727 (2000). 113. G. Zong, L. Xiana and G. Lua, Tetrahedron Lett. 48, 3891-3894 (2007). 114. T. Gunnlaugsson, D. A. Mac Dónaill and D. Parker, Chem. Commun. 2000, 93-94 (2000). 115. T. Gunnlaugsson, D. A. Mac Dónaill and D. Parker, J. Am. Chem. Soc. 123, 12866-12876 (2001). 116. M. de Sousa, B. de Castro, S. Abad, M. A. Miranda and U. Pischel, Chem. Commun. 2006, 2051-2053 (2006). 117. L. Li, M.-X. Yu, F. Y. Li, T. Yi and C. H. Huang, Colloids Surf. A 304, 49-53 (2007). 118. 55 V. Luxami and S. Kumar, New J. Chem. 32, 2074-2079 (2008). – 24 – 119. J. H. Qian, X. H. Qian, Y. F. Xu and S. Y. Zhang, Chem. Commun. 2008, 4141-4143 (2008). 120. E. Pérez-Inestrosa, J.-M. Montenegro, D. Collado, R. Suau and J. Casado, J. Phys. Chem. C 111, 6904-6909 (2007). 121. P. Remón, R. Ferreira, J.-M. Montenegro, R. Suau, E. Pérez-Inestrosa and U. Pischel, ChemPhysChem 10, 2004-2007 (2009). 122. A. Coskun, E. Deniz and E. U. Akkaya, Org. Lett. 7, 5187-5189 (2005). 123. D. Jiménez, R. Martínez-Máñez, F. Sancenón, J. V. Ros-Lis, J. Soto, A. Benito and E. García-Breijo, Eur. J. Inorg. Chem. 2005, 2393-2403 (2005). 124. W. Sun, C. H. Xu, Z. Zhu, C. J. Fang and C. H. Yan, J. Phys. Chem. C 112, 16973-16983 (2008). 125. Z. X. Li, L. Y. Liao, W. Sun, C. H. Xu, C. Zhang, C. J. Fang and C. H. Yan, J. Phys. Chem. C 112, 5190-5196 (2008). 126. Y. Zhou, H. Wu, L. Qu, D. Zhang and D. Zhu, J. Phys. Chem. B 110, 15676-15679 (2006). 127. W. Sun, Y.-R. Zheng, C.-H. Xu, C.-J. Fang and C.-H. Yan, J. Phys. Chem. C 111, 11706- 11711 (2007). 128. U. Pischel and B. Heller, New J. Chem. 32, 395-400 (2008). 129. J. Andreasson, S. D. Straight, S. Bandyopadhyay, R. H. Mitchell, T. A. Moore, A. L. Moore and D. Gust, J. Phys. Chem. C 111, 14274-14278 (2007). 130. M. Amelia, M. Baroncini and A. Credi, Angew. Chem. Int. Ed. 47, 6240-6243 (2008). 131. E. Perez-Inestrosa, J. M. Montenegro, D. Collado and R. Suau, Chem. Commun. 2008, 1085-1087 (2008). 132. J. Andreasson, S. D. Straight, T. A. Moore, A. L. Moore and D. Gust, J. Am. Chem. Soc. 130, 11122-11128 (2008). 133. D. Margulies, C. E. Felder, G. Melman and A. Shanzer, J. Am. Chem. Soc. 129, 347-354 (2007). 134. M. Suresh, A. Ghosh and A. Das, Chem. Commun. 2008, 3906-3908 (2008). 135. E. Katz, I. Willner, Chem. Commun. 2005, 5641-5643 (2005). 136. M. N. Chatterjee, E. R. Kay and D. A. Leigh, J. Amer. Chem. Soc. 128, 4058-4073 (2006). 137. R. Baron, A. Onopriyenko, E. Katz, O. Lioubashevski, I. Willner, S. Wang and H. Tian, Chem. Commun. 2006, 2147-2149 (2006). – 25 – 138. E. Katz and I. Willner, Electrochem. Commun. 8, 879-882 (2006). 139. F. Galindo, J. C. Lima, S. V. Luis, A. J. Parola and F. Pina, Adv. Funct. Mater. 15, 541-545 (2005). 140. A. Bandyopadhyay and A. J. Pal, J. Phys. Chem. B 109, 6084-6088 (2005). 141. D. Fernandez, A. J. Parola, L. C. Branco, C. A. M. Afonso and F. Pina, J. Photochem. Photobiol. A 168, 185-189 (2004). 142. F. Pina, A. Roque, M. J. Melo, I. Maestri, L. Belladelli and V. Balzani, Chem. Eur. J. 4, 1184-1191 (1998). 143. G. Will, J. S. S. N. Rao and D. Fitzmaurice, J. Mater. Chem. 9, 2297-2299 (1999). 144. J. Hiller and M. F. Rubner, Macromolecules 36, 4078-4083 (2003). 145. F. Pina, J. C. Lima, A. J. Parola and C. A. M. Afonso, Angew. Chem. Int. Ed. 43, 1525- 1527 (2004). 146. R. Stadler, S. Ami, C. Joachim and M. Forshaw, Nanotechnology 15, S115-S121 (2004). 147. A. P. De Silva, Y. Leydet, C. Lincheneau and N. D. McClenaghan, J. Phys. Cond. Mat. 18, S1847-S1872 (2006). 148. A. Adamatzky, IEICE Trans. Electron. E87C, 1748-1756 (2004). 149. G. Bell and J. N. Gray, in: Beyond Calculation: The Next Fifty Years of Computing, edited by P. J. Denning and R. M. Metcalfe, Chapter 1, Pages 5-32 (Copernicus/Springer, NY, 1997). 150. L. Fu, L. C. Cao, Y. Q. Liu and D. B. Zhu, Adv. Colloid Interf. Sci. 111, 133-157 (2004). 151. K. P. Zauner, Critical Rev. Solid State Mater. Sci. 30, 33-69 (2005). 152. Unconventional Computation, Lecture Notes in Computer Science, Vol. 5715, edited by C. S. Calude, J. F. Costa, N. Dershowitz, E. Freire and G. Rozenberg (Springer, Berlin, 2009). 153. S. G. Shiva, Introduction to Logic Design. 2nd edition (Marcel Dekker, New York, 1998). 154. D.-H. Qu, Q.-C. Wang and H. Tian, Angew. Chem. Int. Ed. 44, 5296-5299 (2005). 155. J. Andréasson, S. D. Straight, G. Kodis, C.-D. Park, M. Hambourger, M. Gervaldo, B. Albinsson, T. A. Moore, A. L. Moore and D. Gust, J. Am. Chem. Soc. 128, 16259-16265 (2006). 156. J. Andréasson, G. Kodis, Y. Terazono, P. A. Liddell, S. Bandyopadhyay, R. H. Mitchell, T. A. Moore, A. L. Moore and D. Gust, J. Am. Chem. Soc. 126, 15926-15927 (2004). – 26 – 157. M. V. Lopez, M. E. Vazquez, C. Gomez-Reino, R. Pedrido and M. R. Bermejo, New J. Chem. 32, 1473-1477 (2008). 158. D. Margulies, G. Melman and A. Shanzer, J. Am. Chem. Soc. 128, 4865-4871 (2006). 159. O. Kuznetz, H. Salman, N. Shakkour, Y. Eichen and S. Speiser, Chem. Phys. Lett. 451, 63- 67 (2008). 160. F. Pina, M. J. Melo, M. Maestri, P. Passaniti and V. Balzani, J. Am. Chem. Soc. 122, 4496- 4498 (2000). 161. G. Ashkenazi, D. R. Ripoll, N. Lotan and H. A. Scheraga, Biosens. Bioelectron. 12, 85-95 (1997). 162. S. Sivan and N. Lotan, Biotechnol. Prog. 15, 964-970 (1999). 163. S. Sivan, S. Tuchman and N. Lotan, Biosystems 70, 21-33 (2003). 164. A. S. Deonarine, S. M. Clark and L. Konermann, Future Generation Computer Systems, 19, 87-97 (2003). 165. R. Unger and J. Moult, Proteins 63, 53-64 (2006). 166. M. N. Stojanovic, D. Stefanovic, T. LaBean and H. Yan, in: Bioelectronics: From Theory to Applications, edited by I. Willner and E. Katz, Chapter 14, Pages 427-455 (Wiley-VCH, Weinheim, 2005). 167. A. Saghatelian, N. H. Volcker, K. M. Guckian, V. S. Y. Lin and M. R. Ghadiri, J. Am. Chem. Soc 125, 346-347 (2003). 168. G. Ashkenasy and M. R. Ghadiri, J. Am. Chem. Soc. 126, 11140-11141 (2004). 169. Z. Ezziane, Nanotechnology 17, R27-R39 (2006). 170. K. Rinaudo, L. Bleris, R. Maddamsetti, S. Subramanian, R. Weiss and Y. Benenson, Nature Biotechnol. 25, 795-801 (2007). 171. M. N. Win and C. D. Smolke, Science 322, 456-460 (2008). 172. A. Ogawa and M. Maeda, Chem. Commun. 2009, 4666-4668 (2009). 173. M. N. Stojanovic, T. E. Mitchell and D. Stefanovic, J. Am. Chem. Soc. 124, 3555-3561 (2002). 174. Y. Benenson, Curr. Opin. Biotechnol. 20, 471-478 (2009). 175. H. Lederman, J. Macdonald, D. Stefanovic and M. N. Stojanovic, Biochemistry 45, 1194- 1199 (2006). 176. K. Schlosser and Y. Li, Chemistry & Biology 16, 311-322 (2009). – 27 – 177. J. Elbaz, B. Shlyahovsky, D. Li and I. Willner, ChemBioChem 9, 232-239 (2008). 178. T. Li, E. Wang and S. Dong, J. Am. Chem. Soc. 131, 15082-15083 (2009). 179. M. Moshe, J. Elbaz and I. Willner, Nano Lett. 9, 1196-1200 (2009). 180. I. Willner, B. Shlyahovsky, M. Zayats and B. Willner, Chem. Soc. Rev. 37, 1153-1165 (2008). 181. M. Soreni, S. Yogev, E. Kossoy, Y. Shoham and E. Keinan, J. Amer. Chem. Soc. 127, 3935-3943 (2005). 182. J. Xu and G. J. Tan, J. Comput. Theor. Nanosci. 4, 1219-1230 (2007). 183. M. Kahan, B. Gil, R. Adar and E. Shapiro, Physica D 237, 1165-1172 (2008). 184. N. H. E. Weste and D. Harris, CMOS VLSI Design: A Circuits and Systems Perspective (Pearson/Addison-Wesley, Boston, 2004). 185. J. Wakerly, Digital Design: Principles and Practices (Pearson/Prentice Hall, Upper Saddle River, NJ, 2005). 186. U. Alon, An Introduction to Systems Biology. Design Principles of Biological Circuits (Chapman & Hall/CRC Press, Boca Raton, Florida, 2007). 187. R. S. Braich, N. Chelyapov, C. Johnson, P. W. K. Rothemund, L. Adleman, Science 296, 499-502 (2002). 188. Y. Setty, A. E. Mayo, M. G. Surette and U. Alon, Proc. Natl. Acad. USA 100, 7702-7707 (2003). 189. N. E. Buchler, U. Gerland and T. Hwa, Proc. Natl. Acad. USA 102, 9559-9564 (2005). 190. O. Pretzel, Error-Correcting Codes and Finite Fields (Oxford Univ. Press, Oxford, 1992). 191. B. Arazi, A Commonsense Approach to the Theory of Error Correcting Codes (MIT Press, Cambridge, MA, 1988). 192. K. Glinel, C. Dejugnat, M. Prevot, B. Scholer, M. Schonhoff and R. V. Klitzing, Colloids Surf. A 303, 3-13 (2007). 193. S. K. Ahn, R. M. Kasi, S. C. Kim, N. Sharma and Y. X. Zhou, Soft Matter 4, 1151-1157 (2008). 194. I. Tokarev and S. Minko, Soft Matter 5, 511-524 (2009). 195. P. M. Mendes, Chem. Soc. Rev. 37, 2512-2529 (2008). 196. M. Pita, S. Minko and E. Katz, J. Mater. Sci.: Materials in Medicine 20, 457-462 (2009). 197. I. Willner, A. Doron and E. Katz, J. Phys. Org. Chem. 11, 546-560 (1998). – 28 – 198. V. I. Chegel, O. A. Raitman, O. Lioubashevski, Y. Shirshov, E. Katz and I. Willner, Adv. Mater. 14, 1549-1553 (2002). 199. E. Katz, L. Sheeney-Haj-Ichia, B. Basnar, I. Felner and I. Willner, Langmuir 20, 9714- 9719 (2004). 200. X. Wang, Z. Gershman, A. B. Kharitonov, E. Katz and I. Willner, Langmuir 19, 5413-5420 (2003). 201. I. Luzinov, S. Minko and V. V. Tsukruk, Prog. Polym. Sci. 29, 635-698 (2004). 202. S. Minko, Polymer Rev. 46, 397-420 (2006). 203. I. Tokarev, V. Gopishetty, J. Zhou, M. Pita, M. Motornov, E. Katz and S. Minko, ACS Appl. Mater. Interfaces 1, 532-536 (2009). 204. M. Motornov, J. Zhou, M. Pita, V. Gopishetty, I. Tokarev, E. Katz and S. Minko, Nano Lett. 8, 2993-2997 (2008). 205. M. Motornov, J. Zhou, M. Pita, I. Tokarev, V. Gopishetty, E. Katz and S. Minko, Small 5, 817-820 (2009). 206. M. Krämer, M. Pita, J. Zhou, M. Ornatska, A. Poghossian, M. J. Schöning and E. Katz, J. Phys. Chem. B 113, 2573-2579 (2009). 207. J. Zhou, T. K. Tam, M. Pita, M. Ornatska, S. Minko and E. Katz, ACS Appl. Mater. Interfaces 1, 144-149 (2009). 208. X. Wang, J. Zhou, T. K. Tam, E. Katz and M. Pita, Bioelectrochemistry 77, 69-73 (2009). 209. T. K. Tam, J. Zhou, M. Pita, M. Ornatska, S. Minko and E. Katz, J. Am. Chem. Soc. 130, 10888-10889 (2008). 210. T. K. Tam, M. Ornatska, M. Pita, S. Minko and E. Katz, J. Phys. Chem. C 112, 8438-8445 (2008). 211. L. Amir, T. K. Tam, M. Pita, M. M. Meijler, L. Alfonta and E. Katz, J. Am. Chem. Soc. 131, 826-832 (2009). 212. T. K. Tam, M. Pita, M. Ornatska and E. Katz, Bioelectrochemistry 76, 4-9 (2006). 213. Y. Shirokane, K. Ichikawa and M. Suzuki, Carbohydrate Res. 329, 699-702 (2000). – 29 – FIGURES Figure 1. Left: The identity “gate” mapping digital 0 and 1 to the same values. Right: A possible response curve. – 30 – Figure 2. The response function corresponding to the reaction A A C + → , see Equation (7), for three different values of the parameter p. All three curves are concave. – 31 – + →(cid:34) Figure 3. The response function corresponding to the reaction A C , where the omitted reactant is not considered a variable input, see Equation (12), for three different pairs of values of the parameters a and b. All three curves are convex. – 32 – Figure 4. A desirable sigmoid response for filtering: the central inflection region is narrow and positioned away from both logic 0 and 1, and the slopes at and near both logic values are very small (ideally, they should be zero). – 33 – Figure 5. Smoothly varying response surfaces of the AND gate realized by the reaction A B C + → , see Equation (15), for two choices of the parameters α and β defined in Equation (16). The upper panels give the front view, whereas the lower panels offer the back view of the same surfaces. – 34 – Figure 6. Same as in Figure 5, but with fast-reaction parameters (large α and β). This case illustrates the emergence of a response surface with non-smooth features: formation of a ridge (here symmetric, along the diagonal), and also shrinking of the region for which the value of the (cid:71) ∇ gradient near the point (0,0) remains small. Note that the gradient at the origin, x y 0, 0 = = is zero for all the surfaces shown in both figures. Similar nonuniformities set in all along the ridge region, including near the logic (1,1). The emergence of an (off-diagonal) ridge can already be seen in the right panels in Figure 5, which correspond to a relatively large value of β. z x y ( , ) , – 35 – Figure 7. Left: Schematic of a one-sided sigmoid-type behavior. Right: A desirable two-sided sigmoid response surface for AND gates. – 36 – Figure 8. The three-AND-gate network, with separately varied inputs constant) in an experimental19 realization. x 1,2,3 (and 1y kept – 37 –
1210.2111	2	1210	2013-03-25T21:33:48	Design principles and fundamental trade-offs in biomimetic light harvesting	[ "physics.bio-ph", "physics.chem-ph", "quant-ph" ]	Recent developments in synthetic and supramolecular chemistry have created opportunities to design organic systems with tailored nanoscale structure for various technological applications. A key application area is the capture of light energy and its conversion into electrochemical or chemical forms for photovoltaic or sensing applications. In this work we consider cylindrical assemblies of chromophores that model structures produced by several supramolecular techniques. Our study is especially guided by the versatile structures produced by virus-templated assembly. We use a multi-objective optimization framework to determine design principles and limitations in light harvesting performance for such assemblies, both in the presence and absence of disorder. We identify a fundamental trade-off in cylindrical assemblies that is encountered when attempting to maximize both efficiency of energy transfer and absorption bandwidth. We also rationalize the optimal design strategies and provide explanations for why various structures provide optimal performance. Most importantly, we find that the optimal design strategies depend on the amount of energetic and structural disorder in the system. The aim of these studies is to develop a program of quantum-informed rational design for construction of organic assemblies that have the same degree of tailored nanoscale structure as biological photosynthetic light harvesting complexes, and also have the potential to reproduce their remarkable light harvesting performance.	physics.bio-ph	physics	Design principles and fundamental trade-offs in biomimetic light harvesting Mohan Sarovar Scalable and Secure Systems Research, Sandia National Laboratories, MS 9158, 7011 East Avenue, Livermore, California 94550 USA E-mail: [email protected] K. Birgitta Whaley Berkeley Center for Quantum Information and Computation, Berkeley, California 94720 USA Department of Chemistry, University of California, Berkeley, California 94720 USA PACS numbers: 87.15.A-, 87.15.bk, 81.07.Nb Abstract. Recent developments in synthetic and supramolecular chemistry have created opportunities to design organic systems with tailored nanoscale structure for various technological applications. A key application area is the capture of light energy and its conversion into electrochemical or chemical forms for photovoltaic or sensing applications. In this work we consider cylindrical assemblies of chromophores that model structures produced by several supramolecular techniques. Our study is especially guided by the versatile structures produced by virus-templated assembly. We use a multi-objective optimization framework to determine design principles and limitations in light harvesting performance for such assemblies, both in the presence and absence of disorder. We identify a fundamental trade-off in cylindrical assemblies that is encountered when attempting to maximize both efficiency of energy transfer and absorption bandwidth. We also rationalize the optimal design strategies and provide explanations for why various structures provide optimal performance. Most importantly, we find that the optimal design strategies depend on the amount of energetic and structural disorder in the system. The aim of these studies is to develop a program of quantum-informed rational design for construction of organic assemblies that have the same degree of tailored nanoscale structure as biological photosynthetic light harvesting complexes, and also have the potential to reproduce their remarkable light harvesting performance. 3 1 0 2 r a M 5 2 ] h p - o i b . s c i s y h p [ 2 v 1 1 1 2 . 0 1 2 1 : v i X r a Design principles and fundamental trade-offs in biomimetic light harvesting 2 Interest in the molecular mechanisms underlying photosynthetic light harvesting has recently escalated, fueled in part by the potential of engineering biomimetic solar energy harvesting technologies. The biomimetic approach aims to reproduce properties of the light harvesting complexes (LHCs) found in biology by using solid-state or organic components engineered at the nanoscale. Natural LHCs are remarkably efficient at all of the primary stages of photosynthesis: light capture, energy transfer, free carrier generation, and charge transfer [1]. Furthermore, LHCs perform these tasks in a manner that is robust to varying external and internal conditions. Reproducing such efficiencies and robustness would revolutionize our energy production capabilities. This constitutes part of the tremendous appeal of biomimetic approaches to designing photovoltaic technologies. However, photosynthetic light harvesting also has useful lessons for development of other technologies. In particular, efficient photon capture and conversion to charge is also an important component of sensor technologies, raising the question whether these might also benefit from biomimetic approaches. Biomimetic light harvesting may be viewed as one component, the "front end", of artificial photosynthesis, which seeks to generate energy rich materials or fuels from sunlight. Recent years have seen impressive achievements in several aspects of artificial photosynthesis, with progress in a diverse range of platforms ranging from molecular to semiconductor systems [2]. In this work we focus on light harvesting, the initial stage of any photosynthetic unit, and address the question of how the collection of light energy may be optimized by a biomimetic LHC and what tradeoffs are involved in achieving this goal, within a program of quantum-informed rational design. In order to construct a functional LHC from the bottom-up we require a detailed mapping between structural features and motifs, and the light harvesting or sensing capabilities of the composite system. Establishing such a mapping is complicated by the wide variety of different structures evidenced by natural LHCS. Despite the detailed variations in structure, however, some general motifs do emerge. In particular, the majority of LHCs consist of densely packed pigment-protein complexes in which the light-absorbing chromophores (e.g. chlorophyll and carotenoid molecules) are arranged with specific relative orientations and locations within scaffolds provided by proteins ‡. This complexity indicates that a sophisticated understanding of how the nanoscale structure influences the light harvesting function is necessary in order to introduce precision and accuracy into the biomimetic approach. In particular, recent studies of the behavior of the natural LHCs consisting of pigment-protein complexes have revealed evidence of quantum dynamical effects in the electronic energy transfer through the complex, indicating that quantum effects hitherto neglected in analysis of energy transport may play a role in the high quantum efficiency and need to be accounted for ‡ Exceptions to this general feature are provided by the antenna complexes of photosynthetic bacteria that are adapted to survive under conditions of very weak illumination, e.g., green sulfur bacteria, which consist of very large numbers (103 − 104) of chromophores without a protein scaffold [3, 4]. Design principles and fundamental trade-offs in biomimetic light harvesting 3 [5]. Theoretical modeling of the optical and electronic properties of large-scale pigment protein structures is a key component of obtaining this mapping and building a deeper understanding of the structure-function relationships in light harvesting. Much recent theoretical analysis has focused on understanding the quantum efficiency of energy transfer in biological LHCs -- e.g. Refs. [6, 7, 8, 9, 10, 11, 12]. Detailed modeling, including incorporation of quantum coherent effects such as exciton delocalization and chromophore-protein interactions, has allowed rationalization of the energy transfer times seen in experiments and the high quantum efficiencies typical of LHCs. Theoretical simulations have identified the dynamical parameters necessary for optimal energy transfer in model LHC systems and shown that a delicate balance between quantum coherent and incoherent dynamics appears to characterize this optimal energy transport regime [13, 14]. Calculations exploring the landscape of these parameters for a small, well characterized LHC, the Fenna-Matthews-Olson (FMO) complex, have further indicated that this component of the green sulfur bacteria photosynthetic system operates in just such an optimal regime [10, 12]. This would suggest that the design of optimized biomimetic LHCs should engineer the nanoscale structure so as to achieve strong inter-pigment couplings that compete with decoherent processes such as vibrational relaxation. Since strong interpigment coupling is an important requirement for quantum coherence in electronic energy transfer, a related underlying question for the biomimetic LHC program is whether such quantum coherence is coincidental or whether it is essential to the light harvesting function. In this paper we take a different approach to the study of optimality in light harvesting by incorporating an important perspective from the field of multi-objective optimization. Our starting point is the recognition that light harvesting, whether by natural or artificial systems, is not uniquely focused on achieving a high quantum efficiency for conversion of light to charge carriers, but is inherently a multi-objective optimization with several key objectives that must be simultaneously taken into account. In biological systems there are a wide array of objectives, not all of which necessarily have the same weight or status. One could say that the most important objective for a biological system is survival, which will ultimately dictate the changes in design features in response to environmental changes. In constructing biomimetic LHCs, we have a simpler task in that while the number of objectives are still greater than one, they are probably fewer and more equivalent in rank than in the biological case. For example, for photovoltaic applications one will likely want not only to maximize energy transfer and free-carrier generation efficiencies, but also to maximize the spectral width of absorption. In contrast, for a sensor the second objective may be to maximize the sensitivity to a particular wavelength rather than the spectral width of absorption. It is well known that in the presence of such multiple objectives one can have competition between them, which results in the emergence of families of optimal solutions that negotiate the trade-offs between the competing objectives in different ways [15]. Design principles and fundamental trade-offs in biomimetic light harvesting 4 The questions posed in this work are thus two-fold. First, are there fundamental trade-offs involved in light harvesting? Second, if so, how can these be negotiated by engineering the structure of light harvesting complexes? We address these questions here by explicit calculation of the simultaneous optimization of the two desiderata mentioned above for a light harvesting system that might be used for energy conversion; we seek to simultaneously optimize the efficiency of excitonic transport and the spectral width of absorption. We do this by studying a prototypical biomimetic light harvesting antenna and requiring that it absorb photons in as wide a spectral window as possible while also efficiently transporting the resulting excitation energy to regions of charge separation. The ultimate limits of spectral width are dictated by choice of pigments; that is, pigment transition energies largely determine the width of absorption profiles. Therefore, one method for increasing the absorption profile width is to include pigments with as many different transition energies as possible. However, this creates an energetically disordered aggregate with typically reduced transport efficiency. Thus there is a trade- off, or competition, between the two objectives. One way to negotiate this conflict is to use molecular aggregation; by employing strong Coulombic coupling, aggregates of pigments can broaden or sharpen absorption profiles as exemplified by the classic H- and J-aggregates absorption profiles [16]. In this work we shall explore the extent to which this provides an effective technique for negotiating the trade-off between efficiency and spectral-width. We pause to mention some previous work examining the optimization and design of molecular structure in the context of light harvesting. Fetisova conducted some early and far-sighted studies into the relationship between structure and function in light harvesting complexes and strategies for optimization of structure, e.g. [17, 18, 19]. Much of this work was conducted well before experimental evidence for dynamical quantum coherence and therefore considered classical models of light harvesting dynamics. More recently, Fingerhut et al have employed structural optimization to design synthetic centers for ultra-fast charge separation for artificial photosynthesis [20]. The charge separation dynamics was modeled classically using Marcus theory, and notably, they examined a multi-objective optimization landscape and considered trade-offs between the quantum efficiency of charge separation and other objectives. Knoester and co- workers have examined the excitonic and optical properties of general cylindrical aggregates of identical chromophores (with and without disorder) in several works, including Refs. [21, 22]. Finally Noy et al [23], and more recently Scholes et al [24], have compiled summaries of insights gained from studying natural LHCs and discussed how they aid the design of artificial light harvesting systems. The remainder of this paper is structured as follows. Section 1 introduces the physical systems forming the cylindrical assemblies of chromophores that constitute the focus of our study. Section 2 then outlines our theoretical models of the structure and electronic excitation dynamics in these systems. Section 3 presents the results of Design principles and fundamental trade-offs in biomimetic light harvesting 5 our multi-objective optimization studies in terms of achievable objectives and tradeoffs encountered. This is followed in section 4 by a detailed analysis of the structural and excitonic properties of the chromophore assemblies that optimize the objectives. Then in section 5 the preceding study and analysis is condensed into a set of design principles for light harvesting complexes, particularly cylindrical antennas. Finally, we conclude with a discussion and an outline of future work in section 6. 1. Cylindrical chromophore assemblies We focus in this work on cylindrical molecular assemblies as a prototypical architecture for biomimetic LHCs. Many systems self-assemble or can be templated into cylindrical structures and several of these have been studied as candidates for artificial light harvesting systems. Examples include carbocyanine molecules with hydrophobic and hydrophilic side groups and porphyrin derivatives that aggregate into cylindrical structures [25, 26]. There are also examples of cylindrical molecular aggregates in natural LHCs, the most prominent one being the chlorosome complexes of green sulfur bacteria [3, 4]. One particularly promising realization of a cylindrical aggregate is not formed from the direct aggregation of chromophores, but rather, from virus-templated assembly. Virus-templated chromophore assemblies are supramolecular complexes constructed by attaching chromophores to protein coats of viruses that then self-assemble into large, regular structures. The self-assembling protein coats are used as rigid scaffolds that guide the synthetic organization of chromophores. Structures templated using the tobacco-mosaic, M13 and MS2 viruses, amongst others, have been demonstrated for use in light harvesting [27, 28, 29], drug delivery [30], and battery technologies [31, 32]. This method of templated self-assembly presents a particularly promising path towards producing engineered assemblies of chromophores with controlled nanoscale structure. The templated assembly allows for a greater degree of customization than direct aggregation of chromophores because the nanoscale structure can be controlled by the choice of chromophores, templating protein, and of chromophore-protein linker groups. Templated assembly has the further advantage that pigments which do not naturally self-aggregate can be used. Also, it should be noted that the virus-templated assembly process creates a protein-pigment structure and not a direct molecular aggregate; as noted above, the former are more common in natural LHCs. We consider here templated assemblies based on use of the tobacco mosaic virus (TMV) as a scaffold. The protein of this virus can self-assemble into stacked disks or cylinders, depending on assembly conditions such as solution pH [33]. From a practical stand-point TMV assemblies are advantages because the assembly process is very well known and large quantities can be reliably produced. Various laboratories have demonstrated covalent attachment of pigments to various sites on mutated TMV protein monomers [27, 28] (see Fig. 1), and hence shown that the inter-pigment distances can Design principles and fundamental trade-offs in biomimetic light harvesting 6 be tuned by choice of attachment site. This is advantageous for studying the impact of quantum coherent effects on light harvesting, because coherent dynamics will be more prevalent in densely packed structures where inter-pigment couplings are strong. Thus the ability to vary the inter-pigment distance controllably over a finely tuned and large range of values, with other features of the protein scaffold held constant, allows the synthesis of chromophore arrays with and without the potential for such coherent dynamics. (a) Templated assembly using TMV (b) Variety of sites on a TMV protein monomer at which mutations can be introduced to facilitate covalent attachment of chromophores Figure 1. The self-assembling TMV protein provides a scaffold for producing an array of chromophores with well defined inter-chromophore distances. 2. Structural and dynamical model 2.1. Structural model Our multi-objective optimization studies are made for an idealized but versatile model for cylindrical arrays of chromophores that is shown in Fig. 2. The model consists of N disks, each with M chromophores attached at specific sites, as determined by an implicit protein scaffold or template. Chromophore location on sequential disks may be off-set by a variable amount and the chromophores have variable orientation but they are restricted to all have the same orientation relative to their disk. This model allows us to simulate TMV-templated chromophore assemblies in cylindrical and disk- Design principles and fundamental trade-offs in biomimetic light harvesting 7 like structures, which may have vertical or helical stacking of chromophores along the cylinder. The cylindrical array of chromophores will function as an antenna absorbing photons and transporting the resulting photo-excitation to one of its ends. This end of the cylinder (referred to as the "bottom" of the cylindrical aggregate) interfaces with a surface or electron acceptor and charge separation occurs at this interface. For the TMV-templated structures a distinct chromophore whose energy levels match well with the remainder of the chromophores and the surface work function can be attached at the end to facilitate the electron transfer event. In this first work, we will not be concerned with optimizing these subsequent charge separation and transfer events but rather with the optimization of the preceding energy transfer to the separating interface. In the present calculations we assume that the following degrees of freedom (design parameters) can be tuned: (i) the transition energy of the chromophores on each disk (i.e., each of the N disks has a distinct energy, but the M chromophores on any single disk have the same transition energy), (ii) all chromophores attach to the TMV protein scaffold with the same orientation and this orientation defines a transition dipole that is specified by two angles, θ a tangential angle and φ a radial angle, (iii) there can be a degree of misalignment between neighboring disks and this defines a helical angle, θh, which allows introduction of a helical twist along the cylinder coordinate. See Fig. 2 for representations of these angles. This results in a total of N + 3 controllable degrees of freedom, which may be used to produce a wide range of chromophore array structures that range from complete alignment along any axis, to helical arrays with variable numbers of helical strands, e.g., M = 1 with θh > 0 corresponds to a single helical strand, M = 2 with θh > 0 to a double helix, etc. TMV can template a 1-helix at the appropriate conditions, however we will not study this structure in this work, and will instead focus on M = 17 as specified below. In addition, we note that the chlorosome of green sulfur bacteria is assumed to have a structure that is an M -helix [4] and therefore this model is flexible enough to capture artificial and natural light harvesting structures. The calculations presented here make explicit comparison of two TMV-templated aggregates based on the stacked disk morphology, which are designed to have very different inter-chromophore distances. Upon aggregation each TMV "disk" consists of 17 protein monomers. Assuming complete functionalization (attachment of chromophores) this results in M = 17 chromophores per disk. Each chromophore is attached to the disk at a specific location, which is determined by its binding site on a protein monomer: this can be varied by prior treatment of the protein, allowing variable attachment location and hence variable inter-chromophore distance, see Fig. 1. We compare here TMV103 and TMV123, where the numerical labels refer to the protein monomer site where pigments are attached. The 103 site is within the pore of the assembled TMV protein and attachment here results in a disk of chromophores of radius 25 A, implying an average neighboring pigment separation distance (within a disk) of about 10 A, and thus strong coupling of chromophores. In contrast, attachment at site 123 results in Design principles and fundamental trade-offs in biomimetic light harvesting 8 Figure 2. Reduced description of chromophore assembly resulting from stacked disk TMV structure, in which only the chromophore properties are variable and the TMV protein scaffold is held constant. There are M chromophores attached to each disk and N disks. All chromophores are assumed to be attached with the same orientation and this orientation defines a transition dipole for each chromophore which is specified by two angles, θ and φ. Finally, an angle θh specifies the degree of misalignment between neighboring disks, and allows for a helical twist. a disk of chromophores of radius 40 A and an average neighboring pigment separation distance of 14 − 15 A, which leads to weaker electronic couplings. The vertical distance between disks remains the same for both structures: ∼ 20 A. While TMV103 can achieve a greater density of pigments, we note, however, that the coupling between pigments on neighboring disks can nevertheless still be the same for both structures because the distance between disks is the same. A question we seek to answer in the following analysis is whether the dense packing within disks and any resultant increase in coherent dynamics of TMV103 is advantageous for any aspect of light harvesting. We note that current experimental techniques do not allow independent tuning of all the above design parameters for TMV-templated cylindrical assemblies. For instance, it is difficult to restrict a disk to only containing chromophores of a given species (transition energy) when there are several species on the whole cylinder. However, it is important to choose a large number of independent degrees of freedom in order to explore the landscape of optimal light harvesting given near-ideal resources. The figures of merit obtained can therefore be considered upper bounds on what is currently experimentally achievable. In a forthcoming publication we will examine the optimization landscape for a more modest set of design parameters. N disks !"#$%&'&($&%)"()%"+,-"Design principles and fundamental trade-offs in biomimetic light harvesting 9 2.2. Dynamical model The TMV-templated cylindrical chromophore assemblies are complex biomolecular systems. In order to efficiently compute and compare their light harvesting properties we must resort to an effective description of the electronic degrees of freedom. In analogy with the conventional modeling of natural light harvesting complexes [5, 34] we employ the Frenkel Hamiltonian to describe coherent dynamics: Jij(i(cid:105)(cid:104)j + j(cid:105)(cid:104)i) Ei i(cid:105)(cid:104)i + (cid:88) (cid:88) He = (1) i i,j where i(cid:105) denotes an electronic excitation localized on pigment i in the complex, Ei is the transition energy of pigment i and Jij denotes the Coulombic coupling of pigments i and j. For simplicity, we assume that each pigment has only one dominant transition in the wavelength region of interest. This model can be easily generalized to pigments with more than one transition. In this work we treat the electronic coupling in the dipole- dipole coupling approximation. This approximation is generally valid for inter-pigment separations greater than ∼ 12A [35, 36], and while we will model some cases where the inter-pigment distance is slightly smaller than this, the error due to the dipole-dipole approximation in those cases will not affect our conclusions significantly. The eigenstates of He are the excitonic states in the complex, whose decomposition in terms of the site basis {i(cid:105)} we denote: ek(cid:105) = (cid:88) Ui,k i(cid:105) (2) i where U is the matrix that diagonalizes He. In addition to this coherent dynamics, the coupling of pigment electronic degrees of freedom to protein degrees of freedom, and the resultant decoherent and dissipative dynamical effects must be incorporated into the simulation. There is little experimental information on the vibrational dynamics of the protein degrees of freedom in virus-templated assemblies and hence we employ a model that makes minimal assumptions. The electronic-vibrational coupling is chosen to be linear and the vibrational degrees of freedom are modeled harmonically, resulting in the interaction Hamiltonian: (cid:88) i(cid:105)(cid:104)i(cid:88) HI = cξ,i(aξ,i + a † ξ,i) (3) i ξ Hv =(cid:80) i,ξ ωξ,ia where aξ,i denotes the annihilation operator for mode ξ coupled to the excited state of pigment i, and the vibrational "bath" is described by a free harmonic Hamiltonian: † ξ,iaξ,i. We model the harmonic bath using an over-damped Brownian oscillator model [37] in the high temperature limit. The spectral density is taken to be Ohmic with Lorenz-Drude regularization: J(ω) = 2λγω ω2+γ2 , with reorganization energy λ = 100cm−1, and relaxation time γ−1 = 100fs. The true vibrational and solvent dynamics of TMV-templated assemblies likely has a more complex spectral density than this structureless model. However, in the absence of experimental data we use this Design principles and fundamental trade-offs in biomimetic light harvesting 10 simple spectral density with minimal assumptions, noting that its generic features and parameter values are similar to the spectral densities used for simulation of electronic energy transfer in natural LHCs [34]. In order to describe the effective dynamics of the electronic degrees of freedom that are of interest, we average over the harmonic bath degrees of freedom. There are numerous methods for doing this averaging, at various levels of approximation. In this work we use the modified Redfield formalism [38, 39] to derive an effective equation of motion for the excitonic populations of the complex once the vibrational degrees of freedom have been averaged. In effect the modified Redfield approach derives a dynamical equation for the exciton populations: dP (t) dt = RP (t) (4) where R is the modified Redfield rate matrix and P (t) is a vector of excitonic populations of length n + 2, where n = N × M is the total number of chromophores in the complex, and we have included two additional entries to track population lost via exciton loss (recombination) and excitation trapping (due to localization at the interface and subsequent charge separation). The exciton loss and trapping rates are chosen phenomenologically since there are no experimentally measured values for TMV- templated chromophore aggregates). See Appendix B for details. An important feature of this approach is that while the dynamics of off-diagonal elements of the exciton density matrix, which correspond to the excitonic coherences, are not explicitly followed in time, the components of the interaction Hamiltonian that generate them are included perturbatively in the calculation of the population transfer rate matrix elements, see Eq. (42) of [38]. We employ the modified Redfield formalism here for several reasons. First, it has been shown to be reasonably accurate across a wide range of parameter regimes, varying from strong to weak system-bath coupling, and is thus preferred to the standard Redfield approach outside of the regime of weak system-bath coupling [38]. The approach is particularly effective at modeling energy transfer in systems with structural or energetic disorder that is larger than, or comparable to, the magnitude of bath induced reorganization effects [39]. Second, it is an extremely efficient method of dynamical propagation and this efficiency will be critical for the numerically intensive optimizations we undertake. The primary drawback of the modified Redfield model is that it only allows the explicit dynamical simulation of excitonic populations and not of coherences. Given that the excitonic populations will nevertheless be perturbatively influenced by the coherences (as described above), this is not a significant drawback for our purposes since we are primarily interested in asymptotic efficiency of transport and spectral width of absorption, both of which can be extracted from excitonic populations (see Appendix B). Finally, we note that using the symmetries of the cylindrical stacked disk structure together with the dipole-dipole approximation implies that we only have to consider a Design principles and fundamental trade-offs in biomimetic light harvesting 11 limited domain for the two angles that define the transition dipole of a chromophore: θ ∈ [0, π/2], φ ∈ [−π/2, π/2], as well as for the helical angle θh ∈ [−π/M, π/M ]. The Hamiltonians resulting from all other choices of these angles can be replicated by a choice within these domains -- e.g. the coupling between disks with all dipoles pointed up and all dipoles pointed down yields the same excitonic structure. Therefore, in order to improve the efficiency, we allow the angular design variables to vary within these restricted domains. 3. Optimized light harvesting in cylindrical chromophore assemblies 3.1. Multi-objective optimization and Pareto fronts The multi-objective optimizations described here employ N = 10 disks in the cylindrical aggregate, with the energy of the chromophores on any disk allowed to vary in the range 400 − 450nm. We assume full functionalization of the TMV protein monomers and hence there are M = 17 chromophores per disk. That is, TMV naturally assembles into disks with 17 protein monomers [27], and assuming each protein monomer is functionalized with a chromophore, this results in M = 17 chromophores per disk. All chromophores are assumed to have the same transition dipole strength, µ = 3 Debye. When the two angles defining the orientation of the transition dipole of the chromophore, θ, φ and the helical angle θh, are also included, this yields 13 design parameters that define the Hamiltonian of the electronic degrees of freedom of the chromophore assembly. We seek to optimize over this design parameter space in order to find the structures that are optimal for two objectives: transport efficiency and spectral width of absorption. The maximization of these objectives is performed here using a genetic optimization algorithm that evaluates the energy transfer efficiency and absorption spectral width for each member of its evolving population. Appendix A provides more detail on the specifics of the genetic optimization algorithm. Assemblies with and without disorder are considered, where the disorder can be both structural and energetic. For the simulation of systems with disorder, we average the objectives over 504 instances of disorder at each design parameter configuration. The disorder is introduced as independent, Gaussian distributed perturbations of the relevant parameters, i.e., of each chromophore's transition energy (with variance 2nm), transition dipole orientation angles θ, φ (with variance 0.2 radians), and helical angle θh (with variance 0.2 × 2π/M radians). Note that the introduction of disorder will break the equivalence of the M chromophores on a single disk. In the field of multi-objective optimization, a critical concept is the Pareto front. This is the locus or curve in objective space formed by all the solutions to the optimization program that optimally negotiate the trade-offs between multiple objectives, where this means that increasing the value of a particular objective (for a maximization program), leads to the decrease in the value of one or more of the other Design principles and fundamental trade-offs in biomimetic light harvesting 12 objectives. The shape of the Pareto front reveals the amount of competition between the various objectives. In cases where there is significant competition between different objectives, one must choose a trade-off solution that suitably compromises between the competing objectives. See Appendix A for more details on Pareto fronts and multi- objective optimization. Our genetic optimization is designed to converge on the Pareto front for the problem of light harvesting using a cylindrical antenna, with the dual objective functions of efficiency of excitonic transport and spectral width of absorption. 3.2. Pareto fronts for TMV103 and TMV123 The Pareto fronts for simultaneous optimization of spectral-width and efficiency for TMV103 and TMV123 are shown in Fig. 3, for both calculations with and without disorder, First consider the fronts for the ideal structure in the case of no disorder (black dots and red triangles for TMV103 and TMV123, respectively). The efficiencies and spectral widths achievable are very similar for both the densely packed (TMV103) and sparsely packed (TMV123) structures, and the shape of the trade-off curve is very similar too. The presence of curvature in both of these Pareto fronts indicates that there is a trade-off between achieving high spectral width and high efficiency. In particular, we see that for low spectral width absorption, both of these idealized cylindrical structures can achieve near unit efficiency of electronic energy transport. The similarity of these two fronts at all except the largest spectral widths suggests that in the absence of disorder there is generally little advantage achieved by dense packing and strong coupling of pigments; the same efficiencies and spectral widths can be achieved with dense and sparse packing. The exception is the regime of high spectral width, where there appears to be a crossover to and from a regime in which the densely packed TMV103 shows a better simultaneous optimization of the two objectives, consistently achieving a higher efficiency for a given spectral bandwidth (in the range 100-120 nm), In general, however, the results without disorder show that the negotiation of the tradeoff between the two objectives is comparable for densely and sparsely packed structures. With the introduction of structural and energetic disorder, several marked changes occur to the Pareto fronts. Firstly, both fronts collapse in the horizontal direction, reflecting the result that much smaller spectral widths are now achievable. There is also a small but noticeable amount of collapse in the vertical direction, indicating that somewhat smaller efficiencies are achievable with disorder. In addition, the Pareto fronts for TMV103 and TMV123 separate and become distinguishable. These generic changes can be rationalized in term of energetic and structural features, which we will do below. For low spectral width absorption, the densely packed TMV103 (blue squares) shows a slight enhancement in efficiency over the sparsely packed TMV123 (green crosses), although both are still very close to the optimal values obtained with the ordered structures. It is noteworthy that the best efficiencies achievable are comparable for TMV123 and TMV103, and somewhat surprisingly, not very different from the best Design principles and fundamental trade-offs in biomimetic light harvesting 13 Figure 3. Pareto fronts in the efficiency-spectral width objective space for TMV103 and TMV123 structures with and without structural and energetic disorder. The statistical variation in the points with disorder is negligible (we confirmed this) at the level of sampling performed and therefore error bars have not been included for clarity. The inset is a zoom into the congested high efficiency region. efficiencies achieved in the absence of disorder. This indicates that efficiency of energy transfer in cylindrical structures can be robust to disorder. In contrast to this robust behavior of the efficiency, the achievable spectral widths are however now much smaller in the presence of disorder. Here the dense packing of TMV103 is clearly beneficial and allows significantly greater spectral width to be attained. In general terms, the overall shapes of the Pareto fronts in the presence of disorder are similar for TMV103 and TMV123, and show that a more drastic trade-off between efficiency and spectral bandwidth is required than in the absence of disorder. Thus, to increase spectral width beyond ∼ 68nm (∼ 75nm) for TMV123 (TMV103) requires a significant sacrifice in efficiency of energy transport. The generic nature of the Pareto fronts obtained for the two TMV-templated chromophore assemblies with and without disorder indicate that the trade-off between efficiency and spectral width is fundamental light harvesting antennas and independent of the presence of disorder. Beyond this result however, several subtle differences are apparent between the detailed behavior of the densely and sparsely packed systems. In particular, the densely packed TMV103 appears to be better able to preserve both a high efficiency and high absorption bandwidth in the presence of disorder. This suggests that the dense packing and ensuing quantum coherence may serve to provide robustness of function against sources of disorder. In the following for such cylindrical 4050607080901001101201300.50.550.60.650.70.750.80.850.90.951Spectral width (nm)Efficiency TMV103, no disorderTMV123, no disorderTMV103, disorderTMV123, disorder5060700.960.98 TMV103, no disorderTMV123, no disorderTMV103, disorderTMV123, disorderDesign principles and fundamental trade-offs in biomimetic light harvesting 14 section, we explore these factors in more detail by examining the parameters defining the structures on the Pareto fronts. 4. Design variables emerging from optimal structures To understand the Pareto fronts in Fig. 3 and to appreciate how the objectives are maximized and the conflicts negotiated, we now examine the performance metrics in parameter space. We use two complementary summaries of the structural information and present them in Figs. 4 and 5 for the case of no disorder, and Figs. 6 and 7 for the case with disorder. The first summary in both cases (Figs. 4 and 6) presents the parameter variations for structures on the Pareto front, while the second (Figs. 5 and 7) presents a detailed analysis of the excitonic and dynamical features of extremal structures on the Pareto fronts in Fig. 3. In the next two subsections we will interpret the data in these summaries, but first we describe in detail the content of the figures. Figs. 4 and 6 summarize the values of the 13 design parameters for structures on the Pareto front, for TMV103 and for TMV123 without (Fig. 4) and with (Fig. 6) disorder. The x-axis on all plots in these figures varies over the integer parameter s, which indexes structures on the Pareto fronts in Fig. 3 from left to right, i.e., starting with the most efficient (smallest spectral width) and ending with the least efficient (largest spectral width). There are three panels of plots displayed in each subfigure of Figs. 4 and 6. The left panels, (i) show a color-coded representation of the energies of the N = 10 disks for each structure on the corresponding Pareto front. The middle panels, (ii) show the two orientational angles θ, φ and the helical angle θh for each structure on the corresponding Pareto front. The third panels, (iii) show the ratio of coupling between neighboring chromophores on a single disk (intra-disk) to the coupling between closest neighboring chromophores on adjacent disks (inter-disk), which we define as Λ ≡ Jintra-disk/Jinter-disk. This ratio is important because it indicates the direction of strong electronic coupling -- either within disks (large values of the ratio) or between disks and along the cylinder (small values of the ratio). It is important to appreciate that the electronic coupling (J) by itself does not dictate the rate of energy transfer between two chromophores (or a group of chromophores). The energy difference between them also plays a role. However, the value of Λ informs us about the direction of dominant coupling and thus will be used in the interpretation of results below. Figs. 5 and 7 present the excitonic structure and dynamics of the extremal structures on each of the four Pareto fronts shown in Fig. 3. The extremal structures are defined as the solutions on each of the Pareto fronts that achieve the maximum of one of the objectives at the expense of the other: these are precisely the structures indexed by minimal and maximal values of s. Examining these extremal structures elucidates the design principles involved in maximizing the objectives. Fig. 5 examines extreme structures in TMV103 and TMV123 in the absence of disorder, and Fig. 7 examines extremes structures for TMV103 and TMV123 in the presence of disorder. Design principles and fundamental trade-offs in biomimetic light harvesting 15 Each subfigure in these plots is divided into two panels; the left panel shows details of the most efficient structure and the right panel shows details of the structure that achieves the greatest spectral width. Between these two panels, in the center of the figure we show the corresponding optimal orientations of the chromophore transition dipoles in the two extremal structures. Note that these orientations are shown for only two of the N = 10 disks, the orientations on the remaining disks are identical to these. The structure giving the most efficient transport is on top and is to be viewed together with the left panel. The structure giving the greatest spectral bandwidth is below and is to be viewed together with the right panel. Considering now the left and right panels of Figs. 5 and 7, each of these shows three plots. In all of these plots the x-axis indexes the excitons (sorted from lowest to highest energy) for the corresponding extremal structure. Various properties that quantify the delocalization and energetics of the excitons and the resulting modified Redfield rates for energy transfer are shown as a function of the exciton index, as follows. (i) The top plot shows two measures that quantify the optical accessibility and delocalization of the excitons, respectively. The green line shows the magnitude of the excitonic dipole in units of the individual chromophore transition dipole magnitude (3 Debye), which constitutes a measure of the optical accessibility of the excitons. The blue line shows the inverse participation ratio (IPR) of the exciton n is the probability of exciton between disks , defined as: dIPRk = k being localized on disk n which is a sum of the probabilities of exciton k being i Ui,k2 where localized on any of the chromophores composing disk n: pk the sum is over chromophores i that compose disk n and Ui,k are elements of the transformation matrix that defines the site-basis to exciton-basis transformation, Eq. (2). The disk inverse participation ratio dIPR measures the extent to which the excitons are delocalized over multiple disks; a dIPR value of d indicates that the exciton has significant amplitude over chromophores on d disks. n = (cid:80) 2 where pk 1(cid:80)10 n=1 pk n (ii) The middle plot shows the energy and the spatial location of the excitons. The blue line denotes the energy of the exciton. The green line gives the disk number where most of the amplitude of the exciton is localized; i.e. arg maxn pk n for any exciton indexed by k. This quantity constitutes a measure of the approximate spatial location of each exciton. Oscillations in this localization parameter (as a function of k) are a heuristic (iii) The bottom plots show the non-zero elements of the modified Redfield rate matrix R describing exciton dynamics in the structure. The magnitude of each element is color coded. These rates are normalized by the largest rate in the matrix and so are all between 0 and 1. This plot nicely illustrates the domains of connectivity in excitonic/energy space. Thus, if two groups of excitons are connected by large matrix elements in R, then there will be significant population exchange between them. Individual domains are formed by groups of strongly coupled excitons that Design principles and fundamental trade-offs in biomimetic light harvesting 16 (a) TMV103 without disorder (b) TMV123 without disorder Figure 4. Design variables for structures on the Pareto fronts without disorder, for (a) TMV103 and (b) TMV123. The x-axis on all plots shows the structure index s on the Pareto front. Left panels (i): color coded disk energies for the N = 10 disks. Central panels (ii): transition dipole orientation angles θ, φ and helical angle θh, with the upper subplot showing the variation of efficiency (green line) and spectral width (blue line) as a function of s. Right panels (iii): ratio of intra- to inter-disk dipole-dipole coupling, Λ. are only weakly coupled to other excitons. In the next section we will refer to these sets of plots in Figs. 5 and 7 to interpret and explain the details of the parametric trends shown in Figs. 4 and 6. We note that for the disordered cases, Figs. 6 and 7, the parameters, structures, and excitonic details shown are before the introduction of disorder. That is, the physical structures, and consequently excitonic details, shown in these figures are perturbed by each instance of disorder that is sampled over. 4.1. Optimal structures in the absence of disorder We first compare the optimal values of the 13 design variables for TMV103 and TMV123 without disorder, Fig. 4. We see that the trend in variance of energies of disks is similar for both structures -- i.e. the variance increases steadily as spectral width increases. Thus the incorporation of either distinct chromophores or distinct local energetic environments for these will be necessary to achieve large spectral width in the 01020304050600123456s (high efficiency to low efficiency)(cid:82)4090140Width (nm)0.50.751Eff010203040506000.511.5s (high efficiency to low efficiency)Angles (rad) (cid:101)(cid:113)(cid:101)h!"#$!"""#$!""#$010203040500.50.60.70.80.911.11.21.31.4s (high efficiency to low efficiency)(cid:82)0100200Width (nm)0.50.751Eff0102030405000.511.5s (high efficiency to low efficiency)Angles (rad) (cid:101)(cid:113)(cid:101)h!"#$!"""#$!""#$Design principles and fundamental trade-offs in biomimetic light harvesting 17 (a) TMV103 without disorder (b) TMV123 without disorder Figure 5. Analysis of excitonic structure and dynamics for the extremal structures on the TMV103 and TMV123 Pareto fronts (without disorder ). See main text (beginning of section 4) for explanation of the plots. ordered structures. Furthermore the energy profiles in Figs. 4(a)(i) and 4(b)(i) show that the most efficient structures have little heterogeneity in terms of pigment energies, whereas in the opposite extreme, the high-spectral width structures have highly varied pigment energies (although the bottom disk where the trap is situated always has the lowest energy chromophores). The structures in the middle of the Pareto front, i.e., those that achieve a balance between efficiency and spectral width, have a very specific distribution of pigment energies that defines an energy funnel that gradually transitions from high energy pigments at the top to low energy pigments at the bottom of the cylinder. Turning to the choice of angular design variables, it is evident that the tangential angle θ has the most influence on light harvesting function. For TMV103 (Fig. 4(a)(ii)), the most efficient structures have θ ≈ 0 and this angle transitions to a tilted optimal value θ ≈ 0.6 as we move along the Pareto front. For TMV123, there is a similar transition from θ ≈ 0 for the most efficient structure at the left extremal of the Pareto e!ciencyspectral widthe!ciencyspectral widthDesign principles and fundamental trade-offs in biomimetic light harvesting 18 front, to θ ≈ 0.4 for the extremal structure with largest spectral width, although for this more sparse distribution of chromophores, the corresponding transition happens much faster (i.e., at smaller s) and the tilted orientation is preferred for most structures on the Pareto front. The radial angle for all structures on the Pareto front for both TMV103 and TMV123 is restricted to be in the range φ < 0.2. Therefore having transition dipoles that are tilted into or out of the disk plane does not seem to present any advantage for either efficient transport or spectral width. Within this small range the radial angle can fluctuate and still yield similar values of efficiency and spectral width. Finally, the helical angle θh (whose limits are smaller: −π/17 ≤ θh ≤ π/17) can also fluctuate significantly without affecting either objective. This suggests that in the absence of disorder, the objectives are fairly robust to both the radial and helical angles (at least in the ranges φ < 0.2, −π/17 ≤ θh ≤ π/17). To understand the behavior of Pareto structures with respect to the tangential angle, θ, it is useful to also look at the behavior of the ratio of intra-disk and inter-disk dipole couplings Λ, which is shown in Fig. 4(iii). It is evident that in the absence of disorder, this ratio drops significantly as s increases and that it shows a transition from Λ > 1 to Λ (cid:46) 1 at about the same point as θ transitions from zero to ∼ 0.6 (∼ 0.4) for TMV103 (TMV123). This transition in Λ as a function of s indicates that strong coupling within disks is preferred for improved efficiency, while strong coupling between adjacent disks is preferred for enhanced spectral width. Consequently the optimal structures show a transition in the direction of the dominant coupling as one moves along the Pareto front. To confirm that θ is indeed responsible for this change in the direction of the dominant coupling within the cylindrical array, we plot the parametric dependence of the dipole-dipole coupling between two neighboring pigments on the same disk (Jintra-disk) and two neighboring pigments on adjacent disks (Jinter-disk), for TMV103 and TMV123 in Fig. 8 . We show here only the parametric dependence on the tangential (θ) and helical (θh) angles, since these dependencies are the most relevant. As can be seen from this figure, the coupling between pigments on the same disk dominates for most of parameter space for both TMV103 and TMV123. However, for a region around θ ≈ 0.6, this intra-disk coupling becomes very small and is dominated by the inter-disk coupling. This region where the inter-disk coupling dominates is larger for TMV123 because the distance between pigments on the same disk for this structure is larger. So we conclude that the spatial direction of electronic coupling, which is measured by the value of the ratio Λ, can be controlled by the tangential angle of the transition dipole orientation. This analysis raises the question as to what is the functional reason for transitioning from dominant coupling within disks to dominant coupling between disks as we progress along the Pareto front from most efficient to largest spectral width, i.e., as we increase s? At first glance, this appears somewhat counter-intuitive, since one might expect that strong coupling between disks would lead to efficient excitation transfer, and conversely Design principles and fundamental trade-offs in biomimetic light harvesting 19 that strong excitonic coupling within a disk would lead to spectral broadening. However, the true optimal solutions reflect a more subtle balance of the light harvesting properties possible in this design variable space and geometry. Fig. 8 shows that in both TMV103 and TMV123, due to the close proximity of pigments within a disk, it is difficult for the coupling between the disks to dominate over the intra-disk coupling. There is only a small region around θ ≈ 0.6 where this happens and in this region both intra-disk and inter-disk couplings are small ((cid:46) 10 cm−1). In particular, the inter-disk coupling at this angular configuration is dominated by the reorganization energy λ = 100 cm−1 and hence the transport down the cylinder will be Forster-like and on slow timescales. Therefore the strategy of simply maximizing coupling between disks to have maximum energy transfer down the cylinder will not work because of the geometric dimensions in these systems. The alternative strategy that is converged on by the optimization of transport efficiency in our calculations, is to have large intra-disk coupling and smaller inter-disk coupling, while maintaining only a slight energy gradient. The slight energy gradient implies that neighboring disks contain almost resonant chromophores (recall that there is no disorder at this stage) and hence even the small coupling between neighboring disks creates excitons that are delocalized across these disks. We can confirm this strategy by examining the extremely efficient structures in 5. For TMV103 the delocalization of excitons by this strategy is corroborated by the large dIPR values and the oscillation of the disk number in which the excitons have primary spatial locality (both in the left panel of Fig. 5(a)). This network of moderately delocalized excitons with overlapping spatial locality results in a dense modified Redfield rate matrix (bottom panel of Fig. 5(a)) and efficient population transfer. The most efficient structure in TMV123 utilizes a similar strategy as TMV103, except that the extent of exciton delocalization across disks is a little greater due to the smaller energy gradient (in this case, all disks are almost the same energy). For small values of s, TMV123 settles on values of θ such that the inter-disk and intra-disk couplings are comparable. This creates significant energetic overlap between delocalized excitons that are distant and results in efficient long-range (multi-chromophoric) Forster transfer [40]. These observations are evidenced in the left panel of Fig. 5(b) that shows the significant delocalization of excitons in the most efficient TMV123 structure, and the dense modified Redfield matrix describing the efficient long-range transfer of energy. We now consider the structures with largest spectral width, shown in the right panels of Fig. 5. Here the strategy for optimizing spectral width in TMV103 and TMV123 is identical. The first element, as already noted is a varied distribution of transition energies: the main mechanism by which spectral width is enhanced is simply by having pigments that absorb at various energies. In addition to this, it is interesting that the angular configurations for the structures maximizing spectral width converge on a coupling ratio of Λ (cid:46) 1. This results in the inter-disk coupling dominating or being the same order as the intra-disk coupling. To understand this, we note that Design principles and fundamental trade-offs in biomimetic light harvesting 20 even a small intra-disk coupling is sufficient to create delocalized excitons within disks (because all chromophores on the same disk are resonant), whereas a larger inter-disk coupling is required since chromophores on neighboring disks may be energetically different. Therefore, in order to create a large cluster of coupled chromophores and hence excitons with large delocalization, it is advantageous to maximize the inter-disk coupling while keeping the intra-disk coupling small. This is precisely the strategy utilized in the extremal structures with the largest spectral width, i.e., the large s value structures at the righthand end of the Pareto front. Indeed, the right panels of Fig. 5(a) and 5(b) show that this approach delocalizes excitons over multiple disks (as evidenced by the dIPR and disk population oscillations), which is a signature of electronic coupling of many chromophores over multiple disks. Such electronic coupling of many chromophores results in spectral broadening. It is interesting that the excitons that are most delocalized are the low and high energy ones. This expands the absorption spectrum beyond the limits set by the range of single chromophore transition energies -- i.e. the absorption profiles extend beyond 400nm and 450nm. 4.2. Optimal structures in the presence of disorder We now examine the structures and parameters of the optimal configurations on the Pareto fronts when energetic and structural disorder are included. The presence of disorder invalidates many of the optimization strategies outlined above for disorder- free systems. Firstly, the most striking observation for the optimal design variables of the disordered systems, is the presence of a strong energy funnel in both Figs. 6(a)(i) and 6(b)(i). To achieve any efficiency in the presence of disorder an energy funnel is absolutely necessary, and in both cases (TMV103 and TMV123), a shallow energy funnel amounting to a gradual decrease in pigment energies as the disks approach the bottom layer is more effective at generating efficient transport. With the exception of these most efficient structures, the energetic profiles for the remaining structures on the Pareto front for both TMV103 and TMV123 are remarkably similar. The optimal design strategy for the non-extremal structures appears to be to have discrete decreases in energy every couple of disks, leading to a more step-like energy gradient than the smooth gradient preferred by the most efficient structures. Turning to the choice of angular design parameters, consider the case of TMV123 first (Fig. 6(b)(ii)). Unlike the disorder-free case, where a variety of angle choices (especially for φ and θh) yielded similar objectives, in the presence of disorder a few optimal configurations emerge. In particular, the most efficient structures have θ = θh ≈ 0. At the other extreme, the structures with the largest spectral width have θ ≈ π/2 and θh ≈ 0. The radial angle is constrained to be φ < 0.25 and is close to zero for nearly all structures on the Pareto front. As with the disorder-free case, we find that the tangential angle has the biggest influence on the objectives. The radial and helical angles have only a slight influence, as long as they are constrained to be ≈ 0. For Design principles and fundamental trade-offs in biomimetic light harvesting 21 (a) TMV103 with disorder (b) TMV123 with disorder Figure 6. Design variables for structures on the Pareto fronts with disorder, for (a) TMV103 and (b) TMV123. The x-axis on all plots shows the structure index s on the Pareto front. Left panels (i): color coded disk energies for the N = 10 disks. Central panels (ii): transition dipole orientation angles θ, φ and helical angle θh, with the upper subplot showing the variation of efficiency (green line) and spectral width (blue line) as a function of s. Right panels (iii): ratio of intra- to inter-disk dipole-dipole coupling, Λ. disordered TMV103 (Fig. 6(a)(ii)), the tangential angle again has the biggest influence and it transitions from θ ∼ 0.4 to θ ∼ π/2 as s increases. The other angles fluctuate along the Pareto front of TMV103 more than for TMV123, but we cannot discern any pattern in their variation along the Pareto front. Finally, Figs. 6(a)(iii) and 6(b)(iii) show that there is a clear preference for having dominant coupling along the cylinder (low Λ) to optimize efficiency, while dominant coupling within disks is preferred in order to optimize spectral width. In fact the most efficient structures maximize the inter-disk coupling. For TMV103 this is achieved around θ ≈ 0.4, θh = 0 and for TMV123 it is achieved around θ ≈ 0, θh ≈ 0 (see Fig. 8). And similarly, the structures with maximum spectral with maximize the intra-disk coupling, which occurs at θ ≈ π/2 as seen from Fig. 8. We can gain more insight into the parameter landscape by examining the extreme structures on the disorder Pareto fronts. As the analysis below will show, the inability to construct large delocalized excitonic states on any timescale in the presence of significant static disorder dominates the design choices made in optimizing light harvesting. The 010203040500510152025s (high efficiency to low efficiency)(cid:82)506580Width (nm)0.850.921Eff0102030405000.511.5s (high efficiency to low efficiency)Angles (rad) (cid:101)(cid:113)(cid:101)h!"#$!"""#$!""#$02040608010011.522.533.544.555.5s (high efficiency to low efficiency)(cid:82)405570Width (nm)0.850.921Eff02040608000.511.5s (high efficiency to low efficiency)Angles (rad) (cid:101)(cid:113)(cid:101)h!"#$!"""#$!""#$Design principles and fundamental trade-offs in biomimetic light harvesting 22 (a) TMV103 with disorder (b) TMV123 with disorder Figure 7. Analysis of excitonic structure and dynamics for the extremal structures on the TMV103 and TMV123 Pareto fronts (with disorder ). See main text (beginning of section 4) for explanation of the plots. left panels of Figs. 7(a) and 7(b) show that the most efficient structures for TMV103 and TMV123 have similar excitonic structure, showing several distinct features. Firstly, the excitons are mostly localized on a single disk. There is a slight delocalization, especially in TMV103, as evidenced by the dIPR and oscillations in disk population but it is minimal compared to the disorder-free structures. This is a result of the strict energy gradient, with each disk having pigments with a different transition energy and consequently little delocalization across disks. The addition of static disorder, and also dynamic disorder induced by protein fluctuations, will of course further localize the excitons. The structure of the modified Redfield rate matrix for the most efficient TMV103 and TMV123 structures (bottom plots of left panels Fig. 7) is seen to be fairly sparse with small rates between densely coupled domains. The domains are localized on disks and there is larger transfer rates between excitons within a domain than between domains. This localization of excitons into domains and inter-domain dominated transport is reminiscent of multi-chromophoric Forster transfer [40], and we e!ciencyspectral widthe!ciencyspectral widthDesign principles and fundamental trade-offs in biomimetic light harvesting 23 can expect this to provide accurate description of the energy transfer in these structures optimized for efficiency in the presence of disorder. The strategy here will be to have slightly delocalized excitons on disks that are as strongly coupled as possible to excitons on neighboring disks due to maximized inter-disk electronic coupling (at θ ≈ 0.4, θh = 0 for TMV103 and at θ ≈ 0, θh ≈ 0 for TMV123), and a shallow energy gradient which ensures that neighboring disks contain chromophores that are nearly resonant despite the disorder. (a) TMV103 (b) TMV123 Figure 8. Dipole-dipole coupling between two neighboring pigments on one disk (brown-tan surface) and two neighboring pigments on adjacent disks (red-yellow surface) as a function of the tangential transition dipole orientation (θ) and the inter- disk helical (θh) angles (the radial transition dipole orientation φ is fixed here at zero). (a) TMV103, (b) TMV123. The primary difference between these two structres is the density of pigments within a disk. The distance between pigments on adjacent disks for both TMV103 and TMV123 is 20A. The distance between pigments on the same disk is ∼ (2π/17) × 25 = 9.24 Afor TMV103 and ∼ (2π/17) × 40 = 14.78 Afor TMV123. Now we turn to the extremal structures that maximize spectral width in the presence of disorder. These are illustrated with the lower structures shown in the central panels of Fig. 7. There is little obvious difference between these two extremal structures for TMV103 and for TMV123. Both consist of disks that are J-aggregate-like, with dipoles aligned head-to-tail. The energy of each disk is different, except for the first two and last two which have almost identical energies. This creates a series of almost independent J-aggregates (i.e. bright excitons are the lowest energy ones) that absorb at a range of energies §. The reason for this form of the optimal structure for achieving § Note that the excitonic details shown in Fig. 7 are for the ideal design parameters. Each instance of disorder will perturb these and result in a perturbed version of these excitonic details. In particular, the exciton delocalization for the maximum spectral width structures (for TMV103 and TMV123) seems large from the dIPR values shown in the right panels of Fig. 7. However, any instance of disorder will break the symmetries of the ideal design and result in more localized excitons. (cid:239)0.2(cid:239)0.100.10.200.40.81.21.6020406080100120(cid:101) (rad)(cid:101)h (rad)J (cm(cid:239)1)(cid:239)0.2(cid:239)0.100.10.200.40.81.21.6051015202530(cid:101) (rad)(cid:101)h (rad)J (cm(cid:239)1)Design principles and fundamental trade-offs in biomimetic light harvesting 24 maximal spectral width can be understood by examining Fig. 8, which reveals that the maximum possible excitonic coupling between any two chromophores (for TMV103 and TMV123) is an intra-disk coupling that results from θ = π/2, i.e., from dipoles aligned head-to-tail within a disk. This strong coupling is advantageous for competing against disorder and maintaining excitonic coupling in its presence. It is not, however, advantageous for efficient transport down the cylinder since the strong coupling is intra- disk rather than inter-disk, but it is nevertheless advantageous for spectral broadening. We can rationalize this finding with a simple model for the effect of disorder on coupling of any two chromophores. Thus, given excited states of two chromophores with energies E1, E2 and electronic coupling J, the resulting excitonic states have energies (cid:112)(E1 − E2)2 + 4J 2 . e± = E1 + E2 2 ± 2 Assuming that neither are dark states (i.e. have zero net dipole strength), the absorption peaks of the coupled system will be centered around e±. In presence of finite coupling J, these energies can be different from E1 and E2, causing spectral broadening due In the presence of disorder, E1 − E2 can be large, in which to excitonic coupling. case e± will only be significantly different from the original transition energies (E1 and E2) if J is comparably large. This is the reason for the J-aggregation seen in the maximum spectral width structures; this mechanism achieves the strongest coupling and thus enables broadening even in the presence of disorder. The spectral widths achievable with disorder for TMV103 are slightly larger than those achievable for TMV123 because the former system can attain greater intra-disk coupling strengths due to the greater proximity of pigments within a given disk (see Fig. 8). In summary, in the presence of disorder, maximal spectral width is achieved by having each disk effectively act as its own antenna with strong intra-disk coupling present to provide spectral broadening and with almost every disk composed of pigments with distinct transition energies. Interestingly, this optimized structure bears resemblance to multi-junction or tandem solar cells [41, 42], which are also designed to increase the range of wavelengths of utilized photons. It is interesting to note that the strategies found here for optimizing spectral width with and without disorder are only effective in their respective cases. That is, the disorder strategy of creating J-aggregate-like maximally coupled disks would not be effective in the absence of disorder. This is because of the well known property of J-aggregates that they possess red-shifted, narrow lines of absorption because the structural symmetry renders most states optically inaccessible (i.e. results in dark states). Therefore in the absence of disorder the absorption profile would consist of thermally broadened peaks around the J-aggregate absorption lines, which is unlikely to produce a very wide spectrum. However in the presence of disorder, the complete J- aggregate symmetry is broken and the majority of states are no longer dark states, allowing wide bandwidth absorption and effective spectral broadening at the same Design principles and fundamental trade-offs in biomimetic light harvesting 25 time (note that the maximum disorder should be comparable in magnitude to the maximum achievable electronic coupling for this strategy to be effective). Disorder is thus essential to the success of this strategy. Similarly, the disorder-free strategy that is optimal for maximizing spectral width is to use the spectral broadening resulting from excitonic coupling across many chromophores. This strategy is in turn ineffective in the presence of disorder, since this would suppress such couplings and resulting broadening, as seen from the above equation for a two chromophore system. These observations underscore the important point that the optimal strategies for light harvesting can change dramatically depending on the amount of disorder present in the system. Finally, we comment on the effects of changing the non-Hamiltonian components of the system. Although we have not optimized over the uncontrollable degrees of freedom, we have performed multi-objective optimizations for several values of the reorganization energy and several degrees of disorder. The main functional influence of the reorganization energy is to vary the amount of absorption linewidth broadening. For example, when the reorganization energy was doubled to λ = 200cm−1 the achievable absorption linewidths for TMV103 and TMV123 increased. However, the shapes of the Pareto fronts remained the same and the relative advantage that TMV103 has over TMV123 in the presence of disorder was preserved (i.e. TMV103 is able to achieve greater absorption widths while maintaining efficiency than TMV123). The optimal structures also remained similar, with the same patterns of change in design variables when moving across the Pareto fronts. In contrast to changes in reorganization energy, changes in the amount of disorder were seen here to dramatically effect the optimal structures. This is already evident when comparing the Pareto fronts and optimal structures in cases of zero disorder and finite disorder presented above. We find that when the structural and energetic disorder are increased further they can dominate the light harvesting function, with structural optimizations playing less of a role. In such cases, the formation of an energetic gradient is the most effective design principle. We consider the situation that we have presented above to be the most interesting since with this choice of disorder, the fluctuations in electronic coupling energies due the disorder are comparable to the other energy scales in the system (e.g. thermal energy, reorganization energy). This results in many design parameters playing an active role in light harvesting performance and in the most interesting optimization structure. For future work, it would be interesting to incorporate experimentally determined disorder parameters, as they become available, into these multi-objective optimization studies. 5. Design principles for engineering light harvesting antennas Given the understanding we have gained of the structures on the Pareto front and their relative performance, we can now extract several design principles for optimal light harvesting using cylindrical chromophore assemblies. Design principles and fundamental trade-offs in biomimetic light harvesting 26 (i) A trade-off between efficiency and spectral width exists for cylindrical light harvesting antennas. There are limits to maximizing both of these objectives, especially in the presence of disorder. (ii) Optimal structures for efficiency or spectral width change significantly depending on whether energetic and structural disorder is present or not. In the absence of disorder, many choices of the design variables yield similar values of the objective functions. However, when disorder is present only a few choices of the design variables are truly optimal -- i.e., the design variables for neighboring structures on the Pareto front are fairly similar. (iii) An energy funnel is essential regardless of the density of pigments for such structures with a single dedicated sink for excitons. We see the emergence of an energy funnel for both TMV103 and TMV123. A shallow energy funnel is the most effective for transport. (iv) The attachment angle that dictates the tangential angle of the dominant dipole (θ) is the most critical orientation degree of control, followed by the helical angle θh. The radial angle φ should be kept as close to zero as possible since all structures on the Pareto front retain this property. (v) An increased density of pigments does not aid in increasing efficiency or spectral width in the absence of disorder. However, in the presence of disorder, increased density is advantageous since it can lead to stronger electronic couplings which can be used to combat the deleterious effects of disorder. (vi) Simply having access to strong electronic coupling is not sufficient to overcome the deleterious effects of disorder. The direction of this coupling has to be controlled and used effectively. In the case of TMV103 and TMV123 the strongest couplings are intra-disk simply because of the dimensions. These couplings only help to maintain spectral width in the presence of disorder and do not directly generate efficient energy transfer (although other strategies help to generate efficient transfer). We note that in the present system with a full complement of M = 17 chromophores per disk, there does not appear to be any apparent benefit from having a helical twist to the chroomophore arrangement along the cylinder. This may simply reflect the relatively close packing of chromophores in this system and the ability of the other 12 parameters to simultaneously optimize the energy efficiency and spectral width. However it may also reflect the lack of any chiral constraint element at the chromophore level. More extensive calculations with smaller values of M and more refined models of chromophores are needed in order to determine whether additional structural constraints such as a center of chirality on the chromophores are necessary for achieving helical optimal structures for these design objectives. We can now ask how many of these design principles are manifest in natural photosynthetic light harvesting antennae. Such observations are unavoidably speculative Design principles and fundamental trade-offs in biomimetic light harvesting 27 since we do not understand the structure-function relationships in natural antennae sufficiently well at this stage to make general statements. Nevertheless, it is still interesting to compare these emergent design principles to features and motifs seen in natural systems. Firstly, consider chlorosomes, the main light harvesting antennas of green sulfur bacteria, which are cylindrical antennas consisting of very densely packed BChl c, d, or e molecules [3, 4]. Given the above results, it is plausible that the dense packing found in chlorosomes has the primary function of increasing the range of usable photons, an important feature for green sulfur bacteria, which typically live in severely energy limited environments. It is difficult to speculate on the impact of structure on energy transfer efficiency because it is unknown at present how the photo-excitation couples out of the chlorosome complex and hence we cannot model energy extraction simply as trapping sites at the base of the cylinder as done above. One could also speculate that the small energy gradient seen in PS-II [43] is consistent with the above observation above that small energy gradients are beneficial for maximum efficiency of transport in disordered systems. 6. Conclusion Inspired by the cylindrical assemblies of chromophores that can be synthesized by TMV- templated assembly, we have examined the landscape of light harvesting performance In particular, we have studied the trade-offs involved achievable by such structures. in optimizing over multiple objectives relevant to light harvesting. In addition to identifying a fundamental trade-off between optimizing energy transfer efficiency and bandwidth of absorption, our calculations have allowed the development of several design principles to guide the design and construction of such cylindrical assemblies of chromophores. We find that the presence of disorder drastically effects the optimal structural forms and therefore it is essential to characterize the amount of structural and energetic disorder present in typical TMV-templated chromophore assemblies. Experiments to-date on protein-templated systems are insufficient to characterize this degree of disorder, and hence this is a critical task for future experiments. The design principles established here provide guidelines for a program of quantum-informed rational design (QuIRD) for biomimetic light harvesting systems. Our study also lends insight into the structure of biological photosynthetic light harvesting complexes. The multi-objective optimization study clarifies the potential role of strong electronic couplings enabled by high densities of chromophores. In realistic systems with disorder, our study suggests that strong electronic couplings are somewhat beneficial for maintaining efficiency of energy transfer, but more importantly, they are essential for increasing the spectral width of absorption profiles. Therefore the strong electronic couplings recently observed in several biological LHCs -- e.g. Refs. [44, 45, 46, 47] -- may not only aid excitation transport in such systems, but also simultaneously benefit optical measures of light harvesting performance such as spectral Design principles and fundamental trade-offs in biomimetic light harvesting 28 width of absorption. This first multi-objective optimization of structural design for biomimetic light harvesting systems suggests several interesting directions for further work. The first avenue for future work is the consideration of measures of performance other than absorption bandwidth. Although efficiency of energy transfer is likely to be desirable in all applications, other measures of performance in the optical domain could be considered. For example, sensitivity to photons of a particular wavelength could be relevant for the design of sensor technologies. As another avenue of future work, we plan to carry out a multi-objective optimization study of the helical structural that is templated using TMV [33]. The helical assembly could be more stable than the stacked disk assembly considered in this work, however features such as an energy gradient could be more difficult to implement in the helical structures. It would also be interesting to simulate excitation dynamics for the optimal structures on the Pareto front using a non-perturbative technique -- such as the hierarchical equations of motion formalism [48] -- that captures dynamical coherence (time-dependent coherence between excitons). Although the spectral characteristics will be unaffected by such treatments, they will refine the measure of energy transfer efficiency and also give an explicit indication for how important dynamical coherence is to the effectiveness of energy transfer in such structures. Unfortunately, the TMV- templated chromophore assemblies are large (in terms of number of chromophores), and beyond the limit of what is currently computationally feasible using such non- perturbative techniques. However, this direction could become feasible in the future. Finally, we plan to perform multi-optimization studies of TMV-templated chromophore assemblies that are restricted to varying the experimental degrees of freedom that are currently tunable. That is, the design variables will be restricted to the ones that are currently experimentally accessible, so that the optimized designs could be constructed and verified in the immediate future. 7. Acknowledgements We gratefully acknowledge Matthew Francis and Daniel Finley for useful discussions relating to virus-templated chromophore assemblies. We thank Akihito Ishizaki for bringing to our attention the continued fraction solution in Ref. [49]. Financial support for MS and KBW was provided by the DARPA QuBE (Quantum Effects in Biological Environments) program. The views expressed are those of the authors and do not reflect the official policy or position of the Department of Defense or the U.S. Government. Some of the multi-objective optimization were performed using supercomputer time allocated through XSEDE. Sandia National Laboratories is a multi-program laboratory managed and operated by Sandia Corporation, a wholly owned subsidiary of Lockheed Martin Corporation, for the United States Department of Energy's National Nuclear Security Administration under contract DE-AC04-94AL85000. Design principles and fundamental trade-offs in biomimetic light harvesting 29 References [1] R E Blankenship. Molecular mechanisms of photosynthesis. Blackwell Science, 2002. [2] Anthony F Collings and Christa Critchley, editors. Artificial Photosynthesis. From Basic Biology to Industrial Application. Wiley-VCH, Weinheim, FRG, May 2006. [3] V I Prokhorenko, D B Steensgaard, and A. R. Holzwarth. Exciton Dynamics in the Chlorosomal Antennae of the Green Bacteria¡ i¿ Chloroflexus aurantiacus¡/i¿ and¡ i¿ Chlorobium tepidum¡/i¿. Biophysical journal, 79(4):2105 -- 2120, 2000. [4] S Ganapathy, G T Oostergetel, P K Wawrzyniak, M Reus, A G M Chew, F Buda, E J Boekema, D A Bryant, A. R. Holzwarth, and H J M de Groot. Alternating syn-anti bacteriochlorophylls form concentric helical nanotubes in chlorosomes. Proc. Natl. Acad. Sci. USA, 106:8525, 2009. [5] Yuan-Chung Cheng and G. R. Fleming. Dynamics of light harvesting in photosynthesis. Annu. Rev. Phys. Chem., 60:241, 2009. [6] Alexandra Olaya-Castro, Chiu Fan Lee, Francesca Fassioli Olsen, and Neil F Johnson. Efficiency of energy transfer in a light-harvesting system under quantum coherence. Phys. Rev. B, 78:085115, 2008. [7] A. Ishizaki and G. R. Fleming. Theoretical examination of quantum coherence in a photosynthetic system at physiological temperature. Proc. Natl. Acad. Sci. USA, 106:17255, 2009. [8] Patrick Rebentrost, Masoud Mohseni, and Alan Aspuru-Guzik. Role of quantum coherence and environmental fluctuations in chromophoric energy transport. J. Phys. Chem. B, 113:9942, 2009. [9] F. Caruso, A. W. Chin, A. Datta, S F Huelga, and M. B. Plenio. Highly efficient energy excitation transfer in light-harvesting complexes: the fundamental role of noise-assisted transport. J. Chem. Phys., 131:105106, 2009. [10] J. Wu, F. Liu, Y. Shen, J. Cao, and R. J. Silbey. Efficient energy transfer in light-harvesting systems, I: optimal temperature, reorganization energy, and spatial-temporal correlations. New J. Phys., 12:105012, 2010. [11] Mohan Sarovar, Yuan-Chung Cheng, and K Birgitta Whaley. Environmental correlation effects on excitation energy transfer in photosynthetic light harvesting. Phys. Rev. E, 83:011906, 2011. [12] Masoud Mohseni, Alireza Shabani, Seth Lloyd, and Herschel Rabitz. Optimal and robust energy transport in light-harvesting complexes: (II) A quantum interplay of multichromophoric geometries and environmental interactions. arXiv.org, April 2011. [13] Patrick Rebentrost, Masoud Mohseni, Ivan Kassal, Seth Lloyd, and Alan Aspuru-Guzik. Environment-assisted quantum transport. New J. Phys., 11:033003, 2009. [14] M. B. Plenio and S F Huelga. Dephasing-assisted transport: quantum networks and biomolecules. New J. Phys., 10:113019, 2008. [15] RT Marler and JS Arora. Survey of multi-objective optimization methods for engineering. Structural and Multidisciplinary Optimization, 26(6):369 -- 395, 2004. [16] T Kobayashi. J-Aggregates. World Scientific, 1996. [17] Z G Fetisova, A Yu Borisov, and M V Fok. Analysis of structure-function correlations in light- harvesting photosynthetic antenna: Structure optimization parameters. Journal of Theoretical Biology, 112(1):41 -- 75, January 1985. [18] Z G Fetisova. Survival Strategy of Photosynthetic Organisms. 1. Variability of the Extent of Light- Harvesting Pigment Aggregation as a Structural Factor Optimizing the Function of Oligomeric Photosynthetic Antenna. Model Calculations. Molecular Biology, 38(3):434 -- 440, 2004. [19] A A Novikov, A S Taisova, and Z G Fetisova. Optimal spectral coordination of subantennae in natural antennae as an efficient strategy for light harvesting in photosynthesis. Journal of Bioinformatics and Computational Biology, 04(04):887 -- 909, August 2006. [20] Benjamin P Fingerhut, Wolfgang Zinth, and Regina de Vivie-Riedle. The detailed balance limit of photochemical energy conversion. Phys. Chem. Chem. Phys., 12(2):422, 2009. Design principles and fundamental trade-offs in biomimetic light harvesting 30 [21] C Didraga and J Knoester. Optical spectra and localization of excitons in inhomogeneous helical cylindrical aggregates. J. Chem. Phys., 121:10687, 2004. [22] J Knoester. Modeling the optical properties of excitons in linear and tubular J-aggregates. Int. J. Photoenergy, 2006:61364, 2006. [23] D. Noy, C. C. Moser, and P. L. Dutton. Design and engineering of photosynthetic light-harvesting and electron transfer using length, time, and energy scales. Biochim. Biophys. Acta Bioenerg., 1757:90, 2006. [24] Gregory D Scholes, Graham R Fleming, Alexandra Olaya-Castro, and Rienk van Grondelle. Lessons from nature about solar light harvesting. Nature Chemistry, 3(10):763 -- 774, September 2011. [25] Dorthe M Eisele, Jasper Knoester, Stefan Kirstein, Jurgen P Rabe, and David A Vanden Bout. Uniform exciton fluorescence from individual molecular nanotubes immobilized on solid substrates. Nature Nanotechnology, 4(10):658 -- 663, August 2009. [26] S M Vlaming, E A Bloemsma, M Linggarsari Nietiadi, and J Knoester. Disorder-induced exciton localization and violation of optical selection rules in supramolecular nanotubes. J. Chem. Phys., 134(11):114507, 2011. [27] R. A. Miller, A D Presley, and M. B. Francis. Self-assembling light-harvesting systems from synthetically modified tobacco mosaic virus coat proteins. J. Am. Chem. Soc., 129:3104, 2007. [28] M Endo, M Fujitsuka, and T Majima. Porphyrin light-harvesting arrays constructed in the recombinant tobacco mosaic scaffold. Chem. Eur. J., 13:8660, 2007. [29] Yoon Sung Nam, Taeho Shin, Heechul Park, Andrew P Magyar, Katherine Choi, Georg Fantner, Keith A Nelson, and Angela M Belcher. Virus-Templated Assembly of Porphyrins into Light- Harvesting Nanoantennae. J. Am. Chem. Soc., 132(5):1462 -- 1463, February 2010. [30] W Wu, S C Hsiao, Z M Carrico, and M. B. Francis. Genome-free viral capsids as multivalent carriers for Taxol delivery. Angew. Chem. Int. Ed., 48:9493, 2009. [31] K T Nam, D-W Kim, P J Yoo, C-Y Chiang, N Meethong, P T Hammond, Y-M Chiang, and A. M. Belcher. Virus-enabled synthesis and assembly of nanowires for lithium ion battery electrodes. Science, 312:885, 2006. [32] E Royston, A Ghosh, P Kofinas, M T Harris, and J N Culver. Self-assembly of virus-structured high surface area nanomaterials and their application as battery electrodes. Langmuir, 24:906, 2008. [33] A Klug. The tobacco mosaic virus particle: structure and assembly. Phil. Trans. R. Soc. Lond. B, 354:531, 1999. [34] A. Ishizaki, T. R. Calhoun, G. S. Schlau-Cohen, and G. R. Fleming. Quantum coherence and its interplay with protein environments in photosynthetic electronic energy transfer. Phys. Chem. Chem. Phys., 12:7319, 2010. [35] Aurora Munoz-Losa, Carles Curutchet, Brent P. Krueger, Lydia R. Hartsell, and Benedetta Mennucci. Fretting about FRET: Failure of the Ideal Dipole Approximation. Biophys. J., 96(12):4779 -- 4788, 2009. [36] Lei Yang, Stefano Caprasecca, Benedetta Mennucci, and Seogjoo Jang. Theoretical Investigation of the Mechanism and Dynamics of Intramolecular Coherent Resonance Energy Transfer in Soft Molecules: A Case Study of Dithia-anthracenophane. J. Am. Chem. Soc., 132(47):16911 -- 16921, 2010. [37] Shaul Mukamel. Principles of nonlinear optical spectroscopy. Oxford University Press, 1999. [38] M Yang and G. R. Fleming. Influence of phonons on exciton transfer dynamics: comparison of Redfield, Forster, and modified Redfield equations. Chem. Phys., 282:163, 2002. [39] W M Zhang, T Meier, V Chernyak, and S. Mukamel. Exciton-migration and three-pulse J. Chem. Phys., femtosecond optical spectroscopies of photosynthetic antenna complexes. 108:7763, 1998. [40] Seogjoo Jang, Marshall D Newton, and Robert J Silbey. Multichromophoric Forster resonance Design principles and fundamental trade-offs in biomimetic light harvesting 31 energy transfer. Phys. Rev. Lett., 92:218301, 2004. [41] Hector Cotal, Chris Fetzer, Joseph Boisvert, Geoffrey Kinsey, Richard King, Peter Hebert, Hojun Yoon, and Nasser Karam. III -- V multijunction solar cells for concentrating photovoltaics. Energy & Environmental Science, 2(2):174, 2009. [42] Tayebeh Ameri, Gilles Dennler, Christoph Lungenschmied, and Christoph J Brabec. Organic tandem solar cells: A review. Energy & Environmental Science, 2(4):347 -- 363, 2009. [43] R C Jennings, F M Garlaschi, L Finzi, and G Zucchelli. Slow exciton trapping in Photosystem II: A possible physiological role. Photosynth. Res., 47(2):167 -- 173, 1996. [44] Gregory S Engel, Tessa R Calhoun, Elizabeth L Read, Tae-Kyu Ahn, Tomas Mancal, Yuan-Chung Cheng, Robert E Blankenship, and Graham R Fleming. Evidence for wavelike energy transfer through quantum coherence in photosynthetic systems. Nature, 446:782, 2007. [45] Elisabetta Collini, Cathy Y Wong, Krystyna E Wilk, Paul M G Curmi, Paul Brumer, and Gregory D Scholes. Coherently wired light-harvesting in photosynthetic marine algae at ambient temperature. Nature, 463:644, 2010. [46] G. Panitchayangkoon, D Hayes, K A Fransted, J. R. Caram, E Harel, J Wen, R E Blankenship, and G. S. Engel. Long-lived quantum coherence in photosynthetic complexes at physiological temperature. Proc. Natl. Acad. Sci. USA, 107:12766, 2010. [47] Gabriela S Schlau-Cohen, Akihito Ishizaki, Tessa R Calhoun, Naomi S Ginsberg, Matteo Ballottari, Roberto Bassi, and Graham R Fleming. Elucidation of the timescales and origins of quantum electronic coherence in LHCII. Nature Chemistry, March 2012. [48] A. Ishizaki and G. R. Fleming. Unified treatment of quantum coherent and incoherent hopping dynamics in electronic energy transfer: reduced hierarchy equations approach. J. Chem. Phys., 130:234111, 2009. [49] T Takagahara, E Hanamura, and R. Kubo. Stochastic models of intermediate state interaction in second order optical processes -- stationary response II. J. Phys. Soc. Jpn., 43:1 -- 6, 1977. [50] K Deb, A Pratap, S Agarwal, and T Meyarivan. A Fast and Elitist Multiobjective Genetic Algorithm: NSGA-II. IEEE Transactions on Evolutionary Computation, 6(2):1, 2002. [51] K. Sastry. Single and Multiobjective Genetic Algorithm Toolbox in C++. Technical Report 2007016, University of Illinois, Genetic Algorithms Lab, June 2007. [52] V. I. Novoderezhkin, A B Doust, C. Curutchet, G. D. Scholes, and R. van Grondelle. Excitation dynamics in Phycoerythrin 545: modeling of steady-state spectra and transient absorption with modified Redfield theory. Biophysical J., 99:344, 2010. Design principles and fundamental trade-offs in biomimetic light harvesting 32 Appendix A. Multi-objective optimization The optimization of multiple objectives is a commonly encountered problem in engineering and sciences. Given a set of control variables, (cid:126)x ≡ (x1, x2, ..., xN ), the task of multi-objective optimization, in this example a maximization, is: max (cid:126)x [f1((cid:126)x), f2((cid:126)x), ...fM ((cid:126)x)] (A.1) where fi((cid:126)x) are M objectives (cost functions). One could, and usually does, also have equality or inequality constraints on the variables. For convenience we notate (cid:126)f ≡ [f1((cid:126)x), f2((cid:126)x), ...fM ((cid:126)x)]. This optimization over multiple objectives has multiple non- degenerate solutions unlike a single objective optimization which has a single solution (or multiple degenerate ones). The set of solutions are referred to as Pareto points and are the set of points in objective space such that improving any one objective can only be done at the expense of another. Specifically, (cid:126)f is a Pareto point, or is Pareto optimal, if there does not exist another feasible objective vector (cid:126)f(cid:48) such that i ≥ fi ∀i ∈ {1, 2, ..., M}, and f(cid:48) f(cid:48) The set of Pareto points in objective space is also called the Pareto front. For each Pareto point in objective space, (cid:126)f∗, there is a corresponding Pareto point in control variable space, (cid:126)x∗ and is specified by the control variables that achieve (cid:126)f∗. j > fj for at least one j ∈ {1, 2, ..., M}. objectives into one objective, e.g. F ((cid:126)x) = (cid:80)M One approach for solving multi-objective optimizations is to combine the multiple i=1 αifi((cid:126)x) with α ∈ R is a linear combination of objectives. Then a conventional single objective optimization solution method is used to solve this problem. This strategy is acceptable when the acceptable combination of objectives is known a priori. However, in many problems one cannot know ahead of time how to combine the multiple objectives into one. In addition, knowledge of the Pareto front for the full multi-objective problem is valuable for understanding the types of trade-offs involved in the optimization problem. In this work, where we are particularly interested in characterizing the trade-offs involved in light harvesting, the full multi-objective optimization ins critical. There are a variety of methods for solving multi-objective optimizations [15]. In this work we employ an evolutionary algorithm which has the advantage that it requires few assumptions and a priori knowledge of the optimization landscape. We utilize the non-dominated sorting genetic algorithm II (NGSA-II) by Deb et al. [50] implemented in the C++ optimization toolbox written by Sastry [51]. The specific genetic algorithm parameters we use are: (i) Population size: 100 (ii) Number of generations evolved: 100 (iii) Proportion of population replaced in each generation: 0.7 (iv) Crossover probability (with simulated binary crossover): 0.85 (v) Mutation rate (with polynomial mutation): 0.3 Design principles and fundamental trade-offs in biomimetic light harvesting 33 These parameter values were converged upon by experimentation and determining what combination of population size, mutation rate and crossover rate allowed good exploration of the optimization landscape. For each optimization, multiple runs were executed (there were several restarts), each from a different randomly chosen starting population and all the resulting populations were collected. This distribution of starting points added another degree of randomization to ensure that the large optimization landscape was sampled reasonably well. Appendix B. Modified Redfield theory The equations of modified Redfield theory specify rates of exciton population transfer, and define a rate equation for the exciton population dynamics: dP (t) dt = RP (t) (B.1) where R is the modified Redfield rate matrix [38], and P (t) is a vector of exciton populations. In this work we only consider exciton dynamics in the single exciton manifold since this is the most relevant at low to moderate solar irradiance. As derived in Ref. [38], the rate for population transfer from exciton k to k(cid:48) is: Rk,k(cid:48) = 2Re dτ F ∗ k(cid:48)(τ )Ak(τ )Nk,k(cid:48)(τ ) (B.2) (cid:90) ∞ with 0 Fk(cid:48)(τ ) = exp(−i(E0 Ak(τ ) = exp(−i(E0 Nk,k(cid:48)(τ ) = (cid:16) k(cid:48) − λk(cid:48))τ − g∗ k + λk)τ − gkk,kk(τ )) k(cid:48)k(cid:48),k(cid:48)k(cid:48)(τ )) gk(cid:48)k,kk(cid:48)(τ ) − [ gk(cid:48)k,kk(τ ) − gk(cid:48)k,k(cid:48)k(cid:48)(τ ) − 2iλk(cid:48)k,k(cid:48)k(cid:48)] × [ gkk(cid:48),kk(τ ) − gkk(cid:48),k(cid:48)k(cid:48)(τ ) − 2iλkk(cid:48),k(cid:48)k(cid:48)] e2(gkk,k(cid:48)k(cid:48) (τ )+iλkk,k(cid:48)kτ ) (cid:17) Here, E0 energy and λk is the exciton reorganization energy, defined by λk ≡(cid:80)K n. Similarly, λαβ,γδ ≡ (cid:80)K defined as gαβ,γδ(t) ≡(cid:80)K k = Ek−λk is the 0−0 exciton transition energy -- i.e. Ek is the exciton transition n=1 Un,k4λn with Un,k the exciton-site basis transfer coefficients and λn the reorganization energy of site n,γUn,δλn, and the exciton lineshape function is n,γUn,δgn(t) where gn(t) is the single chromophore (site) phonon-induced lineshape function. In this work we assume all chromophores have the same lineshape function, and it derives from a high temperature over-damped Brownian oscillator model of the phonons with spectral density J(ω) = 2λγω ω2+γ2 . In this case, n=1 U∗ n=1 U∗ n,αUn,βU∗ n,αUn,βU∗ (cid:18) 2λ β2γ2 − iλ γ (cid:19)(cid:0)e−γt − 1 + γt(cid:1) g(t) = (B.3) Design principles and fundamental trade-offs in biomimetic light harvesting 34 (cid:90) ∞ The integrals defining the modified Redfield rates are time-consuming to perform numerically for an arbitrary lineshape function, but in the case of the model above, the integral defining the modified Redfield rates becomes of the form: Rk,k(cid:48) ∝ 2Re dτ ec1τ−c2e−γτ 0 (B.4) where ci are k, k(cid:48) dependent, but time-independent, complex coefficients. Integrals of this form can be performed analytically using a continued fraction expansion, as shown by Takagahara et al [49]. We use this continued fraction solution to calculate the modified Redfield rates efficiently. in the Appendix of Ref. Now we detail how to extract a measure of spectral width of absorption and efficiency of transport from the modified Redfield model of exciton dynamics. Appendix B.0.1. Spectral width An expression for the absorption spectrum in terms of the modified Redfield exciton transfer rates is [52]: (cid:17)(cid:111) (cid:88) (cid:110)(cid:90) ∞ 0 (cid:16) i(ω− ωk)t− K(cid:88) Un,k4g(t)− tγk 2 d2 kRe k n=1 dt exp A(ω) = ω (B.5) where the k sum is over all excitons. dk and ωk ≡ Ek/ are the magnitude of the transition dipole and the transition frequency of exciton k, respectively. g(t), as detailed above, is the phonon induced lineshape function, taken to be that of an over-damped Brownian oscillator and the same for all chromophores. Finally, γk is the inverse lifetime k(cid:48)(cid:54)=k Rk(cid:48)k(cid:48)kk. We construct the absorption spectrum according to this expression and normalize it to have maximum 1. Then the width of the absorption spectrum is defined as the sum of frequency intervals for which A(ω) > 0.1. of exciton k, given by a sum of outgoing modified Redfield rates: γk = −(cid:80) Appendix B.0.2. Efficiency Consider the modified Redfield rate equation, Eq. (B.1), which describes the dynamics of the exciton populations. In this work, we use P (t) ≡ (p1(t), p2(t), ..., pK(t), ptrap(t), ploss(t))T , where the first K elements are the populations of the K excitons at time t (in the main text we consider cylindrical TMV assemblies with N disks and M chromophores per disk, and therefore K = N × M ). The remaining two elements in this population vector are the population in the trap (i.e. the fraction of excitons that have undergone charge separation), ptrap(t), and the population lost to recombination, ploss(t). The elements of R that represent rates of transfer between excitonic populations are prescribed by modified Redfield theory as above, and the rates of transfer from exciton k to the trap (Rk→trap) and loss (Rk→loss) channels are defined as: Rk→trap = γtrap (cid:88) n∈disk N Un,k2 Rk→loss = γlossd2 k (B.6) Design principles and fundamental trade-offs in biomimetic light harvesting 35 (cid:80) The motivation behind that k → trap rate is that exciton k has probability n∈disk N Un,k2 of being localized at one of the chromophores on the bottom most disk (disk N ) where the trapping (charge separation) is assumed to take place. Hence the total trapping rate is this probability multiplied by the bare trapping rate, γtrap. In this work we choose 1/γtrap = 4ps. The motivation behind the k → loss rates is that the rate of radiative recombination of an exciton is equal to a bare rate of recombination γloss multiplied by the dipole strength of the exciton. This ensures that bright and dark excitons recombine at different rates. In this work we choose the bare rate of recombination as 1/γloss = 1ns. the trap, η = ptrap(t → ∞). Using a Laplace transform solution of the rate equation We define the efficiency of transport as the asymptotic fraction of population in Eq. (B.1) results in the following expression for this asymptotic population: P (t → ∞) = lim s→0 (sI − R)−1P (0) (B.7) where P (0) is the initial population distribution. We approximate this asymptotic solution by taking a small value of s in the above expression. We have confirmed that each exciton's dipole. That is, pk(0) = d2 our choice of s does not influence the solution, especially the value of η = ptrap(t → ∞). (cid:80)K Our choice of P (0), the initial exciton distribution is dictated by the size of k/N , where N is a normalized so that k=1 pk(0) = 1. This initial state specifies that the probability that an exciton is initially populated is proportional to its transition dipole strength. The initial trap and loss populations are set to zero. Finally, we comment on alternatives to asymptotic efficiency for measures of energy transfer effectiveness. The average trapping time, defined as the time at which the trap population is greater than 1− for some small , is a measure that captures the speed at which the excitation is transferred. This measure could be more useful than asymptotic efficiency when further stages of the light harvesting process (e.g. charge separation) are modeled and optimized. This is because for such a multi-stage optimization it will be important to capture the characteristic timescales of the processes that form the various stages. However, we note that calculating this measure is more computationally intensive than the asymptotic measure of efficiency we use in this work. This is because it requires temporal propagation of the system density matrix or population vector, whereas the asymptotic measure defined above only requires a rate matrix inversion.
1806.01315	2	1806	2019-02-04T22:02:51	Cell Motility Dependence on Adhesive Wetting	[ "physics.bio-ph", "q-bio.CB" ]	Adhesive cell-substrate interactions are crucial for cell motility and are responsible for the necessary traction that propels cells. These interactions can also change the shape of the cell, analogous to liquid droplet wetting on adhesive substrates. To address how these shape changes affect cell migration and cell speed we model motility using deformable, 2D cross-sections of cells in which adhesion and frictional forces between cell and substrate can be varied separately. Our simulations show that increasing the adhesion results in increased spreading of cells and larger cell speeds. We propose an analytical model which shows that the cell speed is inversely proportional to an effective height of the cell and that increasing this height results in increased internal shear stress. The numerical and analytical results are confirmed in experiments on motile eukaryotic cells.	physics.bio-ph	physics	Cell Motility Dependence on Adhesive Wetting Yuansheng Cao,1, ∗ Richa Karmakar,1, ∗ Elisabeth Ghabache,1 Edgar Gutierrez,1 Yanxiang Zhao,2 Alex Groisman,1 Herbert Levine,3 Brian A. Camley,4 and Wouter-Jan Rappel1, † 1Department of Physics, University of California, San Diego, La Jolla, California 92093, USA 2Department of Mathematics, The George Washington University, Washington, DC 20052, USA 3Department of Bioengineering, Center for Theoretical Biological Physics, Rice University, Houston, Texas 77005, USA 4Department of Physics & Astronomy and Department of Biophysics, Johns Hopkins University, Baltimore, Maryland 21218, USA Adhesive cell-substrate interactions are crucial for cell motility and are responsible for the neces- sary traction that propels cells. These interactions can also change the shape of the cell, analogous to liquid droplet wetting on adhesive substrates. To address how these shape changes affect cell migration and cell speed we model motility using deformable, 2D cross-sections of cells in which adhesion and frictional forces between cell and substrate can be varied separately. Our simulations show that increasing the adhesion results in increased spreading of cells and larger cell speeds. We propose an analytical model which shows that the cell speed is inversely proportional to an effective height of the cell and that increasing this height results in increased internal shear stress. The numerical and analytical results are confirmed in experiments on motile eukaryotic cells. INTRODUCTION Migration of eukaryotic cells plays an important role in many biological processes including development[1], chemotaxis[2], and cancer invasion[3]. Cell migration is a complex process, involving external cues, intra-cellular biochemical pathways, and force generation. The adhesive interaction between cells and their extracellular environ- ment is an essential part of cell motility [4] and is generally thought to be responsible for frictional forces necessary for propulsion [5, 6]. These frictional forces are due to the motion of the cytoskeleton network and can be measured by traction force microscopy [7]. On the other hand, adhesive cell-substrate interaction can also lead to cell spreading in both moving and non-moving cells [8 -- 10]. This is similar to the spreading of a liquid droplet during the wetting of an adhesive substrate. The resulting changes of the cell shape can potentially affect cell motility. Experimentally, it is not possible to decouple the effect of adhesion and friction, making it challenging to quantify the relative importance of spreading in cell motility. Here we investigate the dependence of motility on cell-substrate adhesion using a mathematical model in which we can alter the adhesive forces independent of frictional forces. We carry out numerical simulations of this model using the phase field approach, ideally suited for objects with deforming free boundaries [11, 12]. We focus on a 2D vertical cross-section of a migrating cell which captures both cell-substrate interactions and internal fluid dynamics [13]. Our adhesive interactions are based on the phase-field description of wetting [14, 15] and are independent of the molecular details of cell-substrate adhesion. Our simulations, together with an analytical 2D model extended from a previous 1D model [16], generate several nontrivial and testable predictions which are subsequently verified by experiments using motile Dictyostelium discoideum cells. RESULTS Model Our vertical cross-sectional model cell captures the interaction of the cell with the bottom and, possibly, top substrate, as well as the interior of the cell [13] (Fig. 1). This is in contrast to most computational studies of cell motility which model a flat cell that is entirely in contact with the substrate[17 -- 20]. This interior consists of a viscous cytoskeleton and is described as a compressible actin fluid [21] with constant viscosity while cell movement is driven by active stress, located at the front of the cell. Note that we do not consider myosin-based contraction. Furthermore, and following Ref. [21], we neglect the coupling between the actin fluid (representing the cytoskeleton) and the cytoplasm. The latter is assumed to be incompressible, resulting in volume conservation. This type of model which treats the cytoskeleton as an active viscous compressible fluid has been used in several recent studies [22 -- 26]. Friction is caused by the motion of the cytoskeleton relative to the substrate and is taken to be proportional to the actin fluid velocity. To accurately capture cell shape and its deformations, we use the phase field approach in which an auxiliary field φ(r, t) is introduced to distinguish between the interior (φ = 1) and exterior (φ = 0). This approach 9 1 0 2 b e F 4 ] h p - o i b . s c i s y h p [ 2 v 5 1 3 1 0 . 6 0 8 1 : v i X r a 2 FIG. 1: Schematic illustration of a model cell on a substrate. The cross-section of the cell is represented by a phase field φ while the substrate is defined by a field χ. The dynamics of the cytoskeleton network is modeled as an actin fluid with velocity u (red arrow). Forces in the model include the membrane tension, cell-substrate adhesive forces, forces due to active actin polymerization, and cytosolic viscous forces (proportional to u). Actin polymerization is restricted to a narrow region near the substrate at the cell front-half, as indicated by the yellow band. Additional model details are given in main text and in Supplemental Material. allows us to efficiently track the cell boundary which is determined by φ(r, t) = 1/2 [12, 20, 24, 25, 27, 28]. In our model, boundary motion is driven by fluid flow which is determined by adhesion, friction, membrane forces and active protrusion. The cell is placed on a substrate which is parallel to the x direction, and polarized in one direction. As described in earlier work [24, 25, 29], the evolution of the cell's shape is determined by the phase field dynamics: ∂φ(r, t) ∂t = −u · ∇φ(r, t) + Γ(∇2φ − G(cid:48)/ + c∇φ), (1) where the advection term couples the velocity field of the actin fluid, u, to the phase field, is the width of the boundary, Γ is a relaxation coefficient, G is a double-well potential with minima at φ = 1 and φ = 0, and c is the local curvature of the boundary (see Supplemental Material). The actin fluid velocity field is determined by the stationary Stokes equation with an assumption of perfect com- pressibility (zero pressure and neglecting the inertial term because of low Reynolds number) [21, 30]: ∇ · [νφ(∇u + ∇uT )] + F sub + F mem + F area + ∇ · σa = 0, (2) where ν is the viscosity of the cell and where σa is the active stress due to actin polymerization, further detailed below. F sub is the interaction between the cell and substrate and contains both adhesion and friction, F sub = F adh + F f ric. The adhesive force is given by F adh = δH(φ,χ) δφ ∇φ, with the cell-substrate interaction potential: (cid:90) H(φ, χ) = dr2φ2(φ − 2)2W (χ). Here, χ(r) is a constant field which marks the substrate (or ceiling) and continuously changes from χ = 1 (within the substrate) to χ = 0 (out of substrate; Fig. S1). W (χ) is a potential with a negative adhesion energy per unit length controlled by a parameter A such that larger values of A represent a larger adhesive force between cell and substrate. In addition, this potential contains a short-range repulsion that ensures that the cell does not penetrate the substrate. The term φ2(φ − 2)2 is added to ensure that the force peaks within the boundary and vanishes at φ = 0 and φ = 1. The second term in F sub describes the frictional force between the cell and the substrate. Depending on the cell type, these forces can arise from focal adhesions or from non-specific cell-substrate interactions. For simplicity, the frictional force in our cross-sectional model is modeled as a viscous drag proportional to the actin fluid velocity : F f ric = −ξsχu − ξdu, where the first term is the cell-substrate friction, parameterized by the coefficient ξs, and the second term represents a damping force, introduced to increase numerical stability. We have verified that the cell speed changes little when we vary the drag coefficient ξd (Fig. S2). Initially, we will vary both the adhesion energy (which controls spreading) and the frictional drag separately, allowing us to determine its relative contribution to cell motility. We will then Membrane tensionActin polymerizationAdhesionActin fluid φSubstrate χMoving directionu3 FIG. 2: a, Cell shapes for different values of substrate adhesion strength. The blue dots here, and elsewhere, schematically indicate the location of active stress. Scale bar 5µm. b, Cell speed (blue circles) and effective height of a cell (red triangles) as a function of the adhesion strength. examine model extensions which implement dependent adhesion and friction mechanisms (see Fig. 4). The uniform membrane tension F mem and a force arising from cell area conservation F area are introduced as in our previous work [25, 27]. The latter force results in cell shapes with roughly constant area. More details of these forces, along with details of the simulation techniques for Eqns. (1&2) are given in Supplemental Material. As a consistency check, we have simulated cells without any propulsive force and have verified that the resulting static shapes agree well with shapes obtained using standard energy minimization simulations [3] (Fig. S3). Polarization in our model is introduced through the polarization indicator ρa which is steering the actin polymer- ization. For simplicity, we have chosen ρa = 1 at the front half and ρa = 0 at the rear half of the cell. This corresponds to different actin promoter (e.g., Rac or Cdc42) distributions at the front and back induced by internal or external signaling. We assume the density of newly-made actin filaments is uniform at the cell membrane so we do not track the evolution of the actin density. In more complicated models[24, 25] the actin can diffuse or be advected but we do not include them here to keep the model simple. Following earlier work [13, 32, 33], we assume that protrusions that are generated by actin polymerization only occur at the front of the cell and close to the substrate. Our formulation of the active stress σa incorporates these assumptions. Specifically, we introduce a field ψ(r) with width λ and located a distance away from the substrate (Fig. S1). By making the active stress proportional to G(ψ)φρa(r), we restrict possible protrusions to a narrow band parallel to the substrate and in the cell front. This band is schematically shown in yellow in Fig. 1. In addition, we localize the stress to the interface by multiplying the expression of σa by the factor ∇φ2. This is schematically shown as blue dots in Figs. 2 and 3. The expression for the stress is then given by: (3) Here, ηa is the protrusion coefficient, and n = ∇φ/∇φ is the normal to the cell boundary. Note that our model does not include any possible feedback between substrate and stress generation. σa = −ηaG(ψ)φρa(r)∇φ2 n n. Our simulations are carried out as described previously [25] and further detailed in the Supplemental Material where we also list the full set of equations. As initial conditions, our simulations start with polarized cells in which the distribution of ρa is asymmetric. The cell's speed is tracked by vc = dxc/dt with xc the cell mass center xc =(cid:82) xφd2r/(cid:82) φd2r and simulations are continued until a steady state has been achieved. Parameters values used in the simulations are given in Table S1. Simulation results and analysis We first investigate how cells move on a single substrate with different adhesion energies. For this, we solve the phase field equations for different values of the adhesion parameter A. Examples of resulting cell shapes are shown in Fig. 2 while an example of the actin fluid velocity field is shown in Fig. S4. We find that with increasing adhesion strength, cells spread more and thus become thinner, similar to the spreading of a droplet on surfaces with increasing 0204000.10.20.3357baAdhesion 10 pN20 pN30 pNAdhesion (pN)Cell Speed (μm/s)Effective Height (μm)wettability (Fig. 2a). Our simulations reveal that the cell speed (i.e., the velocity parallel to the substrate) keeps increasing as the adhesion increases, without any indication of saturation (Fig. 2b). This is perhaps surprising, as our physical intuition suggests that adhesion and friction go hand in hand, with larger adhesion corresponding to higher friction. In our simulations, however, adhesion and friction are independent and can be separately adjusted. To provide insights into the relation between adhesion, cell shape and speed, we consider a simplified version of Eq. (2), similar to the 1D model examined in Ref. [16]. Since only asymmetric stress will contribute to the cell's speed [34], we only need to take into account the viscosity, friction and active stress in the equation: 4 (4) where σvis = ∇u + ∇uT , ξ is a friction coefficient taken to be spatially homogeneous, and σa is the active stress which is 0 outside the cell. Boundary conditions include a steady cell shape n · vc = n · u, zero net traction force (cid:82) ξud2r = 0, and zero parallel stress σvis · t = 0, where n, t are the normal and tangential unit vector, respectively. ν∇ · σvis − ξu + ∇ · σa = 0, It is in general not possible to solve Eq. 4 in a arbitrary geometry. However, for the special case of a fixed-shape rectangular cell with length L and height H occupying x ∈ [−L/2, L/2], y ∈ [0, H] we can solve for the cell speed vc (see the Supplemental Material). By averaging the stress over the vertical direction and following Carlsson's one-dimensional solution [16], we find: (cid:90) L/2 −L/2 vc = − 1 4νH σa xx sinh(κx) sinh(κL/2) dx, where κ =(cid:112)ξ/(2ν) determines the spatial scale of the decay of a point stress source [16] and where σa to the normalized active stress dipole 1/(LH) ×(cid:82) xσxxdx. xxdy (see also the Supplemental Material). From this solution it is clear that asymmetric active stress distribution will lead to cell motion. When κL/2 (cid:28) 1, corresponding to a highly viscous cytoskeleton [35], the speed is proportional In the phase field model, the active stress in Eq. 3 is a negative bell shape function located at the front tip of the cell. This active stress can be approximated by σxx = −λβδ[x− (L/2)−] where β is the active stress strength and where the stress is assumed to be located just inside the cell (see the Supplemental Material and Ref. [16]). Substituting this into Eq. 5, we find xx =(cid:82) H 0 σa (5) vc = λβ 4νH , (6) which shows that the cell speed scales inversely with the height of the cell, and that this scaling is independent of the cell length. Of course, a real cell will not be rectangular, and in the Supplemental Material we show that the cell speed scales with the average height for a more complex-shaped cell (Fig. S5). This suggests that the cell speed can be parameterized using an effective height Heff, which can be computed by averaging the height over the cell length: Heff = (1/L)(cid:82) H(x)dx. In Fig. 2b we see that Heff is monotonically decreasing when adhesion increases. The inverse relation between cell speed and effective height qualitatively agrees with the above analysis. Interestingly, the above found relation between cell speed and cell height does not depend on the way the cell's effective height is altered. To verify this, we also simulated cells in confined geometries in which they are "squeezed" between two substrates, as shown in Fig. 3a (an example of a cell with the actin fluid velocity field can be found in Fig. S4). Consistent with our analytical results, we find that as the chamber height is reduced, the cell's speed increases while the cell's effective height decreases (Fig. 3b). Furthermore, changing the adhesive strength on the top and bottom substrate while keeping the distance between them fixed will also affect the cell shape and its effective height (Fig. 3c and Fig. S4). Our simulations show that a difference in the top and bottom adhesion leads to an asymmetric cross-section and that the cell's effective height reaches a maximum for equal top and bottom adhesion (Fig. 3d). Consistent with Eq. 6, our simulations show that the cell speed reach a minimum for substrates with equal adhesive strengths (Fig. 3d). Our results can be explained by realizing that cells contain a cytoskeleton network that can be described as a compressible viscous actin fluid. This actin fluid contains "active" regions which are confined to a layer with fixed width of λ, and "passive" regions that are outside these active regions. Active stress is only generated within this active region. Large viscosity will make the cell speed independent of cell length (see Eq. 5 and Ref. [16]). However, this viscosity also leads to dissipation due to internal shear stress: passive regions are coupled to the active regions through vertical shear interactions, resulting in dissipation. This dissipation increases with increasing cell height, as can also be seen in the velocity profile shown in Fig. S6, and thinner cells will move faster. We have tested this explanation by carrying out additional simulations. In one set, we simulated cells moving in chambers of varying height while keeping the ratio of the size of the active stress layer λ and cell height constant. Consistent with our 5 FIG. 3: a. Cell shapes for different chamber heights. The adhesion strength of the top and bottom substrate is fixed at 30 pN. Scale bar: 5µm. b. Corresponding cell speed (blue circles) and effective height (red triangles) as a function of chamber height. c. Cell shapes in a chamber with adhesive top and bottom substrates, with the top substrate adhesion fixed to 30 pN. Scale bar: 5µm. d. Cell speed (blue circles) and effective height (red triangles) as a function of adhesion strength of the bottom substrate (chamber height=6µm). FIG. 4: a, Cell crawling speed dependence on adhesion strength and friction coefficient (normalized by ξs =1 Pa s/µm). Cell speed is visualized using the colormap. The dashed lines correspond to different dependencies of the friction on adhesion (white: constant friction, black: linear dependence, red: exponential dependence). b, Cell speed for unconfined cells as a function of adhesion for linear (black line) and exponential (red line) dependence on adhesion (for parameters see main text). c, Speed of confined cells as function of bottom substrate adhesion strength for linear (black line) and exponential (red line) dependence of friction on adhesion (top substrate adhesion=30 pN, chamber height=6µm). theoretical predictions, the speed of these cells is independent of the chamber height (Fig. S7). In addition, we have simulated cells in which the active stress region spans the entire front. Again in line with our theoretical insights, the cell speed was found to be largely independent of the chamber height (Fig. S8). In our simulations, we have kept the friction coefficient constant and have thus ignored any potential link between adhesion and friction. This is likely appropriate for Dictyostelium cells but may not be valid for mammalian cells that have integrin mediated focal adhesions. The exact dependence of friction on adhesion is complicated and poorly understood [36, 37]. Our model, however, can easily be extended to explore the entire phase space of friction and adhesion. To illustrate this, we compute the speed of a cell crawling on a single substrate by sampling a broad range of adhesion strengths (A = 10 pN to A = 40 pN) and friction coefficients (ξs = 1 Pa s/µm to ξs = 103 Pa s/µm) while keeping all other parameters fixed. The resulting cell speeds are shown in Fig.4a using a color map. As expected, cells stall when adhesion is low and friction is high (dark blue region) while the highest cell speed occurs for large adhesion and a relatively broad range of low friction values (yellow region). Different dependencies between friction and adhesion correspond to different trajectories through the two- dimensional phase space of Fig. 4a. The results we have presented so far correspond to traversing the phase space along the white dashed line in Fig. 4a. The black dashed line in this figure, on the other hand, represents a linear dependence between friction and adhesion (ξs = ξb + ξlA/Al with ξb =1 Pa s/µm, ξl =5 Pa s/µm, and Al =1 pN) 030600.10.140.184567468100.050.10.150.2468Chamber Height (μm)Cell Speed (μm/s)Effective Height (μm)b4 μm6 μm 8 μmaChamber HeightTop 30 pNcBottom 0 pNBottom 30 pNBottom 60 pNCell Speed (μm/s)Effective Height (μm)Bottom Adhesion (pN)d020406000.050.10.15102030400.020.060.1ExpLinExpLin10203040012300.10.2Cell speed (μm/s)Adhesion (pN)Adhesion (pN)Log10(ξs, norm )abμm/scAdhesion (pN)Cell speed (μm/s)UnconfinedConfined6 FIG. 5: Experimental tests of the numerical and theoretical predictions. a. Schematic side view of the microfluidics chamber. Cells are placed in a confined chamber with variable height. The top substrate is composed of PDMS while the bottom substrate is either composed of PDMS or coated with less adhesive PEG. Cells are guided by a chemoattractant (cAMP) gradient of strength 0.45 nM/µm in the chamber. b. Cell speed in the gradient direction for varying chamber height, indicated along the x-axis, and top/bottom substrate composition, indicated by the label. For both PDMS/PDMS and PDMS/PEG substrate compositions, the cell speed decreases as chamber height is increased. Furthermore, for fixed chamber height, cells move faster when the bottom substrate is less adhesive (i.e., PDMS/PEG). P-value < 10−5 with unpaired t-test. Error bars represent the standard error of the mean. c. Scan of the cell area profile at the top and bottom of the chamber using the fluorescent membrane marker Car1-RFP (h = 5µm). Cells with PEG-coated bottom substrates (PDMS/PEG) show asymmetric shapes whereas cells with PDMS bottom substrates (PDMS/PDMS) show symmetric shapes. Scale bar, 10µm. d. Ratio of top and bottom contact area under different conditions (label indicates bottom substrate composition). P-value < 10−5 with Wilcoxon rank-sum test. while the red dashed line represents an exponential dependence (ξs = ξb +ξe exp (A/Ae) with ξb =1 Pa s/µm, ξe =1 Pa s/µm, and Ae =7 pN). For these two adhesion-friction dependencies, we have computed the cell speed for unconfined (Fig.4b) and confined cells (Fig.4c). For friction that depends linearly on adhesion, the speed of unconfined cells continues to increase as adhesion increases (black line in Fig.4b). This is very similar to the results we obtained for constant friction (cf. Fig. 2b). For exponential friction, the speed of unconfined cells initially increases for increasing adhesion. As adhesion increases, however, friction becomes more and more dominant, and cell's speed reaches a maximum, followed by a decrease (red line in Fig.4b). This bi-phasic dependence of adhesion is consistent with a variety of experiments[22, 38 -- 40]. For confined cells and a linear friction-adhesion relationship, the dependence of the cell speed on the adhesive strength of the bottom substrate is shown in Fig. 4.c (black line). Again, the results are very similar to our previously studied, constant friction case (cf. Fig.3d): cell speed reaches a minimum when the top and bottom adhesion strength are equal. Not surprisingly, the dependence of cell speed on bottom adhesion is different for the exponential relationship. Here, friction becomes dominant when adhesion increases, resulting in a cell speed that continuously decreases. These results show that friction plays a relatively small role in determining cell speed unless friction ξs increases over orders of magnitude when adhesion A changes by small amounts. 5μm7μm10μm00.51abcd Bottom area / top area*******PDMS/PDMSPDMS/PEG1.55μm7μm10μm02468PDMS/PDMSPDMS/PEG*******Vx (μm/min)1012GlassPDMSPEG or PDMSSide viewgradient chamberchemo-attractantno chemo-attractanthVxPDMSTop (z=h)Bottom (z=0)PDMSPEGPDMSExperimental results 7 To test the above predictions, we performed motility experiments of Dictyostelium discoideum cells. Importantly, these cells, unlike mammalian cells, do not make integrin mediated focal adhesions and their substrate adhesion is likely to be mediated by direct physiochemical factors such as van der Waals attraction [41]. Experiments are carried out in microfluidic devices, as shown in Fig. 5a and modified from earlier work [42] (see also Supplemental Material and Fig. S9). Cells are moving in chambers with height h and with substrates that have variable adhesive properties. A constant cAMP gradient is established by diffusion so that cells preferably move in one direction (denoted as the x direction). Note that the constant signal polarizing the cell in one direction is consistent with our model of a constantly-polarized cell. Dictyostelium cells move by extending actin filled protrusions called pseudopods which can extend over a significant distance from the substrate. As a consequence, our confined cells occlude the entire space between two substrates. This was verified explicitly by labeling the cell with a fluorescent membrane marker and creating confocal z-stacks (Fig. S10). The results also demonstrate that in the case of symmetric adhesion the outline of the cell does not change appreciably as one moves from one to the other substrate. Furthermore, using LimE as a fluorescent marker, we have verified that the level of actin polymerization is largest near the substrates (Fig. S10 C). This observation is in agreement with earlier experiments of Dictyostelium cells migrating in a narrow channel [33] which revealed significantly larger levels of LimE fluoresence near the channel walls. The top substrate of the chamber consists of Polydimethylsiloxane (PDMS) and the bottom substrate is either made of PDMS or is coated with a thin layer of Polyethylene glycol (PEG) gel. Cells moving on these PEG-coated substrates have vastly reduced adhesion, as reported in earlier studies [43]. We measure the average speed of the cell Vx in the direction of the chemoattractant gradient, both as a function of the height of the chamber and for different substrate compositions (Fig. 5b). Furthermore, to quantify the effect of the adhesive properties of the substrates on migrating cells, we measure the contact area of the cell on both top and bottom substrates of the chamber using confocal microscopy (Fig. 5c and d). More adhesive substrates will result in more cell spreading and thus larger contact areas. Our theoretical predictions for cells in confinement are that decreased height increases speed, and that cells in asymmetric adhesion are faster than cells in symmetric adhesion. Both of these qualitative predictions are observed in our experiments. First, our experiments show that cell speed is significantly affected by the height of the chamber (Fig. 5b). Cells in chambers of height h = 10µm move markedly slower than cells in chambers with h = 7µm which, in turn, have smaller speed than cells in chambers with h = 5µm. The trend of slower motion in deeper chambers holds for both PDMS and PEG coated bottom substrates. Furthermore, we have verified that these results do not depend on the steepness of the gradient (Fig. S11). These observations are fully consistent with our numerical and theoretical predictions (Fig. 3). In addition, our experiments show that cells moving in a chamber with unequal top and bottom adhesion are markedly asymmetric (Fig. 5c), consistent with past results that showed that Dictyostelium cells only weakly adhere to PEG. Specifically, the contact area of cells on PEG coated substrates is significantly smaller than the contact area on PDMS substrates and the resulting asymmetry can be quantified by the ratio of bottom and top contact area. Cells with PDMS on top and bottom and for h =5µm and h =7µm have ratios close to 1 indicating that the shape is symmetric. In contrast, cells moving in chambers with these values of h that have a PEG bottom have ratios that are much smaller than 1, indicating a more asymmetric cell shape. For the largest value of h (h =10µm) cell preferentially attach to the top PDMS substrate, resulting in negligible contact area at the bottom PEG substrate and ratios close to 0. For this chamber height, the ratio for PDMS substrates is larger than one since cells are loaded on the bottom substrate and cannot fully attach to the top substrate. Importantly, quantifying the cell speed for the different chambers reveals that cells in the symmetric PDMS/PDMS condition move slower than cells in the asymmetric PDMS/PEG condition (Fig. 5b). Again, these experimental results are fully consistent with our theoretical and numerical predictions and show that cell shape, and more specifically its effective height, can significantly affect motility speed (Fig. 3). DISCUSSION AND CONCLUSION In this study, we examined how cell shape can affect cell speed using simulations, analytics, and experiments. We should stress that our experiments can only be compared to the simulations on a qualitative level. Values for the model parameters are not precisely known, and our model cell is not fully three-dimensional. Nevertheless, separating the frictional and adhesive force in the model provides clear insights into the role of adhesion and cell shape in determining 8 cell speed. This separation also makes it challenging to compare our results to previous studies that investigated the effects of cell-substrate interactions on cell speed. For example, a recent study using fish keratocyte cells [22] found that cell spreading increases with adhesion strength (measured by the concentration of adhesive molecules). These experiments also revealed a biphasic speed dependence on adhesion such that cell speed increases between low and intermediate adhesion strengths and decreases between intermediate and high adhesion strengths. These results are similar to earlier experimental studies, and have previously been interpreted in terms of minimal models without cell shape [39, 44, 45]. Our results suggest that the increase of cell speed with increased adhesion found in these experiments might be attributed to cell spreading and a lower effective height. The observed decrease in cell speed following a further increase in adhesion can then be explained by a larger relative role of frictional forces. Likewise, our experimental results suggest that our experiments operate in a regime where substrate friction is less important than the internal viscosity and hence the major effect of the substrate modification is the change in adhesion. Our numerical and experimental results indicate that changing cell morphology through confinement can also significantly alter the migration speed, with decreasing chamber heights resulting in increased cell speeds. Comparison with other cell types is challenging as cells might change their behavior following confinement. A recent study using normal human dermal fibroblast cells, for example, found that slow mesenchymal cells can spontaneously switch to a fast amoeboid migration phenotype under confinement [45]. This phenotypic transition makes it difficult to directly compare those observations with our results and further investigation is needed to determine how different cell types behave in confinement. We should point out that the simple scaling of cell speed dependence on cell height (Eq. 6) is based on the assumption of localized active stress (the numerator) and uniform cytoskeleton viscosity (the denominator) in the entire cross section. As shown in our experimental work and in previous studies [33], F-actin is localized close to the substrate, in support of the first assumption. It is currently unclear whether the second approximation is valid for Dictyostelium or other cells. Presumably, in cells with a clear segregation of actin cortex and cytoplasm, a large viscosity contrast could be present. Nevertheless, our arguments might still hold, as long as passive regions are coupled to the active regions through shear interactions (one example is the model in Ref.[32]). In this case, passive regions will still slow down the cell, but the relation between cell speed and shape will be more complicated and will have contributions from regions with different viscosity. To address this more general case, a full three-dimension model with viscosity contrast between the cortex and cytoplasm is necessary and will be part of future extensions. In summary, we show how adhesion forces result in cell spreading and that the accompanying shape changes can result in larger velocities. Key in this result is the existence of a narrow band of active stress that has a smaller spatial extent than the height of the cell. As a result, the dissipation due to the shear stress between this active band and the remainder of the cell increases as the effective height of the cell increases. In our model, we have assumed a cell motility model corresponding to stable flat protrusions. The conclusion that cell speed scales inversely with the effective height is also valid for other cell motility models as long as the active propulsion region has limited spatial extent. For example, replacing the constant active stress by an oscillating stress, similar to protrusion-retraction cycles seen in amoeboid cells, does not change the qualitative results (Fig. S12). Further extensions of our model could include focal adhesive complexes (to model a broader range of eukaryotic cell types) and different types of actin structures in different parts of the cell. These extensions can then be used to further determine the role of adhesion in cell motility. ∗ These two authors contributed equally. † Electronic address: [email protected] [1] A. Munjal and T. Lecuit, Development 141, 1789 (2014). [2] V. Kolsch, P. G. Charest, and R. A. Firtel, J Cell Sci 121, 551 (2008). [3] D. Wirtz, K. Konstantopoulos, and P. C. Searson, Nature Reviews Cancer 11, 512 (2011). [4] B. Geiger, J. P. Spatz, and A. D. Bershadsky, Nature reviews Molecular cell biology 10, 21 (2009). [5] G. Charras and E. Sahai, Nature reviews Molecular cell biology 15, 813 (2014). [6] C. E. Chan and D. J. Odde, Science 322, 1687 (2008). [7] A. K. Harris, P. Wild, and D. Stopak, Science 208, 177 (1980). [8] T. Frisch and O. Thoumine, Journal of biomechanics 35, 1137 (2002). [9] K. Keren, Z. Pincus, G. M. Allen, E. L. Barnhart, G. Marriott, A. Mogilner, and J. A. Theriot, Nature 453, 475 (2008). [10] C. A. Reinhart-King, M. Dembo, and D. A. Hammer, Biophysical journal 89, 676 (2005). [11] J. Kockelkoren, H. Levine, and W.-J. Rappel, Physical Review E 68, 037702 (2003). [12] F. Ziebert and I. S. Aranson, PloS one 8, e64511 (2013). [13] E. Tjhung, A. Tiribocchi, D. Marenduzzo, and M. Cates, Nature communications 6, 5420 (2015). 9 [14] Y. Zhao, S. Das, and Q. Du, Physical Review E 81, 041919 (2010). [15] W. Mickel, L. Joly, and T. Biben, The Journal of chemical physics 134, 094105 (2011). [16] A. Carlsson, New journal of physics 13, 073009 (2011). [17] A. St´ephanou, E. Mylona, M. Chaplain, and P. Tracqui, Journal of theoretical biology 253, 701 (2008). [18] A. Mogilner, Journal of mathematical biology 58, 105 (2009). [19] M. Buenemann, H. Levine, W. J. Rappel, and L. M. Sander, Biophys J 99, 50 (2010). [20] S. Alonso, M. Stange, and C. Beta, PloS one 13, e0201977 (2018). [21] B. Rubinstein, M. F. Fournier, K. Jacobson, A. B. Verkhovsky, and A. Mogilner, Biophysical journal 97, 1853 (2009). [22] E. L. Barnhart, K.-C. Lee, K. Keren, A. Mogilner, and J. A. Theriot, PLoS biology 9, e1001059 (2011). [23] J. S. Bois, F. Julicher, and S. W. Grill, Physical review letters 106, 028103 (2011). [24] D. Shao, H. Levine, and W.-J. Rappel, Proc Natl Acad Sci U S A 109, 6851 (2012). [25] B. A. Camley, Y. Zhao, B. Li, H. Levine, and W.-J. Rappel, Physical Review Letters 111, 158102 (2013). [26] T. L. Goff, B. Liebchen, and D. Marenduzzo, arXiv preprint arXiv:1712.03138 (2017). [27] D. Shao, W.-J. Rappel, and H. Levine, Physical Review Letters 105, 108104 (2010). [28] S. Najem and M. Grant, Physical Review E 88, 034702 (2013). [29] T. Biben, K. Kassner, and C. Misbah, Physical Review E 72, 041921 (2005). [30] J. S. Bois, F. Julicher, and S. W. Grill, Physical review letters 106, 028103 (2011). [31] K. A. Brakke, Experimental mathematics 1, 141 (1992). [32] K. Kruse, J. Joanny, F. Julicher, and J. Prost, Physical biology 3, 130 (2006). [33] O. Nagel, C. Guven, M. Theves, M. Driscoll, W. Losert, and C. Beta, PloS one 9, e113382 (2014). [34] H. Tanimoto and M. Sano, Biophysical journal 106, 16 (2014). [35] A. R. Bausch, F. Ziemann, A. A. Boulbitch, K. Jacobson, and E. Sackmann, Biophys. J. 75, 2038 (1998). [36] M. Srinivasan and S. Walcott, Physical Review E 80, 046124 (2009). [37] S. Walcott and S. X. Sun, Proc Natl Acad Sci U S A 107, 7757 (2010). [38] A. Huttenlocher, M. H. Ginsberg, and A. F. Horwitz, The Journal of cell biology 134, 1551 (1996). [39] S. P. Palecek, J. C. Loftus, M. H. Ginsberg, D. A. Lauffenburger, A. F. Horwitz, et al., Nature 385, 537 (1997). [40] M. L. Gardel, B. Sabass, L. Ji, G. Danuser, U. S. Schwarz, and C. M. Waterman, J cell Biol 183, 999 (2008). [41] W. F. Loomis, D. Fuller, E. Gutierrez, A. Groisman, and W.-J. Rappel, PloS one 7, e42033 (2012). [42] M. Skoge, H. Yue, M. Erickstad, A. Bae, H. Levine, A. Groisman, W. F. Loomis, and W.-J. Rappel, Proc Natl Acad Sci U S A 111, 14448 (2014). [43] T. Tzvetkova-Chevolleau, E. Yoxall, D. Fuard, F. Bruckert, P. Schiavone, and M. Weidenhaupt, Microelectronic Engi- neering 86, 1485 (2009). [44] P. A. DiMilla, K. Barbee, and D. A. Lauffenburger, Biophys. J. 60, 15 (1991). [45] Y.-J. Liu, M. Le Berre, F. Lautenschlaeger, P. Maiuri, A. Callan-Jones, M. Heuz´e, T. Takaki, R. Voituriez, and M. Piel, Cell 160, 659 (2015). [46] B. A. Camley, Y. Zhang, Y. Zhao, B. Li, E. Ben-Jacob, H. Levine, and W.-J. Rappel, Proc Natl Acad Sci U S A 111, 14770 (2014). [47] Y. Zhao, Y. Ma, H. Sun, B. Li, and Q. Du, arXiv preprint arXiv:1712.01951 (2017). [48] K. Keren, Z. Pincus, G. M. Allen, E. L. Barnhart, G. Marriott, A. Mogilner, and J. A. Theriot, Nature 453, 475 (2008). [49] H. Levine and W.-J. Rappel, Phys Today 66 (2013), 10.1063/PT.3.1884. [50] D. Fuller, W. Chen, M. Adler, A. Groisman, H. Levine, W.-J. Rappel, and W. F. Loomis, Proc Natl Acad Sci U S A 107, 9656 (2010). [51] M. Sussman, in Methods in cell biology, Vol. 28 (Elsevier, 1987) pp. 9 -- 29. [52] S. Paliwal, P. A. Iglesias, K. Campbell, Z. Hilioti, A. Groisman, and A. Levchenko, Nature 446, 46 (2007). [53] M. Skoge, M. Adler, A. Groisman, H. Levine, W. F. Loomis, and W.-J. Rappel, Integrative Biology 2, 659 (2010). 1 (S1) Supplemental Material for "Cell Motility Dependence on Adhesive Wetting" PHASE FIELD MODEL OF CELL MOTILITY The equations for the phase-field cross section model are: ∂φ(r, t) = −u · ∇φ(r, t) + Γ(∇2φ − G(cid:48)/ + c∇φ) ∂t (S2) Here, φ describes the field of the cell. The double-well potential is defined as G = 18φ2(1 − φ)2 and the curvature is computed as c = −∇ · (∇φ/∇φ) while Γ is a relaxation coefficient. The force terms are explicitly explained below. ∇ · [νφ(∇u + ∇uT )] + F sub + F mem + F area + ∇ · σa = 0. The substrate force contains the cell-substrate adhesion and friction: F sub = F adh + F f ric, where F f ric = −ξsχu − ξdu, F adh = δH(φ, χ) δφ ∇φ. Here, u is the velocity field of the actin fluid and ξs, ξd are the cell-substrate friction coefficient and damping coefficient, respectively. χ is the field describing the substrate, and H(φ, χ) is the interaction potential between the cell and substrate. The The cell moves either on top of a plain substrate or between a top and bottom substrate. The location of these substrates is given by a field χ(y) with a boundary width of δ (Fig. S1). Here, χ = 1 indicates the substrate into which the cell cannot penetrate, and χ = 0 indicates the region accessible to the cell. In our simulations, the substrate is parallel to the x direction and, for the case of a single substrate located at y = yB, χ is written as For a chamber with a parallel top substrate located at y = yT this becomes χ(y) = 1 2 − 1 2 tanh{3(y − yB)/δ}, χ(y) = 1 2 + 1 2 tanh{3[y − (yT + yB)/2 − (yT − yB)/2]/δ}. Given φ and χ, the interaction potential is: H(φ, χ) = (cid:90) dr2φ2(φ − 2)2W (χ), where W (χ) contains an attractive term, corresponding to adhesion, and a repulsive term, corresponding to the non-penetrability of the substrate. For the bottom substrate, we use (S3) while the potential for the top substrate has an identical form with replaced by −. Here, A is the adhesion energy per unit length, g is a parameter that measures the penalty of overlap between cell and substrate [S1], and G is a double-well potential G(χ) = 18χ2(1 − χ)2. The energy function χ(y + ), + δ W (χ) = −2A G(χ) g 2 (cid:90) (cid:90) γ H(φ, χ) = φ2(φ − 2)2W (χ)d2r corresponds, in the sharp interface limit, to an adhesive energy equal to −Al where l is the length of the cell in contact with the substrate. Note that the inclusion of the φ2(φ−2)2 results in a force that only vanishes outside the membrane [S2]. In our simulations we take δ = /2. For this choice of δ we simulated cells without any propulsive force. The resulting static shapes can be directly compared to standard energy minimization simulations. Fig. S3 shows that the phase field shapes agree well with shapes obtained using Surface Evolver, a simulation tool that evolves surfaces toward minimal energy by a gradient descent method [S3]. The contribution from both the tension and bending of the membrane is captured by F mem. In our simulation we ignore the bending term since it contributes little to the shape of cell. The tension energy is given by [S4, S5]: Hten = [∇φ2 + G(φ) ]d2r, 2 δφ ∇φ. Area conservation is introduced via F area = Ma((cid:82) φdr2 − A0)∇φ with A0 the resulting in F mem = δHten prescribed area size, and Ma a parameter which controls the strength of the area constraint [S4]. The active stress term in our model, σa = −ηaG(ψ)φρa∇φ2 n n, is similar to our earlier work [S6] but only acts near the substrate. This is accomplished through the addition of the term G(ψ) = 18ψ2(1 − ψ)2, where ψ, for the bottom substrate, takes on the form 2 ψB(y) = 1 2 + 1 2 tanh{3[yB + ( + λ/2) − y]/λ}. A similar expression is used for the top substrate. The inclusion of G(ψ) results in active stresses confined to a band with width λ and located a distance away from the substrate (Fig. S1). Note that vertical height of the active stress is controlled by λ and that(cid:82) G(ψ)dy = λ/2. Three examples of the velocity fields obtained numerically are shown in Fig. S4, corresponding to the cell motion on single substrate, confined in channels and confined in channels with asymmetric adhesion (Fig. 2 and Fig. 3 in main text). The retrograde flow patterns are similar to previous studies in[S6]. NUMERICAL METHODS The equation for φ is stepped by uniform time step ∆t = 2 × 10−3s in a forward Euler scheme so that φ at time step n + 1 is obtained from φ at time step n: φ(n+1) = φ(n) − ∆tu + ∆tΓ[∇2φ(n) − G(cid:48)(φ(n)) + c(n)∇φ(n)], Here, c(n) = −∇· (∇φ(n)/∇φ(n)) is computed using a finite difference method and all other differentiation operators are computed using a fast Fourier spectral method. Simulations were carried out on a 256 × 256 grid of size 50µm × 50µm. Model parameters, modified from [S5, S6], are listed in Table S1. The velocity field u is updated every time step by a semi-implicit Fourier spectral method after updating φ as detailed in [S5]. The equation is iterated as: ξ0uk+1 − ν φ∇2uk+1 = ∇ · [νφ∇uk + ν(φ − φ)∇uT k ] − ξsχuk + F , where φ = 2, and F represents the terms in the Stokes equation that are independent of the iteration step k. The iteration will continue until maxuk+1 − uk maxuk < 0.1, or until a maximal number of iterations (here chosen to be 20) is reached. ANALYTICAL RESULTS As stated in the main text, we aim to analytically solve Eq. S1&S2, where several simplifications have to be made. First, we are trying to find the steady-state solutions, so the cell shape will not change with time. Thus we drop Eq. S1 and, instead, put boundary conditions for Eq. S2. In accordance with our simulations, we choose slip boundary conditions, similar to[S7]. The boundary condition for the steady-state cell shape is u · n = (cid:126)vc · n, where (cid:126)vc is the cell's mass of center velocity, which is our target to solve, and n is the normal unit vector of the boundary. The cell's boundary is free so the parallel stress at the boundary is zero t · σvis = 0, where t is the tangential unit vector of the boundary. Notice that the active stress is always constrained inside the cell so it will not enter any boundary conditions. The total force of the cell exerted on substrate should be balanced which gives a zero net traction force condition (cid:90) ξ(r)ud2r = 0, where ξ(r) is the friction coefficient at different locations. To get analytical expressions, we neglect the spatial heterogeneity in friction and simply take ξ(r) = ξ. This simplification does not change the central feature of our main result (the cell's speed is inversely related to the cell's height). Second, we only take into account the viscosity, friction and active stress because they are directly related to the cell motion. The adhesion, area conservation and membrane forces only contribute to the cell's shape, which is implicitly included in the boundary conditions. Thus we get a simplified equation for Eq. S2: 3 Integrating the above equation and using the zero traction force condition, we obtain (cid:72) (νσvis + σa) · ndl = 0. As ν∇ · σvis − ξu + ∇ · σa = 0. (S4) the active stress σa is constrained inside the cell, this will lead to a condition equivalent to the zero traction force condition (cid:73) n · σvisdl = 0, which is the zero traction force condition we used below. Notice that a fixed cell shape has to be given in order to apply the boundary conditions. Since we only care about the cell's mass of center velocity (cid:126)vc, and not the full solution for u, we will next show how (cid:126)vc can be obtained without knowing u. Analytical solution of the rectangular model cell Here we wish to solve the Eq. S4 for a rectangular fixed cell shape x ∈ [−L/2, L/2], y ∈ [0, H] with an unknown cell speed vc (notice we put the x-direction as cell moving direction so vc is a scalar). The boundary conditions are ux(x = ±L/2) = vc, uy(y = 0, H) = 0, and(cid:82) udxdy = 0. Integrating the Stokes equation, we get(cid:82) d2r(ν∇ · σvis + ∇ · σa) =(cid:72) (νσvis · n + σa · n)dl = ξ(cid:82) ud2r = 0. Note that the active stress σa should be constrained within the cell [S7] resulting in the zero net traction force condition (cid:72) σvis · ndl = 0. This means (cid:82) [σvis −L/2)]dy = (cid:82) [∂xuxx=L/2 − ∂xuxx=−L/2]dy = 0 and (cid:82) [σvis xx (x = xy (y = 0)]dx = 0 due to the rectangular The tangential vector t can be determined by the normal vector tx = − ny, ty = nx. The zero-parallel stress xx (x = L/2) − σvis xy (y = H) − σvis shape. condition t · σvis = 0 results in nyσvis xx − nxσvis xy = 0, nyσvis xy − nxσvis yy = 0. For rectangular boundaries, these conditions lead to at all boundaries. σvis xy = 0, (S5) Since the cell is moving along x-direction, only ux is relevant and we can integrate the 2D Stokes equation in the y-direction. With the condition of σvis xy = 0, we obtain a 1D Stokes equation: where ux =(cid:82) H xx =(cid:82) H 2ν ∂2 ux ∂x2 − ξ ux + ∂ σa xx ∂x = 0, (S6) xxdy. The corresponding boundary conditions are ux(L/2) = ux(−L/2) = vcH and ∂x uxx=L/2 = ∂x uxx=−L/2. This is exactly the same problem as in reference [S7]. Using standard Green's function methods, we obtain: 0 uxdy, and σa 0 σa ux(L/2) = − 1 4ν and, since ux = vc at boundaries x = ±L/2, we obtain (cid:90) L/2 −L/2 σa xx sinh(κx) sinh(κL/2) dx, vc = ux(L/2) H , (S7) as reported in the main text (Eq. 5). If the active stress is confined in a band with width λ, i.e.,(cid:82) H 4 0 σa xxdy = λf (x), the cell's speed vc will scale as: (S8) xv − ξv + f(cid:48)(x) = 0 where v0 is a constant, corresponding to the boundary velocity determined by the 1D problem 2ν∂2 with homogeneous boundary conditions. Notice that this scaling does not depend on the vertical position of the active stress. Therefore, our model will give the same cell speed independent of the type of active stress (actin polymerization, myosin contraction), as long as the integrated active stress is the same. vc = , λv0 H Effective height for non-rectangular cells In the above section, the speed of a rectangular cell was determined exactly. Actual cells are, of course, not rectangular but obtaining a solution for cells with more complex shapes is challenging. Nevertheless, insight can be obtained by considering a cell composed of two rectangles, one positioned at [−L, 0] × [0, H2] and one positioned at (0, L] × [0, H1] (H1 < H2 (see Fig. S5). We take the active stress to be located at the latter (right) rectangle. This problem has the same boundary conditions as above, with two additional continuity conditions: ux(x = 0+) = ux(x = 0−), ∂xuxx=0+ = ∂xuxx=0− . 0 uxdy and u2 =(cid:82) H2 To simplify the problem, we introduce the new variables u1 =(cid:82) H1 (cid:90) H1 condition we have: (cid:90) H2 (cid:90) H2 u2(0−) = ux(x = 0−)dy = ux(x = 0−)dy + Together with u1(L) = H1vc, u2(−L) = H2vc we get 0 H1 0 0 uxdy. Using the continuity ux(x = 0+)dy = (H2 − H1)vc + u1(0+). ∂xuxx=−Ldy = ∂xuxx=0dy + 0 = 0. u20−L + u1L (cid:90) H2 (cid:90) H2 H1 (cid:90) H1 (cid:90) H1 0 ∂xuxx=Ldy. The zero traction force will give(cid:90) H2 0 (cid:90) H2 ∂xu20 = such that Combining with the stress continuity we obtain 0 0 H1 ∂xu20−L + ∂xu1L 0 = 0. ∂xuxx=0dy = ( dy + dy)(∂xuxx=0) = ∂xu10 + ∂xu2−L − ∂xu1L. (S9) (S10) (S11) Notice that Eq. S10 and Eq. S11 have clear physical meanings, namely flow conservation and force balance, respec- tively. It is convenient to introduce the net flow C and net force F on each rectangle: 0 = −F, 0 = −C, ∂xu20−L = F, ∂xu1L u20−L = C, u1L and, using the zero-parallel stress condition, we obtain the 1D version of the problem for the right and left rectangle: 2ν∂2 xu2 − ξu2 = 0, 2ν∂2 xu1 − ξu1 + ∂xσa = 0, 0 σxxdy. u1 can be solved by superposition of two parts: u1 with homogeneous boundary conditions and active stress, and u1 with inhomogeneous boundary conditions but zero active stress. After substituting u1 = u1 + u1, we obtain 2ν∂2 x u1 − ξ u1 + ∂xσa = 0, u1L 0 = 0, ∂x u1L 0 = 0, with σa =(cid:82) H1 and 2ν∂2 x u1 − ξ u1 = 0, u1L 0 = −C, ∂x u1L 0 = −F. We can then solve for u1, u1 and u2 and obtain the boundary velocity: H1vc = u1(L) = va − C 2 − αF 2κ , H2vc = u2(−L) = − C 2 + αF 2κ , and the boundary stresses: 5 (S12) ∂xu2−L = − F 2 where κ =(cid:112)ξ/2ν, α = coth(κL/2). va and πa are the boundary speed and boundary stress from the homogeneous , ∂xu1L = πa − F 2 , ∂xu10 = πa + , ∂xu20 = − ακC 2 , F 2 − ακC 2 (S13) + ακC 2 F 2 + ακC 2 equation of u1, which are constants. To calculate vc, we have to determine C and F . Eq. S12 gives one condition H1/H2 = u1(L)/u2(L) and an additional condition from the stresses in Eq. S13 is needed. Unfortunately, there is no simple relation between the four equalities in Eq. S13 since the stress continuity equation cannot be defined at the boundary at x = 0 and between y = H1 and y = H2. Instead, we assume that the ratio of the integrated stress at x = 0 satisfies (∂xu10)/(∂xu20) = β. Then, we have: vc = κvaα2(1 + β) + κva(1 − β) − 2απa κ[α2(1 + β)(H1 + H2) + (β − 1)(H1 − H2)] . (S14) Note that when H1 = H2, corresponding to β = 1, this result gives the same scaling as for the simple rectangular shape. With α (cid:29) 1, corresponding to a highly viscous cytoskeleton, we have vc ≈ va H1 + H2 − 2πa κα(1 + β)(H1 + H2) , (S15) which clearly shows that the cell speed is scaling inversely with the average height (H1 + H2)/2. TEST OF MODEL PREDICTIONS The above analysis indicates that the ratio of the height of stress band λ and the cell height H determines the cell speed. Thus, cells with equal ratio should have similar speeds. To test this explicitly, we simulated cells in chambers with heights varying between h = 4µm and h = 10µm, constraining the cell's height, while keeping the ratio λ/h = 0.5. Cells shapes for three different chamber heights are shown in Fig. S7a while the cell speed and effective height as a function of chamber height are shown in Fig. S7b and c, respectively. Clearly, the results from Fig. S7b show that the speed of the cell is independent of the chamber height, consistent with our model prediction. In addition, our derived expression predicts that if λ = H, corresponding to an active stress region that spans the entire height of the cell, the cell speed should be independent of the chamber height. To verify this, we performed simulations of confined cells with the active stress at the entire cell front. To this end, we no longer constrain the stress to a narrow band and, instead, use σa = −φρa(1 − χ)∇φ2 n n. We introduce the factor of 1 − χ to prevent protrusion in the region where the cell and substrate overlap, something that is excluded from occurring in other models when the band restricts protrusion. Resulting cell shapes for different chamber heights are shown in Fig. S8a. In Fig. S8b, we plot the cell speed as a function of the chamber height and in the Fig. S8 we plot the effective height. As expected, the cell speed changes little as the chamber height is varied, again consistent with our predictions. OSCILLATORY PROTRUSIONS Results in the main text are for cells with constant active stress, resulting in constant cell shapes. Such constant shapes are applicable to fish keratocytes, fast moving cells that maintain their morphology [S8]. Other cell types, how- ever, including neutrophils and Dictyostelium discoideum cells [S9], move in a more time-dependent way, with repetitive and short-lived protrusions called pseudopods. To determine the dependence of cell speed on chamber height for these 6 types of cells we introduce an oscillatory modulation to the active stress: σa = −φρaG(ψ) sin(2πt/T )∇φ2 n n. Here, T is the period of the oscillation cycle which can be varied. Results from additional simulations show that the cell speed gets larger as the substrate adhesion is increased (Fig. S12a). This dependence on adhesion was found to be largely independent of the period and is similar to the one found for model cells with constant stress (Fig. 2b). Also consistent with our results in the main text (Fig. 2c), the effective height is again inversely related to the adhesion strength. EXPERIMENTS Cell culture and preparation Wild type Dictyostelium discoideum (AX4) cells were transformed with a construct in which the regulatory region of actin 15 drives genes encoding a fusion of GFP to LimE (∆ coil LimE-GFP) and a gene encoding a fusion of RFP to Coronin (LimE GFP/corA RFP)[S10]. Cells were transformed with the plasmid pDM115 cAR1-RFP (Hygromycin resistance) to visualize the membrane. Cells were grown in a shaker, containing 35.5g HL5 media ( R(cid:13)FORMEDIUM)/L of DI water[S11] in a shaker. When cells reached their exponential phase (1 − 2 × 106 cells/mL), they were harvested by centrifugation, washed in KN2/Ca buffer (14.6 mM KH2PO4, 5.4 mM Na2HPO4, 100 µM CaCl2, pH 6.4), and resuspended in KN2/Ca at 107 cells/mL. The washed cells were developed for 5h with pulses of 50 nM cAMP added every 6 min. Microfluidic device The design of microfluidic device used in the study is similar to the design of the devices that were previously used to study gradient sensing in yeast[S12] and chemotaxis in Dictyostelium[S13, S14]. The microfluidic device (Fig. S9) consists of a lithographically fabricated silicone (polydimethylsiloxane, PDMS, Sylgard 184) chip and a cover glass substrate (with either PDMS or hydrogel coating, see below), against which the chip is sealed using vacuum suction. To this end, the network of liquid-filled microfabricated microchannels of the chip, which are relatively narrow and either 100 or 10 µm deep, is surrounded by a wide (∼6 mm) and deep (∼1 mm) groove, serving as a vacuum cup. When the PDMS chip is placed on a substrate, the application of vacuum to the cup generates a pulling force that instantly seals the liquid-filled microchannels of the chip against the substrate. The application of vacuum also leads to controlled partial collapse of the microchannels, making it possible to reduce the depth of the 10 µm deep microchannels by > 5µm by controlling the level of vacuum. The network of liquid-filled microchannels of the device (Fig. S9) has a single outlet (out), two main inlets, for a C0 =100nM solution of cAMP (in 1) and for buffer (in 2), and an auxiliary inlet for cell loading (in c). The functional region of the device has two mirror-symmetric 100 µm deep, 500 µm wide flow-through channels (Fig. S9), which are connected to the two main inlets and are flanking 3 clusters of 10 µm deep gradient chambers. The flow through the device is driven by applying equal differential pressures of ∼2 kPa between the two main inlets and the outlet. The resulting mean flow velocity in the 500 µm wide flow-through channels is ∼200 µm/s. The gradient chambers are all 70 µm wide and each cluster has 15 identical chambers with equal lengths. The lengths L of the gradient chambers in the upstream, middle, and downstream clusters are 360, 220, or 120 µm, respectively. There is practically no flow through the gradient chamber because of near zero pressure gradient along them, and the diffusion of cAMP from the flow-through channel perfused with the 100 nM solution to the flow-through channel perfused with buffer results in linear concentration profiles of cAMP with gradients of 0.28, 0.45, and 0.83 nM/µm, respectively. In different sets of experiments, the application of different levels of vacuum resulted in the effective depths of the gradient chambers of 10, 7, and 5 µm. Substrate preparation In our experiments, the microfluidic chips were sealed against cover glass substrates with two different types of coating: ∼10 µm thick layer of PDMS of the same type as the material of the chip and ∼3 µm thick layer of 30% polyethylene glycol (PEG) gel. In the former case, the cover glass was a #1.5 thickness 47 mm circle at the bottom of a 50 mm WillCo cell culture dish. A small amount (∼0.2 mL) of PDMS pre-polymer (10:1 mixture of base and curing agent of Sylgard 184 by Dow Corning) was dispensed onto the cover glass. Spin-coating was made at 6000 rpm for 2 min, and PDMS was cured by overnight baking in a 60◦C oven. In the latter case, the cover glass was #2, 7 50x35 mm rectangle. The cover glass was cleaned with water and ethanol, dried, air-plasma treated for 10 s, and then exposed to 3-(Trimethoxysilyl) propyl Methacrylate ( R(cid:13)Aldrich) vapor at 75◦C for 30 min. A 30% PEG pre-polymer solution was prepared by mixing PEG diacrylate (PEG-DA; avg Mn 900, R(cid:13)Aldrich) with a 0.03% aqueous solution VA086 (300 µg dissolved in 1000 µL of DI water) in a 3:7 ratio by volume. VA086 is iLine (365nm) sensitive UV photo-initiator that cross-links PEG-DA molecules (thus, converting a PEG-DA solution into a PEG gel) by binding to the acrylate groups and also links PEG-DA chains to the acrylate groups on the glass surface. An ∼100µL drop of the solution was dispensed onto the center of the cover glass and squeezed to a thin layer by placing an untreated #1.5 thickness, 30 mm diameter round cover glass on top, gently pushing this second cover glass with a pipette tip, and removing the excess solution with a wipe. The cross-linking of PEG-DA was done by exposing it to a total of 2.19 J/cm2 of 365 nm UV (derived from 365nm UV LEDs; ∼365 mW/cm2 for 60 sec). After the round cover glass was removed, the 50x35 mm cover glass had an ∼4 µm thick layer of covalently bonded PEG gel in the middle. Data acquisition and image analysis Differential interference contrast (DIC) images were taken of all gradient chambers on a spinning-disk confocal Zeiss Axio Observer inverted microscope using a 10x objective and a Roper Cascade QuantEM 512SC camera. DIC images were captured every 15 s for 30 min and were used to calculate the speed of the cells. To obtain the shape of the cells, fluorescent images (488 nm and 561 nm excitation) were captured every 2 seconds with a 63X oil objective. To visualize the shape of the cells near the substrates, z-stacks of confocal images were collected. The centroids of all cells were tracked across the gradient chambers from 10X image sequences using Slidebook 6 (Intelligent Imaging Innovations) software. Cells that moved more than 5 frames without encountering another cell were chosen for data analysis. 50 to 100 cell tracks were analyzed in each experiment. Velocity in the gradient direction, Vx(t), was computed using data from frames 45 s apart with Matlab R2016a (The MathWorks, Natick, MA). We have verified that cell speeds were largely independent of their positions within the gradient chambers. Consequently, the average speed was defined as the mean speed of all cells at least 30µm from the sides of the chamber adjacent to the flow-through channels at all recorded times. Cells outlines near the top (PDMS chip) and the bottom (substrate, PDMS or PEG) of the gradient chambers were obtained from confocal fluorescence images at 63X magnification with a custom-made Matlab code, as follows. After removing the average background intensity value, images were binarized using a threshold that was dependent on the cell's maximum intensity. Matlab algorithms were then used to dilate images, to fill possible holes, to erode images, to smooth images, and to provide information (area and outlines) about the connected pixels of the binary image. Finally, using the resulting images, we computed the ratio between the cell contact area at the top and bottom of the chamber and averaged this ratio over three time points for each cell. Statistics and reproducibility Each experiment was carried out four or five times on different days and the data were averaged for N=200-300 cells. Cell speed was found to be approximately normally distributed and p-values were computed with the unpaired t-test. For the area size ratio, the data distribution was not normal, and the Wilcoxon rank-sum test was used to obtain the p-values. The variations of the cell speed with the gradient chamber height and the type of substrate coating (PDMS vs. PEG) followed the same trends in gradient chambers of different lengths, L (cf. Fig. 4d and Fig. S11). Parameter γ A0 Ma Γ ν ξd ξs ηa λ δ g TABLE I: Model Parameters Description Tension Width of phase field Cell area size Cell area conservation strength Phase field relaxation parameter Cell viscosity Damping coefficient Substrate friction coefficient Active protrusion coefficient Width of active stress confinement Width of the substrate phase field Substrate repellent coefficient 8 Value 20 pN 2 µm 120 µm2 20 pN/µm 0.4 µm/s 102 pN s/µm 0.05 Pa s/µm 5 Pa s/µm 103 pN µm2 2 µm 1 µm 5 × 103 pN/µm 9 FIG. S1: Illustration of the substrate field χ, with width δ, together with the protrusion band ψ, with width λ and located a distance of away from the substrate. FIG. S2: Cell speed as a function of adhesion strength for different values of the drag coefficient ξd (in units of Pa s/µm). Cell speed changes little as ξd is increased from 0 to 0.5. FIG. S3: Simulation results of the phase-field method without active force (red line) compared to results obtained using Surface Evolver (white dots). The phase field is plotted using the indicated color scale. (a). Adhesion strength 10 pN. (b). Adhesion strength 20 pN. δε+λ/2λχ(y)ψ(y)Substratey01010203040Adehsion Strength (pN)00.10.20.3Cell Speed (µm/s)ξd=0ξd=0.05ξd=0.500.515μmab5μm10 FIG. S4: Numerical results showing the phase field using a color scale, the outline of the cell in black (defined as φ = 1/2), and the actin fluid velocity (multiplied with the phase field φ) for a cell moving on a single substrate (a), and confined in a channel with equal (b) and unequal substrate adhesion (c). Arrows indicate the direction of the velocity and the arrow length indicates the amplitude of the velocity. FIG. S5: Schematic illustration of the non-deformable cell considered here, consisting of two rectangles of unequal height. Active stress occurs in the right (front) rectangle. FIG. S6: The average speed along the y-direction, defined as 1/L(cid:82) φuxdx, for a cell in a confined chamber with a height of h = 4µm (red line) and h = 6µm (blue line). The vertical shear dissipation increases with increasing cell height. -10010-10010-10010-10010-10010-10010abc0L-LH1H2σa00-10-5051000.020.040.060.084μm6μmy (μm)Average speed (μm/s)11 FIG. S7: Simulation results of cells with a constant ratio 0.5 of the width of active stress band and the chamber height. The cyan dots schematically indicate the active stress sites. (a) Cell shapes for different chamber heights. Scalebar=5 µm (b) Cell speed as a function of chamber height. (c) Effective height as a function of chamber height. FIG. S8: Simulation results of cells with active stress at the entire front, as indicated. (a) Cell shapes for different chamber heights. Scalebar=5µm (b) Cell speed as a function of chamber height. (c) Effective height as a function of chamber height. Chamber Height= 4 μmChamber Height= 6 μmChamber Height= 8 μm468100.050.10.24681046810Cell Speed (μm/s)Chamber Height (μm)Effective Height (μm)Chamber Height (μm)abc4680.10.20.30.4468579Chamber Height= 4 μmChamber Height= 6 μmChamber Height= 8 μmCell Speed (μm/s)Chamber Height (μm)Effective Height (μm)Chamber Height (μm)abc12 FIG. S9: Design of the microfluidic device. The enlarged image is the experimental DIC view using a 10x objective showing gradient chambers and flow chambers with cells. FIG. S10: Experimentally obtained cell shapes and F-actin distribution for a cell moving in a 5 µm high channel. (a) z-stack of a cell containing the fluorescent membrane marker Car1-RFP. The cell extends from top to bottom PDMS substrate. (b) Cell outlines for different z values ranging from 0 (magenta) to 5 µm (cyan). The outline is essentially identical for all z values. (c) Average fluorescence intensity (normalized) of LimE, an F-actin marker, for each confocal slice as a function of z for representative cells in channels with height of 5,7 and 10 µm. All cells examined (N=5) displayed a qualitatively similar pattern with increased intensity close to the substrates. abc0102030x (μm)05101520y (μm)0246810z(μm)0.70.750.80.850.90.951Average intensity normalizedLimE5 μm7 μm10 μm13 FIG. S11: Average cell speed for different chamber heights and substrate composition for channel length L = 120µm, corresponding to a gradient of 0.83 nM/µm, and L = 360µm, corresponding to a gradient of 0.28 nM/µm. FIG. S12: Cell speed and effective height for cells with oscillatory active stress. (a) Average cell speed, computed as moving distance divided by cycle time, and (b) effective height as a function of substrate adhesion strength. Shown are the results for oscillatory stress cycles with two different periods. 5μm7μm10μm00.050.10PDMSPEG5μm7μm10μmPDMSPEGVx (μm/s)abChannel Length = 120μm0.1500.050.10Vx (μm/s)0.15Channel Length = 360μm102030456T=60sT=120s1020300.050.1T=60sT=120sAverage Speed Per Cycle (μm/s)Adhesion Strength (pN)Adhesion Strength (pN)Average Height Per Cycle (μm)ab14 ∗ These two authors contributed equally. † Electronic address: [email protected] [S1] B. A. Camley, Y. Zhang, Y. Zhao, B. Li, E. Ben-Jacob, H. Levine, and W.-J. Rappel, Proc Natl Acad Sci U S A 111, 14770 (2014). [S2] Y. Zhao, Y. Ma, H. Sun, B. Li, and Q. Du, arXiv preprint arXiv:1712.01951 (2017). [S3] K. A. Brakke, Experimental mathematics 1, 141 (1992). [S4] D. Shao, W.-J. Rappel, and H. Levine, Physical Review Letters 105, 108104 (2010). [S5] B. A. Camley, Y. Zhao, B. Li, H. Levine, and W.-J. Rappel, Physical Review Letters 111, 158102 (2013). [S6] D. Shao, H. Levine, and W.-J. Rappel, Proc Natl Acad Sci U S A 109, 6851 (2012). [S7] A. Carlsson, New journal of physics 13, 073009 (2011). [S8] K. Keren, Z. Pincus, G. M. Allen, E. L. Barnhart, G. Marriott, A. Mogilner, and J. A. Theriot, Nature 453, 475 (2008). [S9] H. Levine and W.-J. Rappel, Phys Today 66 (2013), 10.1063/PT.3.1884. [S10] D. Fuller, W. Chen, M. Adler, A. Groisman, H. Levine, W.-J. Rappel, and W. F. Loomis, Proc Natl Acad Sci U S A 107, 9656 (2010). [S11] M. Sussman, in Methods in cell biology, Vol. 28 (Elsevier, 1987) pp. 9 -- 29. [S12] S. Paliwal, P. A. Iglesias, K. Campbell, Z. Hilioti, A. Groisman, and A. Levchenko, Nature 446, 46 (2007). [S13] M. Skoge, M. Adler, A. Groisman, H. Levine, W. F. Loomis, and W.-J. Rappel, Integrative Biology 2, 659 (2010). [S14] M. Skoge, H. Yue, M. Erickstad, A. Bae, H. Levine, A. Groisman, W. F. Loomis, and W.-J. Rappel, Proc Natl Acad Sci U S A 111, 14448 (2014).
1303.7276	1	1303	2013-03-29T01:12:32	The effect of limb kinematics on the speed of a legged robot on granular media	[ "physics.bio-ph", "cond-mat.soft", "physics.flu-dyn", "q-bio.QM" ]	Achieving effective locomotion on diverse terrestrial substrates can require subtle changes of limb kinematics. Biologically inspired legged robots (physical models of organisms) have shown impressive mobility on hard ground but suffer performance loss on unconsolidated granular materials like sand. Because comprehensive limb-ground interaction models are lacking, optimal gaits on complex yielding terrain have been determined empirically. To develop predictive models for legged devices and to provide hypotheses for biological locomotors, we systematically study the performance of SandBot, a small legged robot, on granular media as a function of gait parameters. High performance occurs only in a small region of parameter space. A previously introduced kinematic model of the robot combined with a new anisotropic granular penetration force law predicts the speed. Performance on granular media is maximized when gait parameters minimize body acceleration and limb interference, and utilize solidification features of granular media.	physics.bio-ph	physics	The effect of limb kinematics on the speed of a legged robot on granular media Chen Li1, Paul B. Umbanhowar2, Haldun Komsuoglu3, and Daniel I. Goldman1∗ 1School of Physics, Georgia Institute of Technology, Atlanta, Georgia 30332, USA 2Department of Mechanical Engineering, Northwestern University, Evanston, IL, 60208, USA 3Department of Electrical and Systems Engineering, University of Pennsylvania, Philadelphia, PA 19104, USA ∗Corresponding author. E-mail: [email protected] 3 1 0 2 r a M 9 2 ] h p - o i b . s c i s y h p [ 1 v 6 7 2 7 . 3 0 3 1 : v i X r a Achieving effective locomotion on diverse ter- restrial substrates can require subtle changes of limb kinematics. Biologically inspired legged robots (physical models of organisms) have shown impressive mobility on hard ground but suffer performance loss on unconsolidated granular ma- terials like sand. Because comprehensive limb- ground interaction models are lacking, optimal gaits on complex yielding terrain have been de- termined empirically. To develop predictive mod- els for legged devices and to provide hypothe- ses for biological locomotors, we systematically study the performance of SandBot, a small legged robot, on granular media as a function of gait parameters. High performance occurs only in a small region of parameter space. A previously in- troduced kinematic model of the robot combined with a new anisotropic granular penetration force law predicts the speed. Performance on granular media is maximized when gait parameters mini- mize body acceleration and limb interference, and utilize solidification features of granular media. INTRODUCTION To move effectively over a wide range of terrestrial terrain requires generation of propulsive forces through appropriate muscle function and limb kinematics [1, 2]. Most biological locomotion studies have focused on steady rhythmic locomotion on hard, flat, non-slip ground. On these surfaces kinematic (gait) parameters like limb frequency, stride length, stance and swing dura- tions, and duty factor can change as organisms walk, run, hop and gallop [1]. There have been fewer biological stud- ies of gait parameter modulation on non-rigid and non- flat ground, although it is clear that gait parameters are modulated as the substrate changes during challenges like climbing [3, 4], running on elastic/damped substrates [5], transitioning from running to swimming [6], and running on different preparations of granular media [7]. Even subtle kinematic changes in gait can lead to major dif- ferences in limb function [8]. A major challenge is to de- velop models of limb interaction with complex substrates and to develop hypotheses for how organisms vary gait parameters in response to substrate changes. The RHex class of model locomotors (robots) has proved useful to test hypotheses of limb use in biolog- ical organisms on hard ground [9] and recently on more complex ground with few footholds [10] or the ability to flow [11]. These hexapedal devices model the dynami- cally stable locomotion of a cockroach and were the first legged machines to achieve autonomous locomotion at speeds exceeding one body length/s. In these devices, complexity in limb motion is pared down to a few biologi- cally relevant parameters controlling intra-cycle "stance" and "swing" phases of 1-dof rotating limbs (referred to as"gait" parameters hereafter; see detailed description in Methods and Results). When these gait parameters are appropriately adjusted, RHex shows performance compa- rable in speed and stability to organisms on a diversity of terrain [12]. However, because of the scarcity of exist- ing models of limb interaction with complex substrates, adjustment of the gait parameters is typically done em- pirically [13, 14]. Sand, a granular medium [15], is of particular interest for studies examining the effects of limb kinematics on lo- comotor performance on yielding terrain. In a previous study [11] we found that minor changes in the limb kine- matics of a small RHex-class robot, SandBot (Fig. 1a), produced major changes in its locomotor mode and per- formance (speed) on a granular medium, poppy seeds (see Section 2.2). This sensitivity occurs, in part, be- cause forced granular media remain solid below the yield stress, but can flow like a fluid when the yield stress is exceeded [16]. We tested SandBot on granular media of different yielding properties (set by granular volume frac- tion; see Section 2.1) at various limb frequencies but with the other gait parameters fixed. While there is no fun- damental theory at the level of fluid mechanics that ac- counts for the physics of the solid-fluid transition of gran- ular media or the dynamics of the fluidized regime, em- pirical models of granular penetration force have proved useful to predict SandBot's speed [11]. SandBot's propul- sion is determined by factors that control this transition during limb-ground interaction (limb penetration depth, limb speed, body mass, grain friction, volume fraction, etc.). Using a simplified equation describing the granu- lar penetration force, we developed a kinematic model to explain the locomotion of SandBot (see Section 2.2). In this study, we advance our understanding of the effects of limb kinematics on locomotor performance by testing SandBot with varying gait parameters on sand of fixed yield strength and at fixed limb frequency. We find that robot speed depends sensitively on limb kinemat- ics; while the original model qualitatively captures this sensitivity, the penetration force used in the model and other assumptions need to be modified to explain some important features. Our study not only reveals the spe- cific optimal kinematics for SandBot on granular media, but also advances our understanding of how in general to achieve effective legged locomotion on complex terrain. BACKGROUND AND REVIEW OF PREVIOUS STUDY To understand the effect of limb kinematics addressed here, we first summarize the mechanism of SandBot lo- comotion on granular media (called rotary walking) dis- covered in our previous study [11]. In this section, we discuss the physics of granular media that controls the limb penetration depth (which governs locomotion per- formance) and then review our previous experiments and kinematic model. Physics of Limb-Granular Media Interaction The physics that controls locomotor performance is the relative magnitude of the penetration resistance force (originating in the granular media) and the sum of the external forces (weight, inertial forces). When these bal- ance, the granular media solidifies, allowing the robot to be supported at a fixed limb penetration depth. The previous SandBot study [11] revealed that as the limb (or any simple intruder) vertically penetrates into the medium, the penetration force scales with z, the depth of the intruder below the surface [17], as Fp(z) = k(φ)z, where φ is the volume fraction, the ratio of the solid volume of the granular media to the volume that it occupies (for natural dry sand, 0.55 < φ < 0.64). The constant, k(φ), characterizes the penetration resistance and increases with φ. In this paper we keep φ fixed at ap- proximately the critical packing state [16, 18, 19] (which is close to the as-poured volume fraction) where granular media neither globally dilate nor compact in response to shear (see Section 3). Review of Previous Observations and Model In the previous study of SandBot [11], we fixed intra- cycle limb kinematics (by using gait parameters that produce consistent motion on granular media, see Sec- tion 3) and measured SandBot's average speed vx on poppy seeds as a function of volume fraction and the 2 cycle-averaged limb frequency ω. We observed a sensi- tive dependence of vx on both φ and ω, and developed a kinematic model which explained this dependence and revealed two distinct locomotor modes determined by whether the granular media solidifies during limb-ground interaction. Our kinematic model describes the limb-ground in- teraction of SandBot by considering the motion of just a single limb (Fig. 1b). SandBot has six approxi- mately c-shaped limbs (c-legs) divided into two alternat- ing tripods. C-legs in the same tripod rotate in synchrony and each c-leg rotates about a horizontal axis normal to the robot body. We simplify the multi-leg ground inter- FIG. 1. Mechanism of SandBot locomotion on granular me- dia. (a) SandBot, a six-legged insect inspired robot, moves with an alternating tripod gait. The three arrows indicate the limbs of one tripod. (b) Schematic of single-leg represen- tation of SandBot with mass m = 1/3 SandBot's total mass. With the body contacting the surface, the motor axle is height h = 2.5 cm above the ground. The c-leg is approximately a circular arc (radius R = 3.55 cm, arc span 225 degree). Leg angle θ is measured clockwise about the axle and between the downward vertical and a diameter through the axle. Leg depth z = 2R cos θ − h. (c) Magnitude of penetration force Fp (blue curve) relative to force required for upward motion, m(g + a), (red lines) determines the locomotor mode. The force required for quasi-static movement (mg) is shown for reference. When Fp and m(g + a) intersect rotary walking occurs. When Fp and m(g + a) do not intersect (dashed red curve and above), the robot swims. (d) Schematic of rotary walking. The granular material flows in the intervals [αi, βi] (red arrow) and [βf , αf ] (blue arrow) where Fp < m(g+a) and the c-leg rotates about the axle (red and blue circles). The material is a solid in the interval [βi, βf ] (gray sector; line with arrows in (c)) where Fp exceeds m(g + a) and the c-leg rotates about its center (green circle and arrow), lifting and propelling body forward by step length s = R(sin βf − sin βi). The c-leg is above the ground in the interval [αf , αi + 2π]. Note that [αi, αf ] in (c) is symmetric to vertical (θ = 0) as a result of assuming the force is isotropic (see Section 4.2). action of each tripod to that of a single limb carrying 1/3 the total body mass 3m (2.3 kg), as body weight is approximately uniformly distributed between each c-leg. We also considered a c-leg as a simple intruder ignoring its more complicated geometry, i.e. Fp(z) = kz. The previous study [11] showed that the simple intruder ap- proximation gave approximately the same results as a more realistic treatment in which penetration force was integrated over the submerged leading surface of a c-leg. In this study we use the simple intruder approximation. We make the approximation that SandBot's body is in stationary contact with the surface at the onset forward motion in each cycle [20], and define the c-leg's angular position, θ, as the clockwise angular displacement from the configuration where the center of curvature of the c- leg is directly beneath the axle, see Fig. 1b. During a full rotation, as θ changes from −π to π, the c-leg ini- tially contacts the ground at θ = αi and loses ground contact at θ = αi. Because leg depth can be approxi- mated as z = 2R cos θ − h when the body is in contact with the surface [21], penetration force can be written as (cid:3) (blue curve in Fig. 1c), where R = 3.55 cm and h = 2.5 cm are the radius of the c- leg and the hip height (i.e. distance from c-leg axle to underside of body) respectively. Fp(θ) = 2Rk(cid:2)cos θ − h 2R Of prime importance in determining SandBot's per- formance is the magnitude of the penetration force Fp(θ) relative to the sum of the forces required to the sup- port the body weight and accelerate the body upward m(g + a) (red curve in Fig. 1c), where g is the accel- eration due to gravity and a the acceleration [11]. The relevant acceleration is given by the jump in robot speed when the granular media solidifies, Rω, divided by the characteristic response time of the c-leg interacting with the granular media, ∆t(φ), i.e. a = Rω/∆t. Two distinct locomotor modes are possible depending on whether or not Fp(θ = 0) > m(g + a): 1. Rotary walking -- movement with solidification (see Fig.1b-d): As the c-leg rotates into the ground after initial leg-ground contact at θ = αi, the penetra- tion force increases with increasing depth. In the rotary walking regime the material beneath the c- leg solidifies and leg penetration stops at an angle θ = βi when Fp(βi) = m(g +a), see Fig. 1c,d. Since the frictional force between the c-leg and granu- lar material is insufficient for the leg to roll, the c-leg instead rotates about its center of curvature (green circle and arrow) lifting and advancing the robot in the process. Rotary walking continues until θ = βf , beyond which the c-leg again pene- trates through the material since Fp(θ) < m(g + a) and the body is again in contact with the ground (blue circle and arrow). Rotary walking thus oc- curs over a finite range of leg angle βi < θ < βf or [βi, βf ] (horizontal arrow in Fig. 1c and gray sec- 3 tor in Fig. 1d) where βi and βf are determined by Fp(βi,f ) = 2Rk(cos βi,f − h 2R ) = m(g + a). For a given [βi, βf ], Fig. 1d shows that the robot ad- vances a distance s = R(sin βf − sin βi), where we call s the step size. During one complete gait cy- cle of period T, each alternating tripod advances the robot by s, giving an average robot speed of vx = 2s/T = sω/π. 2. Swimming -- movement without solidification: When Fp(0) < m(g+a) (Fig. 1c, dashed red curve), the granular material beneath the penetrating c-leg never solidifies and rotary walking does not occur, i.e. βi = βf = 0. Instead, the limb constantly slips through the surrounding fluidized granular mate- rial, similar to a swimmer's arm in water, and the robot advances slowly (vx < 1 cm/s). In this regime forward motion occurs when the frictional and inertial (drag) forces generated by the c-legs exceeds the frictional force between the robot body and the surface. The two constants characterizing the interaction with the granular medium, k and ∆t, together with limb fre- quency ω, determine the relative magnitudes of Fp and m(g+a) and consequently control which locomotor mode the robot operates in. Reducing k (by decreasing φ) and/or increasing ω reduces the rotary walking range; in other words, the less compact the granular material is and/or the faster the limbs rotate, the deeper the c-legs have to penetrate before the granular material solidifies and rotary walking begins, and the more susceptible the robot is to entering the slow swimming mode. This sim- ple kinematic model captures the observed sensitive de- pendence of vx on φ and ω, with k(φ) and ∆t(φ) as two fitting parameters. In summary, our previous study of SandBot [11] showed that to locomote effectively on granular media, limbs kinematics that access the solid phase of granular media should be employed. METHODS AND RESULTS The limb kinematics of each tripod during one cy- cle are parameterized by three "gait parameters", see Fig. 2(a,b). The kinematics of both tripods are periodic (with period T ) and offset by half a period T /2 but are otherwise identical. For the conditions in this and previ- ous experiments, a motor controller in the robot ensures that the target kinematics are achieved. Limb kinemat- ics consist of a "swing" phase (orange), which is typically faster, and a "stance" phase (green), which is typically slower, with respective frequencies ωf and ωs. During hard ground locomotion in the RHex-class of Robots (and for animal locomotion in general), "swing" and "stance" phases typically correspond to off-ground 4 In the previous study [11], we used SGK to test vx(φ, ω). Now armed with the understanding of how SandBot moves on granular media gained from this work, we set out to determine the effects of limb kinematics in detail. We set φ = 0.605 and ω = 8 rad/s, and measure SandBot's average speed on granular media as we sys- tematically vary gait parameters, i.e. vx = vx(θs, θ0, dc). We pick ω = 8 rad/s because at this intermediate fre- quency SandBot displays both rotary walking and swim- ming as the clock signal is varied. We pick φ = 0.605 to remove the effect of local volume fraction change which causes a premature transition from rotary walking to swimming [22] and adds to the complexity of the prob- lem. We first test the effect of the extent and location of the slow phase for fixed dc = 0.5, measuring speed vx = vx(θs, θ0). We vary the parameters between 0 ≤ θs ≤ 2 and −2 ≤ θ0 ≤ 2, which are the limits set by the robot's controller. We choose dc = 0.5 because it is close to the dc values of both HGK and SGK. This gave us an easy FIG. 3. Average speed vx of SandBot on granular media (φ = 0.605) as a function of (θs, θ0) for dc = 0.5 and ω = 8 rad/s. (a) The experimental data has a localized region of high speeds with peak vx ≈ 9 cm/s near {θs, θ0} = {1.5, −0.5}.. Circles show that vx for SGK (red) and HGK (blue) matches data for dc = 0.5 despite the formers slightly different dc values, see Section 4.3. Inset: original data from which main figure is interpolated. (b) Predicted vx(θs, θ0) from the kinematic model with Fp = kz captures the single peak but predicts a lower speed for SGK than for HGK contrary to observation and fails to account for the observed peak in speed at θ0 = −0.5. FIG. 2. SandBot's intra-cycle limb kinematics and affects its speed on granular media. (a) Each leg rotation is com- posed of a fast phase (orange) and a slow phase (green). θs and θ0 define the angular extent and center of the slow phase respectively. (b) Leg angle θ as a function of time during one cycle (normalized to T ). θ(t) of the other tri- pod is shifted by T /2 but otherwise identical. dc is the duty cycle of the slow phase, i.e. fraction of the period spent in the slow phase. (c) Instantaneous speed of Sand- Bot on granular media with hard ground clock signal (HGK: {θs, θ0, dc} = {0.85, 0.13, 0.56}; red) and soft ground clock signal (SGK: {θs, θ0, dc} = {1.10,−0.50, 0.45}; blue). With HGK SandBot moves slowly (vx ≈ 2 cm/s) on granular me- dia, but with SGK (red), it advances rapidly (vx ≈ 8 cm/s). and ground-contact phases, respectively. But because during locomotion on granular media this correspondence is not necessarily true, we simply call them fast and slow phases. In practice the fast and slow phases are implicitly defined by the triplet {θs, θ0, dc} where θs is the angular extent of the slow phase, θ0 is the angular location of the center of the slow phase, and dc is the duty cycle of the slow phase (the fraction of the period in the slow phase). Specifying the cycle averaged limb frequency ω fully determines the motion of the limbs in the robot frame. By definition, ωs = θs T (1−dc) , and ω = T dc 2π T . Typically, gait parameters are set so that ωs < ω < ωf , but the reverse is possible when θs becomes large enough and/or dc small enough. , ωf = 2π−θs In the first tests of SandBot on granular media (Fig. 2c), we found that kinematics tuned for rapid stable bouncing motion on hard ground (HGK: {θs, θ0, dc} = {0.85, 0.13, 0.56}) produced little motion on granular me- dia (red curve). Empirical adjustment to soft ground kinematics (SGK: {θs, θ0, dc} = {1.10,−0.50, 0.45}) re- stored effective (walking) locomotion on granular media (blue curve). way to project HGK (dc = 0.56) and SGK (dc = 0.45) onto the vx = vx(θs, θ0) plot (dc = 0.5), assuming that a small change of dc near dc = 0.5 does not affect speed significantly (see Fig. 6a and Section 4.3 which support this assumption)). Measurements of vx = vx(θs, θ0) (Fig. 3a) show a sin- gle sharp peak in speed near {θs, θ0} = {1.5, −0.5}. High speeds only occur within a small island of −1 < θ0 < 0 and θs > 0.5 surrounding the peak; lower speeds fill the remainder of the space. The drop in speed is rapid as θ0 is varied away from the peak, and is less so when θs is varied away from the peak; this is also evident in cross sections through the peak (blue circles in Figs. 3a and Fig. 6a, respectively). Ignoring the effect of dc, the SGK parameters (blue dot) lie close to the peak while the HGK parameters (red dot) are in the low speed re- gion. The optimal gait parameters which we found for SandBot locomotion on poppy seeds at φ = 0.605 and ω = 8 rad/s are: {θs, θ0, dc} = {1.5,−0.5, 0.55}. These gait parameters generate about 20% higher speed than the previously used SGK parameters. Variation of the duty cycle at fixed {θs, θ0} = {1.5,−0.5} also has a substantial influence on speed. Data (blue circles in Fig. 6b) show a well defined peak at dc ≈ 0.5. Speed drops off relatively slowly for dc > 0.55, and more quickly to small (swimming) speeds for dc < 0.5. DISCUSSION Application of Model to Slow Phase Extent and Location Variation To apply our kinematic model to SandBot locomotion with varied limb kinematics, we must consider the effects of variable limb kinematics during limb-ground interac- tion. Depending on the gait parameters during ground contact, the limbs could be rotating in the fast phase, in the slow phase, or in a combination of both. In our pre- vious study, the kinematic model ignored limb frequency variability during ground contact and only considered the robot limb rotating at the constant cycle averaged limb frequency ω. However, as limb kinematics change, the variability of limb frequency in ground contact needs to be taken into account. For our test of vx = vx(θs, θ0) within 0 ≤ θs ≤ 2 and −2 ≤ θ0 ≤ 2 and at dc = 0.5, ωf >> ωs so that only the slow phase can possibly achieve rotary walking, as fast limb rotation results in swimming. In this case ωs (instead of ω) controls acceleration a and thus determines the rotary walking range [23]. For fixed dc, varying [θs, θ0] changes the extent and location of the slow phase; the angular extremes of the slow phase are θi,f = θ0 ± θs 2 , see Fig. 5a. Varying θs also changes ωs which controls the rotary walking range. 5 Therefore varying [θs, θ0] affects where the slow phase overlaps with the rotary walking range. The step length s is given by s = R(sin ψf−sin ψi) where ψi = max(βi, θi) and ψf = min(βf , θf ) if there is overlap or s = 0 if there is no overlap. The larger the overlap, the further the robot moves forward in a cycle. As shown in Section 2.3, the rotary walking range is given by solving the equation Fp(θ) = [βi, βf ] 2Rk(cos βi,f − h 2R ) = m(g + a), with a given by a = mωs ∆t . We can evaluate how [θi, θf ] overlaps with [βi, βf ] to de- termine s, and calculate the robot speed using vx = 2s T = sω π . For fixed ω, speed vx scales with step length s. Figure 3b shows the model prediction of vx using fit- ting parameters k = 210 N/m, ∆t = 0.37 s. Compar- ing prediction with observation (Fig. 3a), the model cap- tures the peak and predicts similar magnitudes of speeds. However the predicted peak is symmetric about θ0 = 0 while the observed peak is symmetric about θ0 = −0.5. Anisotropic Penetration Force Law If the penetration force of the granular material in- creased like kz as assumed in the model, we would ex- pect θ0 = 0 as this value would give the largest overlap between the slow phase and the rotary walking range as determined by the material strength (see scheme in Fig. 5a). To investigate why the robot performs best with θ0 = −0.5 we attached a c-leg to a force/torque sensor and measured the grain resistance as the c-leg was ro- tated through the granular media at ω = 0.35 s−1 (the horizontal rotation axis of the c-leg was positioned the same distance h = 2.5 cm above the grain surface as when it is mounted on the robot). Figure 4b shows a clear asymmetry in the penetration (vertical) force with the measured peak force occurring near β0 = −0.75; pen- etration force during rotation peaks before the intruder reaches the maximum depth. We confirmed that the measured anisotropy in the penetration force is intrin- sic to our granular medium and is not an artifact of the particular shape of the c-leg by additionally rotating a rectangular bar and a sphere into granular media at the same hip height: both objects exhibited a peak force at β0 = −0.75. We speculate that the asymmetry in penetration force during rotation into granular media is a result of the changing limb orientation during rotational intrusion. For vertical penetration (which we considered in the model in the previous study), the intruder is constantly pushing down on the granular material. The grain con- tact network generated in granular material in response to intrusion [24] forms a downward pointing cone which generates a force symmetric to the vertical (θ = 0). In rotational intrusion, however, the direction of intrusion is constantly changing; the direction of the force cone should change as well and correlate with the instanta- 6 inal penetration force law in our model to Fp(θ) = 2Rk(cid:48) {cos [b (θ − β0)] + 1} for F > 0, where β0 = −0.75, and k(cid:48) and b are new fit parameters. Following the same procedure described in Section 4.1, we find the robot speed by calculating [βi, βf ], the overlap between [θi, θf ] and [βi, βf ], and the step size. Figure 5c shows vx predicted by a fit to the model using the anisotropic penetration force law (fitting parameters k(cid:48) = 65 N/m, ∆t = 0.4 s, and b = 0.8). Besides cap- turing the peak behavior of measured speed, the model also captures the shift in peak location to θ0 = −0.5 (i.e. asymmetry to θ0 = 0). For fixed θs, speed is maximal when the center of the slow phase corresponds with the center of the rotary walking range (Fig. 5a). If θ0 is differ- ent from β0 = −0.75, the overlap of the slow phase and FIG. 4. Asymmetry of vx with respect to θ0 = 0 is due to anisotropic penetration force during limb rotation into gran- ular media. (a) For all θs at dc = 0.5, vx(θ, θ0) (Fig. 2a) is maximal (dashed vertical black line) at θ0 = −0.5 (θs = 1.5 shown). Inset: peak location θ0 = −0.5 does not change for dc = 0.8. (b) Vertical penetration force Fp (solid blue curve) during c-leg rotation into poppy seeds reaches maximum at β0 = −0.75 (dashed black line) and is asymmetric to θ0 = 0. Inset: force measurement schematic. Solid blue curve in (a) is prediction from the model with anisotropic penetration force. neous direction of intrusion. We hypothesize that the force during rotational intru- sion is maximal at β0 = −0.75 because for larger angles part of the cone reaches the surface and/or terminates on the horizontal walls of the container and can no longer support the entire grain contact network, thus reducing the maximal yield force. We also note that the angle at which maximal force is developed is close to the angle of repose 0.52 that we measure for the poppy seeds. This angle is the same as the internal slip angle in cohesionless granular material [16] which plays an important role in the formation of the grain contact network, supporting the plausibility of our speculation. To account for the measured angular offset in peak force from vertical (Figs. 2c and 3a), we modify the orig- FIG. 5. Overlap of the slow phase and the rotary walk- ing interval predicted by the anisotropic penetration force law better predicts vx(θs, θ0). (a) Overlap of the slow phase [θi, θf ] and the rotary walking range [βi, βf ] determines step length s and thus speed vx. For the configuration shown, s = R(sin βf − sin θi). θ0 and β0 are centers of the slow (b) In phase and the rotary walking range, respectively. SGK the slow phase (centered at θ0 = −0.50) overlaps nearly completely with the rotary walking range (centered at β0 = −0.75), explaining its high speed as compared to HGK (centered at θ0 = 0.13) which has little overlap. (c) Pre- diction of vx(θs, θ0) from the the model with the anisotropic penetration force for dc = 0.5 captures the asymmetry of vx with respect to θ0 = 0 and predicts higher speed for SGK than for HGK. Fitting parameters: k(cid:48) = 65 N/m, ∆t = 0.4 s and b = 0.8. 7 stood. When there is no tripod overlap (only one tripod with ground contact at any given time) each tripod ad- vances the robot a distance s for a total displacement of 2s per period. This is the case for dc ≤ 0.5. However in the limit of dc = 1 both tripods are simultaneously in the slow phase as the duration of the fast phase is zero. The simultaneous slow phases generate a total displacement of just s [26] instead of 2s without tripod overlap. As a result, the predicted speed at dc = 1 must be halved (i.e. vx = s 2π ). Lacking a way to quantify the tri- T = sω the rotary walking range decreases, which reduces step length and thus speed. In accord with observation, SGK (red dot) lies near the peak while HGK (blue dot) lies in a region of low speeds. Figure 5a,b shows that SGK has higher speed than HGK because the overlap between the slow phase and the rotary walking range (gray sector in Fig. 5a) is significantly larger. At fixed θ0 = β0, increasing the extent of the slow phase (increasing θs) from zero initially increases speed as the extent of the slow phase increases within the ro- tary walking envelope (see Figs. 2a and 6a). However, ωs increases with θs which increases the acceleration and reduces the rotary walking range. For sufficient extent (near θs = 1.5 for the data shown in Figs. 2a and 6a) the slow phase contains the rotary walking range and step length is determined by the latter. Further increase in θs reduce the rotary walking range. Rotary walking is not possible for θs ≥ 2πdc∆t as the material is never strong enough to both support and accelerate the robot. In Fig. 6a the experimental speed is noticeably lower than the model prediction at the largest θs = 2. As we discuss below in regards to variation in dc, this re- duction is a apparently the result of tripod overlap (both tripods simultaneously in ground contact) which occurs for a greater portion of the slow phase for larger θs. (cid:16) 4k(cid:48) m − g (cid:17) ω R Effect of Duty Cycle While the model prediction of vx(dc) (blue curve in Fig. 6b) matches the magnitudes of speeds at intermedi- ate dc ≈ 0.5, it does not quantitatively match the shape of measured speed vs dc. Below dc ≈ 0.5, the model pre- dicts that vx(dc) increases monotonically with increasing dc. This trend is in accord with the experimental ob- servations; however the model prediction is consistently higher than measured speed. Above dc ≈ 0.5, the model predicts that vx(dc) is independent of dc, but the mea- sured speed decreases with increasing dc and is lower than the model prediction. We now discuss possible reasons for the discrepancies at low dc < 0.1 (labeled region I in Fig. 6b), intermediate 0.1 < dc < 0.5 (region II) and high dc > 0.5 (region III). In region I, dc is small so that ωs and thus a become large enough to ensure that the robot is in the swimming mode (i.e. movement without solidification). Here the model assumes rotary walking and predicts zero speed. In experiment, the robot can still advance slowly at each step due to thrust forces from continuously slipping limbs generated by frictional drag [25] and/or inertial move- ment of material. This thrust competes with friction from belly drag, and as in [11], we find that these re- sult in low average speed of vx ≈ 1 cm/sec. The model's overestimate of speed at high dc (region III) is a result of tripod overlap and can be readily under- FIG. 6. Comparison of data and anisotropic force model pre- dictions for vx(θs) and vx(dc). (a) Measured vx(θs, θ0 = −0.5) (blue circles) through the speed maximum (Fig. 2a) deviates from the anisotropic force model prediction (blue curve) at large θs due to limb overlap. (b) Measured vx(dc) (blue cir- cles) for {θs = 1.5, θ0 = −0.5} is maximum at dc ≈ 0.5. The model (solid blue curve) accurately predicts the speed for dc ≈ 0.5 but is inaccurate elsewhere due to contributions of swimming neglected by the model (region I), a decrease in rotary walking range from cratering induced depth reduction and unequal penetration forces developed by c-legs on oppo- site sides of the body (II), and tripod overlap at high dc (III). Model prediction with tripod overlap included (dashed blue line) better matches the data in region III. pod overlap effect, we assume that the reduction of step length from 2s to s is linear with dc for dc > 0.5; the data is in good agreement with this prediction (dashed blue curve in Fig. 5). This reduction in speed is a purely kinematic effect that is inherent to the rotary walking gait at high dc. Two plausible mechanisms explain the model's over- estimate of speed at intermediate dc (region II): hole digging and uneven weight distribution. Lateral obser- vations of the robot kinematics at low dc show that the rapid motion of the c-leg during the slow phase throws significant numbers of particles out of the limb's path which creates a depression. For a deep enough hole, ro- tary walking is impossible due to the reduced penetration depth of the leg below the now lower surface of the de- pression. The second mechanism concerns the model's assumption of uniform weight distribution between the three legs of the tripod. In the dc range where the robot advances slower than predicted, observations show that the robot rotates in the horizontal plane. Rotation oc- curs in this transition region between pure swimming and pure rotary walking because the side of the robot with two c-legs in ground contact undergoes rotary walking while the opposite side is in the swimming mode. Due to the increased gravitational and inertial forces on the single c-leg, the penetration forces are never sufficient to achieve ground solidification under the leg even at the maximum penetration depth. CONCLUSIONS We have built upon our previous experiments and models of a legged robot, SandBot, to explore how changes in limb kinematics affect locomotion on granular media. We found that even when moving on controlled granular media of fixed volume fraction at fixed cycle- averaged limb frequency, speed remains sensitive to vari- ations in gait parameters that control angular extent, an- gular location, and temporal duty factor of the slow phase of the limb cycle. We showed that the assumptions in a previously introduced model (which accurately predicted speed as a function of limb frequency and volume frac- tion) had to be modified to incorporate an anisotropic penetration force during rotational intrusion into gran- ular media as well as changes in acceleration of the leg as gait parameters were varied. With these modifica- tions the model was able to capture speed as a function of angular extent and angular location. The model also indicates that as duty cycle is changed, effects due to si- multaneous limb pairs (tripods) in ground contact, rapid limb impact into sand, and unequal weight distribution on limbs become important. Our experiments and modified model explain why gait parameters that allow the robot to rapidly bounce over hard ground lead to loss of performance on granular me- 8 dia. They demonstrate how the angular extent and lo- cation of the slow phase must be adjusted to optimize interaction with granular media by minimizing inertial force and limb interference, and maximizing the use of solid properties of granular media. Further studies of SandBot guided by our kinematic model should reveal how physical parameters of both robot (mass distribu- tion, limb compliance, limb shape, belly shape) and the environment (grain friction, density, incline angle, grav- ity) control the solid-fluid transition and thus affect the limb-ground interaction and performance. However, ad- vances are required in theory and experimental charac- terization of complex media. Otherwise we must continue to rely on empirical force laws specific to particular ge- ometries, kinematics and granular media. The existence of a speed optimum in gait parameter space implies that control of limb kinematics is critical to move effectively on granular media, whether actively through sensory feedback, and/or passively through me- chanical feedback. Future work should compare these results to investigations of gait optimization on hard ground [13]. The differences in limb kinematics on sand compared to hard ground are intriguing because on hard ground performance is optimized by making the robot bounce. However, this carries with it the risk of yaw, pitch and roll instability due to mismanaged kinetic en- ergy. On granular media such instabilities appear rare; instead most gait parameters (see Fig. 3a) result in little or no forward movement due to mismanaged fluidization of the ground. Thus, our results could have a practical benefit as they suggest strategies for improving the per- formance of current machines [27 -- 29] on variable terrain via new limb and foot designs and control strategies. Finally, an enormous number of organisms contend with sand [30], moving on the surface (or even swim- ming within it [31]). While the observed phenomena and proposed locomotion modes (e.g. rotary walking) appear specific to SandBot and its c-shaped limbs, the underlying principles could apply to locomotion of or- ganisms on yielding substrates. For example, our re- cent work on terrestrial hatchling sea turtle locomotion demonstrates that their effective movement on sand pro- ceeds through solidification of the granular medium [32]. Integrated studies of biological organisms and physical models can provide hypotheses [33] for passive and ac- tive neuro-mechanical [34] control strategies as well as better understanding of energetics [35] for movement on complex terrain. [1] R. McNeill Alexander. Principles of Animal Locomotion. Princeton University Press, 2003. [2] Michael H. Dickinson, Claire T. Farley, Robert J. Full, M. A. R. Koehl, Rodger Kram, and Steven Lehman. How animals move: An integrative view. Science, 288:100, 2000. [3] D. I. Goldman, T. S. Chen, D. M. Dudek, and R. J. Full. Dynamics of rapid vertical climbing in cockroaches reveals a template. Journal of Experimental Biology, 209(15):2990 -- 3000, 2006. [4] B. C. Jayne and D. J. Irschick. Effects of incline and speed on the three-dimensional hindlimb kinematics of a generalized iguanian lizard (dipsosaurus dorsalis). Jour- nal of Experimental Biology, 202(2):143 -- 159, 1999. [5] D. P. Ferris, K. L. Liang, and C. T. Farley. Runners adjust leg stiffness for their first step on a new running surface. Journal of Biomechanics, 32(8):787 -- 794, 1999. [6] A. A. Biewener and G. B. Gillis. Dynamics of mus- cle function during locomotion: Accommodating vari- able conditions. Journal of Experimental Biology, 202(23):3387 -- 3396, 1999. [7] D. I. Goldman, W. L. Korff, M. Wehner, M. S. Berns, and R. J. Full. The mechanism of rapid running in weak sand. Integrative and Comparative Biology, 46:E50 -- E50, 2006. [8] K. Autumn, S. T. Hsieh, D. M. Dudek, J. Chen, C. Chi- taphan, and R. J. Full. Dynamics of geckos running ver- tically. Journal of Experimental Biology, 209(2):260 -- 272, 2006. [9] P. Holmes, R. J. Full, D. Koditschek, and J. Gucken- heimer. The dynamics of legged locomotion: Models, analyses, and challenges. Siam Review, 48(2):207 -- 304, 2006. [10] Joseph C. Spagna, Daniel I. Goldman, Pei-Chun Lin, Daniel E. Koditschek, and Robert J. Full. Distributed mechanical feedback in arthropods and robots simplifies control of rapid running on challenging terrain. Bioin- spiration and Biomimetics, 2:9 -- 18, 2007. [11] Chen Li, Paul B. Umbanhowar, Haldun Komsuoglu, Daniel E. Koditschek, and Daniel I. Goldman. Sensitive dependence of the motion of a legged robot on granular media. Proceedings of the National Academy of Science, 106(9):3029 -- 3034, 2009. [12] J. Schmitt, M. Garcia, R. C. Razo, P. Holmes, and R. J. Full. Dynamics and stability of legged locomotion in the horizontal plane: a test case using insects. Biol. Cybern., 86:343, 2002. [13] J. D. Weingarten, G. A. D. Lopes, M. Buehler, R. E. Groff, and D. E. Koditschek. Automated gait adap- tation for legged robots. volume Vol.3 of 2004 IEEE International Conference on Robotics and Automation (IEEE Cat. No.04CH37508), pages 2153 -- 8, Piscataway, NJ, USA, 2004. IEEE. [14] D. E. Koditschek, R. J. Full, and M. Buehler. Mechanical aspects of legged locomotion control. Arthropod Structure and Development, 33(3):251 -- 272, 2004. [15] H. M. Jaeger, S. R. Nagel, and R. P. Behringer. The physics of granular material. Physics Today, 49(4):32 -- 38, 1996. [16] R.M. Nedderman. Statics and kinematics of granular ma- terials. Cambridge University Press, 1992. [17] G. Hill, S. Yeung, and S. A. Koehler. Scaling vertical drag forces in granular media. Europhysics Letters, 72(1):137 -- 143, 2005. [18] Paul B. Umbanhowar and Daniel I. Goldman. Granu- lar impact and the critical packing state. submitted to Physical Review E Rapid Communications, 2009. [19] A. N. Schofield and C. P. Wroth. Critical State Soil Me- chanics. McGraw-Hill, London, 1968. 9 [20] When tripods move in phase (e.g. dc = 1) this approxi- mation is exact as the body rests on the surface during the swing phase. [21] In our previous study [11] we measured leg depth to the bottom of the c-leg; here we measure it to the point on the c-leg furthest from the motor axle to simplify the expres- sion for penetration force vs. angle. As mg is comparable to kz for this study, the smallest penetration depth where rotary walking begins is close to the maximum possible value of 2R − h so that the simplified expression for the leg depth is nearly identical to the exact value. [22] In rotary walking when s < R, a limb encounters mate- rial disturbed by its previous step. If the initial volume fraction exceeds the critical value φ ≈ 0.605, disturbed granular material dilates to a lower volume fraction af- ter each step; if initially φ < 0.605, disturbed granu- lar material compacts to a higher volume fraction after each step[18]. The dilation of disturbed granular mate- rial above φ ≈ 0.605 results in premature transition from rotary walking to swimming if s < R as the disturbed granular material is weaker which increases penetration and reduces step length. Here we choose φ = 0.605 which ensures that the φ encountered by the leg is unchanged even for s < R. [23] The previous study did not err in considering vx as a function of ω alone since for fixed clock parameters ωs scales with ω. The only difference would be ∆t(φ) → ωs/ω∆t.. [24] Junfei Geng, D. Howell, E. Longhi, R. P. Behringer, G. Reydellet, L. Vanel, E. Cl´ement, and S. Luding. Foot- prints in sand: The response of a granular material to lo- cal perturbations. Physical Review Letters, 87(3):035506, Jul 2001. [25] R. Albert, M. A. Pfeifer, A. L. Barabasi, and P. Schiffer. Slow drag in a granular medium. Physical Review Letters, 82(1):205 -- 208, 1999. [26] For dc = 1, tripods are in phase so that each c-leg needs to provide just 1/6 the required total force com- pared to 1/3 when the tripods act independently which could affect the step size. However, we found that for {θs, θ0, dc} = {1.5,−0.5, 1}, decreasing m from 1/3 to 1/6 the body mass left the expression for s unchanged. [27] A. M. Hoover, E. Steltz, and R. S. Fearing. Roach: An autonomous 2.4 g crawling hexapod robot. IROS, pages 22 -- 26, 2008. [28] R. Playter, M. Buehler, and M. Raibert. Bigdog. In Dou- glas W. Gage Grant R. Gerhart, Charles M. Shoemaker, editor, Unmanned Ground Vehicle Technology VIII, vol- ume 6230 of Proceedings of SPIE, pages 62302O1 -- 62302O6, 2006. [29] M. McKee. Mars rover spirit gets stuck as winter ap- proaches. New Scientist, November, 2007. [30] G.W. Brown. Desert Biology. Academic Press, 1974. [31] Ryan D. Maladen, Yang Ding, Chen Li, and Daniel I. Goldman. Undulatory swimming in sand: subsurface lo- comotion of the sandfish lizard. Science, 325:314, 2009. [32] Nicole Mazouchova, Nick Gravish, Andrei Savu, and Daniel I. Goldman. Utilization of granular solidification during terrestrial locomotion of hatchling sea turtles. Bi- ology Letters, in press, 2010. [33] Robert J. Full and Daniel E. Koditschek. Templates and anchors: Neuromechanical hypotheses of legged locomo- tion on land. J. Exp. Bio., 202:3325 -- 3332, 1999. [34] K. Nishikawa, A. A. Biewener, P. Aerts, A. N. Ahn, 10 H. J. Chiel, M. A. Daley, T. L. Daniel, R. J. Full, M. E. Hale, T. L. Hedrick, A. K. Lappin, T. R. Nichols, R. D. Quinn, R. A. Satterlie, and B. Szymik. Neuromechanics: an integrative approach for understanding motor con- trol. Integrative and Comparative Biology, 47(1):16 -- 54, 2007. Times Cited: 13 Annual Meeting of the Society-for- Integrative-and-Comparative-Biology JAN 04-08, 2006 Orlando, FL. [35] T. M. Lejeune, P. A. Willems, and N. C. Heglund. Me- chanics and energetics of human locomotion on sand. Journal of Experimental Biology, 201(13):2071 -- 2080, 1998. Acknowledgements We thank Daniel Koditschek, Ryan Maladen, Yang Ding, Nick Gravish, and Predrag Cvitanovi´c for help- ful discussion. This work was supported by the Bur- roughs Wellcome Fund (D.I.G., C.L., and P.B.U.), the Army Research Laboratory (ARL) Micro Autonomous Systems and Technology (MAST) Collaborative Technol- ogy Alliance (CTA) under cooperative agreement num- ber W911NF-08-2-0004 (D.I.G. and P.B.U.), and the Na- tional Science Foundation (H.K.). Citation Information Chen Li, Paul B. Umbanhowar, Haldun Komsuoglu, Daniel I. Goldman, The effect of limb kinematics on the speed of a legged robot on granular media, Experimental Mechanics 50, 1383 -- 1393 (2010), DOI: 10.1007/s11340- 010-9347-1
1903.00549	1	1903	2019-03-01T21:47:07	Active Fingering Instability in Tissue Spreading	[ "physics.bio-ph", "cond-mat.soft", "q-bio.TO" ]	During the spreading of epithelial tissues, the advancing tissue front often develops fingerlike protrusions. Their resemblance to traditional viscous fingering patterns in driven fluids suggests that epithelial fingers could arise from an interfacial instability. However, the existence and physical mechanism of such a putative instability remain unclear. Here, based on an active polar fluid model for epithelial spreading, we analytically predict a generic instability of the tissue front. On the one hand, active cellular traction forces impose a velocity gradient that leads to an accelerated front, which is, thus, unstable to long-wavelength perturbations. On the other hand, contractile intercellular stresses typically dominate over surface tension in stabilizing short-wavelength perturbations. Finally, the finite range of hydrodynamic interactions in the tissue selects a wavelength for the fingering pattern, which is, thus, given by the smallest between the tissue size and the hydrodynamic screening length. Overall, we show that spreading epithelia experience an active fingering instability based on a simple kinematic mechanism. Moreover, our results underscore the crucial role of long-range hydrodynamic interactions in the dynamics of tissue morphology.	physics.bio-ph	physics	Active Fingering Instability in Tissue Spreading Ricard Alert,1, 2, ∗ Carles Blanch-Mercader,3, 4 and Jaume Casademunt1, 2 1Departament de F´ısica de la Mat`eria Condensada, Universitat de Barcelona, Av. Diagonal 647, 08028 Barcelona, Spain 2Universitat de Barcelona Institute of Complex Systems (UBICS), Universitat de Barcelona, Barcelona, Spain 3Laboratoire Physico Chimie Curie, Institut Curie, PSL Research University, CNRS, 26 rue d'Ulm, 75005 Paris, France 4Department of Biochemistry, Faculty of Sciences, University of Geneva, 30, Quai Ernest-Ansermet, 1205 Gen`eve, Switzerland 9 1 0 2 r a M 1 ] h p - o i b . s c i s y h p [ 1 v 9 4 5 0 0 . 3 0 9 1 : v i X r a (Dated: March 5, 2019) During the spreading of epithelial tissues, the advancing tissue front often develops finger-like protrusions. Their resemblance to traditional viscous fingering patterns in driven fluids suggests that epithelial fingers could arise from an interfacial instability. However, the existence and physical mechanism of such a putative instability remain unclear. Here, based on an active polar fluid model for epithelial spreading, we analytically predict a generic instability of the tissue front. On the one hand, active cellular traction forces impose a velocity gradient that leads to an accelerated front, which is, thus, unstable to long-wavelength perturbations. On the other hand, contractile intercellular stresses typically dominate over surface tension in stabilizing short-wave- length perturbations. Finally, the finite range of hydrodynamic interactions in the tissue selects a wavelength for the fingering pattern, which is, thus, given by the smallest between the tissue size and the hydrodynamic screening length. Overall, we show that spreading epithelia experience an active fingering instability based on a simple kinematic mechanism. Moreover, our results underscore the crucial role of long-range hydrodynamic interactions in the dynamics of tissue morphology. The spreading of epithelial monolayers by collective cell migration is crucial for tissue morphogenesis, wound heal- ing, and tumor progression. Both in vivo and in vitro, mul- ticellular protrusions called epithelial fingers often appear at the front of spreading tissues (Fig. 1) [1 -- 11]. This epithelial fingering resembles the viscous fingering that occurs via the Saffman-Taylor instability when a viscous fluid displaces a more viscous one [12, 13]. However, the mechanisms of these two phenomena must be different because epithelial monolay- ers are more viscous than the fluid that they displace. Hence, several models of epithelial fingering have been proposed [3]. To induce finger formation, some models directly imple- ment leader cells with a distinct behavior, either via special particles [14] or via a dependence of the magnitude of cell motility forces on the curvature of the tissue front [15, 16]. Other models recapitulate epithelial fingering by introducing alignment between cell motility forces and the tissue veloc- ity field [17]. These models predict a moving front to be stable and a non-moving front to exhibit an instability with an unbounded growth rate for a number of finite wavelengths [18, 19]. Fingers were also observed in the numerical solu- tion of other continuum models of spreading epithelia, either treated as active polar fluids [20] or as active nematics with cell proliferation [21]. Recently, fingering was also found in a parameter range of an active vertex model [22]. Finally, inter- face undulations can emerge from the coupling of chemotactic fields to the mechanics of epithelial spreading [23 -- 26]. Despite the many efforts, the physical mechanism of the fingering instability in epithelia remains a matter of debate. Here, we address this problem by means of a continuum active polar fluid model for epithelial spreading. The model includes hydrodynamic interactions through the tissue, and it imple- ments neither leader-cell behavior nor alignment between cel- lular traction forces and the flow field. Yet, we analytically predict a long-wavelength instability of the moving front that FIG. 1. Fingering in epithelial spreading. Scale bar, 200 µm. Adapted from [6] with permission from Pascal Silberzan. explains epithelial fingering. The instability is based on a generic kinematic mechanism, namely the front acceleration associated to a fixed velocity gradient. In spreading epithelia, the velocity gradient is imposed by active traction forces at the edge of the viscous cell monolayer. The fastest-growing mode has a finite wavelength, typically a few hundreds of microm- eters, consistent with the measured finger spacing [11]. This characteristic wavelength is selected by the long-range hydro- dynamic interactions in the tissue, which are either limited by the tissue size or screened by cell-substrate friction forces. The model also shows that intercellular contractility stabilizes short-wavelength perturbations of the tissue boundary. The stabilizing effect of contractility is typically stronger than that of tissue surface tension. Globally, our analysis shows how, as a result of the flows induced by the traction force field, a mor- phological instability may naturally take place in a spreading cell monolayer. Leader cells could then appear upon the onset of the instability, influencing finger development. Model. -- We base our analysis on a continuum active po- lar fluid model of epithelial spreading, which is thus described in terms of a polarity field (cid:126)p ((cid:126)r, t) and a velocity field (cid:126)v ((cid:126)r, t) [27 -- 29]. We neglect cell proliferation and the bulk elastic- ity of the monolayer, which eventually limit the spreading process [6, 17, 30 -- 32]. Tissue spreading is primarily driven 90min12h12h90minby the traction forces exerted by cells close to the monolayer edge, which polarize perpendicularly to the edge by extending lamellipodia towards free space. In contrast, the inner region of the monolayer remains essentially unpolarized, featuring much weaker and transient traction forces [27, 28]. Hence, we take a free energy for the polarity field that favors the un- polarized state p = 0 in the bulk, with a restoring coefficient a > 0, and we impose a normal and maximal polarity as a boundary condition at the tissue edge. In addition, the polar free energy includes a cost for polarity gradients, with K the Frank constant of nematic elasticity in the one-constant ap- proximation [33]. Altogether, (cid:21) (cid:90) (cid:20) a F = pαpα + K 2 (∂αpβ) (∂αpβ) 2 d3(cid:126)r. (1) We assume that the polarity field is set by flow-independent mechanisms, so that it follows a purely relaxational dynamics, and that it equilibrates fast compared to the spreading dynam- ics [34]. Hence, δF/δpα = 0, which yields (cid:112) L2 c∇2pα = pα, (2) K/a is the characteristic length with which the where Lc = polarity modulus decays from p = 1 at the monolayer edge to p = 0 at the center. Then, force balance imposes ∂βσαβ + fα = 0, (3) where σαβ is the stress tensor of the monolayer, and fα is the external force density acting on it. Since tissue spreading occurs over time scales of several hours (Fig. 1), we neglect the elastic response of the tissue [34]. Thus, we relate tissue forces to the polarity and velocity fields via the following con- stitutive equations for an active polar fluid [35] (see discussion and justification in Ref. [34]): σαβ = η (∂αvβ + ∂βvα) − ζpαpβ, fα = −ξvα + ζipα. (4a) (4b) Here, η is the effective monolayer viscosity, and ξ is the cell- substrate friction coefficient. Respectively, ζ < 0 is the active stress coefficient accounting for the contractility of polarized cells, and ζi > 0 is the contact active force coefficient ac- counting for the maximal traction stress exerted by polarized cells on the substrate, T0 = ζih, with h the monolayer height. Stability of the tissue front. -- To study the stability of the advancing front, we consider a rectangular monolayer typical of in vitro experiments (Fig. 1). Thus, the reference state is the flat front solution with (cid:126)p = p0 x(x) x [34] (dashed lines in Fig. 2a). In addition to a maximal normal polarity at the edges, we impose stress-free boundary condi- tions. For an interface of arbitrary shape, (cid:126)p (x = ±L) = n±, and σ · n±x=±L = (cid:126)0, respectively, where n± is the nor- mal unit vector of the top and bottom interfaces. The tissue width L changes according to dL/dt = (cid:126)v · nx=L. Then, x(x) x and (cid:126)v = v0 2 FIG. 2. Instability of the monolayer front. (a) Sketch of the peristaltic perturbations. Dashed lines indicate the flat, unperturbed interface. The dotted line indicates the symmetry axis of the monolayer. (b) Growth rate of the perturbations. Parameter values are in Table I. motivated by experimental observations (Fig. 1), we intro- duce peristaltic small-amplitude perturbations of the flat inter- face, namely those that modify the monolayer width (Fig. 2a): L(y) = L0 + δL(y). From a linear stability analysis [34], we obtain that the growth rate ω(q) of such perturbations is always real, so that no oscillatory behavior is expected. How- ever, the growth rate evidences a long-wavelength instability of the monolayer front. Moreover, the fastest-growing pertur- bation has a finite wavelength (Fig. 2b). In the following, we analyze the contribution of the different forces to the instabil- ity, which allows us to single out its physical mechanism. Traction forces. -- We first consider a limit case with neither intercellular contractility nor cell-substrate friction, ζ, ξ → 0 [27]. In addition, we also consider that the width of the polarized boundary layer of cells is much smaller than the total tissue width, Lc (cid:28) L0, which is generally the case in experiments [27, 28]. In this limit, since active forces are concentrated at the narrow boundary layer, most of the tissue behaves as a passive viscous fluid, for which ∂xσxx ≈ 0 and σxx ≈ 2η dvx/dx. Therefore, the stress is uniform through- out most of the tissue, with a value given by the stress accu- mulated across the boundary layer, namely σxx ≈ T0Lc/h. Consequently, the velocity gradient is also fixed and uniform, and hence the spreading velocity vx(L0) = V0 = dL0/dt reads V0 ≈ T0Lc 2ηh L0 ≡ L0 τ , (5) where we have used that vx(0) = 0. This result means that, due to the sole action of a constant traction force, the flat front accelerates, consistent with measurements [6, 36] and with the size dependence of tissue wetting [28]. Consequently, the q = 0 perturbation mode is unstable, ω (q = 0) = τ−1 > 0, since any uniform displacement of the advancing front makes it depart from its original velocity. Thus, the instability mech- anism is kinematic in nature: The advanced regions of the front move faster than the trailing regions (Fig. 2a). Thereby, traction forces contribute to destabilize perturbations of all wavelengths (Fig. 3a). Therefore, since other forces such as surface tension stabilize short-wavelength perturbations, the tissue front experiences a long-wavelength instability. (b)<latexit sha1_base64="6Q9PuuxDV4TJqsQaqeM0IXZoWHA=">AAACAXicbVBPS8MwHE39O+e/qkcPBocwL6MTQb3IwIvHCdYN1jLSLN3CkrYkvwql7KZfRi8ievIr+AX8NqazB918p5f3XkLeCxLBNTjOl7WwuLS8slpZq65vbG5t2zu7dzpOFWUujUWsugHRTPCIucBBsG6iGJGBYJ1gfFX4nXumNI+jW8gS5ksyjHjIKQEj9e0DL4wjCInkIssTkBPPvMUoFCquB8d9u+Y0nCnwPGmWpIZKtPv2pzeIaSpZBFQQrXtNJwE/Jwo4FWxS9VLNEkLHZMh6hkZEMu3n0yITfBTGCsOI4en5dzYnUutMBiYjCYz0rFeI/3m9FMJzP+dRkgKLqIkYL0wFhhgXe+ABV6auyAwhVHHzS0xHRBEKZrWqqd+cLTtP3JPGRcO5Oa21LssdKmgfHaI6aqIz1ELXqI1cRNEjekZv6N16sJ6sF+v1J7pglXf20B9YH99rDpeJ</latexit><latexit sha1_base64="6Q9PuuxDV4TJqsQaqeM0IXZoWHA=">AAACAXicbVBPS8MwHE39O+e/qkcPBocwL6MTQb3IwIvHCdYN1jLSLN3CkrYkvwql7KZfRi8ievIr+AX8NqazB918p5f3XkLeCxLBNTjOl7WwuLS8slpZq65vbG5t2zu7dzpOFWUujUWsugHRTPCIucBBsG6iGJGBYJ1gfFX4nXumNI+jW8gS5ksyjHjIKQEj9e0DL4wjCInkIssTkBPPvMUoFCquB8d9u+Y0nCnwPGmWpIZKtPv2pzeIaSpZBFQQrXtNJwE/Jwo4FWxS9VLNEkLHZMh6hkZEMu3n0yITfBTGCsOI4en5dzYnUutMBiYjCYz0rFeI/3m9FMJzP+dRkgKLqIkYL0wFhhgXe+ABV6auyAwhVHHzS0xHRBEKZrWqqd+cLTtP3JPGRcO5Oa21LssdKmgfHaI6aqIz1ELXqI1cRNEjekZv6N16sJ6sF+v1J7pglXf20B9YH99rDpeJ</latexit><latexit sha1_base64="6Q9PuuxDV4TJqsQaqeM0IXZoWHA=">AAACAXicbVBPS8MwHE39O+e/qkcPBocwL6MTQb3IwIvHCdYN1jLSLN3CkrYkvwql7KZfRi8ievIr+AX8NqazB918p5f3XkLeCxLBNTjOl7WwuLS8slpZq65vbG5t2zu7dzpOFWUujUWsugHRTPCIucBBsG6iGJGBYJ1gfFX4nXumNI+jW8gS5ksyjHjIKQEj9e0DL4wjCInkIssTkBPPvMUoFCquB8d9u+Y0nCnwPGmWpIZKtPv2pzeIaSpZBFQQrXtNJwE/Jwo4FWxS9VLNEkLHZMh6hkZEMu3n0yITfBTGCsOI4en5dzYnUutMBiYjCYz0rFeI/3m9FMJzP+dRkgKLqIkYL0wFhhgXe+ABV6auyAwhVHHzS0xHRBEKZrWqqd+cLTtP3JPGRcO5Oa21LssdKmgfHaI6aqIz1ELXqI1cRNEjekZv6N16sJ6sF+v1J7pglXf20B9YH99rDpeJ</latexit>(a)<latexit sha1_base64="13scbnor7kbqrmfLK7qZH9VokR0=">AAACAXicbVBPS8MwHE39O+e/qkcPBocwL6MTQb3IwIvHCdYN1jLSLN3CkrYkvwql7KZfRi8ievIr+AX8NqazB918p5f3XkLeCxLBNTjOl7WwuLS8slpZq65vbG5t2zu7dzpOFWUujUWsugHRTPCIucBBsG6iGJGBYJ1gfFX4nXumNI+jW8gS5ksyjHjIKQEj9e0DL4wjCInkIssTkBPPvMUoFCquk+O+XXMazhR4njRLUkMl2n370xvENJUsAiqI1r2mk4CfEwWcCjapeqlmCaFjMmQ9QyMimfbzaZEJPgpjhWHE8PT8O5sTqXUmA5ORBEZ61ivE/7xeCuG5n/MoSYFF1ESMF6YCQ4yLPfCAK1NXZIYQqrj5JaYjoggFs1rV1G/Olp0n7knjouHcnNZal+UOFbSPDlEdNdEZaqFr1EYuougRPaM39G49WE/Wi/X6E12wyjt76A+sj29pkJeI</latexit><latexit sha1_base64="13scbnor7kbqrmfLK7qZH9VokR0=">AAACAXicbVBPS8MwHE39O+e/qkcPBocwL6MTQb3IwIvHCdYN1jLSLN3CkrYkvwql7KZfRi8ievIr+AX8NqazB918p5f3XkLeCxLBNTjOl7WwuLS8slpZq65vbG5t2zu7dzpOFWUujUWsugHRTPCIucBBsG6iGJGBYJ1gfFX4nXumNI+jW8gS5ksyjHjIKQEj9e0DL4wjCInkIssTkBPPvMUoFCquk+O+XXMazhR4njRLUkMl2n370xvENJUsAiqI1r2mk4CfEwWcCjapeqlmCaFjMmQ9QyMimfbzaZEJPgpjhWHE8PT8O5sTqXUmA5ORBEZ61ivE/7xeCuG5n/MoSYFF1ESMF6YCQ4yLPfCAK1NXZIYQqrj5JaYjoggFs1rV1G/Olp0n7knjouHcnNZal+UOFbSPDlEdNdEZaqFr1EYuougRPaM39G49WE/Wi/X6E12wyjt76A+sj29pkJeI</latexit><latexit sha1_base64="13scbnor7kbqrmfLK7qZH9VokR0=">AAACAXicbVBPS8MwHE39O+e/qkcPBocwL6MTQb3IwIvHCdYN1jLSLN3CkrYkvwql7KZfRi8ievIr+AX8NqazB918p5f3XkLeCxLBNTjOl7WwuLS8slpZq65vbG5t2zu7dzpOFWUujUWsugHRTPCIucBBsG6iGJGBYJ1gfFX4nXumNI+jW8gS5ksyjHjIKQEj9e0DL4wjCInkIssTkBPPvMUoFCquk+O+XXMazhR4njRLUkMl2n370xvENJUsAiqI1r2mk4CfEwWcCjapeqlmCaFjMmQ9QyMimfbzaZEJPgpjhWHE8PT8O5sTqXUmA5ORBEZ61ivE/7xeCuG5n/MoSYFF1ESMF6YCQ4yLPfCAK1NXZIYQqrj5JaYjoggFs1rV1G/Olp0n7knjouHcnNZal+UOFbSPDlEdNdEZaqFr1EYuougRPaM39G49WE/Wi/X6E12wyjt76A+sj29pkJeI</latexit>~p<latexit sha1_base64="OlnMx+5SZTECR3GTBeZ32C9qDp4=">AAAB53icbZDNSsNAFIVv6l+tf1WXbgaL4KokIqgbKbhxWcHYQhvKZHrTjp1JwsykUELfQTciuvJ9fAHfxmnNQlvP6pt7zsA9N0wF18Z1v5zSyura+kZ5s7K1vbO7V90/eNBJphj6LBGJaodUo+Ax+oYbge1UIZWhwFY4upn5rTEqzZP43kxSDCQdxDzijBo7anXHyPJ02qvW3Lo7F1kGr4AaFGr2qp/dfsIyibFhgmrd8dzUBDlVhjOB00o305hSNqID7FiMqUQd5PN1p+QkShQxQyTz9+9sTqXWExnajKRmqBe92fA/r5OZ6DLIeZxmBmNmI9aLMkFMQmatSZ8rZEZMLFCmuN2SsCFVlBl7m4qt7y2WXQb/rH5Vd+/Oa43r4g5lOIJjOAUPLqABt9AEHxiM4Bne4N15dJ6cF+f1J1pyij+H8EfOxzf36Y0m</latexit><latexit sha1_base64="OlnMx+5SZTECR3GTBeZ32C9qDp4=">AAAB53icbZDNSsNAFIVv6l+tf1WXbgaL4KokIqgbKbhxWcHYQhvKZHrTjp1JwsykUELfQTciuvJ9fAHfxmnNQlvP6pt7zsA9N0wF18Z1v5zSyura+kZ5s7K1vbO7V90/eNBJphj6LBGJaodUo+Ax+oYbge1UIZWhwFY4upn5rTEqzZP43kxSDCQdxDzijBo7anXHyPJ02qvW3Lo7F1kGr4AaFGr2qp/dfsIyibFhgmrd8dzUBDlVhjOB00o305hSNqID7FiMqUQd5PN1p+QkShQxQyTz9+9sTqXWExnajKRmqBe92fA/r5OZ6DLIeZxmBmNmI9aLMkFMQmatSZ8rZEZMLFCmuN2SsCFVlBl7m4qt7y2WXQb/rH5Vd+/Oa43r4g5lOIJjOAUPLqABt9AEHxiM4Bne4N15dJ6cF+f1J1pyij+H8EfOxzf36Y0m</latexit><latexit sha1_base64="OlnMx+5SZTECR3GTBeZ32C9qDp4=">AAAB53icbZDNSsNAFIVv6l+tf1WXbgaL4KokIqgbKbhxWcHYQhvKZHrTjp1JwsykUELfQTciuvJ9fAHfxmnNQlvP6pt7zsA9N0wF18Z1v5zSyura+kZ5s7K1vbO7V90/eNBJphj6LBGJaodUo+Ax+oYbge1UIZWhwFY4upn5rTEqzZP43kxSDCQdxDzijBo7anXHyPJ02qvW3Lo7F1kGr4AaFGr2qp/dfsIyibFhgmrd8dzUBDlVhjOB00o305hSNqID7FiMqUQd5PN1p+QkShQxQyTz9+9sTqXWExnajKRmqBe92fA/r5OZ6DLIeZxmBmNmI9aLMkFMQmatSZ8rZEZMLFCmuN2SsCFVlBl7m4qt7y2WXQb/rH5Vd+/Oa43r4g5lOIJjOAUPLqABt9AEHxiM4Bne4N15dJ6cF+f1J1pyij+H8EfOxzf36Y0m</latexit>L(y)<latexit sha1_base64="5+T/HmaMpTWcRwzORxSKoJAMp5o=">AAAB63icbZDNSsNAFIUn/tb6V3XpZrAIdVMSEdSNFNy4cFHB2EIaymR60w6dzISZiRBC30I3IrrybXwB38ZpzUJbz+qbe87APTdKOdPGdb+cpeWV1bX1ykZ1c2t7Z7e2t/+gZaYo+FRyqboR0cCZAN8ww6GbKiBJxKETja+nfucRlGZS3Js8hTAhQ8FiRomxo6A3AG4Ivm3kJ/1a3W26M+FF8Eqoo1Ltfu2zN5A0S0AYyonWgeemJiyIMoxymFR7mYaU0DEZQmBRkAR0WMxWnuDjWCpsRoBn79/ZgiRa50lkMwkxIz3vTYf/eUFm4ouwYCLNDAhqI9aLM46NxNPmeMAUUMNzC4QqZrfEdEQUocbep2rre/NlF8E/bV423buzeuuqvEMFHaIj1EAeOkctdIPayEcUSfSM3tC7I5wn58V5/YkuOeWfA/RHzsc3fFKN6A==</latexit><latexit sha1_base64="5+T/HmaMpTWcRwzORxSKoJAMp5o=">AAAB63icbZDNSsNAFIUn/tb6V3XpZrAIdVMSEdSNFNy4cFHB2EIaymR60w6dzISZiRBC30I3IrrybXwB38ZpzUJbz+qbe87APTdKOdPGdb+cpeWV1bX1ykZ1c2t7Z7e2t/+gZaYo+FRyqboR0cCZAN8ww6GbKiBJxKETja+nfucRlGZS3Js8hTAhQ8FiRomxo6A3AG4Ivm3kJ/1a3W26M+FF8Eqoo1Ltfu2zN5A0S0AYyonWgeemJiyIMoxymFR7mYaU0DEZQmBRkAR0WMxWnuDjWCpsRoBn79/ZgiRa50lkMwkxIz3vTYf/eUFm4ouwYCLNDAhqI9aLM46NxNPmeMAUUMNzC4QqZrfEdEQUocbep2rre/NlF8E/bV423buzeuuqvEMFHaIj1EAeOkctdIPayEcUSfSM3tC7I5wn58V5/YkuOeWfA/RHzsc3fFKN6A==</latexit><latexit sha1_base64="5+T/HmaMpTWcRwzORxSKoJAMp5o=">AAAB63icbZDNSsNAFIUn/tb6V3XpZrAIdVMSEdSNFNy4cFHB2EIaymR60w6dzISZiRBC30I3IrrybXwB38ZpzUJbz+qbe87APTdKOdPGdb+cpeWV1bX1ykZ1c2t7Z7e2t/+gZaYo+FRyqboR0cCZAN8ww6GbKiBJxKETja+nfucRlGZS3Js8hTAhQ8FiRomxo6A3AG4Ivm3kJ/1a3W26M+FF8Eqoo1Ltfu2zN5A0S0AYyonWgeemJiyIMoxymFR7mYaU0DEZQmBRkAR0WMxWnuDjWCpsRoBn79/ZgiRa50lkMwkxIz3vTYf/eUFm4ouwYCLNDAhqI9aLM46NxNPmeMAUUMNzC4QqZrfEdEQUocbep2rre/NlF8E/bV423buzeuuqvEMFHaIj1EAeOkctdIPayEcUSfSM3tC7I5wn58V5/YkuOeWfA/RHzsc3fFKN6A==</latexit>L(y)<latexit sha1_base64="H2hf0sJQ898sqfZpOQgRrqL5NiE=">AAAB5HicbZDNSsNAFIVv/K31r+rSzWAR6qYkIqgbKbhx4aKCsYU2lMn0phk6+WFmIpTQR9CNiK58Il/At3FSs9DWs/rmnjNwz/VTwZW27S9raXlldW29slHd3Nre2a3t7T+oJJMMXZaIRHZ9qlDwGF3NtcBuKpFGvsCOP74u/M4jSsWT+F5PUvQiOop5wBnVxei2MTkZ1Op2056JLIJTQh1KtQe1z/4wYVmEsWaCKtVz7FR7OZWaM4HTaj9TmFI2piPsGYxphMrLZ7tOyXGQSKJDJLP372xOI6UmkW8yEdWhmveK4X9eL9PBhZfzOM00xsxEjBdkguiEFJXJkEtkWkwMUCa52ZKwkErKtDlM1dR35ssugnvavGzad2f11lV5hwocwhE0wIFzaMENtMEFBiE8wxu8WyPryXqxXn+iS1b55wD+yPr4BqXlixw=</latexit><latexit sha1_base64="H2hf0sJQ898sqfZpOQgRrqL5NiE=">AAAB5HicbZDNSsNAFIVv/K31r+rSzWAR6qYkIqgbKbhx4aKCsYU2lMn0phk6+WFmIpTQR9CNiK58Il/At3FSs9DWs/rmnjNwz/VTwZW27S9raXlldW29slHd3Nre2a3t7T+oJJMMXZaIRHZ9qlDwGF3NtcBuKpFGvsCOP74u/M4jSsWT+F5PUvQiOop5wBnVxei2MTkZ1Op2056JLIJTQh1KtQe1z/4wYVmEsWaCKtVz7FR7OZWaM4HTaj9TmFI2piPsGYxphMrLZ7tOyXGQSKJDJLP372xOI6UmkW8yEdWhmveK4X9eL9PBhZfzOM00xsxEjBdkguiEFJXJkEtkWkwMUCa52ZKwkErKtDlM1dR35ssugnvavGzad2f11lV5hwocwhE0wIFzaMENtMEFBiE8wxu8WyPryXqxXn+iS1b55wD+yPr4BqXlixw=</latexit><latexit sha1_base64="H2hf0sJQ898sqfZpOQgRrqL5NiE=">AAAB5HicbZDNSsNAFIVv/K31r+rSzWAR6qYkIqgbKbhx4aKCsYU2lMn0phk6+WFmIpTQR9CNiK58Il/At3FSs9DWs/rmnjNwz/VTwZW27S9raXlldW29slHd3Nre2a3t7T+oJJMMXZaIRHZ9qlDwGF3NtcBuKpFGvsCOP74u/M4jSsWT+F5PUvQiOop5wBnVxei2MTkZ1Op2056JLIJTQh1KtQe1z/4wYVmEsWaCKtVz7FR7OZWaM4HTaj9TmFI2piPsGYxphMrLZ7tOyXGQSKJDJLP372xOI6UmkW8yEdWhmveK4X9eL9PBhZfzOM00xsxEjBdkguiEFJXJkEtkWkwMUCa52ZKwkErKtDlM1dR35ssugnvavGzad2f11lV5hwocwhE0wIFzaMENtMEFBiE8wxu8WyPryXqxXn+iS1b55wD+yPr4BqXlixw=</latexit>y<latexit sha1_base64="jU7T+sj0mx/3agfxHb6HMXgLoPY=">AAAB4XicbZDNSsNAFIVv6l+tf1WXbgaL4KokIqgbKbhx2YKxhTaUyfSmHTqZhJmJEEJfQDciuvKVfAHfxmnNQqtn9c09Z+CeG6aCa+O6n05lZXVtfaO6Wdva3tndq+8f3OskUwx9lohE9UKqUXCJvuFGYC9VSONQYDec3sz97gMqzRN5Z/IUg5iOJY84o8aOOvmw3nCb7kLkL3glNKBUe1j/GIwSlsUoDRNU677npiYoqDKcCZzVBpnGlLIpHWPfoqQx6qBYLDojJ1GiiJkgWbx/Zgsaa53Hoc3E1Ez0sjcf/uf1MxNdBgWXaWZQMhuxXpQJYhIy70tGXCEzIrdAmeJ2S8ImVFFm7FVqtr63XPYv+GfNq6bbOW+0rss7VOEIjuEUPLiAFtxCG3xggPAEr/DmjJxH59l5+Y5WnPLPIfyS8/4FS+OKYQ==</latexit><latexit sha1_base64="jU7T+sj0mx/3agfxHb6HMXgLoPY=">AAAB4XicbZDNSsNAFIVv6l+tf1WXbgaL4KokIqgbKbhx2YKxhTaUyfSmHTqZhJmJEEJfQDciuvKVfAHfxmnNQqtn9c09Z+CeG6aCa+O6n05lZXVtfaO6Wdva3tndq+8f3OskUwx9lohE9UKqUXCJvuFGYC9VSONQYDec3sz97gMqzRN5Z/IUg5iOJY84o8aOOvmw3nCb7kLkL3glNKBUe1j/GIwSlsUoDRNU677npiYoqDKcCZzVBpnGlLIpHWPfoqQx6qBYLDojJ1GiiJkgWbx/Zgsaa53Hoc3E1Ez0sjcf/uf1MxNdBgWXaWZQMhuxXpQJYhIy70tGXCEzIrdAmeJ2S8ImVFFm7FVqtr63XPYv+GfNq6bbOW+0rss7VOEIjuEUPLiAFtxCG3xggPAEr/DmjJxH59l5+Y5WnPLPIfyS8/4FS+OKYQ==</latexit><latexit sha1_base64="jU7T+sj0mx/3agfxHb6HMXgLoPY=">AAAB4XicbZDNSsNAFIVv6l+tf1WXbgaL4KokIqgbKbhx2YKxhTaUyfSmHTqZhJmJEEJfQDciuvKVfAHfxmnNQqtn9c09Z+CeG6aCa+O6n05lZXVtfaO6Wdva3tndq+8f3OskUwx9lohE9UKqUXCJvuFGYC9VSONQYDec3sz97gMqzRN5Z/IUg5iOJY84o8aOOvmw3nCb7kLkL3glNKBUe1j/GIwSlsUoDRNU677npiYoqDKcCZzVBpnGlLIpHWPfoqQx6qBYLDojJ1GiiJkgWbx/Zgsaa53Hoc3E1Ez0sjcf/uf1MxNdBgWXaWZQMhuxXpQJYhIy70tGXCEzIrdAmeJ2S8ImVFFm7FVqtr63XPYv+GfNq6bbOW+0rss7VOEIjuEUPLiAFtxCG3xggPAEr/DmjJxH59l5+Y5WnPLPIfyS8/4FS+OKYQ==</latexit>x<latexit sha1_base64="m0ovzmCR6F2l2+TLGv95woygQtY=">AAAB4XicbZDNSsNAFIVv6l+tf1WXbgaL4KokIqgbKbhx2YKxhTaUyfSmHTrJhJmJWEJfQDciuvKVfAHfxmnNQlvP6pt7zsA9N0wF18Z1v5zSyura+kZ5s7K1vbO7V90/uNcyUwx9JoVUnZBqFDxB33AjsJMqpHEosB2Ob2Z++wGV5jK5M5MUg5gOEx5xRo0dtR771Zpbd+ciy+AVUINCzX71szeQLIsxMUxQrbuem5ogp8pwJnBa6WUaU8rGdIhdiwmNUQf5fNEpOYmkImaEZP7+nc1prPUkDm0mpmakF73Z8D+vm5noMsh5kmYGE2Yj1osyQYwks75kwBUyIyYWKFPcbknYiCrKjL1Kxdb3Fssug39Wv6q7rfNa47q4QxmO4BhOwYMLaMAtNMEHBgjP8AbvzsB5cl6c159oySn+HMIfOR/fSmaKYA==</latexit><latexit sha1_base64="m0ovzmCR6F2l2+TLGv95woygQtY=">AAAB4XicbZDNSsNAFIVv6l+tf1WXbgaL4KokIqgbKbhx2YKxhTaUyfSmHTrJhJmJWEJfQDciuvKVfAHfxmnNQlvP6pt7zsA9N0wF18Z1v5zSyura+kZ5s7K1vbO7V90/uNcyUwx9JoVUnZBqFDxB33AjsJMqpHEosB2Ob2Z++wGV5jK5M5MUg5gOEx5xRo0dtR771Zpbd+ciy+AVUINCzX71szeQLIsxMUxQrbuem5ogp8pwJnBa6WUaU8rGdIhdiwmNUQf5fNEpOYmkImaEZP7+nc1prPUkDm0mpmakF73Z8D+vm5noMsh5kmYGE2Yj1osyQYwks75kwBUyIyYWKFPcbknYiCrKjL1Kxdb3Fssug39Wv6q7rfNa47q4QxmO4BhOwYMLaMAtNMEHBgjP8AbvzsB5cl6c159oySn+HMIfOR/fSmaKYA==</latexit><latexit sha1_base64="m0ovzmCR6F2l2+TLGv95woygQtY=">AAAB4XicbZDNSsNAFIVv6l+tf1WXbgaL4KokIqgbKbhx2YKxhTaUyfSmHTrJhJmJWEJfQDciuvKVfAHfxmnNQlvP6pt7zsA9N0wF18Z1v5zSyura+kZ5s7K1vbO7V90/uNcyUwx9JoVUnZBqFDxB33AjsJMqpHEosB2Ob2Z++wGV5jK5M5MUg5gOEx5xRo0dtR771Zpbd+ciy+AVUINCzX71szeQLIsxMUxQrbuem5ogp8pwJnBa6WUaU8rGdIhdiwmNUQf5fNEpOYmkImaEZP7+nc1prPUkDm0mpmakF73Z8D+vm5noMsh5kmYGE2Yj1osyQYwks75kwBUyIyYWKFPcbknYiCrKjL1Kxdb3Fssug39Wv6q7rfNa47q4QxmO4BhOwYMLaMAtNMEHBgjP8AbvzsB5cl6c159oySn+HMIfOR/fSmaKYA==</latexit>L0<latexit sha1_base64="/Zs7oTQp9nmdd8tBAKpRbLbzyrk=">AAAB43icbZDNSgMxFIXv1L9a/6ou3QSL4KpkRFA3UnDjwkVFxxbaoWTSO21o5ockI5Shb6AbEV35Rr6Ab2NaZ6GtZ/XlnhO45wapFNpQ+uWUlpZXVtfK65WNza3tneru3oNOMsXR44lMVDtgGqWI0TPCSGynClkUSGwFo6up33pEpUUS35txin7EBrEIBWfGju5uerRXrdE6nYksgltADQo1e9XPbj/hWYSx4ZJp3XFpavycKSO4xEmlm2lMGR+xAXYsxixC7eezVSfkKEwUMUMks/fvbM4ircdRYDMRM0M9702H/3mdzITnfi7iNDMYcxuxXphJYhIybUz6QiE3cmyBcSXsloQPmWLc2LtUbH13vuwieCf1izq9Pa01Los7lOEADuEYXDiDBlxDEzzgMIBneIN3J3SenBfn9Sdacoo/+/BHzsc3KH+K1w==</latexit><latexit sha1_base64="/Zs7oTQp9nmdd8tBAKpRbLbzyrk=">AAAB43icbZDNSgMxFIXv1L9a/6ou3QSL4KpkRFA3UnDjwkVFxxbaoWTSO21o5ockI5Shb6AbEV35Rr6Ab2NaZ6GtZ/XlnhO45wapFNpQ+uWUlpZXVtfK65WNza3tneru3oNOMsXR44lMVDtgGqWI0TPCSGynClkUSGwFo6up33pEpUUS35txin7EBrEIBWfGju5uerRXrdE6nYksgltADQo1e9XPbj/hWYSx4ZJp3XFpavycKSO4xEmlm2lMGR+xAXYsxixC7eezVSfkKEwUMUMks/fvbM4ircdRYDMRM0M9702H/3mdzITnfi7iNDMYcxuxXphJYhIybUz6QiE3cmyBcSXsloQPmWLc2LtUbH13vuwieCf1izq9Pa01Los7lOEADuEYXDiDBlxDEzzgMIBneIN3J3SenBfn9Sdacoo/+/BHzsc3KH+K1w==</latexit><latexit sha1_base64="/Zs7oTQp9nmdd8tBAKpRbLbzyrk=">AAAB43icbZDNSgMxFIXv1L9a/6ou3QSL4KpkRFA3UnDjwkVFxxbaoWTSO21o5ockI5Shb6AbEV35Rr6Ab2NaZ6GtZ/XlnhO45wapFNpQ+uWUlpZXVtfK65WNza3tneru3oNOMsXR44lMVDtgGqWI0TPCSGynClkUSGwFo6up33pEpUUS35txin7EBrEIBWfGju5uerRXrdE6nYksgltADQo1e9XPbj/hWYSx4ZJp3XFpavycKSO4xEmlm2lMGR+xAXYsxixC7eezVSfkKEwUMUMks/fvbM4ircdRYDMRM0M9702H/3mdzITnfi7iNDMYcxuxXphJYhIybUz6QiE3cmyBcSXsloQPmWLc2LtUbH13vuwieCf1izq9Pa01Los7lOEADuEYXDiDBlxDEzzgMIBneIN3J3SenBfn9Sdacoo/+/BHzsc3KH+K1w==</latexit>00.050.10.150.20.250.300.010.020.030.040.05Growthrate!(h1)Wavenumberq(µm1)(c)<latexit sha1_base64="gikEPqhu3Mj65Jcm+ICZUaFxUsk=">AAACAXicbVBPS8MwHE39O+e/qkcPBocwL6MTQb3IwIvHCdYN1jLSLN3CkrYkvwql7KZfRi8ievIr+AX8NqazB918p5f3XkLeCxLBNTjOl7WwuLS8slpZq65vbG5t2zu7dzpOFWUujUWsugHRTPCIucBBsG6iGJGBYJ1gfFX4nXumNI+jW8gS5ksyjHjIKQEj9e0DL4wjCInkIssTkBPPvMUoFCqu0+O+XXMazhR4njRLUkMl2n370xvENJUsAiqI1r2mk4CfEwWcCjapeqlmCaFjMmQ9QyMimfbzaZEJPgpjhWHE8PT8O5sTqXUmA5ORBEZ61ivE/7xeCuG5n/MoSYFF1ESMF6YCQ4yLPfCAK1NXZIYQqrj5JaYjoggFs1rV1G/Olp0n7knjouHcnNZal+UOFbSPDlEdNdEZaqFr1EYuougRPaM39G49WE/Wi/X6E12wyjt76A+sj29sjJeK</latexit><latexit sha1_base64="gikEPqhu3Mj65Jcm+ICZUaFxUsk=">AAACAXicbVBPS8MwHE39O+e/qkcPBocwL6MTQb3IwIvHCdYN1jLSLN3CkrYkvwql7KZfRi8ievIr+AX8NqazB918p5f3XkLeCxLBNTjOl7WwuLS8slpZq65vbG5t2zu7dzpOFWUujUWsugHRTPCIucBBsG6iGJGBYJ1gfFX4nXumNI+jW8gS5ksyjHjIKQEj9e0DL4wjCInkIssTkBPPvMUoFCqu0+O+XXMazhR4njRLUkMl2n370xvENJUsAiqI1r2mk4CfEwWcCjapeqlmCaFjMmQ9QyMimfbzaZEJPgpjhWHE8PT8O5sTqXUmA5ORBEZ61ivE/7xeCuG5n/MoSYFF1ESMF6YCQ4yLPfCAK1NXZIYQqrj5JaYjoggFs1rV1G/Olp0n7knjouHcnNZal+UOFbSPDlEdNdEZaqFr1EYuougRPaM39G49WE/Wi/X6E12wyjt76A+sj29sjJeK</latexit><latexit sha1_base64="gikEPqhu3Mj65Jcm+ICZUaFxUsk=">AAACAXicbVBPS8MwHE39O+e/qkcPBocwL6MTQb3IwIvHCdYN1jLSLN3CkrYkvwql7KZfRi8ievIr+AX8NqazB918p5f3XkLeCxLBNTjOl7WwuLS8slpZq65vbG5t2zu7dzpOFWUujUWsugHRTPCIucBBsG6iGJGBYJ1gfFX4nXumNI+jW8gS5ksyjHjIKQEj9e0DL4wjCInkIssTkBPPvMUoFCqu0+O+XXMazhR4njRLUkMl2n370xvENJUsAiqI1r2mk4CfEwWcCjapeqlmCaFjMmQ9QyMimfbzaZEJPgpjhWHE8PT8O5sTqXUmA5ORBEZ61ivE/7xeCuG5n/MoSYFF1ESMF6YCQ4yLPfCAK1NXZIYQqrj5JaYjoggFs1rV1G/Olp0n7knjouHcnNZal+UOFbSPDlEdNdEZaqFr1EYuougRPaM39G49WE/Wi/X6E12wyjt76A+sj29sjJeK</latexit>(d)<latexit sha1_base64="O0E+9/qjS04RCCQxe+K3QeWcbN4=">AAACAXicbVBPS8MwHE39O+e/qkcPBocwL6MTQb3IwIvHCdYN1jLSLN3CkrYkvwql7KZfRi8ievIr+AX8NqazB918p5f3XkLeCxLBNTjOl7WwuLS8slpZq65vbG5t2zu7dzpOFWUujUWsugHRTPCIucBBsG6iGJGBYJ1gfFX4nXumNI+jW8gS5ksyjHjIKQEj9e0DL4wjCInkIssTkBPPvMUoFCquD477ds1pOFPgedIsSQ2VaPftT28Q01SyCKggWveaTgJ+ThRwKtik6qWaJYSOyZD1DI2IZNrPp0Um+CiMFYYRw9Pz72xOpNaZDExGEhjpWa8Q//N6KYTnfs6jJAUWURMxXpgKDDEu9sADrkxdkRlCqOLml5iOiCIUzGpVU785W3aeuCeNi4Zzc1prXZY7VNA+OkR11ERnqIWuURu5iKJH9Ize0Lv1YD1ZL9brT3TBKu/soT+wPr4BbgqXiw==</latexit><latexit sha1_base64="O0E+9/qjS04RCCQxe+K3QeWcbN4=">AAACAXicbVBPS8MwHE39O+e/qkcPBocwL6MTQb3IwIvHCdYN1jLSLN3CkrYkvwql7KZfRi8ievIr+AX8NqazB918p5f3XkLeCxLBNTjOl7WwuLS8slpZq65vbG5t2zu7dzpOFWUujUWsugHRTPCIucBBsG6iGJGBYJ1gfFX4nXumNI+jW8gS5ksyjHjIKQEj9e0DL4wjCInkIssTkBPPvMUoFCquD477ds1pOFPgedIsSQ2VaPftT28Q01SyCKggWveaTgJ+ThRwKtik6qWaJYSOyZD1DI2IZNrPp0Um+CiMFYYRw9Pz72xOpNaZDExGEhjpWa8Q//N6KYTnfs6jJAUWURMxXpgKDDEu9sADrkxdkRlCqOLml5iOiCIUzGpVU785W3aeuCeNi4Zzc1prXZY7VNA+OkR11ERnqIWuURu5iKJH9Ize0Lv1YD1ZL9brT3TBKu/soT+wPr4BbgqXiw==</latexit><latexit sha1_base64="O0E+9/qjS04RCCQxe+K3QeWcbN4=">AAACAXicbVBPS8MwHE39O+e/qkcPBocwL6MTQb3IwIvHCdYN1jLSLN3CkrYkvwql7KZfRi8ievIr+AX8NqazB918p5f3XkLeCxLBNTjOl7WwuLS8slpZq65vbG5t2zu7dzpOFWUujUWsugHRTPCIucBBsG6iGJGBYJ1gfFX4nXumNI+jW8gS5ksyjHjIKQEj9e0DL4wjCInkIssTkBPPvMUoFCquD477ds1pOFPgedIsSQ2VaPftT28Q01SyCKggWveaTgJ+ThRwKtik6qWaJYSOyZD1DI2IZNrPp0Um+CiMFYYRw9Pz72xOpNaZDExGEhjpWa8Q//N6KYTnfs6jJAUWURMxXpgKDDEu9sADrkxdkRlCqOLml5iOiCIUzGpVU785W3aeuCeNi4Zzc1prXZY7VNA+OkR11ERnqIWuURu5iKJH9Ize0Lv1YD1ZL9brT3TBKu/soT+wPr4BbgqXiw==</latexit>3 FIG. 3. Contributions to the instability. Growth rates of shape pertur- bations varying the values of different model parameters. Excluding the varied parameter, other parameter values are in Table I except for ξ, ζ → 0. (a) Traction forces completely destabilize the monolayer front. For this plot, T0 = 0, 0.25, 0.5, 0.75, 1 kPa. (b) Long-range transmission of viscous stresses selects the fastest-growing mode. For this plot L0 = 50, 100, 150, 200, 250 µm. (c) Cell-substrate friction screens hydrodynamic interactions to limit the wavelength of the fingering pattern. For this plot, ξ = 10, 102, 103, 104, 105 Pa·s/µm2. (d) Contractility stabilizes short-wavelength perturbations of the monolayer front. For this plot, −ζ = 0, 10, 20, 30, 40 kPa. FIG. 4. Screening of tissue flows. (a) Flow perturbations induced by short-wavelength shape perturbations (q > π/L0) penetrate a dis- tance given by their wavelength. Thus, the interfacial velocity per- turbation increases with wavelength. In contrast, the penetration of flow perturbations induced by long-wavelength shape perturbations (q > π/L0) is limited by the tissue width 2L0, which entails a de- crease of the interfacial velocity perturbation. Consequently, shape pertubations with a wavelength that matches the monolayer width (q ∼ π/L0) feature the fastest growth (Fig. 3b). Parameter values are in Table I except for ξ, ζ → 0. (b) The wavelength of the fastest- growing mode, λ∗, is proportional to the monolayer semiwidth L0 if L0 (cid:46) λ, with λ = (cid:112)η/ξ the hydrodynamic screening length. For wider monolayers, the selected wavelength is size-independent, be- coming proportional to λ. Parameter values are in Table I except for ζ → 0, and ξ = 10, 102, 103, 104, 105 Pa·s/µm2. tissue front. Therefore, flow perturbations further destabilize the flat front in a wavelength-dependent manner. Symbol Description monolayer height nematic length L0 monolayer half-width h Lc T0 maximal traction −ζ ξ η λ intercellular contractility friction coefficient monolayer viscosity Estimate 200 µm 5 µm [28, 37] 25 µm [27, 28] 0.5 kPa [27, 28] 20 kPa [28] hydrodynamic screening length 0.5 mm ((cid:112)η/ξ) 100 Pa·s/µm2 [38] 25 MPa·s [27, 28] TABLE I. Estimates of model parameters. Viscous stresses. -- The kinematic mechanism explains why long-wavelength modes are unstable. However, it does not explain why the most unstable mode occurs at a finite (cid:112) wavelength (Fig. 3a). In fact, the existence of a peak in the growth rate is due to the transmission of viscous stress across η/ξ (cid:29) L0, the monolayer. In the so-called wet limit λ = corresponding to ξ → 0, viscous stresses transmit through the entire monolayer. Thus, a given perturbation of the front generates a flow perturbation that penetrates a distance of the order of its wavelength, π/q, into the monolayer. At the monolayer edge, the stress-free boundary condition im- poses δσxx(±L) = ∓∂xσ0 xx(±L0) δL. Hence, since δσxx = 2η ∂xδvx in the absence of contractility (ζ → 0), the gradient of the velocity perturbation profile is fixed at the boundary, being positive (negative) for advanced (trailing) regions of the For short wavelengths, π/q < L0, the penetration distance of flow perturbations is shorter than the tissue width. Thus, since the slope of velocity perturbations at the interface is fixed, the longer the wavelength, the larger the interfacial ve- locity perturbation (Fig. 4a). Hence, the growth rate increases with the wavelength (Fig. 3a). In contrast, for long wave- lengths, π/q > L0, the penetration distance of flow perturba- tions is longer than the tissue width. Thus, in this case, the de- cay of flow perturbations becomes nearly linear, with a slope that decreases with increasing wavelength. Consequently, per- turbations of longer wavelength feature a smaller interfacial velocity perturbation (Fig. 4a), and hence they are less unsta- ble (Fig. 3a). In conclusion, in the absence of cell-substrate friction forces, the finite width of the monolayer limits the range of hydrodynamic interactions in the tissue, thus giving rise to the peak in the growth rate at q∗ (cid:112) ∼ π/λ if λ (cid:46) L0 (Fig. 3c). Thus, for suf- Cell-substrate friction forces. -- Cell-substrate friction screens the transmission of viscous stresses over distances larger than λ = η/ξ. Consequently, the peak of the growth rate occurs at q∗ ficiently strong friction, the fingering wavelength is given by the hydrodynamic screening length λ instead of the mono- layer width L0 (Fig. 4b). We estimate λ ∼ 0.5 mm (Ta- ble I). Therefore, the crossover from a viscosity-dominated to a friction-dominated regime of monolayer spreading and fingering should be observable in usual in vitro experiments. Surface tension. -- The monolayer edge presents a sur- face tension γ, namely the work per unit area required to ex- pand it. For a curved interface, surface tension gives rise to a normal force: n± · σ · n±x=±L = −γ (cid:126)∇ · n±x=±L ≈ ∼ π/L0 (Fig. 3b). 0.140.160.180.20.220.240.260.280.30.3200.010.020.030.040.05L0Growthrate!(h1)Wavenumberq(µm1)0.100.10.20.30.40.50.600.010.020.030.040.05T0Growthrate!(h1)Wavenumberq(µm1)00.050.10.150.20.250.300.050.10.150.2⇠Growthrate!(h1)Wavenumberq(µm1)0.10.0500.050.10.150.20.250.300.010.020.030.040.05⇣Growthrate!(h1)Wavenumberq(µm1)(d)<latexit sha1_base64="WO4201SibMC27/h//bd7XU4MxJU=">AAACAXicbVC9TsMwGHT4LeUvwMiARYVUlipFSJStEgtjkQit1ESV4zqtVTuJ7C9IUdQNXgYWhGDiFXgB3ganZICWm853Z8t3QSK4Bsf5spaWV1bX1isb1c2t7Z1de2//TseposylsYhVLyCaCR4xFzgI1ksUIzIQrBtMrgq/e8+U5nF0C1nCfElGEQ85JWCkgX3khXEEIZFcZHkCcuqZtxiFQsX14enArjkNZwa8SJolqaESnYH96Q1jmkoWARVE637TScDPiQJOBZtWvVSzhNAJGbG+oRGRTPv5rMgUn4SxwjBmeHb+nc2J1DqTgclIAmM97xXif14/hbDl5zxKUmARNRHjhanAEONiDzzkytQVmSGEKm5+iemYKELBrFY19ZvzZReJe9a4bDg357V2q9yhgg7RMaqjJrpAbXSNOshFFD2iZ/SG3q0H68l6sV5/oktWeecA/YH18Q1sPJeF</latexit><latexit sha1_base64="WO4201SibMC27/h//bd7XU4MxJU=">AAACAXicbVC9TsMwGHT4LeUvwMiARYVUlipFSJStEgtjkQit1ESV4zqtVTuJ7C9IUdQNXgYWhGDiFXgB3ganZICWm853Z8t3QSK4Bsf5spaWV1bX1isb1c2t7Z1de2//TseposylsYhVLyCaCR4xFzgI1ksUIzIQrBtMrgq/e8+U5nF0C1nCfElGEQ85JWCkgX3khXEEIZFcZHkCcuqZtxiFQsX14enArjkNZwa8SJolqaESnYH96Q1jmkoWARVE637TScDPiQJOBZtWvVSzhNAJGbG+oRGRTPv5rMgUn4SxwjBmeHb+nc2J1DqTgclIAmM97xXif14/hbDl5zxKUmARNRHjhanAEONiDzzkytQVmSGEKm5+iemYKELBrFY19ZvzZReJe9a4bDg357V2q9yhgg7RMaqjJrpAbXSNOshFFD2iZ/SG3q0H68l6sV5/oktWeecA/YH18Q1sPJeF</latexit><latexit sha1_base64="WO4201SibMC27/h//bd7XU4MxJU=">AAACAXicbVC9TsMwGHT4LeUvwMiARYVUlipFSJStEgtjkQit1ESV4zqtVTuJ7C9IUdQNXgYWhGDiFXgB3ganZICWm853Z8t3QSK4Bsf5spaWV1bX1isb1c2t7Z1de2//TseposylsYhVLyCaCR4xFzgI1ksUIzIQrBtMrgq/e8+U5nF0C1nCfElGEQ85JWCkgX3khXEEIZFcZHkCcuqZtxiFQsX14enArjkNZwa8SJolqaESnYH96Q1jmkoWARVE637TScDPiQJOBZtWvVSzhNAJGbG+oRGRTPv5rMgUn4SxwjBmeHb+nc2J1DqTgclIAmM97xXif14/hbDl5zxKUmARNRHjhanAEONiDzzkytQVmSGEKm5+iemYKELBrFY19ZvzZReJe9a4bDg357V2q9yhgg7RMaqjJrpAbXSNOshFFD2iZ/SG3q0H68l6sV5/oktWeecA/YH18Q1sPJeF</latexit>(c)<latexit sha1_base64="YqIKExJwS4JI8MJStNpKqbpqo/g=">AAACAXicbVC9TsMwGHT4LeUvwMiARYVUlipFSJStEgtjkQit1ESV4zqtVTuJ7C9IUdQNXgYWhGDiFXgB3ganZICWm853Z8t3QSK4Bsf5spaWV1bX1isb1c2t7Z1de2//TseposylsYhVLyCaCR4xFzgI1ksUIzIQrBtMrgq/e8+U5nF0C1nCfElGEQ85JWCkgX3khXEEIZFcZHkCcuqZtxiFQsV1ejqwa07DmQEvkmZJaqhEZ2B/esOYppJFQAXRut90EvBzooBTwaZVL9UsIXRCRqxvaEQk034+KzLFJ2GsMIwZnp1/Z3Mitc5kYDKSwFjPe4X4n9dPIWz5OY+SFFhETcR4YSowxLjYAw+5MnVFZgihiptfYjomilAwq1VN/eZ82UXinjUuG87Nea3dKneooEN0jOqoiS5QG12jDnIRRY/oGb2hd+vBerJerNef6JJV3jlAf2B9fANqvpeE</latexit><latexit sha1_base64="YqIKExJwS4JI8MJStNpKqbpqo/g=">AAACAXicbVC9TsMwGHT4LeUvwMiARYVUlipFSJStEgtjkQit1ESV4zqtVTuJ7C9IUdQNXgYWhGDiFXgB3ganZICWm853Z8t3QSK4Bsf5spaWV1bX1isb1c2t7Z1de2//TseposylsYhVLyCaCR4xFzgI1ksUIzIQrBtMrgq/e8+U5nF0C1nCfElGEQ85JWCkgX3khXEEIZFcZHkCcuqZtxiFQsV1ejqwa07DmQEvkmZJaqhEZ2B/esOYppJFQAXRut90EvBzooBTwaZVL9UsIXRCRqxvaEQk034+KzLFJ2GsMIwZnp1/Z3Mitc5kYDKSwFjPe4X4n9dPIWz5OY+SFFhETcR4YSowxLjYAw+5MnVFZgihiptfYjomilAwq1VN/eZ82UXinjUuG87Nea3dKneooEN0jOqoiS5QG12jDnIRRY/oGb2hd+vBerJerNef6JJV3jlAf2B9fANqvpeE</latexit><latexit sha1_base64="YqIKExJwS4JI8MJStNpKqbpqo/g=">AAACAXicbVC9TsMwGHT4LeUvwMiARYVUlipFSJStEgtjkQit1ESV4zqtVTuJ7C9IUdQNXgYWhGDiFXgB3ganZICWm853Z8t3QSK4Bsf5spaWV1bX1isb1c2t7Z1de2//TseposylsYhVLyCaCR4xFzgI1ksUIzIQrBtMrgq/e8+U5nF0C1nCfElGEQ85JWCkgX3khXEEIZFcZHkCcuqZtxiFQsV1ejqwa07DmQEvkmZJaqhEZ2B/esOYppJFQAXRut90EvBzooBTwaZVL9UsIXRCRqxvaEQk034+KzLFJ2GsMIwZnp1/Z3Mitc5kYDKSwFjPe4X4n9dPIWz5OY+SFFhETcR4YSowxLjYAw+5MnVFZgihiptfYjomilAwq1VN/eZ82UXinjUuG87Nea3dKneooEN0jOqoiS5QG12jDnIRRY/oGb2hd+vBerJerNef6JJV3jlAf2B9fANqvpeE</latexit>(b)<latexit sha1_base64="ofq78kasmBdkMGFQrMd6jUe3Fgk=">AAACAXicbVC9TsMwGHT4LeUvwMiARYVUlipFSJStEgtjkQit1ESV4zqtVTuJ7C9IUdQNXgYWhGDiFXgB3ganZICWm853Z8t3QSK4Bsf5spaWV1bX1isb1c2t7Z1de2//TseposylsYhVLyCaCR4xFzgI1ksUIzIQrBtMrgq/e8+U5nF0C1nCfElGEQ85JWCkgX3khXEEIZFcZHkCcuqZtxiFQsX14HRg15yGMwNeJM2S1FCJzsD+9IYxTSWLgAqidb/pJODnRAGngk2rXqpZQuiEjFjf0IhIpv18VmSKT8JYYRgzPDv/zuZEap3JwGQkgbGe9wrxP6+fQtjycx4lKbCImojxwlRgiHGxBx5yZeqKzBBCFTe/xHRMFKFgVqua+s35sovEPWtcNpyb81q7Ve5QQYfoGNVRE12gNrpGHeQiih7RM3pD79aD9WS9WK8/0SWrvHOA/sD6+AZpQJeD</latexit><latexit sha1_base64="ofq78kasmBdkMGFQrMd6jUe3Fgk=">AAACAXicbVC9TsMwGHT4LeUvwMiARYVUlipFSJStEgtjkQit1ESV4zqtVTuJ7C9IUdQNXgYWhGDiFXgB3ganZICWm853Z8t3QSK4Bsf5spaWV1bX1isb1c2t7Z1de2//TseposylsYhVLyCaCR4xFzgI1ksUIzIQrBtMrgq/e8+U5nF0C1nCfElGEQ85JWCkgX3khXEEIZFcZHkCcuqZtxiFQsX14HRg15yGMwNeJM2S1FCJzsD+9IYxTSWLgAqidb/pJODnRAGngk2rXqpZQuiEjFjf0IhIpv18VmSKT8JYYRgzPDv/zuZEap3JwGQkgbGe9wrxP6+fQtjycx4lKbCImojxwlRgiHGxBx5yZeqKzBBCFTe/xHRMFKFgVqua+s35sovEPWtcNpyb81q7Ve5QQYfoGNVRE12gNrpGHeQiih7RM3pD79aD9WS9WK8/0SWrvHOA/sD6+AZpQJeD</latexit><latexit sha1_base64="ofq78kasmBdkMGFQrMd6jUe3Fgk=">AAACAXicbVC9TsMwGHT4LeUvwMiARYVUlipFSJStEgtjkQit1ESV4zqtVTuJ7C9IUdQNXgYWhGDiFXgB3ganZICWm853Z8t3QSK4Bsf5spaWV1bX1isb1c2t7Z1de2//TseposylsYhVLyCaCR4xFzgI1ksUIzIQrBtMrgq/e8+U5nF0C1nCfElGEQ85JWCkgX3khXEEIZFcZHkCcuqZtxiFQsX14HRg15yGMwNeJM2S1FCJzsD+9IYxTSWLgAqidb/pJODnRAGngk2rXqpZQuiEjFjf0IhIpv18VmSKT8JYYRgzPDv/zuZEap3JwGQkgbGe9wrxP6+fQtjycx4lKbCImojxwlRgiHGxBx5yZeqKzBBCFTe/xHRMFKFgVqua+s35sovEPWtcNpyb81q7Ve5QQYfoGNVRE12gNrpGHeQiih7RM3pD79aD9WS9WK8/0SWrvHOA/sD6+AZpQJeD</latexit>(a)<latexit sha1_base64="P/2NV/rEH2AxGcsEXmQ9uouTj8w=">AAACAXicbVC9TsMwGHT4LeUvwMiARYVUlipFSJStEgtjkQit1ESV4zqtVTuJ7C9IUdQNXgYWhGDiFXgB3ganZICWm853Z8t3QSK4Bsf5spaWV1bX1isb1c2t7Z1de2//TseposylsYhVLyCaCR4xFzgI1ksUIzIQrBtMrgq/e8+U5nF0C1nCfElGEQ85JWCkgX3khXEEIZFcZHkCcuqZtxiFQsV1cjqwa07DmQEvkmZJaqhEZ2B/esOYppJFQAXRut90EvBzooBTwaZVL9UsIXRCRqxvaEQk034+KzLFJ2GsMIwZnp1/Z3Mitc5kYDKSwFjPe4X4n9dPIWz5OY+SFFhETcR4YSowxLjYAw+5MnVFZgihiptfYjomilAwq1VN/eZ82UXinjUuG87Nea3dKneooEN0jOqoiS5QG12jDnIRRY/oGb2hd+vBerJerNef6JJV3jlAf2B9fANnwpeC</latexit><latexit sha1_base64="P/2NV/rEH2AxGcsEXmQ9uouTj8w=">AAACAXicbVC9TsMwGHT4LeUvwMiARYVUlipFSJStEgtjkQit1ESV4zqtVTuJ7C9IUdQNXgYWhGDiFXgB3ganZICWm853Z8t3QSK4Bsf5spaWV1bX1isb1c2t7Z1de2//TseposylsYhVLyCaCR4xFzgI1ksUIzIQrBtMrgq/e8+U5nF0C1nCfElGEQ85JWCkgX3khXEEIZFcZHkCcuqZtxiFQsV1cjqwa07DmQEvkmZJaqhEZ2B/esOYppJFQAXRut90EvBzooBTwaZVL9UsIXRCRqxvaEQk034+KzLFJ2GsMIwZnp1/Z3Mitc5kYDKSwFjPe4X4n9dPIWz5OY+SFFhETcR4YSowxLjYAw+5MnVFZgihiptfYjomilAwq1VN/eZ82UXinjUuG87Nea3dKneooEN0jOqoiS5QG12jDnIRRY/oGb2hd+vBerJerNef6JJV3jlAf2B9fANnwpeC</latexit><latexit sha1_base64="P/2NV/rEH2AxGcsEXmQ9uouTj8w=">AAACAXicbVC9TsMwGHT4LeUvwMiARYVUlipFSJStEgtjkQit1ESV4zqtVTuJ7C9IUdQNXgYWhGDiFXgB3ganZICWm853Z8t3QSK4Bsf5spaWV1bX1isb1c2t7Z1de2//TseposylsYhVLyCaCR4xFzgI1ksUIzIQrBtMrgq/e8+U5nF0C1nCfElGEQ85JWCkgX3khXEEIZFcZHkCcuqZtxiFQsV1cjqwa07DmQEvkmZJaqhEZ2B/esOYppJFQAXRut90EvBzooBTwaZVL9UsIXRCRqxvaEQk034+KzLFJ2GsMIwZnp1/Z3Mitc5kYDKSwFjPe4X4n9dPIWz5OY+SFFhETcR4YSowxLjYAw+5MnVFZgihiptfYjomilAwq1VN/eZ82UXinjUuG87Nea3dKneooEN0jOqoiS5QG12jDnIRRY/oGb2hd+vBerJerNef6JJV3jlAf2B9fANnwpeC</latexit>(b)<latexit sha1_base64="ofq78kasmBdkMGFQrMd6jUe3Fgk=">AAACAXicbVC9TsMwGHT4LeUvwMiARYVUlipFSJStEgtjkQit1ESV4zqtVTuJ7C9IUdQNXgYWhGDiFXgB3ganZICWm853Z8t3QSK4Bsf5spaWV1bX1isb1c2t7Z1de2//TseposylsYhVLyCaCR4xFzgI1ksUIzIQrBtMrgq/e8+U5nF0C1nCfElGEQ85JWCkgX3khXEEIZFcZHkCcuqZtxiFQsX14HRg15yGMwNeJM2S1FCJzsD+9IYxTSWLgAqidb/pJODnRAGngk2rXqpZQuiEjFjf0IhIpv18VmSKT8JYYRgzPDv/zuZEap3JwGQkgbGe9wrxP6+fQtjycx4lKbCImojxwlRgiHGxBx5yZeqKzBBCFTe/xHRMFKFgVqua+s35sovEPWtcNpyb81q7Ve5QQYfoGNVRE12gNrpGHeQiih7RM3pD79aD9WS9WK8/0SWrvHOA/sD6+AZpQJeD</latexit><latexit sha1_base64="ofq78kasmBdkMGFQrMd6jUe3Fgk=">AAACAXicbVC9TsMwGHT4LeUvwMiARYVUlipFSJStEgtjkQit1ESV4zqtVTuJ7C9IUdQNXgYWhGDiFXgB3ganZICWm853Z8t3QSK4Bsf5spaWV1bX1isb1c2t7Z1de2//TseposylsYhVLyCaCR4xFzgI1ksUIzIQrBtMrgq/e8+U5nF0C1nCfElGEQ85JWCkgX3khXEEIZFcZHkCcuqZtxiFQsX14HRg15yGMwNeJM2S1FCJzsD+9IYxTSWLgAqidb/pJODnRAGngk2rXqpZQuiEjFjf0IhIpv18VmSKT8JYYRgzPDv/zuZEap3JwGQkgbGe9wrxP6+fQtjycx4lKbCImojxwlRgiHGxBx5yZeqKzBBCFTe/xHRMFKFgVqua+s35sovEPWtcNpyb81q7Ve5QQYfoGNVRE12gNrpGHeQiih7RM3pD79aD9WS9WK8/0SWrvHOA/sD6+AZpQJeD</latexit><latexit sha1_base64="ofq78kasmBdkMGFQrMd6jUe3Fgk=">AAACAXicbVC9TsMwGHT4LeUvwMiARYVUlipFSJStEgtjkQit1ESV4zqtVTuJ7C9IUdQNXgYWhGDiFXgB3ganZICWm853Z8t3QSK4Bsf5spaWV1bX1isb1c2t7Z1de2//TseposylsYhVLyCaCR4xFzgI1ksUIzIQrBtMrgq/e8+U5nF0C1nCfElGEQ85JWCkgX3khXEEIZFcZHkCcuqZtxiFQsX14HRg15yGMwNeJM2S1FCJzsD+9IYxTSWLgAqidb/pJODnRAGngk2rXqpZQuiEjFjf0IhIpv18VmSKT8JYYRgzPDv/zuZEap3JwGQkgbGe9wrxP6+fQtjycx4lKbCImojxwlRgiHGxBx5yZeqKzBBCFTe/xHRMFKFgVqua+s35sovEPWtcNpyb81q7Ve5QQYfoGNVRE12gNrpGHeQiih7RM3pD79aD9WS9WK8/0SWrvHOA/sD6+AZpQJeD</latexit>(a)<latexit sha1_base64="P/2NV/rEH2AxGcsEXmQ9uouTj8w=">AAACAXicbVC9TsMwGHT4LeUvwMiARYVUlipFSJStEgtjkQit1ESV4zqtVTuJ7C9IUdQNXgYWhGDiFXgB3ganZICWm853Z8t3QSK4Bsf5spaWV1bX1isb1c2t7Z1de2//TseposylsYhVLyCaCR4xFzgI1ksUIzIQrBtMrgq/e8+U5nF0C1nCfElGEQ85JWCkgX3khXEEIZFcZHkCcuqZtxiFQsV1cjqwa07DmQEvkmZJaqhEZ2B/esOYppJFQAXRut90EvBzooBTwaZVL9UsIXRCRqxvaEQk034+KzLFJ2GsMIwZnp1/Z3Mitc5kYDKSwFjPe4X4n9dPIWz5OY+SFFhETcR4YSowxLjYAw+5MnVFZgihiptfYjomilAwq1VN/eZ82UXinjUuG87Nea3dKneooEN0jOqoiS5QG12jDnIRRY/oGb2hd+vBerJerNef6JJV3jlAf2B9fANnwpeC</latexit><latexit sha1_base64="P/2NV/rEH2AxGcsEXmQ9uouTj8w=">AAACAXicbVC9TsMwGHT4LeUvwMiARYVUlipFSJStEgtjkQit1ESV4zqtVTuJ7C9IUdQNXgYWhGDiFXgB3ganZICWm853Z8t3QSK4Bsf5spaWV1bX1isb1c2t7Z1de2//TseposylsYhVLyCaCR4xFzgI1ksUIzIQrBtMrgq/e8+U5nF0C1nCfElGEQ85JWCkgX3khXEEIZFcZHkCcuqZtxiFQsV1cjqwa07DmQEvkmZJaqhEZ2B/esOYppJFQAXRut90EvBzooBTwaZVL9UsIXRCRqxvaEQk034+KzLFJ2GsMIwZnp1/Z3Mitc5kYDKSwFjPe4X4n9dPIWz5OY+SFFhETcR4YSowxLjYAw+5MnVFZgihiptfYjomilAwq1VN/eZ82UXinjUuG87Nea3dKneooEN0jOqoiS5QG12jDnIRRY/oGb2hd+vBerJerNef6JJV3jlAf2B9fANnwpeC</latexit><latexit sha1_base64="P/2NV/rEH2AxGcsEXmQ9uouTj8w=">AAACAXicbVC9TsMwGHT4LeUvwMiARYVUlipFSJStEgtjkQit1ESV4zqtVTuJ7C9IUdQNXgYWhGDiFXgB3ganZICWm853Z8t3QSK4Bsf5spaWV1bX1isb1c2t7Z1de2//TseposylsYhVLyCaCR4xFzgI1ksUIzIQrBtMrgq/e8+U5nF0C1nCfElGEQ85JWCkgX3khXEEIZFcZHkCcuqZtxiFQsV1cjqwa07DmQEvkmZJaqhEZ2B/esOYppJFQAXRut90EvBzooBTwaZVL9UsIXRCRqxvaEQk034+KzLFJ2GsMIwZnp1/Z3Mitc5kYDKSwFjPe4X4n9dPIWz5OY+SFFhETcR4YSowxLjYAw+5MnVFZgihiptfYjomilAwq1VN/eZ82UXinjUuG87Nea3dKneooEN0jOqoiS5QG12jDnIRRY/oGb2hd+vBerJerNef6JJV3jlAf2B9fANnwpeC</latexit>00.20.40.60.811.21.400.20.40.60.81⇠Fingeringwavelength⇤(mm)MonolayersemiwidthL0(mm)00.050.10.150.20.250.3050100150200wavelengthFlowperturbationvx(µm/h)Coordinatex(µm)q=8⇡/L0q=4⇡/L0q=⇡/L0q=⇡/(2L0)q=⇡/(4L0)±γ d2δL/dy2. Then, the expansion of the growth rate at long wavelengths reads (cid:20) 1 (cid:21) (cid:16) (qL0)4(cid:17) , (6) ω (q) ≈ 1 τ + 3τ − γ 2ηL0 (qL0)2 + O with τ = 2ηh/(T0Lc). This expression reveals the existence of a critical size Lγ ≈ 3γh/(T0Lc) above which the growth rate curves upwards at q → 0 (ω(cid:48)(cid:48)(0) > 0), hence exhibiting the aforementioned peak at a finite wavelength. Alternatively, if γ > γ∗ ≈ T0LcL0/(3h), surface tension prevents the hy- drodynamic selection of a finite fingering wavelength, which is then only limited by the length of the tissue front. The surface tension of the monolayer could be due to actin cables found along its edge, specially along the sides of ep- ithelial fingers [6, 9, 10]. Traction force measurements sug- gest that the tension of such cables is γ ∼ 0.2 mN/m [10], lower than typical surface tensions of cell aggregates, γ ∼ 1 − 10 mN/m [39 -- 42]. Combining these value ranges with T0 = 0.2 − 0.8 kPa and typical values of h and Lc (Table I), the critical monolayer width for fingering is Lγ ∼ 0.3 − 10 µm. Therefore, we expect surface tension not to play a ma- jor role in the fingering instability in monolayers of typical widths L0 ∼ 0.1 − 1 mm. Intercellular contractility. -- Because it decreases the spreading velocity, intercellular contractility has an additional stabilizing effect on the monolayer front (Fig. 3d). We dis- cuss the effects of a uniform contractility in [34]. Here, we consider an intercellular contractility −ζ concentrated at the polarized boundary layer, which has a size-independent con- tribution to the spreading velocity: (cid:20) ζ (cid:21) V0 ≈ T0Lc 2ηh L0 + Lc 2η T0Lc h 2 − . (7) (cid:112) Consequently, in the limit Lc (cid:28) L0 (cid:28) λ, this contractility has no impact on the growth rate of the uniform mode, ω (0) = τ−1, but it contributes a stabilizing quadratic term to the long- wavelength expansion of the growth rate: (cid:20) 1 3τ (cid:21) ω (q) ≈ 1 τ + + ζL2 c 8ηL2 0 (qL0)2 + O (cid:16) (qL0)4(cid:17) . (8) Thus, as surface tension, contractility defines a critical size −3ζLch/(4T0) above which the growth rate fea- Lζ ≈ tures a finite-wavelength peak. Alternatively, if −ζ > −ζ∗ ≈ 0/(3hLc), contractility supresses the hydrodynamic se- 4T0L2 lection of a finite fingering wavelength. For typical contractilities, −ζ ∼ 1−50 kPa [28], the critical monolayer width for fingering is Lζ ∼ 10 − 100 µm, where we have used T0 = 0.2 − 0.8 kPa and estimates for h and Lc (Table I). Therefore, we do not expect contractility to prevent fingering wavelength selection. However, our estimates give Lγ < Lζ, indicating that contractility typically dominates over surface tension in stabilizing short-wavelength shape per- turbations. Thus, the competition between the destabilizing effect of traction forces and the stabilizing effect of intercel- lular contractility defines the band of unstable modes. 4 Uniformly polarized tissues. -- To consider monolayers with bulk polarity [37], we analyze the morphological sta- bility of a uniformly polarized monolayer [34]. In this case, in the absence of contractility, the spreading velocity is con- stant: V0 = T0/(ξh). Hence, the q = 0 mode is marginally stable, ω(q = 0) = 0. Moreover, contractility stabilizes it, ω(0) < 0. Nevertheless, the viscous effects discussed above still give rise to a peak of the growth rate at a finite wave- length. Therefore, even though uniformly polarized tissues do not feature an accelerating front, they still exhibit a fingering instability for sufficiently small contractility [34]. Conclusions. -- Motivated by the observation of finger- like protrusions during the spreading of epithelial monolay- ers, we studied the stability of the advancing front. Modeling the cell monolayer as an active polar fluid, we showed that ac- tive traction forces are responsible for a long-wavelength in- stability of the monolayer front. Several features distinguish this instability from previous proposals. First, it is generic; it takes place for any value of the active traction force. Second, the wavelength of the fingering pattern is selected by the range of hydrodynamic interactions in the tissue. And third, active intercellular forces stabilize short-wavelength perturbations, typically dominating over surface tension effects. Our analysis identifies the physical mechanism of the insta- bility. Cellular traction forces at the monolayer edge set the velocity gradient in the spreading monolayer. Hence, under the same traction force, a larger monolayer spreads faster [28]. Consequently, when the monolayer front is perturbed, the pro- truding regions of the interface advance faster than the trailing regions, thus making the perturbation grow. Therefore, the instability is based on a simple kinematic mechanism, which takes place generically in viscous fluids that sustain a fixed ve- locity gradient in the direction of spreading. In particular, the same morphological instability should occur in the so-called squeeze flow [43], in which an incompressible fluid is forced to spread by decreasing the gap between two plates. In this case, under perfect slip conditions at the plates, the rate of gap reduction sets the fixed velocity gradient. Regarding spreading epithelia, we conclude that neither leader-cell behavior nor regulation of cell motility by curva- ture or by chemotactic fields are necessary for the fingering instability. Therefore, our results are consistent with the emer- gence of leader cells concomitantly with finger growth [6 -- 8, 10, 11]. However, the viscous rheology of the monolayer is essential for the instability. On the one hand, it underpins the velocity gradient that renders the interface unstable and, on the other, it enables wavelength selection for the finger- ing pattern. Concomitant with the fingering instability, shear stresses give rise to flows transversal to the spreading direc- tion, which might lead to the swirls observed in experiments [6, 7]. Finally, in addition to explaining fingering in tissue spreading, our results also account for the morphological in- stability recently observed during tissue dewetting [28]. Our predictions, such as the absence of a traction force threshold for the instability, and whether the fingering wave- length is given by either the monolayer width or the screening (cid:112) length λ = η/ξ, are experimentally testable. Indeed, con- sistent with our result, recent work has shown that the finger spacing is an intrinsic quantity that coincides with the stress correlation length [11]. To further test our predictions, future experiments could perturb active cellular forces, cell-cell and cell-substrate adhesion, and vary monolayer width. Our findings illustrate how hydrodynamic interactions im- pact tissue morphodynamics. In particular, we propose that epithelial fingering can naturally arise from a generic morpho- logical instability in a fluid film driven by interfacial active forces. Thus, our results showcase the relevance of interfa- cial instabilities in driven [44 -- 46] and active [47 -- 53] fluids for tissue spreading. We thank Xavier Trepat for a critical reading of the manuscript, and the members of his lab for discussions. We thank Jordi Ort´ın for discussions. R.A. acknowledges support from Fundaci´o "La Caixa". R.A. and J.C. acknowledge the MINECO under project FIS2016-78507-C2-2-P and General- itat de Catalunya under project 2014-SGR-878. ∗ [email protected]; Present address: Princeton Center for Theoretical Science and Lewis-Sigler Institute for Integra- tive Genomics, Princeton University, Princeton NJ 08544, USA [1] S. R. K. Vedula, A. Ravasio, C. T. Lim, and B. Ladoux, Physi- ology 28, 370 (2013). [2] T. B. Saw, S. Jain, B. Ladoux, and C. T. Lim, Cell. Mol. Bioeng. [3] V. Hakim and P. Silberzan, Reports Prog. Phys. 80, 076601 [4] B. Ladoux and R.-M. M`ege, Nat. Rev. Mol. Cell Biol. 18, 743 8, 3 (2014). (2017). (2017). [5] T. Omelchenko, J. M. Vasiliev, I. M. Gelfand, H. H. Feder, and E. M. Bonder, Proc. Natl. Acad. Sci. U. S. A. 100, 10788 (2003). [6] M. Poujade, E. Grasland-Mongrain, A. Hertzog, J. Jouanneau, P. Chavrier, B. Ladoux, A. Buguin, and P. Silberzan, Proc. Natl. Acad. Sci. U. S. A. 104, 15988 (2007). [7] L. Petitjean, M. Reffay, E. Grasland-Mongrain, M. Poujade, B. Ladoux, A. Buguin, and P. Silberzan, Biophys. J. 98, 1790 (2010). [8] M. Reffay, L. Petitjean, S. Coscoy, E. Grasland-Mongrain, F. Amblard, A. Buguin, and P. Silberzan, Biophys. J. 100, 2566 (2011). [9] J. K. Klarlund, Proc. Natl. Acad. Sci. U. S. A. 109, 15799 (2012). [10] M. Reffay, M. C. Parrini, O. Cochet-Escartin, B. Ladoux, A. Buguin, S. Coscoy, F. Amblard, J. Camonis, and P. Sil- berzan, Nat. Cell Biol. 16, 217 (2014). [11] M. Vishwakarma, J. Di Russo, D. Probst, U. S. Schwarz, T. Das, and J. P. Spatz, Nat. Commun. 9, 3469 (2018). [12] P. G. Saffman and G. Taylor, Proc. R. Soc. A 245, 312 (1958). [13] J. Casademunt, Chaos 14, 809 (2004). [14] N. Sep´ulveda, L. Petitjean, O. Cochet, E. Grasland-Mongrain, P. Silberzan, and V. Hakim, PLoS Comput. Biol. 9, e1002944 (2013). [15] S. Mark, R. Shlomovitz, N. S. Gov, M. Poujade, E. Grasland- Mongrain, and P. Silberzan, Biophys. J. 98, 361 (2010). [16] V. Tarle, A. Ravasio, V. Hakim, and N. Gov, Integr. Biol. 7, 5 1218 (2015). [17] M. Basan, J. Elgeti, E. Hannezo, W.-J. Rappel, and H. Levine, Proc. Natl. Acad. Sci. U. S. A. 110, 2452 (2013). [18] J. Zimmermann, M. Basan, and H. Levine, Eur. Phys. J. Spec. Top. 223, 1259 (2014). [19] D. Nesbitt, G. Pruessner, and C. F. Lee, Phys. Rev. E 96, [20] P. Lee and C. W. Wolgemuth, PLoS Comput. Biol. 7, e1002007 062615 (2017). (2011). [21] A. Doostmohammadi, S. P. Thampi, T. B. Saw, C. T. Lim, B. Ladoux, and J. M. Yeomans, Soft Matter 11, 7328 (2015). [22] D. L. Barton, S. Henkes, C. J. Weijer, and R. Sknepnek, PLOS Comput. Biol. 13, e1005569 (2017). [23] G. Y. Ouaknin and P. Z. Bar-Yoseph, Biophys. J. 97, 1811 (2009). [24] M. Salm and L. M. Pismen, Phys. Biol. 9, 026009 (2012). [25] M. H. Kopf and L. M. Pismen, Soft Matter 9, 3727 (2013). [26] M. Ben Amar and C. Bianca, Sci. Rep. 6, 33849 (2016). [27] C. Blanch-Mercader, R. Vincent, E. Bazelli`eres, X. Serra- Picamal, X. Trepat, and J. Casademunt, Soft Matter 13, 1235 (2017). [28] C. P´erez-Gonz´alez, R. Alert, C. Blanch-Mercader, M. G´omez- Gonz´alez, T. Kolodziej, E. Bazellieres, J. Casademunt, and X. Trepat, Nat. Phys. 15, 79 (2019). [29] R. Alert and J. Casademunt, Langmuir , acs.langmuir.8b02037 (2018). [30] X. Serra-Picamal, V. Conte, R. Vincent, E. An´on, D. T. Tambe, E. Bazellieres, J. P. Butler, J. J. Fredberg, and X. Trepat, Nat. Phys. 8, 628 (2012). [31] P. Recho, J. Ranft, and P. Marcq, Soft Matter 12, 2381 (2016). [32] S. Yabunaka and P. Marcq, Phys. Rev. E 96, 022406 (2017). [33] P.-G. de Gennes and J. Prost, The Physics of Liquid Crystals, 2nd ed. (Oxford University Press, 1993). [34] See Supplementary Material for a justification of the model and details of the linear stability analysis, which includes Refs. [54 -- 79]. [35] D. Oriola, R. Alert, and J. Casademunt, Phys. Rev. Lett. 118, [36] P. Rosen and D. S. Misfeldt, Proc. Natl. Acad. Sci. U. S. A. 77, 088002 (2017). 4760 (1980). [37] X. Trepat, M. R. Wasserman, T. E. Angelini, E. Millet, D. A. Weitz, J. P. Butler, and J. J. Fredberg, Nat. Phys. 5, 426 (2009). [38] O. Cochet-Escartin, J. Ranft, P. Silberzan, and P. Marcq, Bio- phys. J. 106, 65 (2014). [39] R. A. Foty, G. Forgacs, C. M. Pfleger, and M. S. Steinberg, Phys. Rev. Lett. 72, 2298 (1994). [40] G. Forgacs, R. A. Foty, Y. Shafrir, and M. S. Steinberg, Bio- phys. J. 74, 2227 (1998). [41] K. Guevorkian, M.-J. Colbert, M. Durth, S. Dufour, F. Brochard-Wyart, Phys. Rev. Lett. 104, 218101 (2010). and [42] T. V. Stirbat, A. Mgharbel, S. Bodennec, K. Ferri, H. C. Mer- tani, J.-P. Rieu, and H. Delanoe-Ayari, PLoS One 8, e52554 (2013). [43] J. Engmann, C. Servais, and A. S. Burbidge, J. Nonnewton. [44] S. M. Troian, E. Herbolzheimer, S. A. Safran, and J. F. Joanny, Fluid Mech. 132, 1 (2005). Europhys. Lett. 10, 25 (1989). [45] F. Melo, J. F. Joanny, and S. Fauve, Phys. Rev. Lett. 63, 1958 [46] M. Ben Amar and L. J. Cummings, Phys. Fluids 13, 1160 (1989). (2001). [47] A. Callan-Jones, J.-F. Joanny, and J. Prost, Phys. Rev. Lett. 100, 258106 (2008). [48] S. Sankararaman and S. Ramaswamy, Phys. Rev. Lett. 102, [59] E. Theveneau and R. Mayor, Cell. Mol. Life Sci. 70, 3481 68 (2016). 6 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). [67] T. B. Saw, A. Doostmohammadi, V. Nier, L. Kocgozlu, S. Thampi, Y. Toyama, P. Marcq, C. T. Lim, J. M. Yeomans, and B. Ladoux, Nature 544, 212 (2017). [68] K. Kruse, J. F. Joanny, F. Julicher, J. Prost, and K. Sekimoto, Eur. Phys. J. E 16, 5 (2005). [69] F. Julicher, in New Trends in the Physics and Mechanics of Bi- ological Systems, edited by M. Ben Amar, A. Goriely, M. M. Muller, and L. Cugliandolo (Oxford University Press, 2011) Chap. 4. [70] M. C. Marchetti, J. F. Joanny, S. Ramaswamy, T. B. Liverpool, J. Prost, M. Rao, and R. A. Simha, Rev. Mod. Phys. 85, 1143 (2013). [71] J. Prost, F. Julicher, and J.-F. Joanny, Nat. Phys. 11, 111 (2015). [72] T. Wyatt, B. Baum, and G. Charras, Curr. Opin. Cell Biol. 38, [73] N. Khalilgharibi, J. Fouchard, P. Recho, G. Charras, and A. Kabla, Curr. Opin. Cell Biol. 42, 113 (2016). [74] J. Ranft, M. Basan, J. Elgeti, J.-F. Joanny, J. Prost, and F. Julicher, Proc. Natl. Acad. Sci. U. S. A. 107, 20863 (2010). [75] D. A. Matoz-Fernandez, K. Martens, R. Sknepnek, J. L. Barrat, and S. Henkes, Soft Matter 13, 3205 (2017). [76] R. Etournay, M. Popovi´c, M. Merkel, A. Nandi, C. Blasse, B. Aigouy, H. Brandl, G. Myers, G. Salbreux, F. Julicher, and S. Eaton, Elife 4, e07090 (2015). [77] M. Krajnc, S. Dasgupta, P. Ziherl, and J. Prost, Phys. Rev. E 98, 022409 (2018). [78] F. Julicher and J. Prost, Eur. Phys. J. E 29, 27 (2009). [79] C. Blanch-Mercader and J. Casademunt, Soft Matter 13, 6913 (2017). 118107 (2009). [49] N. Sarkar and A. Basu, Eur. Phys. J. E 35, 115 (2012). [50] N. Sarkar and A. Basu, Eur. Phys. J. E 36, 86 (2013). [51] A. Nagilla, R. Prabhakar, and S. Jadhav, Phys. Fluids 30, 022109 (2018). [52] J. J. Williamson and G. Salbreux, Phys. Rev. Lett. 121, 238102 [53] M. Bogdan and T. Savin, R. Soc. Open Sci. 5, 181579 (2018). [54] R. Mayor and C. Carmona-Fontaine, Trends Cell Biol. 20, 319 [55] B. Stramer and R. Mayor, Nat. Rev. Mol. Cell Biol. 18, 43 (2018). (2010). (2017). [56] R. A. Desai, L. Gao, S. Raghavan, W. F. Liu, and C. S. Chen, J. Cell Sci. 122, 905 (2009). [57] A. A. Khalil and P. Friedl, Integr. Biol. 2, 568 (2010). [58] G. F. Weber, M. A. Bjerke, and D. W. DeSimone, Dev. Cell 22, 104 (2012). (2013). 420 (2016). 17, 97 (2016). [60] B. Ladoux, R.-M. M`ege, and X. Trepat, Trends Cell Biol. 26, [61] R. Mayor and S. Etienne-Manneville, Nat. Rev. Mol. Cell Biol. [62] J. Zimmermann, B. A. Camley, W.-J. Rappel, and H. Levine, Proc. Natl. Acad. Sci. U. S. A. 113, 2660 (2016). [63] L. Coburn, H. Lopez, B. J. Caldwell, E. Moussa, C. Yap, R. Priya, A. Noppe, A. P. Roberts, V. Lobaskin, A. S. Yap, Z. Neufeld, and G. A. Gomez, Mol. Biol. Cell 27, 3436 (2016). [64] B. Smeets, R. Alert, J. Pesek, I. Pagonabarraga, H. Ramon, and R. Vincent, Proc. Natl. Acad. Sci. U. S. A. 113, 14621 (2016). [65] R. Vincent, E. Bazelli`eres, C. P´erez-Gonz´alez, M. Uroz, X. Serra-Picamal, and X. Trepat, Phys. Rev. Lett. 115, 248103 (2015). [66] P. Marmottant, A. Mgharbel, J. Kafer, B. Audren, J.-P. Rieu, Supplementary Material for "Active Fingering Instability in Tissue Spreading" ACTIVE POLAR FLUID MODEL OF EPITHELIAL SPREADING Force balance 7 Here, we briefly justify the description of epithelial spread- ing in terms of our continuum active polar fluid model. Polarity dynamics The outwards polarization of cells at the monolayer edge is likely due to contact inhibition of locomotion, a cell-cell in- teraction whereby cells repolarize in opposite directions upon contact [54, 55]. In fact, this interaction is mediated by cell- cell adhesion, with front-rear differences in cadherin-based junctions acting as a cue for the repolarization [1, 56 -- 60]. Although originally proposed for mesenchymal cells, contact inhibition of locomotion is being increasingly recognized to play a key role in orchestrating the collective migration of ep- ithelial monolayers [1, 3, 54, 59 -- 64]. In a cohesive mono- layer, this interaction naturally leads to polarization of cells at the edge towards free space, leaving the inner region of the monolayer unpolarized. Such a polarity profile, in turn, explains the localization of traction forces at the edge and the build-up of tension at the center of epithelial monolay- ers [62, 63]. Therefore, we assume the polarity field (cid:126)p ((cid:126)r, t) to be set by an autonomous cellular mechanism such as contact inhibition of locomotion, which polarizes cells within a time scale τCIL ∼ 10 min [58, 64]. Hence, (cid:126)p ((cid:126)r, t) should remain essentially independent of flows in the monolayer, which oc- cur over a longer time scale given by the strain rate, at least of order τs ∼ 100 min [27, 65]. Consequently, taking a phe- nomenological approach, we propose the polarity field to fol- low a purely relaxational dynamics given by ∂pα ∂t = − 1 γ1 δF δpα , (S1) where F [(cid:126)p ] is the coarse-grained free energy functional for the polarity field (Eq. (1)), and γ1 is a kinetic coefficient (the rotational viscosity for the angular degree of freedom). With respect to the most general dynamics of the polarity field in an active polar fluid, Eq. (S1) neglects polarity advection and corotation, as well as flow alignment and active spontaneous polarization effects. Thus, using Eq. (1), the dynamics of the polarity is given by (cid:0) (cid:1) . ∂tpα = 1 γ1 −apα + K∇2pα (S2) In the limit of fast polarity dynamics compared to the spread- ing dynamics, the polarity field is always at equilibrium, ∂tpα = 0, adiabatically adapting to the shape of the mono- layer. Under this approximation, the polarity field is given by (S3) L2 c∇2pα = pα, where we have defined the characteristic length Lc ≡ of the polar order in the monolayer. (cid:112) K/a σαβ = σs 0 = ∂βσαβ + fα, Flows in cell monolayers occur at very low Reynolds num- bers. Therefore, inertial forces are negligible, and hence mo- mentum conservation reduces to the force balance condition (S4) where σs αβ are the symmetric and antisymmetric parts of the deviatoric stress tensor, and fα is the external force density. Respectively, σE,s αβ is the symmetric part of the Ericksen tensor. This tensor generalizes the pressure P to include anisotropic elastic stresses associated to the orien- tational degrees of freedom in liquid crystals [33]: αβ and σa αβ + σE,s αβ + σa αβ σE αβ = −P δαβ − ∂f ∂ (∂βpγ) ∂αpγ, (S5) where f is the Frank free energy density, namely the integrand of Eq. (1). Thus, the orientational contribution to the Ericksen tensor is of second order in gradients of the polarity field, and hence we neglect it, so that the stress tensor reads αβ + σa σαβ = σs αβ − P δαβ. (S6) Then, the pressure is related to the cell number surface den- sity ρ by the equation of state of the monolayer. For the sake of an estimate, we assume the simplest form for an equation of state, P (ρ) = B (ρ − ρ0) /ρ0, where B is the bulk modu- lus of the monolayer, and ρ0 is a reference density defined by P (ρ0) = 0. Taking the pressure origin at the monolayer edge, ρ0 ∼ 2·103 cells/mm2 [28, 37]. In turn, density differences in the monolayer are, at most, ρ−ρ0 ∼ 6·103 cells/mm2 [28, 37]. Then, the monolayer is expected to be highly compressible because area changes can in principle be accommodated by changes in height, resisted only by the shear modulus of the tissue. Hence, we estimate the bulk modulus of the mono- layer by typical shear moduli of cell aggregates, which are in the range G ∼ 102−103 Pa [40, 41, 66]. Thus, the pressure in the monolayer should be P (cid:46) 30 − 300 Pa. In fact, isotropic compressive stresses (pressures) of ∼ 50 Pa were shown to induce cell extrusion [67]. In conclusion, if tissue spreading is not dominated by cell proliferation [17, 31, 32], the magni- tude of the pressure in the monolayer is expected to be much smaller than the tensile stress (tension) induced by traction forces, as measured by monolayer stress microscopy, which is of the order of kPa [28, 37]. Hence, we neglect pressure: αβ. αβ + σa σαβ = σs (S7) Now, for a nematic medium, the antisymmetric part of the stress tensor is given by σa αβ = 1/2 (pαhβ − hαpβ), where hα = −δF/δpα is the molecular field. From Eq. (S1), the adiabatic approximation for the polarity dynamics, ∂tpα = 0, implies hα = 0. Therefore, the antisymmetric part of the stress tensor vanishes under this approximation, σa αβ = 0. Thus, the stress tensor reduces to σαβ = σs (S8) αβ. Constitutive equations Next, constitutive equations must be given to specify the deviatoric stress tensor σs αβ and the external force fα in terms of the polarity and velocity fields. The generic constitutive equations of an active liquid crystal are provided by active gel theory [68 -- 71]. Here, based on the previous assumptions for the dynamics of the polarity field, we propose a simplified version of the generic constitutive equations of an active polar gel to describe epithelial spreading. First, the spreading occurs on timescales of the order of τs ∼ 100 min [27, 65], at which the tissue should have a fluid rheology. This time scale is much slower than the turnover time scales of proteins in the cytoskeleton or in cell- cell junctions, which are of the order of tens of minutes at most [72, 73]. Intra- or intercellular processes such as cytoskele- tal reorganizations or cell-cell slidings dissipate energy over these time scales, so that elastic energy may only be stored in the tissue at shorter times. In addition, other processes such as cell division, death, and extrusion [74, 75], as well as cell shape fluctuations [66, 76] and topological rearrenge- ments [76, 77] also fluidize the tissue. Therefore, to describe the slow spreading dynamics, we will not consider the elastic response of the tissue at short time scales. Then, in the viscous limit, the constitutive equations for the internal stress and the interfacial force of an active polar medium are: (cid:18) σs αβ = 2ηvαβ + ν1 2 +(cid:0)¯η d vγγ + ¯ν1 d pγhγ − ¯ζ − ζ pαhβ + hαpβ − 2 d (cid:19) (cid:1) δαβ, pγhγδαβ (cid:48) pγpγ − ζqαβ (S9) fα = −ξvα + νi pα + ζipα, (S10) where, qαβ = pαpβ − pγpγ/d δαβ is the traceless symmetric nematic order parameter tensor, with d the system dimension- ality, and vα is the velocity of the fluid with respect to the substrate. The coefficients η and ¯η are the shear and bulk vis- cosities of the medium, ζ is the anisotropic active stress coef- ficient, and ¯ζ and ζ(cid:48) are two isotropic active stress coefficients. Finally, ξ, νi, and ζi are the corresponding interfacial ver- sions of the viscosity (viscous friction), flow alignment (po- lar friction), and active stress (active force) coefficients. The constitutive equation for the internal stress, Eq. (S9), is that of an active polar gel with a variable modulus of the polarity [69]. In turn, the constitutive equation for the interfacial force, Eq. (S10), is less conventional [78], but it was derived from a mesoscopic model of an active polar gel [35]. Now, the adiabatic approximation for the polarity dynamics implies pα = hα = 0, so that flow alignment terms contribute neither to the stress tensor nor to the interfacial force. Next, we assume that polarized cells generate much larger active stresses than unpolarized cells. Hence, we neglect the active stress coefficient ¯ζ in front of ζ and ζ(cid:48). Then, assuming that the two-dimensional fluid layer is compressible, we take ζ = 8 ζ(cid:48) d = 2ζ(cid:48) and 2η = ¯η d = 2¯η for simplicity. Under these simplifications, and using Eq. (S8), the constitutive equations reduce to σαβ = η (∂αvβ + ∂βvα) − ζpαpβ, (S11) (S12) which close the set of equations defining the active polar fluid model of the spreading of an epithelial monolayer. fα = −ξvα + ζipα, LINEAR STABILITY ANALYSIS Here, we give the details of the linear stability analysis of the tissue front. First, we explicitly write down the equations of the model in Cartesian coordiates, which are most conve- nient for the rectangular geometry of the monolayer (Fig. 2a). Thus, the equation for the polarity field, Eq. (2), reads (cid:0)∂2 (cid:0)∂2 L2 c L2 c x + ∂2 y x + ∂2 y (cid:1) px = px, (cid:1) py = py. Respectively, the force balance equation Eq. (3) reads ∂xσxx + ∂yσxy = ξvx − T0/h px, ∂xσyx + ∂yσyy = ξvy − T0/h py, where the components of the stress tensor are given by σxx = 2η ∂xvx − ζp2 x, σxy = σyx = η (∂xvy + ∂yvx) − ζpxpy, σyy = 2η ∂yvy − ζp2 y. (S15a) (S15b) (S15c) Next, we obtain the flat front solution in rectangular geom- etry, which is the reference state of the linear stability anal- ysis. The long (y) axis of the rectangle is much longer than the short (x) axis. Hence, we assume translational invariance along the long axis of the monolayer [27]. Moreover, traction forces are mainly perpendicular to the monolayer boundary [37], so that we take the polarity field along the x direction: x (x) x, where the superindex indicates the zeroth order (cid:126)p = p0 in the perturbations of the front. Now, imposing symmetry as well as maximal polarity and stress-free boundary conditions, x (L0) = 1 and σ0 xx (L0) = 0, one obtains the polarity and p0 velocity profiles: p0 x (x) = sinh (x/Lc) sinh (L0/Lc) , (S16) (S13a) (S13b) (S14a) (S14b) v0 x (x) = ζ + (cid:20) ¯λ 2η 2ζ ¯λ2 4¯λ2 − L2 c − ζ (cid:20) c coth (L0/Lc) T0Lc¯λ2/h ¯λ2 − L2 (cid:2)2 + csch2 (L0/Lc)(cid:3)(cid:21) sinh(cid:0)x/¯λ(cid:1) cosh(cid:0)L0/¯λ(cid:1) (cid:21) ξ sinh (L0/Lc) Lc + 4¯λ2 − L2 c sinh (2x/Lc) sinh (L0/Lc) − T0Lc/h ¯λ2 − L2 c sinh (x/Lc) , (S17) (cid:112) 2η/ξ = √2 λ is a redefined hydrodynamic where ¯λ = screening length, and L0 is the semi-width of the monolayer, which changes according to dL0/dt = v0 x (L0). Next, we introduce peristaltic small-amplitude perturba- tions of the flat interface of the monolayer (Fig. 2a): L (y) = L0 + δL (y) . (S18) Under these perturbations, the polarity and velocity fields take the form px (x, y) = p0 x (x) + δpx (x, y) , py (x, y) = δpy (x, y) , In turn, boundary conditions must keep imposing a normal and maximal polarity, as well as vanishing normal and shear stresses at the interface, which is now curved. To this end, we define the normal and tangential vectors of each interface, n± = ± cos θ x + sin θ y ≈ ± x − dδL t± = ∓ sin θ x + cos θ y ≈ ± dy dδL dy y, x + y, (S21a) (S21b) where θ is the angle between the normal directions of the flat and perturbed interfaces, and the ± index stands for the top and bottom interfaces, respectively (Fig. 2a). Thus, the bound- ary conditions for the polarity read (cid:126)p · t±(cid:12)(cid:12)x=±L = 0. (S22) (cid:126)p · n±x=±L = 1, For the x-component, the conditions imply px (±L) ≈ ±1. This expands into px (±L) = p0 ≈ p0 x (±L0) ± ∂xp0 x (±L) + δpx (±L) x (±L0) δL + δpx (±L) ≈ ±1, which yields δpx (±L) = ∓∂xp0 x (±L0) δL 9 stress gives σxx (±L) = 0 which, after expanding as previ- ously, leads to δσxx (±L) = ∓∂xσ0 xx (±L0) δL (S27) for the stress perturbation. In turn, the condition on the shear stress directly gives δσxy (±L) = 0. (S28) Next, we decompose all perturbations in their Fourier modes, identified by the wave number q: (cid:90) ∞ (cid:90) ∞ (cid:90) ∞ −∞ −∞ δ L (q, t) eiqy dq 2π , δ pα (x, q, t) eiqy dq 2π δvα (x, q, t) eiqy dq 2π , , (S29a) (S29b) (S29c) δvα (x, y, t) = −∞ In terms of the Fourier modes, the equations for the polarity components read In turn, the components of the force balance equation, once the constitutive relation is introduced, read L2 c L2 c (cid:0)∂2 x − q2(cid:1) δ px = δ px, (cid:0)∂2 x − q2(cid:1) δ py = δ py. (cid:19) (cid:18) +(cid:2)T0/h − 2ζ(cid:0)∂xp0 (cid:18) +(cid:2)T0/h − ζ(cid:0)∂xp0 x − 2q2 − ∂2 x + p0 (cid:1)(cid:3) δ px − iqζ p0 (cid:19) (cid:1)(cid:3) δ py = 0. x − q2 − δvx + iqη ∂xδvy 1 λ2 x∂x iqη ∂xδvx + η x + p0 x∂x 1 λ2 δvy η 2∂2 (S30a) (S30b) x δ py = 0, (S31a) (S31b) vx (x, y) = v0 x (x) + δvx (x, y) , (S19) vy (x, y) = δvy (x, y) . (S20) δL (y, t) = δpα (x, y, t) = (S23) (S24) The boundary conditions must also be translated into the Fourier domain, reading δ px (±L) = ∓∂xp0 δ py (±L) = −iq δ L, (S32) x (±L0) δ L, as a boundary condition on the polarity perturbation. For the y-component of the polarity perturbation, the boundary con- dition imposes δpy (±L) = − dδL dy . (S25) Then, the boundary conditions on the stress read t± · σ · n±(cid:12)(cid:12)x=±L = 0, (S26) n± · σ · n±x=±L = 0, which respectively ensure vanishing normal and shear stress at the interfaces. Here, for simplicity, we neglect interfacial tension and bending rigidity, which would contribute stabiliz- ing terms to the growth rate. The condition on the normal xx (±L0) δ L, δσxx (±L) = ∓∂xσ0 (S33) Then, the four coupled differential equations Eqs. (S30) and (S31) are analytically solved for δ pα (x, q) and δvα (x, q). From the Fourier modes of the velocity field, the perturbed spreading velocity V can be computed as δσxy (±L) = 0. V = (cid:126)v · nx=L =(cid:2)(cid:126)v 0 · n + δ(cid:126)v · n(cid:3) x=L x (L0) + ∂xv0 x (L0) δL + δvx (L0) , (S34) ≈ v0 so that δV (y) = V (y) − V0 = ∂xv0 x (L0) δL (y) + δvx (L0, y) . (S35) Thus, the growth rate ω (q) of the tissue shape perturbations follows from δV (y) e −iqydy = dδ L (q) dt = ω (q) δ L (q) . (S36) (cid:90) ∞ −∞ δ V (q) = Hence, ω (q) = ∂xv0 x (L0) + δvx (L0, q) δ L (q) . (S37) The expression of the resulting growth rate is omitted here due to its length. Finally, note that, in our free-boundary problem, the amplitude of the front perturbations does not grow expo- nentially in time. This is because the growth rate depends on time through the monolayer width L0 (t). Consequently, Eq. (S36) yields (cid:20)(cid:90) t (cid:48)(cid:21) δ L (q, t) = δ L (q, 0) exp (cid:48) ω (q, t ) dt . (S38) 0 NEMATIC ELASTICITY In this section, we discuss the effects of the nematic elastic- ity of the polarity field on the growth rate of front shape pertur- bations. Front perturbations distort the polarity field, generat- ing polarity gradients along the tissue front. The elastic energy cost of these transversal polarity gradients is larger for shorter- wavelength perturbations. Thus, to minimize the polar free energy Eq. (1), polarity perturbations decay more steeply for shorter wavelengths. From Eq. (2), the Fourier components of polarity perturbations obey L2 c that their decay length is (cid:96)c(q) = Lc[1 + (qLc)2]−1/2, which decreases with q. Similarly to Eq. (5), the corresponding ve- locity gradient perturbation is then proportional to T0(cid:96)c(q). Therefore, by reducing the size of the polarized boundary layer, (cid:96)c(q) ≤ Lc, nematic elasticity causes a decrease of the growth rate with decreasing wavelength. However, this effect is only significant at wavelengths shorter than the ne- matic length Lc (qLc > 1), which is typically smaller than the size of the fingers. x − q2(cid:1) δ pα = δ pα, so (cid:0)∂2 UNIFORM INTERCELLULAR CONTRACTILITY In this section, we discuss the effects of a uniform intercel- lular contractility term, characterized by the coefficient −¯ζ in Eq. (S9), which is neglected in the Main Text. Like the in- tercellular contractility −ζ, which is localized at the polarized boundary layer of the tissue, a uniform contractility has a sta- bilizing effect on the tissue front (Fig. S1). However, unlike its polarity-related counterpart −ζ, the uniform contractility −¯ζ has a size-dependent contribution to the spreading velocity: (cid:21) (cid:20) T0Lc h V0 ≈ 1 2η 10 FIG. S1. Growth rate of shape perturbations for different values of the uniform intercellular contractility −¯ζ = 0,−1,−2,−3,−4 kPa, which has a stabilizing effect on perturbations of all wavelengths. The other parameter values are in Table I in the Main Text except for ξ, ζ → 0. in the limit Lc (cid:28) L0 (cid:28) λ. Then, the expansion of the growth rate at long wavelengths reads (cid:21) (cid:20) 1 3τ ω (q) ≈ 1 τ + ¯ζ 2η + + ¯ζ 3η (qL0)2+O (cid:16) (qL0)4(cid:17) , (S40) where τ = 2ηh/(T0Lc). This expression shows that a uni- form contractility does not only affect the growth rate of finite- wavelength perturbations but that it also decreases the growth rate of the uniform mode, ω(0). If the contractility is suffi- ciently small to allow the tissue to spread (V0 > 0), −¯ζ < T0Lc/h, the tissue front remains unstable to long-wavelength perturbations. However, unlike the polarity-related contractil- ity −ζ, if the uniform contractility −¯ζ induces the retraction of the monolayer front, it also prevents its fingering instability (Fig. S1). LINEAR STABILITY ANALYSIS FOR UNIFORMLY POLARIZED TISSUES In this section, we address situations in which the epithe- lial monolayer is polarized not only in a boundary layer but also in its bulk [37]. To analyze the effects of bulk polar- ity, we consider the simplest possible situation, namely a uni- formly polarized epithelium. The state of uniform polariza- tion, however, presents two issues. First, it may be unstable in the bulk [79]. Here, we assume that the active contractility and traction force coefficients are small enough not to trig- ger this bulk instability. Second, because the polarity points outward at both fronts of the monolayer (Fig. 2), a spreading monolayer cannot have a uniform polarity. Following previ- ous works [18, 19], we avoid this issue by considering the tis- sue to be bounded by a comoving wall at x = 0. In this case, we can assume a fixed modulus of the polarity field, (cid:126)p = 1. Hence, the polar free energy is left only with Frank elasticity: (cid:90) K 2 + ¯ζ L0 (S39) F = (∂αpβ)(∂αpβ) d3(cid:126)r. (S41) −0.4−0.3−0.2−0.100.10.20.300.010.020.030.040.05−¯ζGrowthrateω(h−1)Wavenumberq(µm−1)Consequently, the equilibrated polarity field, which fulfills δF/δpα = 0, is a solution of ∇2pα = 0. (S42) 11 The rest of the model is unchanged with respect to the case with only a polarized boundary layer analyzed in the Main Text. Thus, the force balance is still given by Eqs. 3-4, and the boundary conditions still impose a normal polarity and a vanishing stress at the monolayer front: (cid:126)p(x = L) = n and σ · nx=L = 0, respectively, with n being the normal unit vector of the tissue front. Moreover, we impose the following boundary conditions at the comoving wall: (cid:126)p(x = 0) = x, and (cid:126)v(x = 0) = T0/(ξh) x. Thus, the reference state of the stability analysis is the flat front solution with (cid:34) cosh(cid:0)x/¯λ(cid:1)(cid:35) sinh(cid:0)x/¯λ(cid:1) p0 x = 1, v0 x(x) = , (S43) where, as previously defined, ¯λ = 2η/ξ. The linear stability analysis can thus be performed as be- fore. In the present case, instead of Eq. (S30), the equations for the Fourier modes of the polarity perturbations read ζ ¯λ 1 ξ T0 h + (cid:112) (cid:0)∂2 x − q2(cid:1) δ px = 0, (cid:0)∂2 x − q2(cid:1) δ py = 0. (cid:19) (S44a) (S44b) (S45a) (S45b) Respectively, the components of the force balance equation, Eq. (S31), reduce to (cid:18) η 1 λ2 δvx + iqη ∂xδvy 2∂2 + (T0/h − 2ζ∂x) δ px − iqζ δ py = 0, x − q2 − (cid:18) (cid:19) 1 λ2 ∂2 x − 2q2 − + (T0/h − ζ∂x) δ py = 0. δvy iqη ∂xδvx + η Finally, Eqs. (S32) and (S33), now become the boundary conditions, previously given by δ px(0) = 0, δ py(0) = 0, δ px (L) = 0, δ py (L) = −iq δ L, (S46a) (S46b) δvx(0) = 0, δvy(0) = 0, δσxx (L) = ∓∂xσ0 δσxy (L) = 0. xx (L0) δ L, (S47a) (S47b) Then, Eqs. (S44) and (S45) are analytically solved for δ pα(x, q) and δvα(x, q). Hence, the growth rate of front per- turbations is computed using Eq. (S37). The resulting expres- sion is omitted here due to its length. However, the growth rate is plotted in Fig. S2, which shows how it changes under variation of different parameters. From these results, we con- clude that, as for tissues with only a polarized boundary layer, uniformly polarized tissues also display an active fingering in- stability. However, in contrast to the boundary layer case, in FIG. S2. Contributions to the instability for a uniformly polar- ized tissue. Growth rates of front perturbations varying the val- ues of different model parameters. Excluding the varied parameter, other parameter values are in Table I in the Main Text except for ζ → 0. (a) Even though the uniform mode (q = 0) is marginally stable, traction forces destabilize the monolayer front. For this plot, T0 = 0, 0.25, 0.5, 0.75, 1 kPa. (b) Long-range transmission of viscous stresses selects the fastest-growing mode. For this plot L0 = 50, 100, 150, 200, 250 µm. (c) Cell-substrate friction screens hydrodynamic interactions to limit the wavelength of the fingering pattern. For this plot, ξ = 102, 5· 102, 2.5· 103, 1.25· 104, 6.25· 104 Pa·s/µm2. (d) Contractility has a stabilizing contribution on the monolayer front. For small contractility, a band of unstable modes at finite wavelength remains. A sufficiently large contractility stabilizes the front. For this plot, −ζ = 0, 1, 2, 3, 4 kPa. the absence of contractility, uniformly polarized tissues spread at a constant velocity V0 = T0/(ξh). As a consequence, the q = 0 mode, is now marginally stable. Nonetheless, all other modes are destabilized by the interplay between active trac- tion forces and viscous stresses. As for the case with an active boundary layer, front perturbations give rise to flow perturba- tions that destabilize the flat front (Fig. S2a). Hence, the active fingering instability is robust to the presence of bulk polarity in the tissue. Moreover, the role of hydrodynamic interactions is un- changed with respect to the active boundary layer case. As in that case, the finite range of hydrodynamic interactions, given by the smallest between the monolayer width L0 and η/ξ determines the hydrodynamic screening length λ = the wavelength of the most unstable mode. Thus, for polarized tissues, varying the monolayer width L0 and the cell-substrate friction coefficient ξ modifies the growth rate in a similar way as for tissues with only a polarized boundary layer (compare Fig. S2b-c to Fig. 3b-c). (cid:112) Finally, the effect of the contractility in uniformly polarized tissues is a bit different than in unpolarized tissues. As for tissues with only boundary polarity, contractility has a stabi- lizing effect on the monolayer front. However, for uniformly polarized tissues, the active contractile stress spans through- (d)<latexit sha1_base64="WO4201SibMC27/h//bd7XU4MxJU=">AAACAXicbVC9TsMwGHT4LeUvwMiARYVUlipFSJStEgtjkQit1ESV4zqtVTuJ7C9IUdQNXgYWhGDiFXgB3ganZICWm853Z8t3QSK4Bsf5spaWV1bX1isb1c2t7Z1de2//TseposylsYhVLyCaCR4xFzgI1ksUIzIQrBtMrgq/e8+U5nF0C1nCfElGEQ85JWCkgX3khXEEIZFcZHkCcuqZtxiFQsX14enArjkNZwa8SJolqaESnYH96Q1jmkoWARVE637TScDPiQJOBZtWvVSzhNAJGbG+oRGRTPv5rMgUn4SxwjBmeHb+nc2J1DqTgclIAmM97xXif14/hbDl5zxKUmARNRHjhanAEONiDzzkytQVmSGEKm5+iemYKELBrFY19ZvzZReJe9a4bDg357V2q9yhgg7RMaqjJrpAbXSNOshFFD2iZ/SG3q0H68l6sV5/oktWeecA/YH18Q1sPJeF</latexit><latexit sha1_base64="WO4201SibMC27/h//bd7XU4MxJU=">AAACAXicbVC9TsMwGHT4LeUvwMiARYVUlipFSJStEgtjkQit1ESV4zqtVTuJ7C9IUdQNXgYWhGDiFXgB3ganZICWm853Z8t3QSK4Bsf5spaWV1bX1isb1c2t7Z1de2//TseposylsYhVLyCaCR4xFzgI1ksUIzIQrBtMrgq/e8+U5nF0C1nCfElGEQ85JWCkgX3khXEEIZFcZHkCcuqZtxiFQsX14enArjkNZwa8SJolqaESnYH96Q1jmkoWARVE637TScDPiQJOBZtWvVSzhNAJGbG+oRGRTPv5rMgUn4SxwjBmeHb+nc2J1DqTgclIAmM97xXif14/hbDl5zxKUmARNRHjhanAEONiDzzkytQVmSGEKm5+iemYKELBrFY19ZvzZReJe9a4bDg357V2q9yhgg7RMaqjJrpAbXSNOshFFD2iZ/SG3q0H68l6sV5/oktWeecA/YH18Q1sPJeF</latexit><latexit sha1_base64="WO4201SibMC27/h//bd7XU4MxJU=">AAACAXicbVC9TsMwGHT4LeUvwMiARYVUlipFSJStEgtjkQit1ESV4zqtVTuJ7C9IUdQNXgYWhGDiFXgB3ganZICWm853Z8t3QSK4Bsf5spaWV1bX1isb1c2t7Z1de2//TseposylsYhVLyCaCR4xFzgI1ksUIzIQrBtMrgq/e8+U5nF0C1nCfElGEQ85JWCkgX3khXEEIZFcZHkCcuqZtxiFQsX14enArjkNZwa8SJolqaESnYH96Q1jmkoWARVE637TScDPiQJOBZtWvVSzhNAJGbG+oRGRTPv5rMgUn4SxwjBmeHb+nc2J1DqTgclIAmM97xXif14/hbDl5zxKUmARNRHjhanAEONiDzzkytQVmSGEKm5+iemYKELBrFY19ZvzZReJe9a4bDg357V2q9yhgg7RMaqjJrpAbXSNOshFFD2iZ/SG3q0H68l6sV5/oktWeecA/YH18Q1sPJeF</latexit>(c)<latexit sha1_base64="YqIKExJwS4JI8MJStNpKqbpqo/g=">AAACAXicbVC9TsMwGHT4LeUvwMiARYVUlipFSJStEgtjkQit1ESV4zqtVTuJ7C9IUdQNXgYWhGDiFXgB3ganZICWm853Z8t3QSK4Bsf5spaWV1bX1isb1c2t7Z1de2//TseposylsYhVLyCaCR4xFzgI1ksUIzIQrBtMrgq/e8+U5nF0C1nCfElGEQ85JWCkgX3khXEEIZFcZHkCcuqZtxiFQsV1ejqwa07DmQEvkmZJaqhEZ2B/esOYppJFQAXRut90EvBzooBTwaZVL9UsIXRCRqxvaEQk034+KzLFJ2GsMIwZnp1/Z3Mitc5kYDKSwFjPe4X4n9dPIWz5OY+SFFhETcR4YSowxLjYAw+5MnVFZgihiptfYjomilAwq1VN/eZ82UXinjUuG87Nea3dKneooEN0jOqoiS5QG12jDnIRRY/oGb2hd+vBerJerNef6JJV3jlAf2B9fANqvpeE</latexit><latexit sha1_base64="YqIKExJwS4JI8MJStNpKqbpqo/g=">AAACAXicbVC9TsMwGHT4LeUvwMiARYVUlipFSJStEgtjkQit1ESV4zqtVTuJ7C9IUdQNXgYWhGDiFXgB3ganZICWm853Z8t3QSK4Bsf5spaWV1bX1isb1c2t7Z1de2//TseposylsYhVLyCaCR4xFzgI1ksUIzIQrBtMrgq/e8+U5nF0C1nCfElGEQ85JWCkgX3khXEEIZFcZHkCcuqZtxiFQsV1ejqwa07DmQEvkmZJaqhEZ2B/esOYppJFQAXRut90EvBzooBTwaZVL9UsIXRCRqxvaEQk034+KzLFJ2GsMIwZnp1/Z3Mitc5kYDKSwFjPe4X4n9dPIWz5OY+SFFhETcR4YSowxLjYAw+5MnVFZgihiptfYjomilAwq1VN/eZ82UXinjUuG87Nea3dKneooEN0jOqoiS5QG12jDnIRRY/oGb2hd+vBerJerNef6JJV3jlAf2B9fANqvpeE</latexit><latexit sha1_base64="YqIKExJwS4JI8MJStNpKqbpqo/g=">AAACAXicbVC9TsMwGHT4LeUvwMiARYVUlipFSJStEgtjkQit1ESV4zqtVTuJ7C9IUdQNXgYWhGDiFXgB3ganZICWm853Z8t3QSK4Bsf5spaWV1bX1isb1c2t7Z1de2//TseposylsYhVLyCaCR4xFzgI1ksUIzIQrBtMrgq/e8+U5nF0C1nCfElGEQ85JWCkgX3khXEEIZFcZHkCcuqZtxiFQsV1ejqwa07DmQEvkmZJaqhEZ2B/esOYppJFQAXRut90EvBzooBTwaZVL9UsIXRCRqxvaEQk034+KzLFJ2GsMIwZnp1/Z3Mitc5kYDKSwFjPe4X4n9dPIWz5OY+SFFhETcR4YSowxLjYAw+5MnVFZgihiptfYjomilAwq1VN/eZ82UXinjUuG87Nea3dKneooEN0jOqoiS5QG12jDnIRRY/oGb2hd+vBerJerNef6JJV3jlAf2B9fANqvpeE</latexit>(b)<latexit sha1_base64="ofq78kasmBdkMGFQrMd6jUe3Fgk=">AAACAXicbVC9TsMwGHT4LeUvwMiARYVUlipFSJStEgtjkQit1ESV4zqtVTuJ7C9IUdQNXgYWhGDiFXgB3ganZICWm853Z8t3QSK4Bsf5spaWV1bX1isb1c2t7Z1de2//TseposylsYhVLyCaCR4xFzgI1ksUIzIQrBtMrgq/e8+U5nF0C1nCfElGEQ85JWCkgX3khXEEIZFcZHkCcuqZtxiFQsX14HRg15yGMwNeJM2S1FCJzsD+9IYxTSWLgAqidb/pJODnRAGngk2rXqpZQuiEjFjf0IhIpv18VmSKT8JYYRgzPDv/zuZEap3JwGQkgbGe9wrxP6+fQtjycx4lKbCImojxwlRgiHGxBx5yZeqKzBBCFTe/xHRMFKFgVqua+s35sovEPWtcNpyb81q7Ve5QQYfoGNVRE12gNrpGHeQiih7RM3pD79aD9WS9WK8/0SWrvHOA/sD6+AZpQJeD</latexit><latexit sha1_base64="ofq78kasmBdkMGFQrMd6jUe3Fgk=">AAACAXicbVC9TsMwGHT4LeUvwMiARYVUlipFSJStEgtjkQit1ESV4zqtVTuJ7C9IUdQNXgYWhGDiFXgB3ganZICWm853Z8t3QSK4Bsf5spaWV1bX1isb1c2t7Z1de2//TseposylsYhVLyCaCR4xFzgI1ksUIzIQrBtMrgq/e8+U5nF0C1nCfElGEQ85JWCkgX3khXEEIZFcZHkCcuqZtxiFQsX14HRg15yGMwNeJM2S1FCJzsD+9IYxTSWLgAqidb/pJODnRAGngk2rXqpZQuiEjFjf0IhIpv18VmSKT8JYYRgzPDv/zuZEap3JwGQkgbGe9wrxP6+fQtjycx4lKbCImojxwlRgiHGxBx5yZeqKzBBCFTe/xHRMFKFgVqua+s35sovEPWtcNpyb81q7Ve5QQYfoGNVRE12gNrpGHeQiih7RM3pD79aD9WS9WK8/0SWrvHOA/sD6+AZpQJeD</latexit><latexit sha1_base64="ofq78kasmBdkMGFQrMd6jUe3Fgk=">AAACAXicbVC9TsMwGHT4LeUvwMiARYVUlipFSJStEgtjkQit1ESV4zqtVTuJ7C9IUdQNXgYWhGDiFXgB3ganZICWm853Z8t3QSK4Bsf5spaWV1bX1isb1c2t7Z1de2//TseposylsYhVLyCaCR4xFzgI1ksUIzIQrBtMrgq/e8+U5nF0C1nCfElGEQ85JWCkgX3khXEEIZFcZHkCcuqZtxiFQsX14HRg15yGMwNeJM2S1FCJzsD+9IYxTSWLgAqidb/pJODnRAGngk2rXqpZQuiEjFjf0IhIpv18VmSKT8JYYRgzPDv/zuZEap3JwGQkgbGe9wrxP6+fQtjycx4lKbCImojxwlRgiHGxBx5yZeqKzBBCFTe/xHRMFKFgVqua+s35sovEPWtcNpyb81q7Ve5QQYfoGNVRE12gNrpGHeQiih7RM3pD79aD9WS9WK8/0SWrvHOA/sD6+AZpQJeD</latexit>(a)<latexit sha1_base64="P/2NV/rEH2AxGcsEXmQ9uouTj8w=">AAACAXicbVC9TsMwGHT4LeUvwMiARYVUlipFSJStEgtjkQit1ESV4zqtVTuJ7C9IUdQNXgYWhGDiFXgB3ganZICWm853Z8t3QSK4Bsf5spaWV1bX1isb1c2t7Z1de2//TseposylsYhVLyCaCR4xFzgI1ksUIzIQrBtMrgq/e8+U5nF0C1nCfElGEQ85JWCkgX3khXEEIZFcZHkCcuqZtxiFQsV1cjqwa07DmQEvkmZJaqhEZ2B/esOYppJFQAXRut90EvBzooBTwaZVL9UsIXRCRqxvaEQk034+KzLFJ2GsMIwZnp1/Z3Mitc5kYDKSwFjPe4X4n9dPIWz5OY+SFFhETcR4YSowxLjYAw+5MnVFZgihiptfYjomilAwq1VN/eZ82UXinjUuG87Nea3dKneooEN0jOqoiS5QG12jDnIRRY/oGb2hd+vBerJerNef6JJV3jlAf2B9fANnwpeC</latexit><latexit sha1_base64="P/2NV/rEH2AxGcsEXmQ9uouTj8w=">AAACAXicbVC9TsMwGHT4LeUvwMiARYVUlipFSJStEgtjkQit1ESV4zqtVTuJ7C9IUdQNXgYWhGDiFXgB3ganZICWm853Z8t3QSK4Bsf5spaWV1bX1isb1c2t7Z1de2//TseposylsYhVLyCaCR4xFzgI1ksUIzIQrBtMrgq/e8+U5nF0C1nCfElGEQ85JWCkgX3khXEEIZFcZHkCcuqZtxiFQsV1cjqwa07DmQEvkmZJaqhEZ2B/esOYppJFQAXRut90EvBzooBTwaZVL9UsIXRCRqxvaEQk034+KzLFJ2GsMIwZnp1/Z3Mitc5kYDKSwFjPe4X4n9dPIWz5OY+SFFhETcR4YSowxLjYAw+5MnVFZgihiptfYjomilAwq1VN/eZ82UXinjUuG87Nea3dKneooEN0jOqoiS5QG12jDnIRRY/oGb2hd+vBerJerNef6JJV3jlAf2B9fANnwpeC</latexit><latexit sha1_base64="P/2NV/rEH2AxGcsEXmQ9uouTj8w=">AAACAXicbVC9TsMwGHT4LeUvwMiARYVUlipFSJStEgtjkQit1ESV4zqtVTuJ7C9IUdQNXgYWhGDiFXgB3ganZICWm853Z8t3QSK4Bsf5spaWV1bX1isb1c2t7Z1de2//TseposylsYhVLyCaCR4xFzgI1ksUIzIQrBtMrgq/e8+U5nF0C1nCfElGEQ85JWCkgX3khXEEIZFcZHkCcuqZtxiFQsV1cjqwa07DmQEvkmZJaqhEZ2B/esOYppJFQAXRut90EvBzooBTwaZVL9UsIXRCRqxvaEQk034+KzLFJ2GsMIwZnp1/Z3Mitc5kYDKSwFjPe4X4n9dPIWz5OY+SFFhETcR4YSowxLjYAw+5MnVFZgihiptfYjomilAwq1VN/eZ82UXinjUuG87Nea3dKneooEN0jOqoiS5QG12jDnIRRY/oGb2hd+vBerJerNef6JJV3jlAf2B9fANnwpeC</latexit>00.020.040.060.080.10.120.140.160.1800.010.020.030.040.05L0Growthrate!(h1)Wavenumberq(µm1)0.0500.050.10.150.20.2500.010.020.030.040.05T0Growthrate!(h1)Wavenumberq(µm1)00.020.040.060.080.10.120.1400.010.020.030.040.05⇠Growthrate!(h1)Wavenumberq(µm1)0.30.250.20.150.10.0500.050.10.1500.010.020.030.040.05⇣Growthrate!(h1)Wavenumberq(µm1)out the monolayer. As a consequence, instead of just affecting short-wavelength perturbations, contractility now decreases the growth rate of all perturbation modes. Hence, for small contractility, the longest and shortest-wavelength modes be- come stable, leaving a band of unstable modes at intermedi- ate wavelengths. Therefore, small contractilities do not abro- gate the fingering instability. A sufficiently large contractility, however, is able to stabilize all modes, thus suppressing the fingering instability (Fig. S2d). 12
1605.03434	2	1605	2017-05-30T13:33:25	Slip of grip of a molecular motor on a crowded track: Modeling shift of reading frame of ribosome on RNA template	[ "physics.bio-ph", "q-bio.SC" ]	We develop a stochastic model for the programmed frameshift of ribosomes synthesizing a protein while moving along a mRNA template. Normally the reading frame of a ribosome decodes successive triplets of nucleotides on the mRNA in a step-by-step manner. We focus on the programmed shift of the ribosomal reading frame, forward or backward, by only one nucleotide which results in a fusion protein; it occurs when a ribosome temporarily loses its grip to its mRNA track. Special "slippery" sequences of nucleotides and also downstream secondary structures of the mRNA strand are believed to play key roles in programmed frameshift. Here we explore the role of an hitherto neglected parameter in regulating -1 programmed frameshift. Specifically, we demonstrate that the frameshift frequency can be strongly regulated also by the density of the ribosomes, all of which are engaged in simultaneous translation of the same mRNA, at and around the slippery sequence. Monte Carlo simulations support the analytical predictions obtained from a mean-field analysis of the stochastic dynamics.	physics.bio-ph	physics	epl draft Slip of grip of a molecular motor on a crowded track: Modeling shift of reading frame of ribosome on RNA template Bhavya Mishra1, Gunter M. Schutz2 and Debashish Chowdhury1 1 Department of Physics, Indian Institute of Technology, Kanpur 208016, India. 2 Institute of Complex Systems II, Forschungszentrum Julich, 52425 Julich, Germany. 7 1 0 2 y a M 0 3 ] h p - o i b . s c i s y h p [ PACS 87.16.-b -- Subcellular structure and processes PACS 05.40.-a -- Fluctuation phenomena, random processes, noise, and Brownian motion PACS 05.70.Ln -- Nonequilibrium and irreversible thermodynamics Abstract -- We develop a stochastic model for the programmed frameshift of ribosomes synthesizing a protein while moving along a mRNA template. Normally the reading frame of a ribosome decodes successive triplets of nucleotides on the mRNA in a step-by-step manner. We focus on the programmed shift of the ribosomal reading frame, forward or backward, by only one nucleotide which results in a fusion protein; it occurs when a ribosome temporarily loses its grip to its mRNA track. Special "slippery" sequences of nucleotides and also downstream secondary structures of the mRNA strand are believed to play key roles in programmed frameshift. Here we explore the role of an hitherto neglected parameter in regulating -1 programmed frameshift. Specifically, we demonstrate that the frameshift frequency can be strongly regulated also by the density of the ribosomes, all of which are engaged in simultaneous translation of the same mRNA, at and around the slippery sequence. Monte Carlo simulations support the analytical predictions obtained from a mean-field analysis of the stochastic dynamics. Key words: ribosome traffic, master equation, programmed frame shift. 2 v 4 3 4 3 0 . 5 0 6 1 : v i X r a Introduction. -- A protein is a linear hetero-polymer made of a sequence of monomeric subunits, called amino acids, each of which is linked to its immediate neighbor by a peptide bond. Nature normally uses 20 different types of amino acids to make proteins in living cells. The partic- ular sequence of the types of amino acids in a protein is determined by the sequence of nucleotides, the monomeric subunits, of the corresponding template mRNA molecule. The actual synthesis of the protein, as directed by the mRNA template, is carried out by a molecular machine [1], called ribosome [2 -- 4] and the process is referred to as translation (of genetic message). Translation is broadly divided into three stages: Initiation, elongation and termi- nation. Elongation of the growing protein by the ribosome takes place in a step-by-step manner, the addition of each amino acid monomer to it is accompanied by a forward stepping of the ribosome on its mRNA template by one codon, each codon being a triplet of nucleotides. Thus, a ribosome is also a molecular motor [2, 4] that exploits the mRNA template as its track and moves forward along it, by three nucleotides in each step, converting chemical energy into mechanical work. At each position of the ribosome its "reading frame" decodes a triplet of nucleotides on the mRNA template and then slides to the next triplet as the ribosome steps forward by a codon. This reading frame is established in the initiation stage of translation and must be main- tained faithfully during the course of normal elongation of the protein. However, in all kingdoms of life, on many template mRNA strands there are some special "slippery" sequences of nucleotides where a ribosome can lose its grip on its track, resulting in a shift of its reading frame either backward or forward by one or more nucleotides. These processes are referred to as ribosomal frameshift [5,6]. The most commonly occurring, and extensively studied, cases correspond to a shift of the reading frame backward or for- ward by a single nucleotide on the mRNA track. These are referred to as -1 frameshift and +1 frameshift, respectively, and will be the main focus of our study in this letter. The rate of frameshift at any arbitrary position on the mRNA track has been found to be negligibly small. In contrast to such a random frameshift, a "programmed" frameshift at a specific location on the mRNA track is known to occur with much higher rate and have important p-1 Bhavya Mishra1 Gunter M. Schutz2 Debashish Chowdhury1 -1 reading frame. Thus it is evident that the composition and length of the slippery sequence, its distance from the downstream pseudoknot, the energetic stability of the pseuodoknot and the kinetics of its unfolding and refolding etc. collectively encode the program that determines not only the spatial location and timing of the frame shift but also its frequency or efficiency. A combination of all these stimulators and signals are believed to alter the normal free energy land- scape of the ribosome, thereby affecting the stability of various intermediate states as well as the kinetics of tran- sitions among them that favor frameshift over the other alternative pathways [5, 6]. Several competing models have been developed to account for the mechanisms of stimu- lation, regulation and control of frameshift; the models differ in (a) their hypothesis as to the sub-step of the mechano-chemical cycle in which the frameshift is assumed to occur, and (b) the assumed structural, energetic and kinetic cause of the slippage [12]. Instead of committing at present to any specific structural model for frameshift, we capture the effects of the slippery sequence and the downstream pseudoknot by physically motivated generic alterations of the kinetic rates in a reduced minimal model of the elongation kinetics. The main aim of this letter is to demonstrate how an hitherto neglected control parameter, namely the average inter-ribosome separation, or equivalently the mean number density of the ribosomes, on the mRNA track can up- or down-regulate the efficiency of ribosomal frameshift in vivo. In the past indications for the interplay of the inter-ribosome separation and the kinetics of unwinding of the pseudoknot have emerged from experimental studies of frameshift in vivo [13] as well as in experiments with synthetic mRNA secondary structures in vitro [14]. Because of the superficial similarities between the simul- taneous movement of multiple ribosomes along a single mRNA track and vehicular traffic along a single-lane high- way, the former is often referred to as ribosome traffic. Various aspects of this traffic-like phenomenon have been modelled by totally asymmetric simple exclusion process (TASEP) [15] and its biologically motivated extensions. The TASEP, an abstract model of self-driven interacting particles, is one of the simplest models of driven non- equilibrium systems in statistical physics. Since the pseu- doknot acts as a bottleneck against the flow of ribosome traffic, the spatio-temporal organization of the ribosomes exhibit some of the key characteristics of TASEP with quenched defects; these include, as we show here, phase segregation of the ribosomes. In contrast to the earlier TASEP-based models of trans- lation [16 -- 28] the model introduced here treats individual nucleotides, rather than triplets of nucleotides (codons), as the basic unit of the mRNA track. Moreover, unlike the earlier TASEP-based models of translation, we do not focus on the phase diagram of the model. Instead,the most important result of our investigation is the up- and down-regulation of ribosomal frameshift caused by the long Fig. 1: (Color online) A schematic representation of -1 pro- grammed ribosomal frameshift (PRF) from initial reading frame to -1 reading frame. biological functions. In this letter we consider, exclusively, programmed frameshift and ignore random frameshift al- together. After suffering a programmed frameshift, the ribosome resumes its operation but decodes the template using the shifted reading frame (see Fig. (1)) thereby pro- ducing a "fusion" protein. The classic example of such a fusion product of -1 frame shift is the gag-pol fusion protein of the human immunodeficiency virus (HIV) [5, 6]. Programmed frameshift, a mode of non-conventional trans- lation [7] happens to be one of the several modes of genetic recoding [6]. A common feature of these recoding phe- nomena is the context-dependent temporary alteration of the readout of mRNA. Understanding the principles fol- lowed by nature for encoding and decoding genetic message would be incomplete without understanding the causes and consequences of all such dynamic recoding. Programmed -1 frameshift requires two key ingredients: (a) A slippery sequence (usually about seven nucleotide long) on the mRNA, and (b) a downstream secondary structure (usually a pseudoknot [8, 9]) of the mRNA [10]. Normally the pseudoknot on the mRNA template is located about 6 nucleotides downstream from the slippery sequence. In order to enter the segment of mRNA template that forms the pseudoknot, a ribosome has to unwind the secondary structure. Thus, undoubtedly, the pseudoknot acts as a roadblock against the forward movement of the ribosome that suffers a long pause on the slippery site. However, such long pauses are necessary, but may not be sufficient, for inducing -1 frameshift of the ribosomes because not all types of road blocks on mRNA can induce frameshift [11]. Moreover, not every ribosome suffers frameshift at a given slippery sequence. Even more intriguing is the fact that the same pseudoknot, that puts such an insurmountable hurdle on the path of a ribosome in the 0 reading frame, allows it to pass through more easily after the shift to the p-2 XXXYYYZOne CodonNucleotidesSlippery SequencetRNAXXXYYYZSlippery SequencetRNATitle Fig. 2: (Color online) A schematic depiction of the model where a secondary structure of the mRNA (a cartoon of a pseudoknot) is shown. The mapping of the tortuous contour of the mRNA template onto a linear (one-dimensionl) chain is shown explic- itly. Each site of the chain represents a single nucleotide. The first seven sites of the segment II form a "slippery" sequence while the next 2-3 nucleotides would correspond to the spacer region whereas the remaining part of the segment II remains folded in the pseudoknot. From the specific site L1 + 1 pro- grammed -1 frameshift can take place. The segments I and III are the segments of the mRNA that are located before and after, respectively, of the pseudoknot. Fig. 3: (Color online) Frameshift flux Jf s is plotted against the parameter Ws/W . In the inset, the probability of coverage of the site i = L1, which is targetted for successful -1 frameshift, is plotted also against Ws/W . The theoretical predictions derived under MFA are drawn by the dashed curve (blue). The numerical data obtained from MC simulations have been plotted with dots; the curve connecting these dots (red) serves merely as a guide to the eye. The initiation rate α = 100s−1, termination rate β = 10s−1 and W = 83 s−1. The system size is N = L + (cid:96) − 1 = 1000 + (cid:96) − 1, with the rod size (cid:96) = 18, L1 = 399, L2 = 450. queue that stretches upstream from the pseudoknot to the slippery sequence and beyond. We also suggest new appli- cations of existing experimental techniques to test some of the new predictions made in this paper on the basis of our theoretical calculations. Model. -- In our model mRNA is treated as a linear chain (also called a lattice) of sites, each of which rep- resents a single nucleotide. Each nucleotide is identified with an integer index j. The total length of the chain is L + (cid:96) − 1, in the units of nucleotide length, although only the segment starting from j = 1 to j = L gets translated by the ribosomes. A specific site S on this chain denotes the second nucleotide of the slippery sequence [6]. A ribosome is modelled as a rigid rod of length (cid:96) also in the units of nucleotide length, i.e., each ribosome can cover simulta- neously (cid:96) successive nucleotides on its mRNA template. According to the convention followed consistently through- out this paper, the instantaneous position of a ribosome on the mRNA template is denoted by the integer index that labels the leftmost site covered by the ribosome at that instant of time. We make a clear distinction between the terms occupied and covered: A site j is occupied by a ribosome, i.e., its instantaneous position is j, if the ribo- some is decoding the triplet of nucleotides j, j + 1, j + 2, while all the sites j, j + 1, ...j + (cid:96) − 1 remain covered by it simultaneously at that instant. In this generic model developed here for ribosomal frameshift the lattice is divided into three segments. The segment I is ranging from site 1 to site L1 − 1 (1 ≤ j ≤ L1 − 1), segment II from site L1 to site L2 (L1 ≤ j ≤ L2) and segment III from site L2 + 1 to site L + (cid:96) − 1 (L2 + 1 ≤ j ≤ L + (cid:96) − 1) (see Fig. 2). The segment II represents the stretch of the mRNA template that is folded in the form of the pseudoknot. All the numerical data presented here have been obtained with the choice L1 = 399 and L2 = 450. Inside segment II seven nu- cleotides, from site j = L1 to j = L1 + 6 represent a slippery sequence whose second site (i.e., j = L1 + 1) is the special site from where the ribosomal frameshift is assumed to take place. The next 2-3 nucleotides would correspond to the spacer region between the slippery sequence and the pseudoknot while the remaining nucleotides of segment II form the pseudoknot. The choice of the numerical value 41 nucleotide for the length of the mRNA in the pseudoknot is only a typical one that lies between estimated lengths of the smallest and largest pseudoknots [9]. Translation initiation is captured in this model as follows: If the first (cid:96) sites of the lattice are empty, a ribosome can occupy the position j = 1 (and, thus, cover the sites 1, 2, ..., (cid:96)). This event occurs with a rate (i.e., probability per time unit) α. Similarly, termination of translation is described as the detachment of the ribosome from the lattice when its position is L, i.e., it covers the last (cid:96) sites of the lattice that are labelled by L, L + 1, ..., L + (cid:96) − 1; the rate of this event is β. p-3 Slippery SequenceXXXYYYαβWsWtWfsWWSegment I Segment II Segment III1 L1-1 L1 L2 L2+1 L+l-1Pseudoknot Z Bhavya Mishra1 Gunter M. Schutz2 Debashish Chowdhury1 Fig. 4: (Color online) Occupational density profile in steps of 3 nucleotides. The dashed line (blue) is the theoretical prediction under MFA and the solid line (red) connecting the discrete symbols (cid:12) is obtained from MC simulation with initiation rate α = 100s−1, termination rate β = 10s−1, W = 83s−1, Wf s = 13.3s−1 and Ws = 0.00074s−1. System size N = L + (cid:96) − 1 = 1000 + (cid:96) − 1 with the rod size (cid:96) = 18; the special site is located at i = 400. The inset shows the occupational density profile of a small region from i = 390 to i = 420 in steps of 1 nucleotide; the dashed line (blue) with ∗ and the solid line (red) connecting the discrete symbols (cid:12) have been obtained from MF theory and MC simulation, respectively. During the elongation stage a ribosome moves forward by three nucleotides upon successful completion of each elongation cycle. However, at the site S, which represents the second nucleotide of the slippery sequence, a ribosome can slip backward on its track by one single nucleotide; the rate of this -1 frameshift event, that we define below, is normally much less than unity. So far as the termination of translation by a frameshifted ribosome is concerned, we assume, for the sake of simplicity, that the ribosome detaches from the lattice when its position is L − 1, i.e., it covers the sites L − 1, L, ..., L + (cid:96) − 2. Thus, in this model the full length of a protein synthesized using the non-shifted frame and that of a fusion protein would consist of the same number of amino acids. In this scenario the overall rate of translation of each codon in the original (unshifted) frame is captured by an "effective rate" W of "hopping" of a ribosome by three steps in the forward direction, i.e., from position j to j + 3. In principle, the effective rate W can be expressed in terms of the actual rates of the individual transitions among the five distinct mechano-chemical states in the elongation stage (for details please see the supplementary information given in [29]). Next we assume that the effective hopping rate W in the pseudoknot segment II gets reduced exponentially to the value Ws, i.e., Ws/W = γ = exp(−b ∆ G) Fig. 5: (Color online) Net forward flux (J) plotted against the coverage density ρc for three different values of the jump rate ratio γ = Ws/W . System size is N = 1000 + (cid:96) − 1, with the rod size (cid:96) = 18, and W = 83 s−1. where −∆ G is a dimensionless parameter. This choice is motivated by the plausible identification ∆ G = (∆G)/(kBT ) where ∆G is the free energy barrier against forward movement of the ribosome by a single nucleotide within the pseudoknot region. Since a typical pseudoknot is not a mere hairpin, the effects of its structural complexity on the effective barrier is captured by the parameter b in the exponential. For the numerical data plotted graphically in this paper b = 3 was chosen. As ∆ G → 0 the choice of the form of Ws in (1) implies Ws → W ; this is consistent with the fact that in this limit the difference between the segment II and the other two segments disappears. In the opposite limit ∆ G → ∞, Ws → 0 as would be expected on physical grounds that it is practically impossible to unzip an extremely stiff psudoknot. The rate of frameshift Wf s depends upon two factors: (a) Strength of the pseudo- knot, expressed by ∆ G, and (b) the frequency Wf s0 which is related to the breaking the bonds between tRNA and codons. Motivated by the above physical considerations we make our next postulate: We assume that the rate of -1 frameshift in the slippery region is given by Wf s = Wf s0 exp(a ∆ G) (2) where a is the parameter that indirectly captures the com- plexity of pseudoknot structure. Note that in the limit ∆ G → 0, Wf s → Wf s0 which is a non-vanishing (but, pre- sumably small) rate of frameshift caused by the slippery sequence in the absence of a downstream psudoknot. More- over, the form (2) is based on the assumption of a sharp increase of Wf s with increasing ∆ G. For the numerical data plotted graphically in this paper a = 3 was chosen. [8]. (1) Similarly, the detachment (premature termination) rate p-4 Wt of a rod from the special site is assumed to be this site. Thus, Jf s is given by Wt = W exp(−c ∆ G) (3) Jf s = Wf sp(L1L1 + 1)P (L1 + 1). (4) Title Rods can detach prematurely from the special site only if the nearest neighbour site from the rightmost part of the rod in the forward direction is already covered by another rod. In an alternative version of this model, defined in the supplementary material, the rods can detach prematurely if the nearest neighbor of the left most part of the rod in backward direction, i.e., the site targetted for -1 frameshift, is already covered by another rod. The choice of the form (3) is motivated by our postulate that (a) in the absence of the pseudoknot (i.e., in the limit ∆ G → 0) both the forward hopping and premature detachment at the slippery site are equally probable, and that (b) for very stiff pseudoknots (i.e., ∆ G → ∞) practically a ribosome stalls (no forward movement because Ws → 0) and, therefore, no possibility of premature detachment [30]. Although, in principle, the two parameters a and c in (1) and (3) are not necessarily equal, we use a = c just for the sake of simplicity. The kinetics is implemented by the following rules: (a) A new rod can attach at site 1, if and only if all initial (cid:96) sites are empty. The rate of attachment at site 1 is α. (b) If there is a rod at site i = L or i = L − 1, then it can detach from the lattice, and the rate of detachment is β. (c) Inside segment I and III rods can jump in forward direction by +3 nucleotides only if the target site is empty and the rate of forward jump is W . (d) Inside segment II there is one special site i = L1 + 1. Except at this special site a rod can jump forward, by step size +3, if the target site is empty and the rate of forward jump in segment II is Ws. (e) From the special site i = L1 + 1 the following movements are possible: (e1) A Rod can jump forward with rate Ws, with step size +3, if the target site is empty. (e2) A Rod can slip back with step size −1 with rate Wf s, if the target site is empty. (e3) A Rod can detach, with rate Wt, from the lattice if the site L1 + 1 + (cid:96) (i.e., the site immediately in front of its forward edge) is occupied by another rod. In the analytical treatment of this model p(ji) is the conditional probability of finding the site j empty, given that the site i is already occupied, where j is the target site when a rod tends to move by ±1 nucleotides. Similarly, q(ki) is the conditional probability of finding the site k empty, given that the site i is already occupied, where k denotes the target site when the rod tends to move by 3 nucleotides. P (i, t) is the occupational probability and it is defined as the probability of finding the left edge of the rod at site i at time t. In this work we are interested in the frameshift flux (Jf s) from the special point which is defined as the total number of ribosomes that undergo frameshift per unit time from We can get the value of P (i, t) at each individual site i and time t by solving master equations under mean field approximation (MFA). At site i = 1 one has = α[1 − (cid:96)(cid:88) P (a, t)] dP (i, t) dt a=1 − W q(i + (cid:96) + 2i)P (i, t), (5) since a new rod can attach only when all initial (cid:96) sites are empty. Therefore a summation up to (cid:96) is taken in the gain part of Eq. 5. At the special site i = L1 + 1 one gets dP (i, t) dt = Wsq(i + (cid:96) − 1i − 3)P (i − 3, t) − [Wsq(i + (cid:96) + 2i) + Wf sp(i − 1i)]P (i, t) − WtP (i + (cid:96), t)P (i, t), and at site i = L1 dP (i, t) dt = Wsq(i + (cid:96) − 1i − 3)P (i − 3, t) + Wf sp(ii + 1)P (i + 1, t) − Wsq(i + (cid:96) + 2i)P (i, t). (6) (7) At the site of termination of translation dP (i, t) dt = W P (i − 3, t) − P (i, t)β, (8) where i = L for non-frameshifted ribosomes where i = L−1 in case of frameshifted ribosome. Note that when the rod is located at i = L − 4 or at i = L − 3 it does not face any exclusion at its target site L − 1 or L, respectively; consequently, there is no factor of q in the gain term on the right hand side of (8). For the same reason, for all the sites i ≥ L − (cid:96) + 1, exclusion effect appears neither in the gain terms nor in the loss terms in the master equations dP (i, t) dt = W P (i − 3, t) − W P (i, t). (9) At all other sites in segment I and III dP (i, t) dt = W q(i + (cid:96) − 1i − 3)P (i − 3, t) − W q(i + (cid:96) + 2i)P (i, t). In segment II dP (i, t) dt = Wsq(i + (cid:96) − 1i − 3)P (i − 3, t) − Wsq(i + (cid:96) + 2i)P (i, t). (10) (11) p-5 Bhavya Mishra1 Gunter M. Schutz2 Debashish Chowdhury1 high densities are neglected under MFA. From Eq. (4), Jf s depends on two factors, (a) the pa- rameter Wf s which, in turn, reflects the complexity of the pseudoknot and (b) the variable p(L1L1 + 1) which indi- cates the availability of an empty target site for frameshift. As the psudoknot strength ∆ G increases the parameter Wf s also increases (equivalently, γ decreases), thereby caus- ing an increase of Jf s. However, the increase of ∆ G also suppresses Ws, leading to the increasingly dense queue of ribosomes just behind the special site, i.e., a concomitant decrease of p(L1L1 + 1). Therefore, with further increase of ∆ G (i.e., decrease of γ) the flux Jf s attains its maximum beyond which it decreases sharply because the target site becomes inaccessible at such high density of ribosomes. In Fig. (4), we show the occupational density profile for j = 1, 4, 7, 10, . . . ) for (in steps of 3 nucleotides i.e. Wf s = 13.3s−1 and Ws = 0.00074s−1. For this value of Ws (cid:28) W the entire segment II behaves as a static bottleneck which creates a high density region in segment I and a low density region in segment III. In the inset we show the occupational density profile (in steps of 1 nucleotide, i.e., for j = 1, 2, 3, . . . ) for a small region from i = 390 to i = 420. Since a rod can jump only in steps of 3 nucleotides, it can occupy only those sites which are of the form 3k + 1, where k = 1, 2, 3, . . . . Everywhere else occupational density will always remain zero behind the special point. After passing through the special site, a rod can occupy site 3k + 1 (non-frameshifted rod) as well as site 3k (frameshifted rod) Thus, the structure observed in the occupational density profile is merely an artefact of the hopping of each rod, normally, by three sites at a time. In Fig. (5), we show the variation in flux J with coverage density ρc for three different values of γ. The flat top of the flux is a well-known feature of the TASEP with a blockage [31, 32]. Another interesting quantity that characterizes PRF is the fraction φ of the proteins synthesized that are actually fusion of two proteins "conjoined at birth" because of PRF. Let nf s be the total number of ribosomes that undergo frameshift from the special point and detach from the site L − 1 (i.e. the designated site of termination of transla- tion by frameshifted ribosomes) after completing synthesis of a fusion protein without suffering premature detach- ment. Similarly, nnf s is the total number of ribosomes that complete synthesis of a full length single protein with- out undergoing frameshift and detach from the designated termination site L. In terms of nf s and nnf s, the fraction φ is defined as (12) φ = nf s nf s + nnf s Thus, φ is also a measure of the efficiency of programmed -1 frameshift. The fraction φ is plotted against the ratio Ws/W in Fig.6. The deviation of the mean-field prediction from the corresponding computer simulation data increases with decreasing Ws/W . This deviation is caused by the increasing crowding of ribosomes where correlations, which Fig. 6: (Color online) The fraction of proteins that are 'fusion' of two proteins arising from PRF are plotted as a function of the jump rate ratio γ = Ws/W . System size is N = 1000+ (cid:96)− 1, with the rod size (cid:96) = 18, and W = 83 s−1 . Results. -- In this work we have followed two different approaches for analyzing the model to obtain quantitative results. All the effective rate constants W , Ws and Wf s chosen for calculation were obtained using the relation between the effective rate constant and basic rate constants of the 5-state original model whose numerical values are given in Table. 1 of the supplementary information given in [29]. In our first approach we have obtained the steady state occupational density profile by solving the Eqs. (5) - (11) iteratively, under mean field approximation (MFA), by the standard Runge-Kutta integration scheme until the steady state was attained. In our second approach we carried out Monte-Carlo simulations (MCS) of the same mechanisms as in our theoretical model. In both the approaches a very high initiation rate α ≈ 100s−1 and very low termination rate β ≤ 10s−1 have been chosen to ensure a high density of the rods on the track. Realistically, typical values of (cid:96) would be about 30 nu- cleotides. In order to study our model with (cid:96) = 30, we would need a proportionately large values of L1, L2 and L. A comparison of our preliminary test results obtained for those realistic sizes with those for the shorter values reported in this manuscript showed no qualitative differ- ence.Therefore, with the shorter values of the sizes, namely, L1 = 399, L2 = 450, L = 1000, we got lot more data for averaging which are reported in this paper. In Fig. (3), we show the variation in frameshift flux Jf s with jump rate ratio γ = Ws/W . The theoretical results predicted under MFA agree well with the corresponding data from MC simulation in the regime where the cover- age density of the ribosomes is sufficiently low. However, with increasing coverage density increasing deviation of the MFA from the MC data is observed. This is not un- expected because the correlations that are significant at p-6 are neglected in MFA, are quite significant. Supplementary Information. -- Title The fraction φ expectedly vanishes in the limit Ws = W . As Ws/W decreases frameshits become more likely which is reflected in the increase of φ. However, as the ratio Ws/W decreases further the number of frameshift events nf s be- gins to decrease because of the increasing unavilability of an empty target site that is needed for -1 frameshift. But, concomitantly nnf s also decreases because of the inability to unwind a stiffer pseudoknot. Consequently, φ saturates as γ = Ws/W → 0. Conclusion. -- Programmed ribosomal frameshift is one of the most prominent modes of recoding of genetic information. In this paper we have demonstrated that the density of ribosomes at and around the slippery sequence is an important parameter that determines the frequency of programmed ribosomal -1 frameshift. In Figs. (3) and (6) we have demonstrated the effects by varying the stiffness of the pseudoknot that, in turn, controls the ribosome density. The suppression of -1 programmed frameshift by a trail- ing ribosome in a dense ribosome traffic on a mRNA track is similar to the suppression of diffusive backtracking of a RNA polymerase (RNAP) motor by another trailing very closely on a DNA track [33]. Unlike the slippery se- quence and the pseudoknot on the mRNA template, this hitherto neglected parameter may control the frequency of -1 frameshift dynamically because the density of the ribosomes on the mRNA can be up- or down-regulated by several different signals and pathways. For laboratory experiments, the stiffness of a pseudoknot can be varied artificially [14] . Using such synthetic mRNA strands our theoretical prediction can be tested experi- mentally by a combination of ribosome profiling technique [34, 35] (for measuring the ribosome density) and FRET (for the frequency of frameshift) [36]. From the perspective of TASEP-based modeling of ribo- some traffic on mRNA template, our work is a significant progress. In all the earlier models of this type each site on the lattice (chain) corresponds to a single codon, i.e., a triplet of nucleotides. In contrast, in the models developed here each site represents a single nucleotide; this modifi- cation was necessary to capture -1 frameshift where the ribosome steps backward on its mRNA track by a single nucleotide. We intend to use this prescription for modeling the mRNA track in our future model of programmed +1 frameshift. The pseudoknot segment of the mRNA track is, effectively, an extended "blockage" against the forward movement of the ribosomes. Not surprisingly, the density profile (see Fig.4) and the net flux (see Fig.5) of the ri- bosomes display the well known characteristics of TASEP with static blockage [31, 32]. Acknowledgments. -- This work has been supported by J.C. Bose National Fellowship (DC), "Prof. S. Sampath Chair" Professorship (DC) and by UGC Junior Research Fellowship (BM). 5-state model of elongation cycle. A ribosome consists of two interconnected subunits called large subunit (LSU) and small subunit (SSU). A tRNA molecule can transit through the intersubunit space. The three binding sites for the tRNA, arranged sequentially from the entry to the exit along its path, are denoted by A, P and E, respectively. One end of the tRNA molecule that interacts with the SSU constitutes the anticodon that binds with the codon on the mRNA track by complementary base-pairing. The other end of the tRNA, that interacts with the LSU, is aminoacylated, i.e., charged with the corresponding amino acid monomer. The elongation cycle can be broadly divided into a se- quence of three major processes: aatRNA selection, peptide bond formation, and translocation. However, aatRNA se- lection itself proceeds in two steps: (i) Preliminary selection based on the differences in the codon-anticodon binding free energies for cognate and non-cognate aatRNA, (ii) kinetic discrimination between cognate and near-cognate aatRNA that is usually referred to as kinetic proofreading. Once an aatRNA passes through these two-step "molec- ular identification" screening, it is accommodated at the A-site and the system awaits the onset of the next major process, namely, peptide bond formation. The third major process, namely translocation, also consists of multiple sub-steps during which the two ribosomal subunits execute well orchestrated movements on the template mRNA while concomitant movement of the tRNAs take place on the triplet binding sites E,P,A on the ribosome. An elongation cycle is captured in our model by a multi- step kinetic process involving N distinct mechano-chemical states. The number N = 5 and the allowed transitions among them, as shown schematically in Fig. (7), capture all the major steps in the elongation cycle that have been established so far by structural and kinetic measurements [1]. The rates of the allowed transitions indicated by the arrows on this diagram are also shown symbolically next to the corresponding arrows. In principle, frameshift can Fig. 7: Schematic diagram for conversion of a 5-state model into a single state model p-7 n-1nn+1n+2n+3n+412345n-1nn+1n+2n+3n+411112345111WbfWbrWpWh1Wh2WrWfWfWrWh1WpWbfWbrWh2WWBhavya Mishra1 Gunter M. Schutz2 Debashish Chowdhury1 take place from any of the five states of this kinetic model. Therefore, corresponding to the 5-state model of elongation cycle, one can envisage at least five different kinetic models of -1 frameshift; the only difference between these different models is the step in which frameshift is assumed to take place. The details of all these five different models are given elsewhere [37]. Although the slippery sequence and the downstream pseudoknot have been implicated in the -1 frameshift the exact step of the longation cycle where it occurs and the structural dynamics causing this frameshift remain con- troversial. Therefore, in this reduced minimal model the five distinct mechano-chemical states are collapsed onto a single effective state labelled by the instantaneous position of the ribosome. Rate Exp. Num.Value Wf Wr Wh1 Wp Wbf Wbr Wh2 100s−1 25s−1 80s−1 80s−1 25s−1 25s−1 60s−1 Source Pape T. et al. [39] Pape T. et al. [39] Rodnina M.V. et al. [40] Pape T. et al. [39] Sharma A.K. et al. [20] Sharma A.K. et al. [20] Wen J.D. et al. [41] Table 1: Parameters in the 5-state model of elongation cycle and their numerical values Reducing the 5-state model to a 1-state model. By using effective rate constant method we have converted the five state model into an equivalent one state generic model [38] Fig. (7). The effective rate constants are, W4 = Wbf W2 = Wf Wh2 Wh2 + Wbr Wh1 Wh1 + Wr and [W1→1∗ ]−1 = [W ]−1 = 1 W2 + 1 Wh1 + 1 Wp + 1 W4 + 1 Wh2 . (13) (14) (15) Calculation of the conditional probabilities p and q. Consider N rods, each of size (cid:96), distributed over a periodic lattice of L sites. We first calculate the conditional probability p(i − 1 i) that the site i − 1 is not covered, given that a ribosome is located at the site i ((i.e., covers the sites i, i + 1, ..., i + (cid:96) − 1). In order to do so we follow [42, 43] and map the process to the so-called zero-range process (ZRP) [44]. In this mapping one labels the N rods consecutively by integers α = 1, 2, . . . , N and the size of the gap between two consecutive rods, i.e. the number of empty sites between them, becomes a particle occupation number nα on site α of the ZRP. The total number of particles in the ZRP Fig. 8: Schematic diagram for Model B. is thus NZRP = L − N (cid:96), corresponding to the number of vacant sites in the TASEP, and the number of sites in the ZRP is given by LZRP = N . This means that the particle number density ρ = N/L of the TASEP is related to the particle number density c = NZRP /LZRP of the ZRP by c = 1/ρ− (cid:96). It is convenient to work in the grand canonical ensemble where the probability of finding n ZRP-particles on (any) given site is Pn(z) = zn(1 − z) and z = c/(1 + c) [43]. Under this mapping the conditional probability p(i−1 i) becomes the probability of a finding at least on particle (gap size > 0) on a site in the zero-range process, i.e., p(i − 1 i) =(cid:80)∞ n=1 Pn(z). Thus p(i − 1 i) = z = 1 − (cid:96)ρ 1 − ((cid:96) − 1)ρ . (16) Next we calculate the conditional probability q(i + (cid:96) + 2 i) which is the conditional probability that the site i + (cid:96) + 2 is not covered, given that the site i is already occupied (i.e., the sites i, i+1, i+2, ..., i+(cid:96)−1 are covered). In the language of the ZRP this is the probability of finding any occupation Fig. 9: Frameshift flux Jf s for model B. In inset, coverage probability at target site i = L1 for frameshift. Dashed line is for theoretical values under MFA and dashed line with (cid:125) is for MC simulation with initiation rate α = 100s−1, and termination rate β = 10s−1. System size N = L + (cid:96) − 1 = 1000 + (cid:96) − 1 and rod size (cid:96) = 18. p-8 Segment I Segment II Segment III αβWWsWtWfsWsXXXYYYZ Slippery sequenceSecondary structure 1 L1-1 L1 L2 L2+1 L+l-1 number larger than 2, i.e., q(i + (cid:96) + 2 i) =(cid:80)∞ (cid:19)3 (cid:18) 1 − (cid:96)ρ Thus q(i + (cid:96) + 2 i) = z3 = n=3 Pn(z). 1 − ((cid:96) − 1)ρ . (17) Models with interactions between rods that are more de- tailed than a hard-core repulsion can be treated with the methods of [45]. Model B. The model B differs from model A only in the rule for detachment of a ribosome. More specifically, a ribosome can detach from the special site, with rate Wt, if and only if the target site under possible -1 frameshift is already covered by another ribosome (see Fig. (8)). Fig. 10: Occupational density profile in steps of 3nts for model B. The dashed line with ∗ is the theoretical values under MFA and the solid line with (cid:12) is for MC simulation with initiation rate α = 100s−1, termination rate β = 10s−1, Wf s = 13.3s−1 and Ws = 0.00074s−1. System size L + (cid:96) − 1 = 1000 + (cid:96) − 1 and rod size (cid:96) = 18. The inset shows the occupational density profile of a small region from i = 390 to i = 420 in steps of 1 nt. The dashed line with ∗ is for theoretical MF calculation and the solid line with (cid:12) is for MC simulation. The master equations for model B are similar to those for Model A, except at the special site (i = L1 + 1); at this special site the master equation is dP (i, t) dt = Wsq(i + (cid:96) − 1i − 3)P (i − 3, t) − [Wsq(i + (cid:96) + 2i) + Wf sp(i − 1i)]P (i, t) − WtP (i − (cid:96), t)P (i, t). (18) All the conditional probabilities for model B are identical to those for Model A. For this model the frameshift flux Jf s is given by Title In Fig. (10) we show the coverage density profile. We observe a discontinuous jump between the high- and low- density phases at the bottleneck. Both the results are very similar to the corresponding results reported for model A in the main text of this letter. REFERENCES [1] Frank J. (Editor), Molecular machines in Biology: Work- shop of the Cell (Cambridge University Press, New York) 2010. [2] Chowdhury D., Phys. Rep., 529 (2013) 1. [3] Chowdhury D., Biophys. J., 104 (2013) 2331. [4] Kolomeisky A. B., J. Phys. Condens. Matter, 25 (2013) 463101. [5] Farabaugh P. J., Programmed Alternative Reading of the Genetic Code. (Springer) 1997. [6] Atkins J. F. and Gesteland R. F. (Editors), Recoding: Expansion of Decoding Rules Enriches Gene Expression. (Springer) 2010. [7] Firth A. E. and Brierley I., J. Gen. Virol., 93 (2012) 1385. [8] Brierley I. et al., Nat. Rev. Microbiol., 5 (2007) 598. [9] Giedroc D. P. and Cornish P. V., Virus Res, 139 (2009) 193. [10] Kim H. K. et al., Proc. Natl. Acad. Sci. USA, 111 (2014) 5538. [11] Brierley I. et al., Biochem. Soc. Trans., 36 (2008) 684. [12] Tinoco I. jr et al., Biopolymers, 99 (2013) 1147. [13] Lopinski J. D.et al., Mol. Cell Biol., 20 (2000) 1095. [14] Tholstrup J. et al., Nucleic Acids Res., 40 (2012) 303. [15] Schutz G. M., Phase Transitions and Critical Phenomena, edited by Domb C. and Lebowitz J. L., Vol. 19 (Academic Press) 2001. [16] Chowdhury D. et al., Phys. Rep., 329 (2000) 199. [17] Schadschneider A. et al., Stochastic Transport in Complex Systems: From molecules to vehicles (Elsevier) 2011. [18] Chowdhury D. et al., Phys. of Life Rev., 2 (2005) 318. [19] Sharma A. K. and Chowdhury D., Phys. Biol., 8 (2011) 026005. [20] Sharma A. K. and Chowdhury D., J. Theor. Biol., 289 (2011) 36. [21] Chou T. et al., Rep. Prog. Phys., 74 (2011) 116601. [22] Zia R.K.P. et al., J. Stat. Phys., 144 (2011) 405. [23] von der Haar T., Comput. Struct. Biotechnol. J., 1 (2012) e201204002. [24] Appert-Rolland C. et al., Phys. Rep, 593 (2015) 1. [25] Chou T. and Lakatos G., Phys. Rev. Lett., 93 (2004) 198101. [26] Lakatos G. and Chou T., J. Phys. A, 36 (2003) 2027. [27] Mitarai N. and Pedersen S., Phys. Biol., 10 (2013) 056011. [28] Turci F. et al., Phys. Rev. E, 87 (2013) 012705. [29] Mishra B., and Chowdhury D., Schutz, G.M. Jf s = Wf sp(L1L1 + 1)P (L1 + 1, t). (19) arXiv:1605.03434 (2016). Results for model B. The results obtained for Model B are also similar to the corresponding results for Model A. In Fig. (9) we show the variation of frameshift flux Jf s with the jump rate ratio γ. In the inset we show the variation of the coverage density at the target site i = L1 with the variation of the ratio γ. [30] Chen J. et al., Nature, 512 (2014) 328. [31] Janowsky S.A. and Lebowitz J.L., Phys. Rev. A, 45 (1992) 618. [32] Schutz G. M., J. Stat. Phys., 71 (1993) 471. [33] Sahoo M. and Klumpp S., EPL, 96 (2011) 60004. [34] Ingolia N. T., Nat. Rev. Genet., 15 (2014) 205. [35] Ingolia N. T., Cell, 165 (2016) 22. p-9 Bhavya Mishra1 Gunter M. Schutz2 Debashish Chowdhury1 [36] Tinoco I. jr and Gonzalez R. L. jr, Genes Dev., 25 (2011) 1205. [37] Mishra B., Ph.D. Thesis, IIT, Kanpur, 2016 (in prepara- tion) . [38] Cleland W. W., Biochemistry, 14 (1975) 3220. [39] Pape T. et al., EMBO J, 17 (1998) 7490. [40] Rodnina M. V. et al., EMBO J, 14 (1995) 2613. [41] Wen J. D. et al., Nature, 452 (2008) 598. [42] Shaw L.B. et al., Phys. Rev. E, 68 (2003) 021910. [43] Schonherr G. and Schutz, G.M., J. Phys. A: Math. Gen., 37 (2004) 8215. [44] Evans M.R. and Hanney T., J. Phys. A:Math. Gen., 38 (2005) R195. [45] Buschle J., Maass P. and Dieterich W., J. Stat. Phys., 99 (2000) 273. p-10
1606.01272	2	1606	2016-06-07T19:22:20	Bypassing damaged nervous tissue	[ "physics.bio-ph", "q-bio.NC" ]	It is shown the principal possibility of bypassing damaged demyelinated portions of the nervous tissue, thereby restoring its normal function for the passage of action potentials.	physics.bio-ph	physics	Bypassing damaged nervous tissue Department of Mechanical and Aerospace Engineering, Princeton University, Princeton, NJ, 08544 e-mail: [email protected] M.N. Shneider It is shown the principal possibility of bypassing damaged demyelinated portions of the nervous tissue, thereby restoring its normal function for the passage of action potentials. Violations of the nerve fiber integrity, such as partial demyelination and micro-ruptures, leads to malfunction of the nervous system and its components [1-5]. For example, a damage or loss of the myelin coating of neurons, which can be caused by various diseases slows down or even blocks action potentials. This results in a variety of disorders, such as sensory impairment, multiple sclerosis, blurred vision, movement control difficulties, as well as problems with bodily functions and reactions [1-6]. In this letter, it is shown that if a partially demyelinated neuron cell remains alive and functional, appropriate stimulation of the axon away from the damaged area could lead to the normal passage of the action potential through bypassing of the demyelinated area. Such stimulation can be a local change in membrane potential induced by the current in saline, similar to that in neuron synchronization [7]. This can be done, for example, by changing the local probe potential, inducing a current in the saline, resulting in charging of the membrane at a node of Ranvier until the potential difference on the membrane becomes sufficient to initiate an action potential outside of the demyelinated area. Then it becomes possible for further propagation of the action potential along the undamaged segments of the axon. As a result, activity of the nerve will be fully or partly restored. Fig. 1. Schematic of a neuron with the characteristic elements of the myelinated axon. The demyelinated portions with damaged segments between n and n+k nodes of Ranvier are also shown. The arrows indicate the bypass shunting of the demyelinated section. To examine the action potentials on myelinated nerve fibers, we employ the Goldman–Albus model of a toad neuron [8]. In this model, which has been used by many authors, and reproduced also in [9], a myelinated nerve fiber is represented as a leaky transmission line with uniform internode sections with a length LM and an outer diameter dM (myelin sheath inclusive) separated by nodes of Ranvier with a length LR and outer diameter dR, which is equal to the diameter of an axon inside the myelin coating. The voltage U(x,t) across the membrane in the internode region is found as a solution to equation [8], which is the equation of potential diffusion: U  t   Ua 2  z  2  bU , (1)  d R m k / d M R i d ( R )2/  2 ; a  , R 1  CR m 1 b  /1 C 1  1 , kk 2 Where . Here /1 CR 1 1 M d / R k ln/1  d ln2 are constants and R capacitance and resistance per unit length; is the myelin resistance times unit length; iR is the axoplasm specific resistance. All parameters and constants are the same as in [8]. The diameters of the myelinated sections of the axon and nodes of Ranvier are The action potential in the nodes of Ranvier 70RU mV, is defined by the Frankenhaeuser–Huxley equation [9,7]. , with the corresponding resting potential , are the myelin μm, respectively. 15Md 9Rd U AP )(t and Fig. 2a shows an example of the action potential propagation along an undamaged myelinated axon, calculated in the framework of approximations and data used in [8]. At the model calculations of the damaged axon we assumed for definiteness that demyelination occurs between 12 and 22 nodes of Ranvier. In our calculations we mimic neuron demyelination by assuming a myelin sheath of a smaller thickness. We assume, as an example, that the diameter of myelinated section in demyelinated areas μm. That is, in demyelinated areas the capacitance per unit length increases sharply, and the resistance - decreases. Fig. 2b shows the computed results of the action potential in the damaged demyelinated axon. Under these assumed conditions, demyelination causes complete blockage of the action potential. d 025.1 225.9   d M R The conditions for action potential propagation can be restored by shunting of the demyelinated section (nodes 11 and 23 for assumed parameters of our model) (Fig. 2c). In this case, the action potential "skips" the demyelinated area when the maximum potential difference across the membrane reaches ~ -32 mV at the 23rd node. That is, the process develops on the 23rd node of Ranvier as if a pulse with amplitude crU 38 mV and duration of a few milliseconds is applied. This corresponds to approximately 35% of the maximum potential difference across the 11th node of Ranvier membrane. In this case, the action potential propagates down the axon, skipping the demyelinated site. t=0.5 ms 1 1.5 2 2.5 3 3.5 4 ) V m ( m V  0 10 20 30 Nodes 40 50 60 40 20 0 -20 -40 -60 -80 t=0.5 ms 1 demeyelinited between 12 and 22 nodes 1.5 2 2.5 0 10 20 40 50 30 Nodes 60 40 20 0 -20 -40 -60 -80 ) V m ( U 40 20 0 -20 -40 -60 -80 t=0.5 ms 1 1.5 2 2.5 3 3.5 4 1.5 2 0 10 20 40 50 30 Nodes 60 ) V m ( U Fig. 2. Saltatory propagation of action potential in normal myelin fibers (a); Blocking the action potential in the demyelinated area (b); Bypass of the demyelinated area (between 12 and 22 nodes of Ranvier) (c). Thus, we have shown that it is possible to bypass damaged demyelinated portions of nervous tissue, thus restoring its normal function. All this is true not only for a single axon, but also for a neuron ensemble, as in the spinal cord, for example. The results presented in this paper, require experimental verification to evaluate the possibility of its implementation in biomedical practice. To observe the propagation, complete blockage or bypass of the action potential may be possible by detecting the second harmonic generation on a sample of nerve tissue, as suggested in [8,11]. These results demonstrate that the second- harmonic response can serve as a local probe for the state of the myelin sheath, providing a high- contrast detection of neuron demyelination. It should be noted that it is also possible to block the action potential in an undamaged nerve fiber, leading to reversible anesthesia, while maintaining, at the nodes of Ranvier, a negative voltage below the threshold in some intermediate sections of the axon. References 1. W. McDonald, The effects of experimental demyelination on conduction in peripheral nerve: a histological and electrophysiological study. II. Electrophysiological observations. Brain, 86, 501-524 (1963) 2. S.G. Waxman Conduction in myelinated, unmyelinated and demyelinated fibers. Arch Neurol., 34, 585-590 (1977) 3. M. Rasminsky, T.A. Sears, Internodal conduction in undissected demyelinated nerve fibers. J Physiol (Lond) 227, 323-350 (1972) 4. S.G. Waxman, J.D. Kocsis, J.A. Black, Pathophysiology of demyelinated axons. In: S.G. Waxman, J.D. Kocsis, P.K. Stys (eds) The axon. (Oxford University Press, New York, 1995) 438-461 5. D. I. Stephanova, M. Chobanova, Action potentials and ionic currents through paranodally demyelinated human motor nerve fibers: computer simulations, Biol. Cybern. 76, 311-314 (1997) 6. S. Love, Demyelinating diseases, J. Clin. Pathol. 59, 1151–1159 (2006) 7. M.N. Shneider, M. Pekker, Correlation of action potentials in adjacent neurons, Phys. Biol. 12, 066009 (2015) 8. L. Goldman, J.S. Albus, Computation of impulse conduction in myelinated fibers; theoretical basis of the velocity-diameter relation, Biophys. J. 8, 596 (1968) 9. M.N. Shneider, A.A. Voronin, A.M. Zheltikov, Modeling the action-potential-sensitive nonlinear-optical response of myelinated nerve fibers and short-term memory, J. Appl. Phys. 110, 094702 (2011) 10. B. Frankenhaeuser, A.F. Huxley, The action potential in the myelinated nerve fibre of Xenopus laevis as computed on the basis of voltage clamp data, J. Physiol. (London) 171, 302 (1964) 11. M.N. Shneider, A.A. Voronin, and A.M. Zheltikov, Action-potential-encoded second- harmonic generation as an ultrafast local probe for nonintrusive membrane diagnostics, Phys. Rev. E 81, 031926 (2010)
1809.09551	1	1809	2018-09-20T05:37:18	Prebiotic Fatty Acid Vesicles through Photochemical Dissipative Structuring	[ "physics.bio-ph" ]	We describe the photochemical dissipative structuring of fatty acids from CO and CO2 saturated water under the solar UVC and UVA photon potential prevalent at Earth's surface during the Archean. Their association into vesicles and their subsequent association with other fundamental molecules of life such as RNA, DNA and carotenoids to form the first protocells is also suggested to occur through photochemical dissipative structuring. In particular, it is postulated that the first vesicles were formed from conjugated linolenic (C18:3n-3) and parinaric (C18:4n-3) acids which would form vesicles stable at the high temperatures (~85 {\deg}C) and the somewhat acidic pH values (6.0-6.5) of the Archean ocean surface, resistant to divalent cation salt flocculation, permeable to ions and small charged molecules, but impermeable to short DNA and RNA, and, most importantly, highly dissipative in the prevailing UVC+UVA regions.	physics.bio-ph	physics	Prebiotic Fatty Acid Vesicles through Photochemical Dissipative Structuring Karo Michaelian1 and Oscar Rodríguez2 1Department of Nuclear Physics and Applications of Radiation, Instituto de Física, UNAM. Cto. Interior de la Investigación Científica, Ciudad Universitaria, Cuidad de México, C.P. 04510, [email protected], Tel: 01-55-525-6225165 2Posgrado en Ciencias Físicas, UNAM. Cto. Interior de la Investigación Científica, Ciudad Universitaria, Cuidad de México, C.P. 04510 [email protected] Abstract We describe the photochemical dissipative structuring of fatty acids from CO and CO2 saturated water under the solar UVC and UVA photon potential prevalent at Earth's surface during the Archean. Their association into vesicles and their subsequent association with other fundamental molecules of life such as RNA, DNA and carotenoids to form the first protocells is also suggested to occur through photochemical dissipative structuring. In particular, it is postulated that the first vesicles were formed from conjugated linolenic (C18:3n-3) and parinaric (C18:4n- 3) acids which would form vesicles stable at the high temperatures (~85 °C) and the somewhat acidic pH values (6.0-6.5) of the Archean ocean surface, resistant to divalent cation salt flocculation, permeable to ions and small charged molecules, but impermeable to short DNA and RNA, and, most importantly, highly dissipative in the prevailing UVC+UVA regions. Resumen Describimos la estructuración disipativa fotoquímica de los ácidos grasos a partir de agua saturada de CO y CO2 bajo el potencial de los fotones solares en el UVC y UVA prevalentes en la superficie de la Tierra durante el Arcaico. También se sugiere que su asociación en vesículas, y su posterior asociación con otras moléculas fundamentales de la vida, tales como ARN, ADN y carotenoides para formar las primeras protocélulas, se produjeron a través de la estructuración disipativa fotoquímica. En particular, se postula que las primeras vesículas se formaron a partir de ácidos linolenicos conjugados (C18: 3n-3) y parináricos (C18: 4n-3) que darían lugar a vesículas estables a altas temperaturas (~ 85 ° C) y valores algo ácidos de pH (6.0-6.5) en la superficie del océano Archeano, resistentes a la floculación de sal de catión divalente, permeables a iones y pequeñas moléculas cargadas, pero impermeables a ADN y ARN corto, y, lo que es más importante, altamente disipativas en los regiones UVC y UVA predominantes. Keywords: Fatty acids, vesicles, dissipative structuring, origin of life, protocells; Ácidos grasos, vesículas, estructuración disipativa, origen de la vida, protocélulas Introduction Theories concerning the origin and evolution of life must necessarily provide a chemical- physical reason for complexation at all biological levels. The traditional ``survival of the fittest'' paradigm at the macroscopic level or the ``chemical stability'' paradigm at the microscopic molecular level are both deficient when it comes to explaining complexation. A more prudent approach would be to consider only what is well established concerning complexation in chemical-physical systems; that the ordering of material out-of-equilibrium occurs exclusively as a response to the dissipation of an externally imposed generalized thermodynamic potential, or in other words, complexation is concomitant with an increase in global entropy production of the system interacting with its environment. Such ordering of material was given the name ``dissipative structuring'' by Ilya Prigogine [1]. The thermodynamic dissipation theory for the origin and evolution of life [2-5] suggests that every incremental complexation in the history of biological evolution; from the formation of complex prebiotic molecules from simpler precursor molecules, their associations into cellular and multi-cellular organisms, to the hierarchical coupling of biotic with abiotic processes of today's biosphere, must necessarily have coincided with an incremental increase in the dissipation of some externally imposed generalized thermodynamic potential [1,6,7]. This theory identified [2] the important thermodynamic potential driving complexation at life's origin as the long-wavelength UVC solar photon potential arriving at Earth's surface during the Archean (see Fig. 1) since these wavelengths have sufficient free energy to break and reform covalent bonds of carbon-based molecules, but not enough energy to disassociate these molecules [2-4]. The UVA region would also have been important in promoting charge transfer reactions useful in dissipative structuring. The UVC wavelengths reached Earth's surface from before the origin of life and lasting for at least 1,000 million years until the end of the Archean [8,9] (Fig. 1), at which time oxygenic photosynthesis overwhelmed natural oxygen sinks, allowing the formation of an atmospheric ozone layer. Figure 1. The wavelengths of peaks in the absorption spectra of many of the fundamental molecules of life (common to all three domains) coincide with an atmospheric window (205-285 nm) predicted in the long wavelength end of the UVC region (100-280 nm) at the time of the origin of life approximately 3.85 Ga and until at least 2.9 Ga (curves black and red respectively). Little light was available in the UVB region (280-315 nm) but much in the UVA region (315-400 nm). By around 2.2 Ga (yellow curve), UVC light at Earth's surface had been extinguished by oxygen and ozone resulting from organisms performing oxygenic photosynthesis. The green curve corresponds to the present surface spectrum. Energy flux values correspond to the sun at the zenith. The font size of the letter indicates approximately the relative size of the molar extinction coefficient of the indicated fundamental molecule (UVC pigment). Image credit; adapted with permission from Michaelian and Simeonov [9]. Elsewhere [5], we have analyzed the autocatalytic photochemical production and proliferation of adenine and single strand DNA at the ocean surface under the UVC+UVA Archean photon potential. In this paper, after describing the general chemical and physical characteristics of fatty acids, we analyze the autocatalytic photochemical production and proliferation of fatty acids and their vesicles at the ocean surface under this same photon potential and give non-equilibrium thermodynamic justification for the association of nucleic acids and other fundamental molecules through encapsulation within these first vesicles. Characteristics of Fatty Acids Natural fatty acids contain a carbon plus hydroxy plus oxygen (carboxyl) head group and a hydrocarbon tail of from 4 to 40 carbon atoms [10] which may be saturated with hydrogen or partially unsaturated (Fig. 2). Fatty acids, fatty alcohols and fatty acid glycerol esters are generally considered to be more relevant to vesicles of early life than the phospholipids composing the cellular wall of most present day organisms because phospholipid biosynthesis is very different between bacteria and archaea, suggesting that their common ancestor was devoid of phospholipid membranes (although this view has been challenged [11,12]). Also, fatty acids are simpler single chain molecules that are more easily produced through abiotic heat activated Fischer-Tropsch or photochemical polymerization of smaller chain hydrocarbons such as ethylene (C2H4) (Fig. 3). Ethylene itself, and other small chain hydrocarbons, can be derived from the reduction of CO2 or CO employing water as the electron donor and either chemically or photochemically catalyzed. Ethylene can also be produced by the UV photolysis of methane which appears to be an important mode of hydrocarbon production in the atmosphere of Titan [13]. Fatty acids are conserved as components of the cell walls in organisms from all three domains of life. On today's ocean surface, and in the Archean fossil record, there is an abundance of even number carbon fatty acids over odd number and this could be explained if indeed such fatty acids are formed by polymerization (telomerization) of ethylene. In particular, there is a predominance of 16 and 18 carbon atom fatty acids in the whole available Precambrian fossil record [14,15] and, in fact, in today's organisms, and in the aerosols obtained above today's ocean surface [16,17]. In general, the longer the hydrocarbon tail and the greater the degree of saturation, the greater the melting temperature; the degree of saturation usually having the greater influence because double carbon bonds introduce a curvature into the hydrocarbon tail preventing them from bonding strongly with neighbors. For example, lipid desaturation by the enzyme desaturase is a key factor in the protection of the photosynthetic apparatus from low-temperature and high-light stresses, as well as being an effective tool for manipulating membrane microviscosity [18]. For example, saturated myristic acid (C14:0) has a melting temperature of 55 °C [19] while linoleic acid with two unsaturated bonds (18:2) has a melting temperature of only -5°C. However, if the unsaturation occurs as conjugated double bonds in the trans configuration near the tail end of the hydrocarbon, such as in parinaric acid (C18:4n-3, see figure 2, then unsaturation is not greatly destabilizing. Parinaric acid has a high melting temperature of 85 °C. Figure 2. Parinaric acid is a conjugated fatty acid of 18 carbon atoms having a melting temperature of (85-86 °C), roughly the temperature of the Archean ocean surface at the origin of life. The 4 conjugated carbon double bonds lead to delocalized electrons, giving it strong absorption over UVB and UVA region of 310 to 340 nm (Fig. 1). Conjugated hydrocarbons have conical intersections allowing for rapid dissipation to heat of the photon-induced electronic excited singlet state energy, making them excellent candidates for Archean dissipative structures. UVC-induced cross linking with other fatty acids at the site of a double bond would convert this conjugated parinaric acid into conjugated linolenic acid (C18:3n-3) with absorption in the UVC at 269 nm (see Fig. 1). Figure 3. Ethylene. A common precursor molecule for the formation of the hydrocarbon tails of the fatty acids through successive polymerization (telomerization). Ethylene can be derived from the reduction of CO2 or CO employing water as the electron donor, either chemically or photochemically catalyzed. Ethylene can also be produced by the UV photolysis of methane which appears to be an important mode of hydrocarbon production in the atmosphere of Titan. Direct photon-induced deprotonation under UVC light (254 nm) of fatty acids can lead to conjugation of the carbon bonds of the acyl tail [20]. The steps involved in the conjugation process are i) photon-induced deprotonation giving rise to carbon-carbon double bonds, ii) double bond migration to give a conjugated diene, triene, or tetraene. The greater the degree of unsaturation of the hydrocarbon tail, the greater the probability of forming conjugated bonds under UVC light [20]. The same UVC light, however, can also cause cross linking between adjacent acyl tails at the site of a double bond which would reduce the average conjugation number [21]. Under the constant UVC flux existing at the origin of life, a stationary state of particular conjugation numbers would therefore have arisen. Under the UV light of the Archean, and according to the postulates of the thermodynamic dissipative structuring [5], the conjugation numbers of fatty acids in this stationary state would be either 2, 3, or 4 since these give rise to peak absorption at 233, 269, and 310-340 nm respectively, which all lie within the Archean surface solar spectrum (Fig. 1). In the following, we will concentrate on vesicles formed from mainly conjugated fatty acids of 18 carbons for the following reasons; 1) fatty acids of 18 (and 16) carbon atoms are predominant in sediments as early as 3.4 Ga and throughout the Archean [14,15] and indeed are the most prevalent sizes in the three domains of today's organisms [15], 2) vesicles containing fatty acids with three conjugated double bonds required for light absorption around the peak in the Archean UVC spectrum at 260 nm (see below) must contain fatty acids of at least 18 carbon atoms in order to be stable at high surface temperatures of the early Archean of around 85 °C, and 3) apart from the telomerization of ethylene, there is a rather simple photochemical route to the production of 18 carbon atom fatty acids from the UVC induced polymerization of highly surface active 9 carbon nonanoic acid C9H18O2 which can be produced through UVC photochemical dissipative structuring of CO, CO2 saturated water [22]. Our postulate for the fatty acids of the first protocells, with chain lengths of 18 carbon atoms and unsaturated and conjugated, is distinct to that previously asserted as being the most plausible prebiotic fatty acids for the first protocells -- of saturated short chain ( 10 carbon atoms) lengths (such as myristoleic acid and its alcohol) because these are the most likely to arise from a Fischer-Tropsch type chemistry [23]. This conclusion, however, ignores UVC induced surface chemistry for synthesis, polymerization, and cross linking (see below) of fatty acids, and the fact that a greater thermal stability of the vesicles then generally believed would have been required because of the high surface temperatures of the early Archean. Inclusion of even longer chain lengths (≥ 20 carbon atoms, for which there is some evidence in the early fossil record (see [14,24] and references therein) increases still the stability with respect to pH range [25] and temperature. Photochemical Synthesis of Fatty Acids A plausible alternative to ocean floor Fischer-Tropsch synthesis of the hydrocarbon chains is ocean surface photochemical synthesis. The fatty acid hydrocarbon tails can be built up from the sequential photon-induced polymerization of an initiator molecule such as ethylene (Fig. 3). Photopolymerization occurs through photon-induced sensitization in the UV region of the spectrum. It generally occurs through direct photon-induced cleavage of the initiator molecule producing a free-radical which subsequently attacks the carbon-carbon double bonds of an existing polymer, thus initiating further polymerization [26]. Polymerization rates are more than two orders of magnitude larger at wavelengths of 254 nm (UVC) than at 365 nm (UVA) [26]. Oxygen acts as a strong inhibitor to polymerization by rapidly reacting with the radical to form a peroxy-based radical which does not promote polymerization [26]. Such an oxidation reaction following hydrolysis is the origin of the carboxyl head group of the fatty acids. The presence of oxygen and the lack of surface UVC light today (Fig. 1) means that hydrocarbon chain polymerization at today's ocean surface [16,17] is only a mere ghost of what it probably was at the origin of life. Indications that ultraviolet light may have played an important role in the formation of hydrocarbons have come from different experiments. Since the early 1960's it was known that irradiation with UVC light of CO2 saturated water containing ferrous salts results in the production of formic acid and formaldehyde [27]. Later, C1 hydrocarbons such as methane, methanol, ethanol and formaldehyde, and formic acid were produced from CO2 and H2O in a photoelectrochemical reactor consisting of a TiO2-coated electrode suspended in CO2 saturated aqueous solution and subjected to UVC light [28,29]. Klein and Pilpel (1973) [30] demonstrated that short chain amphiphiles can be synthesized by a light-dependent reaction from common simple hydrocarbons and an aqueous film of poly aromatic hydrocarbons (PAHs), ubiquous throughout the universe [9], acting as photosensitizers. It has also been shown that imidazole or porphoryins (pyridine) under UVC light can act as a catalyst for the reduction of CO2 to n=2 and longer chain hydrocarbons [31]. Later, Varghese et al. [32] and also Roy et al. [33] identified longer chain hydrocarbons in similar mixtures of CO2 and water irradiated with UV light. In general, and in contrast to the Fischer-Tropsch process, the higher the temperature, the higher the pressure, or the greater the ratio of CO2 to H2O, the larger the quantity of longer chain hydrocarbons obtained. Another plausible route to the formation of fatty acids under a UV environment has been observed Botta et al. [33]. By impinging UV and visible light (185 to 2000 nm) from an xenon lamp on formaldehyde (a product of UVC light on HCN and water) using ZnO and TiO2 as photocatalysts, Botta et al. found that this resulted in fatty acids of chains of from 2 to 5 carbon atoms as well a host of other fundamental molecules of life such as nucleic acids, and amino acids, as well as glycolaldehyde, a precursor to sugars through the formose-reaction. The mechanism of formation of these fatty acids suggested by Botta et al. follows that proposed earlier by Eschenmoser [34] entailing, first the generation of HCN from formaldehyde followed by its oligomerization to diaminomaleonitrile DAMN (detected in their reaction mixture), hydrolysis and successive electron transfer processes. The formation of DAMN from HCN requires the absorption of photons in the long wavelength UVC region and is exactly the same process which we have described as the photochemical dissipative structuring of the purines from UVC light on HCN and water [5] first observed by Ferris and Orgel [35]. The ZnO and TiO2 photocatalysts are suspected of providing electron transfer reactions after absorbing a UV photon. Although these catalysts are rather common and occur naturally and would have been available on the ocean surface (but in a much more diluted phase than that employed by Botta et al.) another electron donor which could have equally catalyzed the reaction is the amino acid histidine or its intermediate known as amino-imidazole-carbon-nitrile (AICN) and those amino acids with a negatively charged R-group; Glu monosodium salt (GluNa), and Asp potassium salt (AspK). These, we have argued [36], were among the first amino acids to have formed complexes with RNA and DNA in order to increase global UVC dissipation as evidenced by their charge transfer absorption spectrum peaking at 270 nm [37] falling in the middle of the Archean UVC atmospheric window, and as evidenced by their chemical affinity their codons and/or anticodons [38]. Here we propose that the long chain, n ≥18, hydrocarbons which became incorporated into the fatty acid vesicles of early life were produced by such surface-sensitized UVC-induced polymerization processes acting on shorter chain hydrocarbons such as ethylene produced through UVC light on CO2 saturated surface water or formaldehyde and HCN saturated water at high temperature (> 85 °C) and possibly a higher than present atmospheric pressure (up to 2 bar [39]). In contradistinction to the Fischer-Tropsch polymerizations operating at very high temperatures and pressures (perhaps occurring at deep sea hydrothermal vents) such photochemical polymerization could only have occurred on the ocean surface where concentrations of fatty acids would have been sufficiently high enough to allow access to the reactive triplet state, and where UVC light was most intense. The experiments of Rossignol et al. [22] suggest that catalysts are not needed for such surface polymerizations, however, the existence of catalyst transition metals for the reduction of CO2 such as Fe, Mn, Co, Ni, Cu, or Zn (particularly Fe) would have been available at the Archean ocean surface [40] and these would have undoubtedly increased the rates of hydrocarbon chain growth. There is again also the possibility that an imidazole, such as the amino acid histidine or its intermediate AICN formed in the process of UVC microscopic dissipative structuring of the purines from HCN [5], could have been the reduction agent. It is well known that imidazole is a strong catalyst that can act as either an acid or a base and furthermore, direct evidence has been obtained for its catalytic function in the formation of phosphatic acids from simpler compounds [41]. Dissipative Structuring Saturated fatty acids do not absorb in the UV except for disassociation at < 180 nm and the carboxyl head group which absorbs with a peak at 207 nm [42]. Under the Archean UVC flux photon-induced deprotonation could lead to a double carbon bond forming at any point on the hydrocarbon tail. A single double carbon bond in the tail will absorb at 210 nm. Migration of the double bonds along the tail is known to occur, leading to conjugated bonds [20]. Two double bonds in a conjugated configuration (diene) will lead to strong absorption at 233 nm [20], those having three in a conjugated configuration (triene) will lead to absorption at 269 nm, while those with 4 (tetraene, see Fig. 2) will lead to absorption at 310-340 nm. All of these latter three absorptions lie within the important UVC+ UVA spectrum arriving at Earth's surface during the Archean (see Fig. 1). Hydrocarbons having conjugated dienes, trienes, or tetraenes almost always have conical intersections [43] allowing rapid dissipation of the electronic excited singlet state energy. Reaching the conical intersection when in the electronic excited state involves a twisting about two C=C bonds and decreasing one of the C-C-C angles producing a kink in the carbon backbone [43]. Therefore, the region of the Archean surface spectrum that structured the CO, CO2 and H2O into fatty acids are the same photons that will be dissipated efficiently by the final photochemical product. These photochemically synthesized and conjugated fatty acid structures can thus be identified as microscopic dissipative structures [5]. The steps involved in the dissipative structuring are thus the following i) UVC-induced reduction of CO2 and CO in water saturated with these to form ethylene, ii) consecutive UVC-induced telomerization of ethylene to form long hydrocarbon tails of an even number of carbon atoms, iii) oxidation and hydrolysis events to stop the growing of the chain and form the carboxyl group, iv) UVC-induced deprotonation of the tails to form a double bond, v) double bond migration to give a conjugated diene, triene, or tetraene which has its own conical intersection (Fig. 4). These dissipatively structured fatty acids are, of course, robust to further photochemical reactions because of the sub picosecond decay times of their electronic excited states, due to their conical intersections. These decay times are too fast to allow appreciable further chemical transformation. Figure 4. Five steps are involved in the photochemical dissipative structuring of -parinaric acid under UVC+UVA light, starting from water saturated with CO2 or CO. i) UVC-induced reduction of CO2 and CO in water saturated with these to form ethylene, ii) UVC-induced telomerization of ethylene to form long hydrocarbon tails of an even number of carbon atoms, iii) oxidation and hydrolysis events to stop the growing of the chain and form the carboxyl head group, iv) UVC- or UVA-induced deprotonation of the tails to form double bonds, v) double bond migration to give a conjugated tetraene -parinaric acid (C18:4n-3) with a conical intersection. We have identified a similar dissipative structuring route for the synthesis of the nucleobase adenine from HCN under the same UVC photon potential [5]. In this case, the steps involve i) a thermal exothermic process leading from 4HCN molecules to the stable HCN tetramer cis- DAMN, ii) a UVC-induced cis to trans transformation through rotation around a carbon-carbon double bond, iii) a UV-A induced tautomerization, and iv) a UVC-induced ring closure. We believe that similar dissipative structuring routes involving different UVC- or UVB-induced processes exist for all of the fundamental molecules of life and this explains their strong absorption in the UVC (Fig. 1), and, in fact, gives a chemical-physical explanation for the origin of life based on non-equilibrium thermodynamics imperatives. Vesicle Formation The most notable problems associated with the proposal of fatty acids conforming walls of the first protocell are that; 1) vesicles of these form only in a narrow alkaline pH range, 2) they have a tendency to aggregate and crystallize in high salt conditions (salt flocculation), leading some to conclude that life must have started in a fresh water environment [44], and 3) a critical vesiculation concentration (CVC) of the fatty acids is required for spontaneous formation of vesicles and this concentration is considerably higher than that required for phospholipid vesicle formation. Recent experiments have shown, however, that the pH range of stability and resistance to salt flocculation of fatty acid vesicles can be greatly increased through covalent cross linking among neighboring chains, which can be induced by either UVC light [21], or by moderate temperatures (~50 -- 70 °C), or by simple aging [21,45]. The same heat treatment applied to non-conjugated unsaturated fatty acids does not appear to have an as large effect on the increase in range of pH stability or stability against salt flocculation. Under all conditions, vesicles of heat treated conjugated linoleic acid showed better stability than non-conjugated linoleic acid [45]. At the high surface temperatures and large UVC flux of the early Archean, considerable cross-linking between fatty acids could be expected, and conjugated fatty acids would thus form robust vesicles under the high temperatures, low pH, and high salt conditions of the Archean ocean surface. Furthermore, mixtures of fatty acids with fatty alcohols and fatty acid glycerol esters of differing lengths provide surprising resistance to salt flocculation [23]. Esterification (replacing the hydrogen of the OH in the carboxyl head group with, for example methyl CH3) of fatty acids which helps with resistance to salt flocculation (see section "Characteristics of Fatty Acids") can be induced by UVC activated transition metal catalysts to provide electrons for the reduction of the fatty acid [46]. UV radiation has also been shown to lead to the formation of aldehydes from fatty acids (through removal of the oxygen in the OH) [22]. There is also another more likely possibility giving rise to vesicle stability over a greater range of pH. Rather than the end to end model of the hydrophobic tails of phospholipids, the observed structure for fatty acids appears to be more like side-by-side [47] with overlap of respective double bonds (see figure 5). This would allow for a natural aging, or temperature- or UVC induced-, cross linking among fatty acids, which makes these vesicles stable over a very wide range of pH values, from 2 to 14 [47]. Figure 5. A vesicle made from variously conjugated (n≥18) fatty acids, for example, -parinaric (C18:4n-3) and conjugated linolenic acid (C18:3n-3), with an encapsulated DNA oligo. The DNA acts as an acceptor molecule for resonant energy transfer from the electronically excited un-conjugated fatty acids which do not have their own conical intersection. The inner and outer layers of the hydrocarbon tails have a side-by-side (rather than end-to-end) orientation which facilitates UVC-induced cross-linking at the overlaps, giving greater stability to the vesicle at the somewhat acid pH values of the Archean ocean surface. The whole system is an efficient photon dissipating system in the UVC+UVA regions of the Archean surface solar spectrum (Fig. 1). Conclusions We have described a plausible photochemical route to the dissipative structuring of the earliest Archean long chain fatty acids and their vesicles from CO2 and CO saturated water under the long wavelength UVC light prevailing at Earth's surface at the origin of life and throughout the Archean. Ethylene would be a common product synthesized by this UVC light incident on the surface water and these would polymerize (telomerize) under the same light into longer chain hydrocarbons of mainly an even number of carbon atoms. These would stop growing after an oxidation or hydrolysis event. They would then become conjugated through the dissipative structuring under the UVC light giving rise to conjugated parinaric acid (C18:4n-3) and linolenic acid (C18:3n-3). Because of the high surface temperatures (~85 °C), only long chain (≥ 18 carbon atoms) with conjugation near the end of the tail (n-3) would bind sufficiently together to form stable vesicles. This is in agreement with the abundance of this size in the fossil record of fatty acids found in sediments of the early Archean. Cross-linking would occur at the overlap of the inner and outer wall hydrocarbon tails at the site of a double bond, making the vesicles stable over low pH ranges (6.0-6.5) and high salt conditions of the Archean ocean surface. Many of the fatty acids would have their own conical intersection allowing the rapid dissipation of the photon-induced electronic excitation energy. These vesicles would then be performing the non-equilibrium thermodynamic function of dissipating into heat the prevailing UVC+UVA photon flux reaching Earth's surface. This was the initial driving force for complexation of material, from synthesis, to proliferation and evolution; all through dissipative structuring. For example, chemical affinity of the fatty acids, mediated through divalent cations, with RNA and DNA, or with the carotenoids, would provide for greater dissipation, especially for those non- conjugated fatty acids which did not have their own conical intersections. Resonant energy transfer would allow the electronic excitation energy of the fatty acids to be dissipated through the conical intersection of the nucleic acid or the carotenoids. This scenario provides a physical- chemical foundation based on photon dissipation for the origin, proliferation and evolution of the fatty acids in conjunction with other fundamental molecules of life. Acknowledgments The authors are grateful for comments on the manuscript by C. Bunge and A. Simeonov. The financial support of DGAPA-UNAM project number IN102316 is greatly appreciated. O.R. further acknowledges DGAPA and CONACyT for providing scholarships during various stages of this work. Bibliography [1] Prigogine, I. Introduction to Thermodynamics Of Irreversible Processes. 3rd Edition, New York: John Wiley & Sons, 1967. [2] Michaelian, K.. "Thermodynamic origin of life", ArXiv 2009 (http://arxiv.org/abs/ 0907.0042) [3] Michaelian, K. "Thermodynamic dissipation theory for the origin of life", Earth Syst. Dynam. 2011, 224, 37-51. [4] Michaelian, K. Thermodynamic Dissipation Theory of the Origina and Evolution of Life: Salient characteristics of RNA and DNA and other fundamental molecules suggest an origin of life driven by UV-C light. Mexico City: Self-published. Printed by CreateSpace, 2016, ISBN:9781541317482. [5] Michaelian, K. "Microscopic dissipative structuring and proliferation at the origin of life", Heliyon. 2017, 3, e00424. [6] Onsager, L. "Reciprocal Relations in Irreversible Processes, I", Phys. Rev. 1931, 37, 405-426. [7] Onsager, L. "Reciprocal Relations in Irreversible Processes, II", Phys. Rev.1931, 38, 2265. [8] Sagan, C. "Ultraviolet Selection Pressure on the Earliest Organisms", J. Theor. Biol. 1973, 39, 195-200. [9] Michaelian, K. and Simeonov, A. "Fundamental molecules of life are pigments which arose and co-evolved as a response to the thermodynamic imperative of dissipating the prevailing solar spectrum", Biogeosciences. 2015, 12, 4913-4937. [10] Johnson, D.W. "A synthesis of unsaturated very long chain fatty acids", Chemistry and Physics of Lipids. 1990, 56 (1), 65-71. ISSN: 0009-3084 [11] Pereto, J., Lopez-Garcia, P. and Moreira, D. "Ancestral lipid biosynthesis and early membrane evolution", Trends Biochem. Sci. 2004, 29, 469-477. [12] Lombard, J., López-García, P. and Moreira, D. "The early evolution of lipid membranes and the three domains of life", Nature Reviews, Microbiology. 2012, 10, 507-515. [13] Bar-nun, A. and Podolak, M. "The photochemistry of hydrocarbons in titan's atmosphere", Icarus. 1979, 38 (1), 115-122. [14] Han, J. and Calvin, M. "Occurrence of fatty acids and aliphatic hydrocarbons in a 3.4 billion-year-old sediment", Nature. 1969, 224 (5219), 576-577. [15] Van Hoeven, W., Maxwell, J.R. and Calvin, M. "Fatty acids and hydrocarbons as evidence of life processes in ancient sediments and crude oils", Geochimica et Cosmochimica Acta. 1969, 33 (7), 877-881. [16] Mochida, M., Kitamori, Y., Kawamura, K., Nojiri, Y. and Suzuki, K. "Fatty acids in the marine atmosphere: Factors governing their concentrations and evaluation of organic films on sea-salt particles", Journal of Geophysical Research: Atmospheres. 2002, 107 (D17), AAC 1-1- AAC 1-10. [17] Wellen, B. A., Lach, E. A. and Allen H. C. "Surface pka of octanoic, nonanoic, and decanoic fatty acids at the air-water interface: applications to atmospheric aerosol chemistry", Phys. Chem. Chem. Phys. 2017, 19, 26551-26558. [18] Kis M, Zsiros O, Farkas T, Wada H, Nagy F, Gombos Z. "Light-induced expression of fatty acid desaturase genes", Proc Natl Acad Sci USA 1998, 95, 4209-4214. [19] Budin, I., Prywes, N., Zhang, N. and Szostak J. W. "Chain-length heterogeneity allows for the assembly of fatty acid vesicles in dilute solutions", Biophysical Journal. 2014, 107, 1582- 1590. [20] Mandal, T. K. and Chatterjee, S. N. "Ultraviolet- and sunlight-induced lipid peroxidation in liposomal membrane", Radiation Research. 1980, 83, 290-302. [21] Bassas, M., Marqués, A. M. and Manresa, "A. Study of the crosslinking reaction (natural and uv induced) in polyunsaturated pha from linseed oil", Biochemical Engineering Journal 2007 40:275{283, 2007. [22] Rossignol, S., Tinel, L., Bianco, A.and Passanant, Mi. "Atmospheric photochemistry at a fatty acid-coated air-water interface", Science. 2016, 353, 699-702. [23] Mansy S. S. and Szostak J. W. "Thermostability of model protocell membranes", PNAS. 2008, 105 (36), 13351-13355. [24] Hahn, J. and Haug, P. "Traces of Archaebacteria in Ancient Sediments", System. Appl. Microbiol. 1986, 7, 178-183. [25] Aveyard, R., Binks, . B.P., Carr, N. and Cross A.W. "Stability of insoluble monolayers and ionization of langmuirblodgett multilayers of octadecanoic acid", Thin Solid Films. 1990, 188 (2), 361-373. [26] Bowman, C. N. and Kloxin, C. J. "Toward an enhanced understanding and implementation of photopolymerization reactions", AIChE J. 2008, 54, 2775-2795. [27] Getoff, N. "Reduktion der kohlensäure in wässeriger lösung unter einwirkung von uv-licht", Zeitschrift für Naturforschung B. 1962, 17 (2), 87-90. [28] Halmann, M. "Photoelectrochemical reduction of aqueous carbon dioxide on p-type gallium phosphide in liquid junction solar cells", Nature. 1978, 275, 115-116. [29] ULMAN, M., TINNEMANS, A. H. A., MACKOR, A., AURIAN-BLAJENI, B. andHALMANN M." Photoreduction of carbon dioxide to formic acid, formaldehyde, methanol, acetaldehyde and ethanol using aqueous suspensions of strontium titanate with transition metal additives", International Journal of Solar Energy. 1982, 1 (3), 213-222. [30] Klein, A. E. and Pilpel, N. "Oxidation of n-alkanes photosensitized by 1-naphthol", J. Chem. Soc Faraday Trans. 1 1973, 69, 1729-1736. [31] Bocarsly, A. B., Gibson, Q.D., Morris, A. J., L'Esperance, R. P., Detweiler, Z. M., Lakkaraju, P. S., Zeitler, E. L. and Shaw T. W. "Comparative study of imidazole and pyridine catalyzed reduction of carbon dioxide at illuminated iron pyrite electrodes", ACS Catalysis. 2012, 2, 1684-1692. [32] Varghese, O. K., Paulose, M., LaTampa., T. J., Grimes, C. A. "High-Rate Solar Photocatalytic Conversion of CO2 and Water Vapor to Hydrocarbon Fuels", Nano Letters 2009, 9 (2), 731-737. [33] Botta, L., Bizzarri, B.M., Piccinino, D., Fornaro, T., Brucato, J.R. and Saladino, R.. "Prebiotic synthesis of carboxylic acids, amino acids and nucleic acid bases from formamide under photochemical conditions", Eur. Phys. J. Plus. 2017, 132, 317. [34] Eschenmoser, A. "On a Hypothetical Generational Relationship between HCN and Constituents of the Reductive Citric Acid Cycle", Chemistry & Biodiversity. 2007, 4 (4), 554- 573. [35] Ferris, J. P. and Orgel, L. E. An "Unusual Photochemical Rearrangement in the Synthesis of Adenine from Hydrogen Cyanide", J. Am. Chem. Soc., 1966, 88, 1074. [36] Mejía, J. and Michaelian, K. "Information encoding in nucleic acids through a dissipation- replication relation", ArXiv:1804.05939 [physics.bio-ph] 2018. [37] Prasad, S., Mandal, I., Singh, S., Paul, A., Mandal, B., Venkatramani, R. and Swaminathan, R.. "Near uv-visible electronic absorption originating from charged amino acids in a monomeric protein", Chem. Sci. 2017, 8, 5416-5433. [38] Yarus, M., Widmann, J. and Knight, R. "RNA-Amino Acid Binding: A Stereochemecal Era for the Genetic Code", J Mol Evol. 2009, 69, 406-429. DOI 10.1007/s00239-009-9270-1 [39] Som, S. M., Catling, D. C., Harnmeijer, J. P., Polivka, P. M. and Buick, R.. "Air density 2.7 billion years ago limited to less than twice modern levels by fossil raindrop imprints", Nature. 2012, 484, 359-362. [40] Hardy. J. T. "The sea-surface microlayer (1982) biology, chemistry and anthropogenic enrichment", Prog. Oceanogr. 1982, 11, 307-328. [41] Epps, D. E. Sherwood, E., Eichberg, J. and Oró, J. "Cyanamide mediated syntheses under plausible primitive earth conditions", Journal of Molecular Evolution. 1978, 11, 279-292. [42] Vicente, A., Antunes, R., Almeida, D., Franco, I. J. A., Hoffmann, S. V., Mason, N. J., Eden, S., Duflot, D., Canneaux, S., Delwiche, J., Hubin-Franskin, M.-J., and Limão-Vieira, P. "Photoabsorption measurements and theoretical calculations of the electronic state spectroscopy of propionic, butyric, and valeric acids", Phys. Chem. Chem. Phys.2009,11 (27),5729-5741. [43] Celani, P., Garavelli, M., Ottani, S., Bemardi, F., Robb, M. A. and Olivucci, M. "Molecular "Trigger" for Radiationless Deactivation of Photoexcited Conjugated Hydrocarbons", J. Am. Chem. Soc. 1995, 117, 11584-11585 [44] Milshteyn, D., Damer, B., Havig, J. and Deamer, D.. "Amphiphilic compounds assemble into membranous vesicles in hydrothermal hot spring water but not in seawater", Life, 2018, 8,11. [45] Fan, Y., Fang, Y., Ma L., and Jiang, H. "Investigation of micellization and vesiculation of conjugated linoleic acid by means of self-assembling and self-crosslinking", J. Surfact. Deterg. 2015, 18, 179-188. [46] Verma P., Kaur, K., Wanchoo, R. K., Toor, A. P. "Esterification of acetic acid to methyl acetate using activated TiO2 under UV light irradiation at ambient temperature", Journal of Photochemistry and Photobiology A: Chemistry. 2017, 336, 170-175. [47] Fan, Y. Fang, Y. and Ma L. "The self-crosslinked ufasome of conjugated linoleic acid: Investigationof morphology, bilayer membrane and stability", Colloids and Surfaces B: Biointerfaces 2014, 123, 8-14.
1507.05655	1	1507	2015-07-20T21:04:05	Effects of macroH2A and H2A.Z on nucleosome structure and dynamics as elucidated by molecular dynamics simulations	[ "physics.bio-ph", "q-bio.BM" ]	Eukaryotes tune the transcriptional activity of their genome by altering the nucleosome core particle through multiple chemical processes. In particular, replacement of the canonical H2A histone with the variants macroH2A and H2A.Z has been shown to affect DNA accessibility and nucleosome stability; however, the processes by which this occurs remain poorly understood. Here, we elucidate the molecular mechanisms of these variants with an extensive molecular dynamics study of the canonical nucleosome along with three variant-containing structures: H2A.Z, macroH2A, and an H2A mutant with macroH2A-like L1 loops. Simulation results show that variant L1 loops play a pivotal role in stabilizing DNA binding to the octamer through direct interactions, core structural rearrangements, and altered allosteric networks in the nucleosome. All variants influence dynamics; however, macroH2A-like systems have the largest effect on energetics. In addition, we provide a comprehensive analysis of allosteric networks in the nucleosome and demonstrate that variants take advantage of stronger interactions between L1 loops to propagate dynamics throughout the complex. Furthermore, we show that post-translational modifications are enriched at key locations in these networks. Taken together, these results provide new insights into the relationship between the structure, dynamics, and function of the nucleosome core particle and chromatin fibers, and how they are influenced by chromatin remodelling factors.	physics.bio-ph	physics	Effects of macroH2A and H2A.Z on nucleosome structure and dynamics as elucidated by molecular dynamics simulations Samuel Bowerman and Jeff Wereszczynski∗ Department of Physics and Center for Molecular Study of Condensed Soft Matter, Illinois Institute of Technology, 3440 S Dearborn St., Chicago, IL 60616, USA E-mail: [email protected] Abstract Eukaryotes tune the transcriptional activity of their genome by altering the nucleosome core particle through multiple chemical processes. In particular, replacement of the canonical H2A histone with the variants macroH2A and H2A.Z has been shown to affect DNA accessibility and nucleosome stability; however, the processes by which this occurs remain poorly understood. Here, we elucidate the molecular mechanisms of these variants with an extensive molecular dynamics study of the canonical nucleosome along with three variant-containing structures: H2A.Z, macroH2A, and an H2A mutant with macroH2A-like L1 loops. Simulation results show that variant L1 loops play a pivotal role in stabilizing DNA binding to the octamer through direct interactions, core structural rearrangements, and altered allosteric networks in the nucleosome. All variants influence dynamics; however, macroH2A-like systems have the largest effect on energetics. In addition, we provide a comprehensive analysis of allosteric networks in the nucleosome and demonstrate that variants take advantage of stronger interactions between L1 loops to propagate dynamics throughout the complex. Furthermore, we show that post-translational modifications are enriched at key locations in these networks. Taken together, these results provide new insights into the relationship between the structure, dynamics, and function of the nucleosome core particle and chromatin fibers, and how they are influenced by chromatin remodelling factors. ∗To whom correspondence should be addressed 1 Introduction Eukaryotes package their genetic code in highly ordered chromatin fibers. The fundamental unit of these structures is the nucleosome core particle (NCP), a complex of ∼147 basepairs of DNA that are wrapped around eight histone proteins (Figure 1). 1 Although they have minimal sequence homology, each core histone has a structural motif of an N-terminal tail, three α-helices connected by two loops (α1-L1-α2-L2-α3), and a C-terminal tail. 1,2 In the assembled NCP, histones are structurally divided into a (H3-H4)2 tetramer that is positioned between two H2A-H2B dimers. The only location of inter-dimer interactions is at the base of the NCP which is formed by the H2A L1 loops, whereas each dimer has two interfaces with the tetramer: the H2A "docking domain" (DD) and the H3-H4 "four helix bundle." 1 -- 4 Cells regulate chromatin stability and DNA accessibility by changing the biochem- ical properties of the NCP. 5 -- 8 One of the primary chromatin remodeling mechanisms is the replacement of H2A or H3 histones with "histone variants." 9 -- 15 These variants have a similar structure and sequence to the canonical histones, however they diverge at key locations that affect inter-histone and DNA-histone contacts. These differences al- ter the structure and stability of the NCP and are therefore implicated in modulating transcriptional activity. For example, the H2A variant macroH2A exists in large pop- ulations in the inactive X chromosome of fe- males but is sparse in active genes. 11,16,17 In contrast, the H2A.Z variant has been linked to both transcriptional activation and re- pression and is enhanced in regulatory re- gions of the genome such as promoters and enhancers. 18,19 Histone variants Figure 1: (left) The nucleosome core particle viewed down the DNA superhelical axis. Co- ordinates are taken from the final snapshot of the first canonical nucleosome simulation. Color code: H3 (red), H4 (blue), H2A (green), H2B (purple), and DNA (grey). (right) The struc- tures of the three L1 loop sequences considered in this study: canonical (top), macroH2A (middle), and H2A.Z (bottom). The canonical loops pos- sess a net negative charge resulting from Glu41, while the macroH2A loops possess a net positive charge from Lys40. The L1 loops of H2A.Z are uncharged, but both macroH2A and H2A.Z loops introduce a larger hydrophobic volume than the canonical. influence chromatin through diverse mechanisms and struc- ture/function relationships. macroH2A is unique among variants in that it possesses multiple domains, including the histone do- main, a 38 residue linker sequence, and a large "macro-domain." 3,20 On its own, the histone domain is sufficient for reducing transcriptional activity in vivo and increasing the stability of the nucleosome complex, even though the crystal structure of an NCP containing this domain shows that variant incorpo- ration causes only minor NCP rearrangements. 11,21 The primary sequence is ∼65% identical to canonical H2A and differs largely from H2A in two important regions: the L1 loops and the docking domain. The canonical 38NYAE41 L1 loop possesses a net negative charge, while 2 in contrast the macroH2A L1 38HPKY41 sequence has a net positive charge and an increased hydrophobicity. Substitutions of the L1 loops in canonical H2A with a macroH2A sequence (the "L1-Mutant") creates nucleosomes with in vitro stabilities and in vivo enrichments that are nearly identical to NCPs containing the complete macroH2A histone domain. 11,21 Therefore, the L1 loops appear to be pivotal in dictating macroH2A's abilities to affect intra- nucleosomal functions. Meanwhile, changes to the docking domains show little effect on in vitro stability, but increase in vivo enrichment. 11,21 The role and mechanisms of the H2A.Z variant remains less well defined, with some experiments showing that H2A.Z increases NCP stability while others have found that it destabilizes the system. Similar to macroH2A, a comparison of the H2A.Z and canonical containing NCP crystal structures show nearly identical overall conformations with the ex- ception of two features. 4 First, the structure of the L1 loops is altered, resulting in increased contacts between the two H2A/H2B dimers which likely helps to stabilize the histone oc- tamer. Second, H2A.Z has fewer hydrogen bonds between the docking domain and H3, which could destabilize the dimer/tetramer interface. This combination of stabilizing one area of the NCP while destabilizing another may account for the disparate experimental results and the multiple functions H2A.Z appears to have. 22 Experiments have revealed a wealth of information about how histone variants affect NCP and chromatin function, yet several questions still remain. For example: how do seemingly minor structural rearrangments affect the stability of the nucleosome? To what extent do changes in the L1 loops propagate through the complex? Do variants influence NCP function through only structural means, or do they take advantage of altered dynamics as well? To address these problems, we have performed extensive molecular dynamics (MD) simulations of four complete NCP systems that include: 1) canonical H2A histones, 2) the macroH2A histone fold domains, 3) "L1-Mutant" H2A histones, and 4) H2A.Z histones. Our results indicate that different sequences in the L1 loops perturb the dynamic and energetic properties in this region of variant containing NCPs. These effects propagate throughout the complex and create subtle, yet important, rearrangements that alter the NCP structure and dynamics through both direct effects and modified allosteric networks. This allows histone variants to influence both the global dynamics and energetics of the NCP, and likely contributes to large-scale structural changes such as DNA breathing and nucleosome opening, as well as inter-NCP interactions in chromatin fibers. 23,24 Materials and Methods System and Simulation Details Simulations of the canonical, macroH2A, and H2A.Z containing nucleosomes were initial- ized from their crystal structures (PDB: 1KX5, 1U35, and 1F66, respectively). 2 -- 4 The L1- Mutant structure was formed using the crystal structure of 1KX5, with the H2A L1 loops mutated from the canonical 38NYAE41 to the macroH2A 38HPKY41 sequence. The systems were neutralized and solvated in a 10 A TIP3P box of 150mM NaCl, creating systems of approximately 250,000 atoms. Each system was simulated three times (see supplemental material for more details). The simulations were done in the NAMD engine (v2.9) using the 3 AMBER12SB fixed point-charge forcefield. 25,26 Monovalent ions were modeled according to Joung and Cheatham. 27 Production simulations were done in the NPT ensemble using stan- dard techniques. 28 -- 31 Coordinates were stored every 2 ps. Visualizations were made using VMD and PyMOL. 32 -- 34 Allosteric Pathways Calculations Allosteric effects were computed with multiple techniques (for specific details, see sup- plemental materials). Per-residue differences in dynamics were determined by calculating the Kullback-Leibler divergence of dihedral angle populations. 35 For these calculations, the canonical populations were used as a reference set. Spurious results were filtered using boot- strapping techniques. Residue-residue correlations were calculated by utilizing the "largest mutual information" method. 36,37 Residue contacts were determined to be when protein Cα or nucleic C1' atoms were within 10 A in at least 70% of the configurations. 38 The mapping of allosteric networks was conducted using the Weighted Implementation of Sub-optimal Pathways approach. 39 The edge-betweenness centrality of residues in the optimal networks were calculated with the NetworkX Python package, with the significance determined by a hypergeometric distribution (see supplemental materials). 40 -- 42 Interaction Energies Interaction energies between the L1 loops were calculated using cpptraj. 43 A cutoff distance of 15 A was used for both van der Waals and electrostatic interactions. Because the L1 loops interact with both protein and DNA, an intermediate dielectric value of 5 was considered (εDNA = 8, εprotein = 4). 44,45 Hydrogen bonds were defined by a separation distance of 3.5 A and an angle of 30◦. DNA binding and complex assembly energies were calculated using the MMPBSA.py function of AMBERTOOLS (v.14). 46 The level of theory was restricted to the Generalized Born Implicit Solvent (igb=5, radii=mbondi2). 47 Coordinates from every 100ps of production simulation were used. The coordinates for the protein constituents were extracted from the nucleosome simulations, but the unbound DNA coordinates were taken from a separate simulation of 147bp of linear B-form DNA in a 150 mM NaCl environment. Error bars in the energies and all other measures are defined by the standard error of the mean, where the number of independent points was determined by the statistical inefficiency of the data set, as computed with the PyMBAR package. 48 Results We performed three independent 250 ns MD simulations for four complete NCP systems: 1) canonical, 2) macroH2A histone fold, 3) "L1-Mutant," and 4) H2A.Z containing NCPs. In each set of simulations, approximately 50 ns was required for the root-mean square de- viations (RMSDs) of the complexes to stabilize (Figures S2-S5) and for the tails to collapse from their initial elongated states. These results are consistent with previous MD simula- tions of the canonical NCP which demonstrated that the overall complex is stable on the 4 hundreds of ns timescale and that the histone tails form strong interactions with the nucle- osomal DNA. 23,49,50 Comparisons between the canonical and variant systems demonstrate that variants have both subtle and large-scale effects on the structure and dynamics of the L1 loops, DNA-histone interactions, and allosteric networks throughout the NCP. Altered Dynamics of L1 Loops Modifications of the L1 loop se- quences in histone variants alter their dynamics and energetics. In the canonical NCP simulations, an average of 0.5 hydrogen bonds were formed between the L1 loops, pri- marily between the carboxamide nitrogen of asparagine and the car- boxylate group of the symmetric glutamate. This is consistent in the L1-mutant (∼0.4) but is reduced to ∼0.2 in macroH2A. In these two systems, the most prevalent hydro- gen bonds were formed between the phenol oxygen of tyrosine and the lone pair of the symmetric histi- dine. In the H2A.Z simulations, L1-L1 hydrogen bonds were almost nonexistent (Table S1). Figure 2: (a) Distance populations for Lys40 to DNA phos- phate show an interaction that is unique to macroH2A-like L1 loops. (b) A representative configuration of the Lys sidechain stretching across the molecule to interact with the dimer's non-associated DNA. This orientation sterically hinders the symmetric loop from forming a similar interac- tion. This interaction contributes significantly to stabiliz- ing DNA-octamer binding in the macroH2A and L1-Mutant systems. Although hydrogen bonds form most frequently in the canonical loops, the net L1 loop interaction energies are more favorable in the variants (Table 1). The close proximity of the negatively charged glutamates in the canonical NCP creates a disfavored electrostatic interaction. However, the L1-Mutant and macroH2A systems avoid an analogous situation through Lys-DNA interactions, which separates the like-charges and creates a more favorable electrostatic configuration. Meanwhile, the lack of charge in the H2A.Z L1 Loops also creates a more favorable electrostatic interaction than the canonical NCP. In addition, the L1 loop rearrangement in the macroH2A and L1-Mutant systems is further stabilized by van der Table 1: Net interaction energies between H2A L1 loops show that macroH2A and H2A.Z loops have more favorable interactions than canonical L1 loops. ƊU is defined as the difference in energies between each variant and the canonical system. Negative values show favorability in the variants. All values are reported in kcal/mol. System Canonical NCP L1 Mutant macroH2A NCP H2A.Z NCP Uelect 6.3 ± 2.0 -0.9 ± 0.4 0.4 ± 1.1 -0.6 ± 0.1 UvdW -9.4 ± 0.8 -13.4 ± 0.8 -12.1 ± 0.9 -6.9 ± 0.6 Utot -3.1 ± 2.2 -14.3 ± 0.9 -11.7 ± 1.0 -7.5 ± 0.6 ƊUelect -- -- -7.2 ± 2.0 -5.9 ± 2.3 -6.9 ± 2.0 ƊUvdW -- -- -4.0 ± 0.8 -2.7 ± 0.9 2.4 ± 1.0 ƊUtot -- -- -11.2 ± 2.4 -8.6 ± 2.3 -4.4 ± 2.3 5 (b) L1-Mutant macroH2A 10 8 Distance ( ) 12 14 16 18 2 4 6 0.6 0.5 0.4 0.3 0.2 0.1 0.0 (a) P(r) Waals interactions. In total, the interaction energies of the L1 loops in the L1-Mutant and macroH2A structures are substantially favored over those of the canonical nucleosome, with respective ƊUtotal values of -11.2 ± 2.4 kcal/mol and -8.6 ± 2.3 kcal/mol. The H2A.Z L1 loop conformations are also more favorable than in the canonical system (ƊUtotal = -4.4 ± 2.3 kcal/mol). The different net charges of the L1 loops influence their interactions with the nucleosomal DNA. In the canonical L1 loops, the negative charge located on Glu41 causes a repulsive force to the negatively charged DNA. However, in the macroH2A and L1-Mutant systems, Lys40 introduces a positive charge into the loop which forms a salt bridge with the DNA basepair across the axis of symmetry (Figure 2). The lysine forming this salt bridge sterically hinders the symmetric lysine residue from doing the same, so the interaction exists in only one dimer. The non-interacting lysine is primarily exposed to solvent while intermittently forming a hydrogen bond with a neighboring histidine. Since the L1 loops of H2A.Z are uncharged, they are not capable of forming similar interactions, and therefore did not make any direct contacts with the DNA. Taken with the re- sults of the L1-L1 loop dynamics, we observe that the macroH2A-like loop sequences stabilize both protein- protein and protein-DNA interactions when compared to both the canonical and H2A.Z histones. System Canonical NCP L1-Mutant p-value macroH2A NCP p-value H2A.Z NCP p-value top 67.1 ± .2 66.0 ± .2 .0003 66.2 ± .3 .0001 66.0 ± .1 .0001 bottom 34.8 ± .1 36.8 ± .4 .0001 36.2 ± .1 .0001 36.1 ± .1 .0001 Separation distances Figure 3: for H2A α2 helix locations show a "bulging" effect in histone variants. The helix is displayed in black while the helix "top" is highlighted in red and the "bottom" in blue. The L1- loops are shown in green for clarity. Shifts in mean separation are on the order of an A, but the changes in pop- ulations are all incredibly significant. Variant Presence Alters Dimer Orientations Reorganization of the L1-loops creates perturbations that affect the dimer orientations in the NCP. For ex- ample, the H2A α2 helix extends across the dimer, with its N-terminal (the "base") at the L1-interface and its C-terminal (the "top") solvent-exposed on the far side of the molecule (Figure 3). Simulation analysis showed that the canonical system exhibited a separation of 67.1 ± 0.2 A between the tops, and 34.8 ± 0.1 A between the bottoms of the H2A α2 helices. In contrast, in each of the variant NCPs there is a "bulging" motion in which the base separation is increased to ∼36.5 A while the top separation is decreased to ∼66.1 A. Although these changes in orientation are only on the order of an A, a t-test indicated that they are all extremely statistically significant (Figure 3). This subtle re-orientation of the dimers alters histone-DNA hydrogen bonding. For example, the guanidine group of H2A Arg29 forms a hydrogen bond with the phosphate group of the 23rd basepair of DNA in all systems. In the canonical NCP, this bond forms in 63% of the configurations, whereas in the L1-mutant 6 it is formed 70% of the time. In addition, the frequency of hydrogen bonding between the 22nd basepair phosphate and the backbone amide of H2B Ser33 increases from 50% in the canonical system to 60% in the L1-Mutant structure. In H2A.Z, the hydrogen bond- ing at these locations increases drastically to 73% and 85%, respectively. Interestingly, the macroH2A nucleosome shows a decreased frequency of both of these interactions (52% and 30%, respectively). The reduced hydrogen bonding in the macroH2A nucleosome is likely a result of sequence deviations in the nearby H2A α1 helix (canonical: 30VH31, H2A.Z: 30IH31, macro: 30ML31). The dimer realignment also affects the hydrogen bonding between protein constituents in the histone core. In the canonical NCP, an average of 14.8 hydrogen bonds are formed between a single dimer and the tetramer, which is in agreement with the ∼15 observed in the crystal structure. This increases to an average of 16.5 in the L1-Mutant. The macroH2A and H2A.Z nucleosomes display an average of 14.7 and 14.4 hydrogen bonds, which are substantially more frequent than the ∼8 observed in the initial configurations. Therefore, the variant dimer reorientation encourages the histones to form hydrogen bonds more frequently than in the crystal structures. Figure 4: Kullback-Leibler Divergence of dihedral angles for the (a) L1-Mutant, (b) macroH2A, and (c) H2A.Z nucleosomes, using the canonical populations as a reference set. The dimers and DNA residues are represented in a rainbow spectrum, where divergence values increase from blue to red. The tetramer is shown in magenta, where the tube radius is wider for larger values. Significant divergences are observed both in the vicinity of and far from the L1-L1 interface. Histone H2A L1 Sequence Influences Dynamics Throughout the Nucleosome The L1 loops not only influence dimer reorientation, but they also perturb the local dy- namics of residues throughout the nucleosome. Calculations of the Kullback-Leibler (KL) divergence between the canonical and variant systems showed the expected disparity in dihedral sampling of residues within the L1 loop region (Figure 4). However, they also high- lighted significant changes in the dynamics of residues that are distant from these loops. In both the L1-mutant and macroH2A systems, the dimer and tetramer constituents of the docking domains have statistically significant KL divergence values, indicating that their local dynamics are different in these systems relative to the canonical NCP. Although this was expected in the macroH2A and H2A.Z systems due to their sequence deviations, the 7 (a) (b) (c) L1-mutant is sequentially identical to the canonical system in these areas. Therefore, the observed difference in dy- namics must be due to allosteric networks that are shifted by the L1 mutation. On the other hand, the dynamics in the histone core of H2A.Z only show small differences from the canonical system, most notably along the H2B α2 helix, while the largest divergences are in the DNA. The L1-mutant and macroH2A variants also have in- creased dynamical correlations between the H2A L1 loops and key portions of the NCP (Table 2 and Figure 5). The strengthened interactions in the L1 loops increase the L1- L1' correlations from 0.42 in the canonical system to >0.67 in each of the variants. In both the canonical and H2A.Z systems, the average correlation between the L1 loops and either docking domain (symmetric - DD, opposing dimer - DD') was 0.36-0.38. However, in the L1-Mutant the average L1-DD and L1-DD' correlations increased to 0.51 for both measurements, which were further increased to 0.56 and 0.60 in the macroH2A structure. Although the variants had an increased correlation between the L1 loops and DNA near the base of the nucleosome, the correlations between the L1 loops and DNA extremities were similar in all four systems. Variant Presence Alters Allosteric Path- ways for Figure 5: Average individ- ual residue correlation with L1 loop residues (a) canoni- cal and (b) L1-Mutant nucleo- somes. Thicker, redder residues are those with stronger average correlations with the L1 loop se- quence. The L1-Mutant nucle- osome shows increased correla- tions near the L1-L1 interface, as well as among H2B-H4 four helix bundle residues. Both systems display appreciable correlations between the L1 loops and dock- ing domain residues in the dimer and tetramer. Histone tails were truncated to improve clarity. The origins of the altered dynamics and correlations in NCPs with variants were probed by computing the opti- mal and suboptimal correlation pathways using with the Weighted Implementation of Suboptimal Paths algorithm 39. Allosteric networks were calculated between the L1 loops and the DNA entry and exit sites, and the tetramer compo- nents of the docking domains for each system. The results revealed that not only are there several networks of dynam- ically coupled residues in the canonical NCP, but that these networks are both modified and strengthened by macroH2A, H2A.Z, and the L1-mutant. The shifts are due to both changes in the NCP hydrogen bonding networks from subtle repositioning of the H2A histones, as well as increased interactions of the L1 loops with one another and with the nucleosomal DNA. In the L1-to-symmetric DNA end pathways, the canonical system utilizes three main routes for information transfer (Figure 6). In the first, networks primarily pass through neighboring H2B Ser33-DNA and H2A α1 helix Arg29-DNA hydrogen bond interactions and into the DNA, whereas in the second the networks enter the DNA through the H2A Arg42- 8 (a) (b) Table 2: Average correlations between L1 loops and relevant regions of the nucleosome core particle for each system. The L1-L1' and L1-DD(') correlations are significantly stronger in the systems possessing the macroH2A L1 loops, while L1 correlations to the DNA extremities are unchanged. The associated docking domain is abbreviated as DD, and the docking domain of the opposing dimer is abbreviated as DD'. System Canonical NCP L1 Mutant macroH2A NCP H2A.Z NCP L1-L1' L1-DD L1-DD' L1-DNA 0.42 0.70 0.73 0.67 0.38 0.51 0.56 0.38 0.36 0.51 0.60 0.38 0.49 0.48 0.48 0.44 DNA hydrogen bond near the intradimer interaction site. The third route for propagation extends along the H2A α2 helix, which passes dynamic information into the DNA basepairs via a Thr76-DNA interaction. The pathways of H2A.Z are similar to the canonical but with more pathways accessing the H2A α1 Arg29-DNA interaction than that of H2B Ser33-DNA. In the L1-Mutant, the increased prevalence of the Arg29-DNA hydrogen bond heavily biases information transfer through this network and increases the strength of this pathway. The decreased Arg29-DNA interaction in the macroH2A nucleosome causes information to be transferred primarily via the H2A Arg42-DNA hydrogen bond, with a significant number of pathways also traversing the H2A α2 helix. The effects of L1-L1' communication transfer are most apparent in the networks between an L1 loop and the DNA end of the opposite symmetry. In the canonical nucleosome, there exist no pathways between L1 loops, therefore networks must pass through indirect routes that include the DNA and histone tails. However, in all of the variant structures information is readily exchanged between the L1 loops, allowing the pathways to immediately cross into the opposite symmetry dimer (Supplemental Figures S12-S15). Once information is passed into this dimer, it follows the typical pathways for L1-to-symmetric DNA end propagation. This results in allosteric networks that are not only stronger, but more direct in the histone variants. Pathways between the L1 loop and docking domains in the same dimer are similar in all systems, but there is a large disparity in the pathways between L1 loops and the docking domain of the other dimer constituent. In the canonical NCP, the majority of paths pass from the L1 loops through the H2B α2 helix into the tetramer portion of the docking domain via the four helix bundle of H2B-H4. The L1-Mutant structure shows an increased number of contacts in this region, creating a more diverse set of pathways between bundle helices. The macroH2A nucleosome displays an alternate route in which pathways instead access the docking domain region via protein-DNA interactions. The H2A.Z system uniquely passes information along the H2B α2 helix of the opposing dimer. Pathways in H2A.Z also access the protein-DNA type route of macroH2A and the four helix bundle route of the canonical and L1-Mutant systems, but at a reduced frequency. 9 Figure 6: The 500 sub-optimal pathways between the L1 loops and symmetrically associated DNA entry for the (a) canonical NCP, (b) L1-Mutant, (c) macroH2A, and (d) H2A.Z projected on simulation snapshots. Also shown is the histogram of pathway distances (e). The L1 and DNA sites are represented as blue spheres, and the pathways are outlined in red with the wider pathways representing those of shorter "distance." Pathways in the variants are shorter, and thus stronger, than in the canonical NCP. 10 Canon Macro L1-Mutant H2A.Z 0.7 0.8 0.9 1.0 1.1 1.2 Distance of Path 40 35 30 25 20 15 10 0.6 05 Number of Paths (e) (b) (c) (a) (d) PTM Targets are Located at Allosteric Hotspots Beyond these specific pathways, dynamic networks exist throughout the NCP. To discern the importance of individual residues on these global networks, the edge-betweenness centrality of nucleosome residues was computed (Figure S8). 41 In the canonical NCP, a majority of the optimal pathways rely heavily on the DNA basepairs and neighboring histone tail lysine and arginine residues to propagate communication throughout the system. In H2A.Z, an increased number of shortest pathways access the L1 loops and the H4 α2 helix, but there remains a heavy reliance on the DNA and histone tails. In the L1-Mutant and macroH2A nucleosomes, dynamic traffic to the four-helix bundle increases. Furthermore, pathways in these systems access L1 residues more frequently than any other region. Interestingly, we find that residues with the highest edge-betweenness scores are more likely to be the sites of post-translational modifications (PTMs). Based on the distribution of centrality scores, we classify residues in the upper tenth percentile as "hotspots" for communication (see supplemental). A comparison of known PTM sites with these allosteric hotspots indicates that PTMs are enriched at these locations, with an enrichment factor of 254% (p-value of 0.0155). When we compare our "hot spot analysis" with known PTM sites, 51 -- 53 we observe a significant population of PTM targets (Figure 7). While PTMs in the histone core are identified more frequently than those in the tails, the most significant subset contains PTMs that have been implicated in affecting mononucleosome stabilities ("monoNCP PTMs"). 51 In relation to the types of PTMs, methylation sites are linked with allosteric hotspots more frequently than phosphorylations or acetylations, likely due to their presence at DNA entry/exit sites and between turns of superhelical DNA. Figure 7: (a) Receiver Operator Curves (ROC) for subsets of PTM sites in the canonical nucleo- some as identified by edge-betweenness centrality ranking. The largest enrichment can be seen in the PTM subset of monoNCP altering PTMs. The core PTMs are also more frequently identified than the tails. (b) ROC for monoNCP altering PTMs across the variant systems. The canonical and H2A.Z systems are shown to depend greater on monoNCP altering PTMs than the L1-Mutant and macroH2A systems for distributing dynamic information. An overall correlation between allosteric hotspots and PTM locations is maintained in the nucleosome variants, however the specific details differ between the systems (Table S2). For example, all four systems show the importance of PTM sites in the H3 histone near 11 0.0 0.0 0.2 0.4 0.6 0.8 non-PTMs Ide ntifie d (%) 1.0 0.0 0.0 0.2 0.4 0.6 0.8 non-PTMs Ide ntifie d (%) 1.0 Canonical L1-Mutant m acroH2A H2A.Z (b) 1.0 0.8 0.6 0.4 0.2 PTMs Identified (%) All PTMs Core PTMs Tail PTMs MonoNCP Alte ring PTMs (a) 1.0 0.8 0.6 0.4 0.2 PTMs Identified (%) the DNA extremities, while histone H4 monoNCP PTM sites in the four-helix bundle are accessed more frequently in variant networks. In general, the canonical system displays the greatest reliance on monoNCP PTMs, then the H2A.Z nucleosome, and finally the L1-Mutant and macroH2A systems, respectively. Structural Stability in Variant Nucleosomes To quantify global NCP dynamics, a full correlation analysis (FCA) was performed on the Cα atoms of the histone core α-helices. 54 Two of the dominant motions identified in this analysis corresponded to the nucle- osome opening motions described by Bohm et al. 24 Projections into the phase-space described by these and other FCA modes showed that all nucleosome systems had similar global dynamics on the hundreds of nanoseconds timescale (see Supplemental Figure S6). However, given that the dynamics of nucleosome opening likely occur on the millisecond timescale, our simulations are far too short to effectively explore the effects of histone variants on large-scale NCP dynam- ics. In contrast to the nucleosome opening motions, the DNA end-to-end separation distance does depend on the identity of the H2A histone (Figure 8). The sampling in the canonical system can be divided into two states: the prominent "compact" state centered around 67 A and the "open" state centered at 71 A. The L1-mutant sampled both the open and closed states, however the percentage of time spent in the open state was reduced from 14% to 11% of the sim- ulation. Both the macroH2A and H2A.Z systems only sampled the closed state. Figure 8: Distance populations for DNA end-to-end spread. The canoni- cal system exists in two states: one cen- tered around 67 A ("compact") and one centered around 71 A ("open"). While the L1-Mutant samples both states, the amount of time spent in the open state is drastically reduced. The macroH2A and H2A.Z nucleosomes exist only in the compact state. Fits are represented in dotted lines. Results of an MM/GBSA analysis indicate that the overall DNA binding energetics are also altered by H2A variants. The DNA binding affinities to the L1-Mutant and macroH2A octamers were 31.0 ± 9.1 kcal/mol and 5.7 ± 9.7 more favorable than binding to the canonical NCP (Table 3). There are two primary contributors to this shift: direct interactions with the L1 loops (ƊƊGL1) and the changes in the DNA configuration (ƊƊGDNA). In the L1-mutant and macroH2A systems, ƊƊGL1 was largely a result of removing the negatively charged Glu41 from the canonical loop and introduction of the Lys40-DNA interaction, which combine for an increase in binding free energy on the order of 10 kcal/mol. The reorientation of the dimers in the macroH2A-like systems also influences a favorable shift in DNA conformation relative to the canonical system (ƊƊGDNA = -9.3 ± 5.6 kcal/mol and -17.0 ± 1.8 kcal/mol for the L1-Mutant and macroH2A nucleosomes, respectively). However, in macroH2A a number of small shifts in the remainder of the NCP oppose binding and therefore make it more comparable to the canonical system, which does not occur in the L1-mutant NCP. The H2A.Z nucleosome does not exhibit the same favorability for DNA binding when compared 12 macroH2A NCP H2A.Z NCP Canonical NCP L1-Mutant NCP 0.30 0.25 0.20 0.10 0.05 0.15 P(r) 0.00 55 60 65 70 Distance ( ) 75 80 Table 3: MM/GBSA calculated binding energies for DNA binding affinity to the histone core in each of the NCP systems. ƊƊG's are referenced against the canonical NCP. The L1 loop sequence and DNA conformations of the variant structures contribute significantly toward favorable binding of DNA in the macroH2A-like systems, relative the canonical NCP. All values are reported in kcal/mol. System Canonical NCP ƊGbinding -428.6 ± 5.6 -459.6 ± 7.2 macroH2A NCP -434.5 ± 7.9 -423.0 ± 6.9 L1 Mutant H2A.Z NCP ƊƊGbinding -- -- -31.0 ± 9.1 -5.7 ± 9.7 5.6 ± 8.9 ƊƊGL1 -- -- -8.7 ± 0.1 -10.7 ± 0.1 -5.8 ± 0.1 ƊƊGDNA -- -- -9.3 ± 5.6 -17.0 ± 1.8 5.8 ± 5.4 -- -- ƊGassembly ƊƊGassembly -618.6 ± 5.8 -679.6 ± 7.5 -637.1 ± 7.9 -616.3 ± 6.9 -60.0 ± 9.5 -18.5 ± 9.8 2.3 ± 9.0 to the canonical octamer, but instead shows a disfavoring shift of 5.6 ± 8.9 kcal/mol. The removal of the negative charge on Glu41 creates a favorable shift of 5.8 ± 0.1 kcal/mol in ƊƊGL1, but this is balanced by the nearly identical free energy penalty in the DNA rearrangement term ƊƊGDNA. Similarly, another MM/GBSA analysis revealed that macroH2A variants modify the en- ergetics of complex assembly. The ƊGassembly of the L1-Mutant and macroH2A nucleosomes were -60.0 ± 9.5 and -18.5 ± 9.8 kcal/mol more favorable than the canonical system. The favorability in the macroH2A-like systems is a result of favorable DNA binding coupled with stronger protein-protein interactions. The ƊGassembly of H2A.Z was in agreement with that of the canonical nucleosome (Table 3). Discussion The simulations and analysis presented here detail a series of mechanisms by which the histone variants macroH2A and H2A.Z influence the dynamics of the nucleosome core par- ticle. The subtle structural rearrangements these variants cause leverage the tightly packed nature of the histone core to influence the global energetics and dynamics of the complex, thus influencing gene expression. Dynamic effects appear to be particularly important, as they allow for the propagation of information through allosteric networks that span large distances. Although our simulations are only able to probe the sub-s timescale, the dy- namic differences observed at the dimer-tetramer and DNA/histone interfaces will likely be amplified on the ms timescale and result in these variants having altered nucleosome opening and DNA breathing motions. These results also offer new insights into biochemical experiments that probed the mech- anism of macroH2A. For example, Nusinow et al. showed that the L1-mutant is enriched in the inactive female X chromosome at nearly the same rate as the complete histone-domain of macroH2A. 11 Point mutations demonstrated that enrichment was significantly increased by the two mutations that introduce additional bulk into the L1 loops, N38H and E41Y, whereas it was decreased by the Y39P mutation, which decreases the size of the L1 loop. Based on our results, we believe that larger sidechains may help encourage the α2 "bulging" motion observed in each of the variant simulations, and therefore make the NCP more variant-like. In another set of experiments, Chakravarthy et al. demonstrated that mutations to the L1 loops modulate the salt-dependent stability of the histone octamer. 21 They showed that 13 in both the L1-mutant and macroH2A-containing system, the histone octamer is stable down to 0.5 M NaCl, whereas the canonical and H2A.Z-containing structures dissociate into dimer and tetramer constituents in solutions below 1.1 M. In agreement with this, we observe a significantly more favorable interaction between the H2A L1 loops in the variant structures than in the canonical structure. Since the L1-L1 interface is the only location of dimer-dimer interaction, stability in this region translates to octamer stability. The mechanisms of H2A.Z remain more elusive. Stability studies have been non-conclusive as some indicate that H2A.Z enhances stability, 55 while others suggest that it destabilizes the nucleosome. Our simulations show H2A.Z nucleosome stabilities that are in agreement with the canonical system, despite their differing dynamics. These systems were constructed with identical sequences, except for H2A composition. Therefore, our findings support a mecha- nism which suggests that H2A.Z by itself has little-to-no effect on NCP stability. Instead, H2A.Z presence may be combined with other factors - such as PTMs or H3 variant presence - in order to alter particle stability. 22,56 Furthermore, the altered dynamics and locations of allosteric networks and hotspots between H2A.Z and canonical nucleosomes may result in different responses to these chromatin remodeling factors. The dimer reordering may also act to recruit transcriptional machinery to chromatin possessing large populations of H2A.Z, such as transcriptional starting sites. Finally, we present a comprehensive analysis of the dynamic networks in the nucleosome. We observe that these networks are strongly affected by the dynamics of the L1 loops which are allosterically linked to a wide number of important regions in the nucleosome core. Using only small changes in their structure, variants are able to modify these networks to affect the function of the NCP. We hypothesize that this is a general mechanism that other chromatin remodeling factors may also utilize. For example, the finding that PTMs are enriched at residues with increased allosteric activity suggests that these perturbations may take advantage of dynamic networks to amplify their effects on chromatin and influence global NCP dynamics. In addition, by altering these networks, variants may be able to tune the responses of nucleosomes to specific PTMs. Future work to study the disparate effects of chromatin remodeling factors on dynamics in the nucleosome is required to fully understand the mechanisms of in vivo gene expression and regulation. Supplementary Information Complete methods; Tables S1-S2; Figures S1-S23. Acknowledgement The authors thank S. Chakravarthy for valuable discussions concerning the work presented here. Research reported in this publication was supported by the National Institute of General Medical Sciences of the National Institutes of Health [grant no. R15GM114758]. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health. This work used the Extreme Science and Engineering Discovery Environment (XSEDE), which is supported by National Science Foundation grant number ACI-1053575. In addition, this research used resources of the 14 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. References (1) Luger, K., Mader,A.W., Richmond,R.K., Sargent,D.F., Richmond,T.J. Crystal structure of the nucleosome core particle at 2.8 A resolution. Nature 1997, 389, 251 -- 260. (2) Davey,C.A., Sargent,D.F., Luger,K., Maeder,A.W. and Richmond,T.J. Solvent Mediated In- teractions in the Structure of the Nucleosome Core Particle at 1.9A Resolution. J. Mol. Biol. 2002, 319, 1097 -- 1113. (3) Chakravarthy,S., Gundimella,S.K., Caron,C., Perche,P.Y., Pehrson,J.R., Khochbin,S. and Luger,K. Structural characterization of the histone variant macroH2A. Mol. Cell. Biol. 2005, 25, 7616 -- 7624. (4) Suto,R.K., Clarkson,M.J., Tremethick,D.J. and Luger,K. Crystal structure of a nucleosome core particle containing the variant histone H2A.Z. Nat. Struct. Biol. 2000, 7, 1121 -- 1124. (5) Osley,M.A., Fleming,A.B. and Kao,C.F. Histone ubiquitylation and the regulation of tran- scription. Results Probl Cell Differ 2006, 41, 47 -- 75. (6) Luger,K., Dechassa,M.L. and Tremethick,D.J. New insights into nucleosome and chromatin structure: an ordered state or a disordered affair? Nat. Rev. Mol. Cell Biol. 2012, 13, 436 -- 447. (7) Bowman,G.D. and Poirier,M.G. Post-translational modifications of histones that influence nucleosome dynamics. Chem. Rev. 2015, 115, 2274 -- 2295. (8) Zhang,Y. and Reinberg,D. Transcription regulation by histone methylation: interplay between different covalent modifications of the core histone tails. Genes and Devel. 2001, 15, 2343 -- 2360. (9) Bonisch,C., Hake,S.B. Histone H2A variants in nucleosomes and chromatin: more or less stable? Nucl. Acids Res. 2012, 40, 10719 -- 10741. (10) Shaytan,A.K., Landsman,D. and Panchenko,A.R. Nucleosome adaptability conferred by se- quence and structural variations in histone H2A-H2B dimers. Curr. Opin. Struct. Biol. 2015, 32C, 48 -- 57. (11) Nusinow,D.A., Sharp,J.A., Morris,A., Salas,S., Plath,K. and Panning,B. The histone domain of macroH2A1 contains several dispersed elements that are each sufficient to direct enrichment on the inactive X chromosome. J. Mol. Biol. 2007, 371, 11 -- 18. (12) Chakravarthy,S., Bao,Y., Roberts,V.A., Tremethick,D. and Luger,K. Structural characteriza- tion of histone H2A variants. Cold Spring Harb. Symp. Quant. Biol. 2004, 69, 227 -- 234. (13) Pehrson,J.R. and Fried,V.A. MacroH2A, a core histone containing a large nonhistone region. Science 1992, 257, 1398 -- 1400. (14) Choo,J.H., Kim,J.D., Chung,J.H., Stubbs,L. and Kim,J. Allele-specific deposition of macroH2A1 in imprinting control regions. Hum. Mol. Gen. 2006, 15, 717 -- 724. 15 (15) Buschbeck,M., Uribesalgo,I., Wibowo,I., Rue,P., Martin,D., Gutierrez,A., Morey,L., Guigo,R., Lopez-Schier,H. and Di Croce,L. The histone variant macroH2A is an epigenetic regulator of key developmental genes. Nat. Struct. Mol. Biol. 2009, 16, 1074 -- 1079. (16) Chadwick,B.P. and Willard,H.F. Cell cycle-dependent localization of macroH2A in chromatin of the inactive X chromosome. J. Cell Biol. 2002, 157, 1113 -- 1123. (17) Costanzi,C., Stein,P., Worrad,D.M., Schultz,R.M. and Pehrson,J.R. Histone macroH2A1 is concentrated in the inactive X chromosome of female preimplantation mouse embryos. Devel- opment 2000, 127, 2283 -- 2289. (18) Guillemette,B., Bataille,A.R., Gevry,N., Adam,M., Blanchette,M., Robert,F. and Gau- dreau,L. Variant histone H2A.Z is globally localized to the promoters of inactive yeast genes and regulates nucleosome positioning. PLoS Biol. 2005, 3, e384. (19) Abbott,D.W., Ivanova,V.S., Wang,X., Bonner,W.M. and Ausio,J. Characterization of the stability and folding of H2A.Z chromatin particles: implications for transcriptional activation. J. Biol. Chem. 2001, 276, 41945 -- 41949. (20) Chakravarthy,S., Patel,A. and Bowman,G.D. The basic linker of macroH2A stabilizes DNA at the entry/exit site of the nucleosome. Nucl. Acids Res. 2012, 40, 8285 -- 8295. (21) Chakravarthy,S. and Luger,K. The histone variant macro-H2A preferentially forms "hybrid nucleosomes". J. Biol. Chem. 2006, 281, 25522 -- 25531. (22) Marques,M., Laflamme,L., Gervais,A.L. and Gaudreau,L. Reconciling the positive and nega- tive roles of histone H2A.Z in gene transcription. Epigenetics 2010, 5, 267 -- 272. (23) Roccatano,D., Barthel,A. and Zacharias,M. Structural flexibility of the nucleosome core par- ticle at atomic resolution studied by molecular dynamics simulation. Biopolymers 2007, 85, 407 -- 421. (24) Bohm,V., Hieb,A.R., Andrews,A.J., Gansen,A., Rocker,A., Toth,K., Luger,K. and Lan- gowski,J. Nucleosome accessibility governed by the dimer/tetramer interface. Nucl. Acids Res. 2011, 39, 3093 -- 3102. (25) Hornak,V., Abel,R., Okur,A., Strockbine,B., Roitberg,A. and Simmerling,C. Comparison of multiple Amber force fields and development of improved protein backbone parameters. Pro- teins 2006, 65, 712 -- 725. (26) Phillips,J.C., Braun,R., Wang,W., Gumbart,J., Tajkhorshid,E., Villa,E., Chipot,C., Skeel,R.D., Kale,L. and Schulten,K. Scalable molecular dynamics with NAMD. J. Comp. Chem. 2005, 26, 1781 -- 1802. (27) Joung,I.S. and Cheatham,T.E. Molecular dynamics simulations of the dynamic and energetic properties of alkali and halide ions using water-model-specific ion parameters. J. Phys. Chem.B 2009, 113, 13279 -- 13290. (28) Ryckaert,J.-P., Ciccotti,G. and Berendsen,H.J. Numerical integration of the cartesian equa- tions of motion of a system with constraints: molecular dynamics of n-alkanes. J. Comp. Phys. 1977, 23, 327 -- 341. 16 (29) Darden,T., York,D. and Pedersen,L. Particle mesh Ewald: An N log (N) method for Ewald sums in large systems. J. Chem. Phys. 1993, 98, 10089 -- 10092. (30) Feller,S.E., Zhang,Y.H., Pastor,R.W. and Brooks,B.R. Constant-pressure molecular-dynamics simulation - the langevin piston method. J. Chem. Phys. 1995, 103, 4613 -- 4621. (31) Olsson,M. H.M., Søndergaard,C.R., Rostkowski,M. and Jensen,J.H. PROPKA3: Consistent Treatment of Internal and Surface Residues in Empirical pKa Predictions. J. Chem. Theor. Comp. 2011, 7, 525 -- 537. (32) Humphrey,W., Dalke,A. and Schulten,K. VMD -- Visual Molecular Dynamics. J. Mol. Graph- ics 1996, 14, 33 -- 38. (33) Stone,J. An Efficient Library for Parallel Ray Tracing and Animation. M.Sc. thesis, Computer Science Department, University of Missouri-Rolla, 1998. (34) Schrodinger, LLC, The PyMOL Molecular Graphics System. 2010, http://www.pymol.org. (35) McClendon,C.L., Hua,L., Barreiro,A. and Jacobson,M.P. Comparing Conformational Ensem- bles Using the Kullback-Leibler Divergence Expansion. J. Chem. Theor. Comp. 2012, 8, 2115 -- 2126. (36) Scarabelli,G. and Grant,B.J. Kinesin-5 allosteric inhibitors uncouple the dynamics of nu- cleotide, microtubule, and neck-linker binding sites. Biophys. J. 2014, 107, 2204 -- 2213. (37) Lange,O.F. and Grubmuller,H. Generalized correlation for biomolecular dynamics. Proteins 2006, 62, 1053 -- 1061. (38) Michaud-Agrawal,N., Denning,E., Woolf,T. and Beckstein,O. MDAnalysis: A Toolkit for the Analysis of Molecular Dynamics Simulations. J. Comp. Chem. 2011, 32, 2319 -- 2327. (39) Van Wart,A.T., Durrant,J., Votapka,L. and Amaro,R.E. Weighted Implementation of Subop- timal Paths (WISP): An Optimized Algorithm and Tool for Dynamical Network Analysis. J. Chem. Theor. Comp. 2014, 10, 511 -- 517. (40) Hagberg,A.A., Schult,D.A. and Swart,P.J. Exploring network structure, dynamics, and func- tion using NetworkX. Proceedings of the 7th Python in Science Conference (SciPy2008). Pasadena, CA USA, 2008, pp 11 -- 15. (41) Brandes,U. A faster algorithm for betweenness centrality. J. Math. Sociol. 2001, 25, 163 -- 177. (42) Rice,J.A. Mathematical Statistics and Data Analysis, 3rd ed., Duxbury Press: Belmont, CA, 2007. (43) Roe,D.R. and Cheatham,T.E. PTRAJ and CPPTRAJ: Software for Processing and Analysis of Molecular Dynamics Trajectory Data. J. Chem. Theor. Comp. 2013, 9, 3084 -- 3095. (44) Cuervo,A., Dans,P.D., Carrascosa,J.L., Orozco,M., Gomila,G. and Fumagalli,L. Direct mea- surement of the dielectric polarization properties of DNA. Proc. Natl. Acad. Sci. USA 2014, 111, E3624 -- 3630. (45) Schutz,C.N. and Warshel,A. What are the dielectric "constants" of proteins and how to vali- date electrostatic models? Proteins 2001, 44, 400 -- 417. 17 (46) Miller,B., McGee,T., Swails,J., Homeyer,N., Gohlke,H. and Roitberg,A. MMPBSA.py: An Efficient Program for End-State Free Energy Calculations. J. Chem. Theor. Comp. 2012, 8, 3314 -- 3321. (47) Onufriev,A., Bashford,D. and Case,D.A. Exploring protein native states and large-scale con- formational changes with a modified generalized born model. Proteins 2004, 55, 383 -- 394. (48) Shirts,M.R. and Chodera,J.D. Statistically optimal analysis of samples from multiple equilib- rium states. J. Chem. Phys. 2008, 129, 124105. (49) Biswas,M., Langowski,J., Bishop,T.C. Atomistic simulations of nucleosomes. WIREs Comput. Mol. Sci. 2013, 3, 378 -- 392. (50) Biswas,M., Voltz,K., Smith,J.C. and Langowski,J. Role of histone tails in structural stability of the nucleosome. PLoS Comput. Biol. 2011, 7, e1002279. (51) Bowman,G.D. and Poirier,M.G. Post-translational modifications of histones that influence nucleosome dynamics. Chem. Rev. 2015, 115, 2274 -- 2295. (52) Khare,S.P., Habib,F., Sharma,R., Gadewal,N., Gupta,S. and Galande,S. HIstome -- a relational knowledgebase of human histone proteins and histone modifying enzymes. Nucl. Acids Res. 2012, 40, D337 -- 342. (53) Hornbeck,P.V., Kornhauser,J.M., Tkachev,S., Zhang,B., Skrzypek,E., Murray,B., Latham,V. and Sullivan,M. PhosphoSitePlus: a comprehensive resource for investigating the structure and function of experimentally determined post-translational modifications in man and mouse. Nucl. Acids Res. 2012, 40, D261 -- 270. (54) Lange,O.F. and Grubmuller,H. Full correlation analysis of conformational protein dynamics. Proteins 2008, 70, 1294 -- 1312. (55) Hoch,D.A., Stratton,J.J. and Gloss,L.M. Protein-protein Frster resonance energy transfer analysis of nucleosome core particles containing H2A and H2A.Z. J. Mol. Biol. 2007, 371, 971 -- 988. (56) Jin,C. and Felsenfeld,G. Nucleosome stability mediated by histone variants H3.3 and H2A.Z. Genes and Devel. 2007, 21, 1519 -- 1529. 18 Supporting information for: Effects of macroH2A and H2A.Z on nucleosome structure and dynamics as elucidated by molecular dynamics simulations Samuel Bowerman and Jeff Wereszczynski∗ Department of Physics and Center for Molecular Study of Condensed Soft Matter, Illinois Institute of Technology, 3440 S Dearborn St., Chicago, IL 60616, USA E-mail: [email protected] Methods System Construction and Simulation Details The canonical nucleosome was initiated from the crystal structure of Daveys et al. (PDB ID: 1KX5). S1 The crystallographic Mn2+ were replaced by physiological Mg2+. Additional Mg2+ ions were added to fill symmetrically suggested voids. The crystallographic waters were also maintained. The L1-Mutant structure was then created from the canonical one by mutating the 38NYAE41 H2A L1 loops to 38HPKY41 sequence of macroH2A. The mutation was done using VMD. The macroH2A system was initialized from the crystal structure solved by Chakravarthy et al. (PDB ID: 1U35). S2 The missing tail segments were constructed using the canonical structure as a reference. Strong similarities in DNA arrangement - particularly at DNA-protein binding sites (Figure S1) - allowed for the 146 basepairs of DNA from the crystal structure to be replaced by the 147 bp (plus Mg2+) of the 1KX5 structure, and H3 residues were mutated to match the sequence of the 1KX5 system. These actions were taken to ensure that differences between the systems were attributable only to H2A sequence divergence. The H2A.Z system was constructed analogously, using the crystal structure of ∗To whom correspondence should be addressed S1 Suto et al. (PDB ID: 1F66). S3 Therefore, each system was composed of 147 palindromic basepairs of α-satellite DNA wrapped around a histone core of Xenopus laevis H3, H4, and H2B with human H2A histones and variants. Histidine states were assigned using PROPKA and the interactive H-Bond Optimizer of the Desmond-Schrodinger package. S4 Each system was simulated three times. Each simulation underwent 10,000 steps of geometric minimization (5,000 steps with protein heavy atoms harmonically restrained by a force constant of 10 kcal/mol/A2 and 5,000 steps without restraints). Heating was done by gradually raising the temperature from 10 to 300 K over 6 ps of simulation in the NVT ensemble. During heating, protein heavy atoms were harmonically restrained with a force constant of 10 kcal/mol/A2. The restraints were then gradually released over 600ps in the NPT ensemble. S5 Each simulation was then conducted for an additional 250ns in the NPT ensemble using a Langevin piston with a 100 fs period and collision frequency of 3 ps-1. The SHAKE algorithm was used to allow for a 2 fs timestep, and long-range electrostatics were calculated using the particle mesh Ewald method. S6,S7 Short range interactions were calculated with a 10 A cutoff, where a switching function was applied at 8 A. It was observed that ∼50ns was required for system equilibration, and so ∼200ns of production data was obtained from each simulation (600ns per system). Allosteric Calculations Residue correlations were calculated using the "largest linear mutual information" method. S8,S9 In this method, the linear mutual information is calculated between all heavy atoms in the system. The residue-wise mutual information values were converted to a Pearson correla- tion coefficient-like value by ri,j = [1 − e(−2Ii,j /3)]1/2, where Ii,j is the largest linear mutual information between any two atoms of the residues i and j. Contact maps were produced in-house using the MDAnalysis package. S10 Two residues were considered to be in contact if their Cα (protein) or C1' (nucleic) atoms were within 10 A for 70% of the configurations. Using the predefined correlation matrices and this contact map, a NetworkX edgelist was formed. S11 The length of each edge was defined by Di,j = −log(ri,j), where ri,j is the correlation value between residues i and j. The optimal paths were calculated using the NetworkX implementation of Dijkstra's algorithm, and the suboptimal paths were determined using Yen's K-Shortest Paths algorithm. The Kullbach-Leibler Divergence of dihedral angles were calculated using the method of McClendon. S12 In this method, each simulation was separated into three blocks (9 blocks per system) with 31,666 configurations (roughly 63.3 ns) per block. Histogram widths were 15 degrees. The Kullbach-Leibler Divergence values were calculated for each variant by using the populations of the canonical system as a reference set. Bootstrapping techniques were employed to calculate the self-divergence of the canonical system. For any residue in a variant system whose divergence value was below this self-divergence threshold, the KL-Divergence value for that residue was set to 0. Edge-betweenness Centrality The importance of a node in a communication network can be defined by its edge-betweenness centrality. S13 In this method, the "shortest" correlation pathway between all residue-pairs is S2 (3) calculated. A residue's edge-betweenness centrality is then defined as the number of shortest paths in which the residue appears: (1) xi (s, t) Xs6=t6=i 1 N C(i) = where N is the total number of paths and xi(s,t) is either 0 (residue i does not exist in path between residues s and t) or 1 (residue i does exist in said path). For visualization purposes, centrality values are normalized such that the minimum centrality is 0 and the maximum centrality is 1, according to the formula: Cnorm (i) = C(i) − Cmin Cmax − Cmin (2) where Cmax and Cmin are the maximum and minimum centrality values in the network. From the plot of normalized centrality value vs percentile (Figure S9), we observe that the difference in centrality between percentile increments is not constant but is large at the upper and lower quartiles and steady in the interquartile region. The inflection point of the upper quartile exists near the tenth percentile, so we have chosen this location as our cut-off for defining allosteric "hotspots." We were able to identify several well-known post-translational modifications (PTMs) in the canonical nucleosome by our betweenness centrality measurement. S14 -- S16 In the canonical system, we observe 6 of 23 mononucleosome altering PTM sites (monoNCP PTMs) in the upper tenth percentile and 12 of 23 in the upper quartile. The significance of observing this subset of residues in each percentile was tested by calculating the pmf of a hypergeometric distribution, S17 pmf (x = k) = (cid:0)K k(cid:1)(cid:0)N−K n−k(cid:1) (cid:0)N n(cid:1) where N is the total number of protein residues (487), n is the percentile population size (n=49 for the upper tenth, and n=122 for the upper quartile), K is the total number of monoNCP PTM residues (23), and k is the number of observed monoNCP PTM sites (k=6 for the upper tenth, and k=12 for the upper quartile). Using these values, the upper tenth percentile observation has a p-value of 0.0155, and the upper quartile p-value is 0.00288. Therefore, the observation that monoNCP PTMs are located at allosteric hotspots is statis- tically significant. Furthermore, we can quantify the presence of monoNCP PTMs at allosteric hotspots by calculating the enrichment factor (EF) of monoNCP PTMs over random selection, (4) N n k K EF = where the variables have the same meaning as for the hypergeometric distribution. We then calculate an EF of 2.54 for monoNCP PTM presence at allosteric hotspots. A plot of EF vs centrality percentile can be found in Figure S11. S3 References (S1) Davey,C.A., Sargent,D.F.; Luger,K., Maeder,A.W. and Richmond,T.J. Solvent Mediated Interactions in the Structure of the Nucleosome Core Particle at 1.9A Resolution. J. Mol. Biol. 2002, 319, 1097 -- 1113. (S2) Chakravarthy,S., Gundimella,S.K., Caron,C., Perche,P.Y., Pehrson,J.R., Khochbin,S. and Luger,K. Structural characterization of the histone variant macroH2A. Mol. Cell. Biol. 2005, 25, 7616 -- 7624. (S3) Suto,R.K., Clarkson,M.J., Tremethick,D.J. and Luger,K. Crystal structure of a nucleosome core particle containing the variant histone H2A.Z. Nat. Struct. Biol. 2000, 7, 1121 -- 1124. (S4) Olsson,M. H.M., Søndergaard,C.R., Rostkowski,M. and Jensen,J.H. PROPKA3: Consistent Treatment of Internal and Surface Residues in Empirical pKa Predictions. J. Chem. Theor. Comp. 2011, 7, 525 -- 537. (S5) Feller,S.E., Zhang,Y.H., Pastor,R.W. and Brooks,B.R. Constant-pressure molecular- dynamics simulation - the langevin piston method. J. Chem. Phys. 1995, 103, 4613 -- 4621. (S6) Ryckaert,J.P., Ciccotti,G. and Berendsen,H.J. Numerical integration of the cartesian equa- tions of motion of a system with constraints: molecular dynamics of n-alkanes. J. Comp. Phys. 1977, 23, 327 -- 341. (S7) Darden,T., York,D. and Pedersen,L. Particle mesh Ewald: An N log (N) method for Ewald sums in large systems. J. Chem. Phys. 1993, 98, 10089 -- 10092. (S8) Lange,O.F. and Grubmuller,H. Generalized correlation for biomolecular dynamics. Proteins 2006, 62, 1053 -- 1061. (S9) Scarabelli,G. and Grant,B.J. Kinesin-5 allosteric inhibitors uncouple the dynamics of nu- cleotide, microtubule, and neck-linker binding sites. Biophys. J. 2014, 107, 2204 -- 2213. (S10) Michaud-Agrawal,N., Denning,E., Woolf,T. and Beckstein,O. MDAnalysis: A Toolkit for the Analysis of Molecular Dynamics Simulations. J. Comp. Chem. 2011, 32, 2319 -- 2327. (S11) Hagberg,A.A., Schult,D.A. and Swart,P.J. Exploring network structure, dynamics, and func- tion using NetworkX. Proceedings of the 7th Python in Science Conference (SciPy2008). Pasadena, CA USA, 2008, pp 11 -- 15. (S12) McClendon,C.L., Hua,L., Barreiro,A. and Jacobson,M.P. Comparing Conformational En- sembles Using the Kullback-Leibler Divergence Expansion. J. Chem. Theor. Comp. 2012, 8, 2115 -- 2126. (S13) Brandes,U. A faster algorithm for betweenness centrality. J. Math. Sociol. 2001, 25, 163 -- 177. (S14) Bowman,G.D. and Poirier,M.G. Post-translational modifications of histones that influence nucleosome dynamics. Chem. Rev. 2015, 115, 2274 -- 2295. (S15) Khare,S.P., Habib,F., Sharma,R., Gadewal,N., Gupta,S., and Galande,S. HIstome -- a rela- tional knowledgebase of human histone proteins and histone modifying enzymes. Nucl. Acids Res. 2012, 40, D337 -- 342. S4 (S16) Hornbeck,P.V., Kornhauser,J.M., Tkachev,S., Zhang,B., Skrzypek,E., Murray,B., Latham,V. and Sullivan,M. PhosphoSitePlus: a comprehensive resource for investigating the struc- ture and function of experimentally determined post-translational modifications in man and mouse. Nucl. Acids Res. 2012, 40, D261 -- 270. (S17) Rice,J.A. Mathematical Statistics and Data Analysis, 3rd ed., Duxbury Press: Belmont, CA, 2007. S5 Figure S1: Comparison of crystallographic DNA arrangment in the canonical (blue), macroH2A (red), and H2A.Z (grey) nucleosomes. As expected, the DNA-histone binding sites show strong agreement in coordination between the three structures. Table S1: Hydrogen bonding at key locations in the nucleosome. For the H-Bonds formed with the DNA, the occupancy of each bond is given. For the other interactions, the average number of hydrogen bonds between each group in a given frame is listed. System Canonical NCP L1 Mutant macroH2A NCP H2A.Z NCP H2A R29-DNA H2B S33-DNA 63% 70% 52% 73% 40% 60% 30% 85% L1-L1 0.5 ± 0.1 0.4 ± 0.1 0.2 ± 0.1 0.1 ± 0.1 Dimer-tetramer 14.8 ± 1.1 16.5 ± 1.0 14.7 ± 0.8 14.4 ± 1.1 S6 Figure S2: Backbone RMSD of three canonical simulations. The simulations were fit to the histone core backbone, and the RMSD calculations were done on the DNA and histone core backbone atoms, excluding tail residues. Figure S3: Backbone RMSD of three L1-Mutant simulations. The simulations were fit to the his- tone core backbone, and the RMSD calculations were done on the DNA and histone core backbone atoms, excluding tail residues. S7 Simulation 1 Simulation 2 Simulation 3 50 100 Time (ns) 150 200 250 3.5 3.0 2.5 2.0 1.5 1.0 0.5 0.0 0 RMSD (eA) Simulation 1 Simulation 2 Simulation 3 50 100 Time (ns) 150 200 250 3.5 3.0 2.5 2.0 1.5 1.0 0.5 0.0 0 RMSD (eA) Figure S4: Backbone RMSD of three macroH2A simulations. The simulations were fit to the his- tone core backbone, and the RMSD calculations were done on the DNA and histone core backbone atoms, excluding tail residues. Figure S5: Backbone RMSD of three H2A.Z simulations. The simulations were fit to the histone core backbone, and the RMSD calculations were done on the DNA and histone core backbone atoms, excluding tail residues. S8 Simulation 1 Simulation 2 Simulation 3 50 100 Time (ns) 150 200 250 4.0 3.5 3.0 2.5 2.0 1.5 1.0 0.5 0.0 0 RMSD (eA) Simulation 1 Simulation 2 Simulation 3 50 100 Time (ns) 150 200 250 4.5 4.0 3.5 3.0 2.5 2.0 1.5 1.0 0.5 0.0 0 RMSD (eA) Figure S6: Dimer dissociation phase space for the nucleosome systems. All systems sample the same region of this space, which suggests that H2A composition has little effect on dimer dissociation in the hundreds of nanoseconds timescale. For reference, the dimer separation event occurs on the millisecond timescale. S9 Figure S7: Pearsified Largest Linear Mutual Information matrices for (a) canonical, (b) L1- Mutant, (c) macroH2A, and (d) H2A.Z nucleosomes. The canonical NCP shows the weakest average correlation across the whole molecule, and the macroH2A variant shows the strongest. The L1-Mutant correlation strengths are similar to macroH2A, while the H2A.Z nucleosome shows correlations that are slightly above the levels of the canonical nucleosome. The steady increase in correlations within the variant systems is likely a result of favorable changes in interhistone interactions. S10 1:0 0:9 0:8 0:7 0:6 0:5 0:4 0:3 0:2 0:1 0:0 CorrelationCoefficient H2B H2A H4 H3 H2B H2A H4 H3 DNA H2B H2A H4 H3 H2B H2A H4 H3 DNA (B) 1200 1000 800 600 400 200 0 Residue (D) 1200 1000 800 600 400 200 0 Residue H2B H2A H4 H3 H2B H2A H4 H3 DNA H2B H2A H4 H3 H2B H2A H4 H3 DNA (A) 1200 1000 800 600 400 200 0 Residue (C) 1200 1000 800 600 400 200 0 Residue Figure S8: Edge-betweenness centrality for (a) canonical, (b) L1-Mutant, (c) H2A.Z, and (d) macroH2A nucleosomes. Brighter, wider regions represent locations that are accessed more fre- quently in the optimal communication pathways of each nucleosome system. The H2A L1-L1 interaction region in the L1-Mutant and macroH2A systems act as communication hubs for al- losteric networks in the nucleosome, whereas the canonical and H2A.Z nucleosomes rely heavier on DNA to propagate information. S11 (a) (c) (b) (d) Figure S9: Centrality scores vs percentile ranking (blue dots) for the canonical nucleosome. The spline fit is represented in red. The drastic change in centrality score in the upper quartile indicates that residues rely heavily on several key residues for information propagation. The inflection point of this trend is located at the tenth percentile. Figure S10: ROC plot for the canonical nucleosome with PTMs separated by modification type. Methylations exist most prevalently as allosteric hotspots, and acetylations are the least prevalent. The early enrichment of methylations is a result of their presence near DNA extremities and between superhelical DNA turns where pathways cross the symmetries. S12 0.0 0.2 0.4 0.6 Percentile 0.8 1.0 1.0 0.8 0.6 0.4 0.2 Centrality Score ALL PTMs Acetylations Methylations Phosphorylations 0.2 0.4 0.6 0.8 1.0 non-PTMs Identified (%) 1.0 0.8 0.6 0.4 0.2 PTMs Identified (%) 0.0 0.0 Figure S11: Enrichment Factor (blue) and monoNCP PTM identification percent (red) as func- tions of centrality percentile in the canonical nucleosome. We observe a strong degree of early enrichment for identifying monoNCP PTMs at allosteric hotspots. At the cutoff of the tenth percentile, we observe an EF of 2.54. Table S2: Protein-normalized centrality values for the 23 monoNCP PTM targets. The percentile rank of each value is also listed. The residues in the upper quartile are listed in bold. Those in the upper tenth percentile are italicized. PTM sites that have significantly smaller centrality values than neighboring sequence residues in the upper quartile are labeled by an () and the value of the neighbor is reported. PTM H3 K4me3 H3 K9ac H3 K14ac H3 K18ac H3 K23ac H3 K36me2,3 H3 Y41ph H3 R42me2a H3 T45ph H3 K56ac H3 S57ph H3 K64ac H3 K115ac H3 T118ph H3 K122ac H4 K16ac H4 S47ph* H4 K77ac H4 K79ac* H4 K91ac H4 R92me H2B K30ar H2B K123ub1 Canonical L1-Mutant macroH2A H2A.Z Centrality 0.05 0.13 0.20 0.27 0.34 0.47 0.46 0.51 0.29 0.17 0.15 0.15 0.15 0.37 0.23 0.21 0.25 0.18 0.31 0.12 0.21 0.94 0.00 Percentile Centrality 0.07 0.19 0.30 0.39 0.51 0.80 1.00 0.79 0.38 0.18 0.16 0.26 0.13 0.43 0.29 0.38 0.28 0.29 0.34 0.38 0.53 0.81 0.00 6.6 27.0 75.6 87.9 92.8 97.5 96.9 98.4 75.0 58.9 39.0 42.7 38.4 93.4 81.9 77.0 85.4 61.8 91.4 23.4 76.4 99.6 0.0 Percentile Centrality 0.05 0.14 0.23 0.30 0.38 0.45 0.67 0.37 0.22 0.22 0.22 0.27 0.06 0.14 0.13 0.16 0.19 0.15 0.56 0.57 0.81 0.26 0.00 6.3 22.6 51.0 73.9 87.3 97.3 100.0 96.7 71.1 20.1 17.9 38.0 11.7 81.9 47.4 70.0 43.9 45.2 63.8 70.0 88.5 97.7 0.0 Percentile Centrality 0.06 0.16 0.26 0.34 0.44 0.69 0.84 0.79 0.28 0.19 0.16 0.21 0.24 0.63 0.39 0.18 0.34 0.25 0.35 0.29 0.33 0.72 0.00 6.2 24.8 53.4 69.8 82.0 86.1 96.9 81.5 52.4 52.4 52.4 63.6 8.5 25.8 23.9 33.9 44.3 28.3 95.0 95.4 99.4 62.4 0.0 Percentile 6.7 27.5 61.2 78.4 88.9 97.7 99.8 99.2 67.0 35.6 28.6 46.1 53.9 96.5 82.7 34.2 78.2 58.0 79.8 68.9 77.6 98.8 0.0 S13 1.0 0.8 0.6 PTMs Identified (%) 0.4 0.2 10-2 Centrality Percentile 10-1 100 12 10 8 6 4 2 0 Enrichment Factor Figure S12: Sub-optimal pathways for L1 loop to DNA exit in the canonical nucleosome. S14 Figure S13: Sub-optimal pathways for L1 loop to DNA exit in the L1-Mutant nucleosome. S15 Figure S14: Sub-optimal pathways for L1 loop to DNA exit in the macroH2A nucleosome. S16 Figure S15: Sub-optimal pathways for L1 loop to DNA exit in the H2A.Z nucleosome. S17 Figure S16: Sub-optimal pathways for L1 loop to the associated docking domain in the canonical nucleosome. The opposing dimer has been removed to improve visualization clarity. S18 Figure S17: Sub-optimal pathways for L1 loop to the associated docking domain in the L1-Mutant nucleosome. The opposing dimer has been removed to improve visualization clarity. S19 Figure S18: Sub-optimal pathways for L1 loop to the associated docking domain in the macroH2A nucleosome. The opposing dimer has been removed to improve visualization clarity S20 Figure S19: Sub-optimal pathways for L1 loop to the associated docking domain in the H2A.Z nucleosome. The opposing dimer has been removed to improve visualization clarity. S21 Figure S20: Sub-optimal pathways for L1 loop to non-associated docking domain in the canonical nucleosome. The opposing dimer has been removed to improve visualization clarity. S22 Figure S21: Sub-optimal pathways for L1 loop to non-associated docking domain in the L1- Mutant nucleosome. The opposing dimer has been removed to improve visualization clarity. S23 Figure S22: Sub-optimal pathways for L1 loop to non-associated docking domain in the macroH2A nucleosome. The opposing dimer has been removed to improve visualization clarity. S24 Figure S23: Sub-optimal pathways for L1 loop to non-associated docking domain in the H2A.Z nucleosome. The opposing dimer has been reduced to only H2B α2 to improve visualization clarity. S25
1610.07774	1	1610	2016-10-25T08:06:15	Bursts of activity in collective cell migration	[ "physics.bio-ph", "cond-mat.stat-mech", "q-bio.CB" ]	Dense monolayers of living cells display intriguing relaxation dynamics, reminiscent of soft and glassy materials close to the jamming transition, and migrate collectively when space is available, as in wound healing or in cancer invasion. Here we show that collective cell migration occurs in bursts that are similar to those recorded in the propagation of cracks, fluid fronts in porous media and ferromagnetic domain walls. In analogy with these systems, the distribution of activity bursts displays scaling laws that are universal in different cell types and for cells moving on different substrates. The main features of the invasion dynamics are quantitatively captured by a model of interacting active particles moving in a disordered landscape. Our results illustrate that collective motion of living cells is analogous to the corresponding dynamics in driven, but inanimate, systems.	physics.bio-ph	physics	a i i "fullversion" - 2018/5/10 - 13:21 - page 1 - #1 Bursts of activity in collective cell migration Oleksandr Chepizhko ∗, Costanza Giampietro †, Eleonora Mastrapasqua ‡, Mehdi Nourazar ∗ , Miriam Ascagni ‡ , Michela Sugni † , Umberto Fascio ‡ , Livio Leggio ‡ , Chiara Malinverno §, Giorgio Scita § , Stephane Santucci ¶ Mikko J. Alava ∗ Stefano Zapperi (cid:107) ∗∗ †† ∗ , Caterina A. M. La Porta † ∗Department of Applied Physics, Aalto University, P.O. Box 11100, FIN-00076 Aalto, Espoo, Finland,†Center for Complexity and Biosystems, Department of Biosciences, University of Milan, via Celoria 26, 20133 Milano, Italy,‡Department of Biosciences, University of Milan, via Celoria 26, 20133 Milano, Italy,§Dipartimento di Scienze della Salute, San Paolo, University of Milan, 20122 Milan, Italy,¶Laboratoire de Physique de l´´Ecole Normale Sup´erieure de Lyon, CNRS and Universit´e de Lyon, 69364 Lyon, France,(cid:107)Center for Complexity and Biosystems, Department of Physics, University of Milan, via Celoria 16, 20133 Milano, Italy,∗∗Institute for Scientific Interchange Foundation, Via Alassio 11/C, 10126 Torino, and ††Istituto di Chimica della Materia Condensata e di Tecnologie per l'Energia, CNR-Consiglio Nazionale delle Ricerche, Via R. Cozzi 53, Milano 20125, Italy Submitted to Proceedings of the National Academy of Sciences of the United States of America [10]. We find that in all these cases the statistical properties of the bursts follow universal scaling laws that are quantitatively similar to those observed in driven disordered systems [28]. Results and discussion Cell front dynamics and activity clusters We perform migra- tion experiments on different cell lines as described in the Ma- terials and Methods section. We extract and record the cell front position as a function of time as shown in Fig. 1 for ex- periments (see Videos S1-S7) and in simulations (Video S8). The fronts show a rough structure with localized bursts of ac- tivity that is reminiscent of elastic lines moving in a pinning landscape [19, 20, 21, 22]. Spatio-temporal velocity fluctua- tions in moving fronts can be effectively quantified by con- structing activity maps, as done previously for moving crack fronts [20] and fluid imbibition through porous media [21]. As discussed in the Materials and Methods section, from each time-lapse movie we construct a spatial map vf (x, y) measur- ing the velocity of the front at position (x, y) (see Fig. S1). Using these maps we can define regions of coordinated activ- ity by introducing a threshold c for the velocity. The maps reported in Fig. 1 vividly illustrate the formation of localized clusters of activity similar to those found in fracture [20] and imbibition [21]. The corresponding distributions of cluster ar- eas are reported in Fig. 2a for experiments on different cell Significance During wound healing and in cancer invasion cells migrate col- lectively driven by active internal forces and invade the available space. Here we show that this motion occurs by intermittent bursts of activity described by universal scaling laws similar to the ones observed in other driven systems where a front propa- gates in response to an external force, such as in fracture and fluid imbibition. Our results demonstrate that living systems display universal non-equilibrium critical fluctuations, induced by cell mutual interactions, that are usually associated to externally driven inanimate media. Reserved for Publication Footnotes Dense monolayers of living cells display intriguing relaxation dynam- ics, reminiscent of soft and glassy materials close to the jamming transition, and migrate collectively when space is available, as in wound healing or in cancer invasion. Here we show that collective cell migration occurs in bursts that are similar to those recorded in the propagation of cracks, fluid fronts in porous media and ferromag- netic domain walls. In analogy with these systems, the distribution of activity bursts displays scaling laws that are universal in different cell types and for cells moving on different substrates. The main features of the invasion dynamics are quantitatively captured by a model of interacting active particles moving in a disordered landscape. Our results illustrate that collective motion of living cells is analogous to the corresponding dynamics in driven, but inanimate, systems. Introduction Collective cell movement depends on intracellular biological mechanisms as well as environmental cues due to the extracel- lular matrix (ECM)[1, 2, 3, 4, 5], mainly composed by collagen which is organized in hierarchical structures, such as fibrils and fibers. The mechanical properties of collagen fibril networks are essential to offer little resistance and high sensitivity to small deformations, allowing easy local remodeling and strong strain stiffening needed to ensure cell and tissue integrity [6]. Wound healing is a typical biological assay to study collec- tive migration of cells under controlled conditions in vitro and is a prototypical experimental method to study active matter [7, 8, 9, 10]. Experiments performed on soluble collagen [11] or other gels [12], micro-patterned [13, 14] and deformable sub- strates [1] show that cell migration is guided by the substrate structure and stiffness [15, 5, 16]. It has been argued that collective migration properties arise from stresses transmitted between neighboring cells [1] giv- ing rise to long-ranged stress waves in the monolayer [17, 18]. Hence the dynamics of an invading cell sheet is ruled by a combination of long-range internal stresses and interactions with the substrate, suggesting an analogy with driven elastic systems moving in a disordered medium such as cracks lines [19, 20], imbibition fronts [21] or ferromagnetic domain walls [22]. The scaling laws in these systems are usually associated with a depinning critical point that has been widely studied by simple models for interface dynamics. Thanks to a combi- nation of numerical simulations [23, 24] and renormalization group theory [25, 23, 26, 27], we now have a detailed picture of the non-equilibrium phase transitions and universality classes in these systems. Here we substantiate the analogy between collective cell migration and depinning by revealing and char- acterizing widely distributed bursts of activity in the collective migration of different types cells (human cancer cells and ep- ithelial cells, mouse endothelial cells) over different substrates (plastic, soluble and fibrillar collagen) and experimental condi- tions (VE-cadherin knock down) and compare the experiments with simulations of a computational model of active particles www.pnas.org/cgi/doi/10.1073/pnas.0709640104 PNAS Issue Date Volume Issue Number 1–17 i i i i i i "fullversion" - 2018/5/10 - 13:21 - page 2 - #2 i i types and for simulations. They all display a power law decay P (S) ∼ S−τ up to a cutoff length S∗. The exponent τ of the power law distribution appears to be similar for different cell types: the fitted values (see Table S1) are all similar within error bars (τ (cid:39) 1.4 − 1.6) and remark- ably close to the values recorded in crack front propagation [20]. As observed in moving cracks fronts, the cluster distri- butions, and in particular the cutoff to the power law decay, depends on the threshold c used to identify the activity (see Materials and Methods and Fig. S1).The value of the cutoff S∗ is in all cases much larger than the typical size of a cell indicating that correlated activity spans multiple cells (Fig. S2). We also provide additional evidence for universality by measuring the relation between width, length and size of clus- ters Fig. 2cd. Also in this case, we find robust scaling across different cell lines and experimental conditions. Cell migration on collagen matrix To study quantitatively the effect of the substrate on the migration capability of cancer cells, we study wound healing under three different cell coat- ing conditions: plastic (i. e. no coating), bovine soluble and fibrillar collagen. The two types of collagen substrates are exactly the same in terms of biochemistry, but differ in the structural organization. In Fig. S3, we show the fibrillar or- ganization of our collagen substrate by scanning electron mi- croscopy (SEM), while soluble collagen substrates display no fibrils or structure. As shown in Fig. 2b, the exponent of the activity cluster distribution is independent on the substrate. The average cluster size, however, is larger for fibrillar colla- gen than for soluble collagen and plastic, independently on the value of c (see Fig 2b and Fig. S4a) To further quantify the role of the substrate in front propagation, we record average position of front as a function of time. As shown Fig. S5a, cell front motion is significantly slower on plastic substrates than on collagen ones. There is no noticeable difference in the average velocity of fronts moving on soluble and fibrillar collagen substrates. Fluctuations do, however, differ as shown by the standard deviation of the front position that is larger for cells moving on fibrillar collagen substrates (Fig S5b) Velocity distributions To better quantify the properties of the different experimental conditions, we resort to particle image velocimetry (PIV, see Materials and Methods) which allows to estimate local velocities treating the cell layer as a fluid. Representative examples of the local velocities for different cell types and for simulations are reported in Fig. 3a which displays the orientation of the velocity field (see also Videos S1-S7). While cells mostly advance towards the empty space there is significant motion also in the transverse and even in the opposite direction. This is summarized in the orientation distribution, revealing small differences due to cell types and substrates (Fig. 3de and S5c). Next, in Fig. 3b and S5d we report the distribution of the absolute value of the veloci- ties extracted from PIV, indicating again quantitative differ- ences between different cell types and substrates. Yet, the general shape of the velocity distribution is similar in all cases as shown in Fig. 3c where the velocities for each experiment are rescaled by their average value. This leads to a collapse of all the distributions, including the results of numerical simu- lations, apart from small deviations in the tails. Effect of VE-cadherin knockdown Internal couplings usually play a key role in avalanche statistics [25, 23, 26, 27]. To un- derstand this point in our context, we assess the role of cell-cell interactions performing the same cell migration assay knock- ing down VE-cadherin in mouse endothelial cells [29]. VE- cadherin is an endothelial specific cadherin molecule located at adherent junctions which regulats adhesion between adjacent endothelial cells. Here we consider cells without VE-cadherin (VEC-null) but still expressing N-cadherin, another important molecule expressed by several types of cells such as neuronal, skeletal and heart muscle cells or fibroblasts [30]. Moreover, N-cadherin is highly expressed in mesenchymal stem cells and cancer cells that are highly invasive and poorly polarized [31]. As shown in Fig. 4a, VEC-null cell fronts move faster, but in a more disordered way (see Fig. S6), than VEC-positive fronts. The distribution of activity cluster areas is very simi- lar in the two cases (Fig. 4a), with slightly larger cutoff size for VEC-positive cells (see Fig. S2). Despite these similarities, VEC-null cells tend to detach from the advancing front and move forward (see Video S9). We have tracked the trajecto- ries of individual cells (Fig. 4c) and shown that they follow a persistent random walk, displaying ballistic motion at short time-scales (Fig. 4d). The combination of faster fronts and in- dividual cells detachment confirms a more invasive phenotype for VEC-null cells [32]. Numerical simulations of an active particle model In order to better understand how avalanche-like fluctuations arise in collective cell migration, we simulate the model of interacting active particle introduced in Ref. [10] (for the details see the Materials and Methods section). The geometry of the system, the size and number of particles are chosen to match the exper- imental images. Particles are first packed into a confined space and then allowed to invade space filled with "surface" parti- cles. For the time scales considered in the present experiments, cell division is negligible and therefore we do not consider it in the model. The main tuning parameters of the model are the inter-particle velocity coupling β, the amplitude of the noise η and the adhesion strength U1. We also take into account the rough structure of the fibrillar collagen substrate by in- cluding a quenched (i.e. time independent) random force field with amplitude σrf and correlation length ζ. Simulations rep- resenting plastic or soluble collagen substrate are performed without quenched disorder (i.e. σrf = 0), with fluctuations arising only from the random initial condition and the time- dependent noise in the dynamics. In addition to the random force field, the main difference between our simulations and Ref. [10] is that we do not consider leader cells. The numerical simulations allow to reproduce with good accuracy all the salient features of collective cell migration re- vealed by our experiments as reported in Figs. 1-3. As already discussed, the numerical simulations reproduce the statistical features of activity bursts (Fig. 2) and cell velocities (Fig. 3). Furthermore, the numerical results illustrate the key role played by the substrate disorder, encoded in its correlation length ζ, in determining the cutoff scale of the cluster size dis- tribution (see Fig. S7). In particular, higher values of ζ corre- spond to larger activity clusters. This explains why cells move in larger activity bursts on fibrillar substrates, where disorder is stronger. Decreasing the velocity coupling β or the adhe- sion strength U1 increases the front velocity (Fig. S8) which should be compared with an analogous result observed when VE-cadherin is knocked down (Fig. 4). The model allows to explore the limits of universality. We observe an activity clus- ter distribution comparable to the experimentally measured one only for relatively small values of β. When β is larger (e.g. β = 60) the exponent τ is significantly smaller, suggest- ing the presence of different universality classes (see Fig. S7 and Table S2 for the results of the fits). Finally, we assess the relevance of interactions (i.e. performing simulations with β = 0 and U0 = U1 = 0) and pinning (i.e. no surface particles and no random field) and find that when either one of the two 2 www.pnas.org/cgi/doi/10.1073/pnas.0709640104 Footline Author i i i i i i "fullversion" - 2018/5/10 - 13:21 - page 3 - #3 i i ingredients is missing the distribution is not a power law any- more (see Fig. S9). When neither pinning and interactions are present, we can not record a well defined front. Universal scalingThe combinations of experimental results and numerical simulations suggests that the activity bursts observed in collective cell migration could be described by uni- versal scaling laws as in non-equilibrium critical phenomena. In order to strengthen this conclusion it is, however, impera- tive to overcome the limitations of power law fitting [33] and focus instead on scaling functions, as indicated by Sethna et al. [28]. Ref. [27, 24] shows that in proximity of a depinning critical point, avalanche distributions P (S) can be written as (cid:104)S2(cid:105) P (S) = 2(cid:104)S(cid:105) and p(s) is a univer- sal scaling function. Here, we follow the same approach with our data and show that indeed all the experimental distri- butions, involving different cell lines and substrate, and those obtained from numerical simulations collapse into a single uni- versal scaling curve as shown in Fig. 5. We thus perform a joint fit of all the distributions with a single scaling function p(s) = s−τ exp(−Cs), yielding τ = 1.58 ± 0.02 as best fit- ting parameter. This result provides a strong test of the uni- versality of the bursty behavior we observe in collective cell migration. (cid:104)S(cid:105) S∗ p(S/S∗), where S∗ = Discussion The collective motion of a cell layer as it invades an empty region has been extensively studied experimentally [7, 8, 9, 10] but the relation with the glassy features observed in confluent layers [34, 35] was not explored. The externally driven motion of disordered and glassy systems typically in- volves intermittent behavior and avalanches [28]. Our work shows that collective bursts of activity are also present in ac- tive matter with cells advancing in avalanches of widely dis- tributed sizes. Similar intermittent activity [36] and scaling behavior has been previously recorded in animal [37] and even human mobility patterns [38] and it is thus intriguing to realize that it exists even at the cellular level. Finally, migration is a key property of tumor cells for tissue invasion and metastasis. Our detailed statistical analysis shows quantitatively how the organization and structure of the substrate, either gel or fib- rillar, affects the way cells move. The final invasion velocity is the same but the motion is different, with larger intermittent fluctuations present on fibrillar substrates. This finding is in- teresting since it implies that cancer cells use different internal mechanisms to migrate depending on the environment, a fact that should be relevant for the metastatic process. Materials and Methods Collagen coating Collagen substrates are prepared from a 3% solution of soluble bovine collagen in 0.01 HCl (code C4243, Sigma), kept in ice until use. Both solubilized ("soluble colla- gen") and re-fibrillated collagen ("fibrillar collagen") are pre- pared in 35 mm Petri dishes. For soluble substrates a sufficient amount of collagen solution is added to a Petri dish so that the bottom is covered, the excess is removed after 15 minutes and the dish is dried at room temperature for at least 2 hours. For fibrillar substrates, a mixture of 3% collagen solution (Code C4243, Sigma), 0.1N NaOH, 10% PBS (8:1:1) is prepared and 0.8mL are pipetted to the petri dish and incubated at 37◦ C overnight to allow the proper gelification/refibrillation. After polymerization, samples are 5-10 micrometers in height. Cell cultureHeLa cell line (ATCC CCL-2) is cultured in DMEM medium supplemented with 10% fetal calf serum, 2 mM L-glutamine, 100 U/ml penicillin, 100 mg/ml strepto- mycin and 0,25 mg/ml amphotericin B (Invitrogen, Milan, Italy) at 37◦C and humidity (95% relative humidity), and CO2 concentration (5%). Cells are seeded in bovine soluble col- lagen, or fibrillar bovine collagen or without collagen-coated dishes and growth up to reach confluence overnight. Endothe- lial cells derived from embryonic stem cells with homozygous null mutation of the VE-cadherin gene (VEC null) [39]. The wild type form of VE-cadherin was introduced in these cells (VEC positive) as described in detail in [40] Endothelial cells isolated from lungs of wild type adult mice and cultured as previously described [29]. Starving medium was MCDB 131 (Invitrogen) with 1% BSA (EuroClone), 2-mM glutamine, 100 U/liter penicillin/streptomycin, and 1-mM sodium pyruvate. The MCF-10A cell line is a non-tumorigenic human mammary epithelial cell line (ATCC CRL10317) grown in DMEM/F12 supplemented with 5%horse serum, 20ng/ml EGF, 0.5mg/ml Hydrocortisone, 100ng/ml Choilera toxin, 10µg/ml Insulin. Migration assay For the migration assay, a wound is intro- duced in the central area of the confluent cell sheet by using a pipette tip and the migration followed by time-lapse imaging. Hela cells were stained with 10 µM Cell tracker green CMFDA (Molecular Probes) in serum-free medium for 30 minutes and then the complete medium was replaced. Mouse endothe- lial cell monolayers were wounded after an overnight starving, washed with PBS, and incubated at 37◦C in starving medium. MCF-10A cell monolayers were wounded after an O/N doxy- cicline induction, washed with PBS, and incubated at 37◦C in fresh media+doxycicline. Time lapse imaging For Hela cells, time-lapse multifield ex- periments were performed using an automated inverted Zeiss Axiovert S100 TV2 microscope (Carl Zeiss Microimaging Inc., Thornwood, NY) with a chilled Hamamatsu CCD camera OrcaII-ER. Displacements of the sample and the image ac- quisition are computer-controlled using Oko-Vision software (from Oko-lab). This Microscope was equipped with a cage incubator designed to maintain all the required environmental conditions for cell culture all around the microscopy work- station, thus enabling to carry out prolonged observations on biological specimens. Cell Tracker and Phase contrast images were acquired with an A-Plan 10x (NA 0.25) objective; the typical delay between two successive images of the same field was set to 10 minutes for 12 hours. For mouse endothelial cells and MCF-10A cells, time-lapse imaging of cell migration was performed on an inverted microscope (Eclipse TE2000- E; Nikon) equipped with an incubation chamber (OKOLab) maintained at 37◦C in an atmosphere of 5% CO2. Movies were acquired with a Cascade II 512 (Photometrics) charge- coupled device (CCD) camera controlled by MetaMorph Soft- ware (Universal Imaging) using a 4X or 10 magnification ob- jective lens (Plan Fluor 10, NA 0.30). Images were acquired every 2 or 5 min over a 24h period. See Table S3 for the time lapse parameters. Scanning Electron Microscopy Collagen substrates are fixed with 2% glutaraldehyde in 0.1 M cacodylate buffer for 2 h at 4C and post-fixed with 1% osmium tetroxyde in 0.1 M sodium cacodylate buffer (2 h, room temperature). After several washings with dH2O, they are dehydrated in an in- creasing ethanol scale and treated with a series of solutions of HMDS (Hexamethyldisilazane) and ethanol in different pro- portions (1:3, 1:1, 3:1 and 100% HMDS), mounted on stubs, covered with pure gold (Agar SEM Auto Sputter, Stansted, UK). The samples are observed under a scanning electron mi- croscope(SEM) (LEO-1430, Zeiss, Oberkochen, Germany). Footline Author PNAS Issue Date Volume Issue Number 3 i i i i i i "fullversion" - 2018/5/10 - 13:21 - page 4 - #4 i i Particle image velocimetry (PIV) The measurements of the velocity field were done using PIVlab app for Matlab [41, 42]. The method is based on the comparison of the intensity fields of two consequent photographs of cells. The difference in the intensity is converted into velocity field measured in px/f rame. Then the velocity is converted to µm/h through coefficient that is specific for each experiment, shown in table S3. Border progression measurement The method of border de- tection is based on the procedures described in [43, 44]. The method is based on the border extraction procedure described in the manual of the Matlab software package (CellSegmentationExample). Front activity maps We analyze the propagation of the cell front, by measuring for each recorded image, and thus, at each time step t, the cell front position y = h(x, t), along the abscissa x. We then construct a local velocity map of such in- terface vf (x, y), by computing the velocity of the cell front at each position of the front (x, y = h(x, t)) during its progres- sion: vf (x, y = h(x, ti)) = h(x, ti+1) − h(x, ti) ti+1 − ti Figure S1a provides such spatial color scale map vf (x, h(x, t)) of the local velocity fluctuations for a typical wound healing experiment using HeLa cells moving on a soluble collagen sub- strate. The various regions of different color levels reveal an intermittent burst-like dynamics on a broad range of length scales. Such a complex dynamics can also be unveiled by representing the spatio-temporal map vf (x, t) giving the in- stantaneous velocity of each point of the cell front during its propagation, as shown in Fig. S1b. Occasional overhangs in the front are eliminated by the maximum value of h for a given x. When the front locally moves backwards, we aver- age all the velocity values at that point. To define areas of correlated activity, we consider the spatial map of the local front velocities vf (x, h(x, t)) and define avalanches as clus- ters of velocities larger than an arbitrary threshold [20, 21]: vf (x, h(x, t)) ≥ c(cid:104)vf(cid:105), where (cid:104)vf(cid:105) is the mean front velocity during the experiment Fig. S1c. Cluster sizes S are given by the area of the clusters, and their shape is characterized by their length l (lateral extension along x) and their width w (extension in the mean direction of propagation y). The cut- off of the cluster size distribution depends slightly on c, but its scaling exponent does not (Fig. S1d). Simulations of the active particle model We simulate collec- tive cell migration using a modified version of the model in- troduced in Ref. [10]. Here we introduce a pinning field and we do not consider a leader cell. The main equation of motion is given by dvi dt (vj − vi) + fij = −αvi + (cid:20) β (cid:88) Ni j (cid:21) + σ(ρi)ηi + Frf (xi) [ 1 ] where the sum is restricted to the nearest neighbors of i, α is a damping parameter, β is the velocity coupling strength, fij = −∇iU (rij) is the force between neighboring cells where [ 2 ] U (r) = U0 exp(−(r/a0)2) + U1(r − a1)2H(r − a1) . Here H(x) is the Heaviside function H(x) = 1 for x > 0 and H(x) = 0 otherwise, U1 is the adhesion and U0 the repulsive strength. The noise term σ(ρi)ηi, here ηi is an Ornstein- Uhlenbeck process with correlation time τ : τ dηi dt = −ηi + ξi, [ 3 ] ξi is a delta-correlated white noise, independent for each cell (cid:104)ξi(t)ξj(t(cid:48))(cid:105) = δijδ(t − t(cid:48)) The amplitude of the noise term σ depends on the density of the neighboring particles ρi: σ(ρi) = σ0 + (σ1 − σ0)(1 − ρi/ρ0). [ 4 ] The neighbors of each particle are defined in the same way as Ref. [10] The neighborhood of a particle i is split into 6 equal sectors. The particle closest to the particle i in each sector (but closer then interaction radius of 100 µm) is cho- sen to be a neighbor. The local density is computed then as ρi = 1/[π((cid:104)d(cid:105)/2)2], where (cid:104)d(cid:105) is the average distance be- tween cell i and its 6 neighbors (for sectors without neighbors the distance is taken to be 100 µm). To model heterogeneous substrates we include an additional random force field Frf (xi). A pair of Gaussian distributed random numbers, represent- ing the two components of the random force, are placed on a square grid of spacing ζ, quantifying the correlation length of the disorder. The value of random force between grid points is obtained through bilinear interpolation. As in the original model, the free surface is modeled by surface particles, that are hindering cells to enter the empty space. The interaction between a surface particle and a cell is modeled as [ 5 ] where U s(r) is U s(r) = As exp(−(r/as)2). A scalar damage variable q is associated with each surface particles and follows the equation ij = −∇iU s(rij) , f s (cid:88) j:rj∈Si ν dqi dt = f s ij . [ 6 ] The values of the parameters that were used for simula- tions match the ones used in Ref. [10]: number of particles N = 4000, size of the box 2L× L, L = 1000 µm, α = 1.42h−1, β, varying, but the value used in most of simulations was β = 10 h−1, τ = 1.39 h−1, σ0 = 150 µmh−3/2, σ1 = 300 µmh−3/2, U0 = 2400 µm2/h2, a0 = 8 µm, U1 = 2h−2, a1 = 35µm, ρ0 = 4.0 × 10−3µm−2. rs the radius of inter- action between a cell and a surface particle, is chosen to be rs = 50µm, ν = 7h−1, As = 2400µm2/h2, as = 8 µm, the threshold value qi = θ after reaching which the surface parti- cle disappears is taken to be θ = 30 µm. The values for the parameters of the random field are also varied. Contrary to Ref. [10], we do not introduce any leader cell. Data fitting We fit the distribution of cluster sizes using the least-square method for binned histograms and the maximum likelihood method (see Fig. S10, Tables S1, S2 and Supporting Information for more details). Acknowledgments OC and SZ acknowledges support from the Academy of Fin- land FiDiPro program, project 13282993. CAMLP and SS thank the visiting professor program of Aalto University where part of this work was completed. SZ is supported by ERC Advanced Grant n. 291002 SIZEFFECTS. MN, MJA are supported by the Academy of Finland through its Centres of Excellence Programme (2012-2017) under project n. 251748. We acknowledge the computational resources provided by the Aalto University School of Science "Science-IT" project, as well as those provided by CSC (Finland). GS is supported by grants from the Associazione Italiana per la Ricerca sul Cancro (AIRC) (n. 10168), Worldwide Cancer Research (AICR-14- 0335), and the European Research Council (Advanced ERC n. 268836) 4 www.pnas.org/cgi/doi/10.1073/pnas.0709640104 Footline Author i i i i i i "fullversion" - 2018/5/10 - 13:21 - page 5 - #5 i i Authors Contributions OC, MN, SS analyzed the data. CAMLP, CG, LL, MA, MS, UF, CM, GS, EM performed the experiments. OC, MN per- formed numerical simulations. CAMLP designed the experi- ments. OC, MJA, SZ designed the model. CAMLP, SZ wrote the manuscript. CAMLP, MJA, SZ coordinated the project. 1. Tambe DT, et al. (2011) Collective cell guidance by cooperative intercellular forces. 38. Song C, Koren T, Wang P, Barabasi AL (2010) Modelling the scaling properties of Nat Mater 10:469–75. human mobility. Nat Phys 6:818–823. 2. Brugues A, et al. (2014) Forces driving epithelial wound healing. Nat Phys 10:683–690. 3. Haeger A, Krause M, Wolf K, Friedl P (2014) Cell jamming: collective invasion of mesenchymal tumor cells imposed by tissue confinement. Biochim Biophys Acta 1840:2386–95. 4. Lange JR, Fabry B (2013) Cell and tissue mechanics in cell migration. Exp Cell Res 319:2418–23. 5. Koch TM, Munster S, Bonakdar N, Butler JP, Fabry B (2012) 3d traction forces in cancer cell invasion. PLoS One 7:e33476. 6. Sacks MS, Sun W (2003) Multiaxial mechanical behavior of biological materials. Annu Rev Biomed Eng 5:251–84. 7. Vedula SRK, Ravasio A, Lim CT, Ladoux B (2013) Collective cell migration: a mech- anistic perspective. Physiology (Bethesda) 28:370–9. 8. Szab´o B, et al. (2006) Phase transition in the collective migration of tissue cells: Experiment and model. Phys Rev E 74:061908. 9. Poujade M, et al. (2007) Collective migration of an epithelial monolayer in response to a model wound. Proc Natl Acad Sci U S A 104:15988–93. 10. Sep´ulveda N, et al. (2013) Collective cell motion in an epithelial sheet can be quan- titatively described by a stochastic interacting particle model. PLoS Comput Biol 9:e1002944. 11. Haga H, Irahara C, Kobayashi R, Nakagaki T, Kawabata K (2005) Collective movement of epithelial cells on a collagen gel substrate. Biophys J 88:2250–6. 12. Ng MR, Besser A, Danuser G, Brugge JS (2012) Substrate stiffness regulates cadherin- dependent collective migration through myosin-ii contractility. J Cell Biol 199:545–63. 13. Saez A, Ghibaudo M, Buguin A, Silberzan P, Ladoux B (2007) Rigidity-driven growth and migration of epithelial cells on microstructured anisotropic substrates. Proc Natl Acad Sci U S A 104:8281–6. 14. Rottgermann PJF, Alberola AP, Radler JO (2014) Cellular self-organization on micro- structured surfaces. Soft Matter 10:2397–404. 15. Oakes PW, et al. (2009) Neutrophil morphology and migration are affected by sub- strate elasticity. Blood 114:1387–95. 16. Metzner C, et al. (2015) Superstatistical analysis and modelling of heterogeneous random walks. Nat Commun 6:7516. 17. Serra-Picamal X, et al. (2012) Mechanical waves during tissue expansion. Nat Phys 8:628–634. 39. Balconi G, Spagnuolo R, Dejana E (2000) Development of endothelial cell lines from embryonic stem cells: A tool for studying genetically manipulated endothelial cells in vitro. Arterioscler Thromb Vasc Biol 20:1443–51. 40. Lampugnani MG, et al. (2002) Ve-cadherin regulates endothelial actin activating rac and increasing membrane association of tiam. Mol Biol Cell 13:1175–89. 41. Thielicke W (2014) The flapping flight of birds: Analysis and application. Ph.D. thesis. 42. Thielicke W, Stamhuis E (2014) Pivlab towards user-friendly, affordable and accurate digital particle image velocimetry in matlab 2. 43. Johnston ST, Simpson MJ, McElwain DLS (2014) How much information can be ob- tained from tracking the position of the leading edge in a scratch assay? Journal of The Royal Society Interface 11. 44. Treloar KK, Simpson MJ (2013) Sensitivity of edge detection methods for quantifying 18. Banerjee S, Utuje KJC, Marchetti MC (2015) Propagating stress waves during epithe- cell migration assays. PLoS One 8:e67389. lial expansion. Phys Rev Lett 114:228101. 19. Maloy KJ, Santucci S, Schmittbuhl J, Toussaint R (2006) Local waiting time fluctua- tions along a randomly pinned crack front. Phys Rev Lett 96:045501. 20. Tallakstad KT, Toussaint R, Santucci S, Schmittbuhl J, Maloy KJ (2011) Local dy- namics of a randomly pinned crack front during creep and forced propagation: an experimental study. Phys Rev E Stat Nonlin Soft Matter Phys 83:046108. 21. Clotet X, Ort´ın J, Santucci S (2014) Disorder-induced capillary bursts control inter- mittency in slow imbibition. Phys Rev Lett 113:074501. 22. Durin G, Zapperi S (2000) Scaling exponents for Barkhausen avalanches in polycrys- talline and amorphous ferromagnets. Phys Rev Lett 84:4075–4078. 23. Leschhorn H, Nattermann T, Stepanow S, Tang LH (1997) Driven interface depinning in a disordered medium. Ann Physik 6:1–34. 24. Rosso A, Le Doussal P, Wiese KJ (2009) Avalanche-size distribution at the depinning transition: A numerical test of the theory. Phys Rev B 80:144204. 25. Narayan O, Fisher DS (1992) Critical behavior of sliding charge-density waves in 4- epsilon dimensions. Phys Rev B 46:11520. 26. Chauve P, Doussal PL, Wiese KJ (2001) Renormalization of pinned elastic systems: how does it work beyond one loop. Phys Rev Lett 86:1785–1788. 27. Le Doussal P, Wiese KJ (2009) Size distributions of shocks and static avalanches from the functional renormalization group. Phys Rev E 79:051106. 28. Sethna J, Dahmen KA, Myers CR (2001) Crackling noise. Nature 410:242–244. 29. Giampietro C, et al. (2015) The actin-binding protein eps8 binds ve-cadherin and mod- ulates yap localization and signaling. J Cell Biol 211:1177–92. 30. Goodwin M, Yap AS (2004) Classical cadherin adhesion molecules: coordinating cell adhesion, signaling and the cytoskeleton. J Mol Histol 35:839–44. 31. Wheelock MJ, Johnson KR (2003) Cadherins as modulators of cellular phenotype. Annu Rev Cell Dev Biol 19:207–35. 32. Giampietro C, et al. (2012) Overlapping and divergent signaling pathways of n-cadherin and ve-cadherin in endothelial cells. Blood 119:2159–70. 33. Clauset A, Shalizi CR, Newman ME (2009) Power-law distributions in empirical data. SIAM review 51:661–703. 34. Angelini TE, et al. (2011) Glass-like dynamics of collective cell migration. Proc Natl Acad Sci U S A 108:4714–9. 35. Park JA, et al. (2015) Unjamming and cell shape in the asthmatic airway epithelium. Nat Mater 14:1040–1048. 36. Weber CA, et al. (2015) Random bursts determine dynamics of active filaments. Pro- ceedings of the National Academy of Sciences 112:10703–10707. 37. Ginelli F, et al. (2015) Intermittent collective dynamics emerge from conflicting im- peratives in sheep herds. Proc Natl Acad Sci U S A 112:12729–34. Fig. 1. Cell front dynamics displays activity bursts. Ex- amples of cell fronts and their time evolution and the corresponding activity maps. Cell fronts are colored according to time. Regions marked by the same color in the activity map move collectively. The scale bars are 100µm. a) HeLa cells moving on soluble collagen substrates. b) Mouse endothelial cells derived from embryonic stem cells (VEC-positive). c) Human mammary epithelial cells (MCF-10A). d) Mouse endothelial cells extracted from lungs (lung-ECs). e) Numerical simulations of the model. Footline Author PNAS Issue Date Volume Issue Number 5 i i i i i i "fullversion" - 2018/5/10 - 13:21 - page 6 - #6 i i Fig. 2. Cell activity bursts statistics displays scaling a) Dis- tributions of the areas of activity clusters display power law scaling with a cutoff. The distributions for different cell types have been shifted for clarity. The slope obtained fitting the distributions is very similar for all cell types. b) Distributions of the areas of activity clusters for HeLa cells moving on different substrates. The scaling exponent is the same but the cutoff size for fibrillar collagen substrates is larger than for plastic and soluble collagen substrates. c) The average cluster length and d) the average cluster width scale with the cluster size with an exponent that is independent on the cell type. 6 www.pnas.org/cgi/doi/10.1073/pnas.0709640104 Footline Author i i i i 100101102103104105S, µm210-810-4100104P(S)HeLa (soluble)lung-ECsMCF-10AVEC-positivesimulation100101102103104S, µm210-810-4100P(S)HeLa (soluble)HeLa (plastic)HeLa (fibrillar)100101102103104105S, µm210-210-1100101102103l, µm100101102103104105S, µm2100101102103w, µm~S0.6~S0.6(a)(b)(c)(d)i i "fullversion" - 2018/5/10 - 13:21 - page 7 - #7 i i Fig. 5. The distribution of activity clusters is universal. The distributions of the areas of activity clusters from Figs. 2 and 4 can all be col- lapsed into a single universal scaling function p(s) = s−τ exp(−Cs) (line) when plotted in terms of the reduced variable s = S/S∗. Fig. 3. Cell velocity distributions display universal statis- tics a) Velocity maps for different cell types and for simulations as obtained from PIV. The colors indicate the orientation of the velocity following the color wheel. The length of the arrows is proportional to the magnitude of the velocity. The scale bars are 100µm. b) The corresponding distributions of velocity magnitudes. c) Scaling the velocities by their magnitude leads to a collapse of all the distributions, apart from small deviations in the tails. d,e) The distribution of the orientations of the velocities. Fig. 4. Knock down of VE-cadherin leads to faster fronts and individual cell motion. a) The time evolution of the front position for mouse endothelial cell expressing (VEC-positive) or not (VEC-null) VE-cadherin. VEC-null cells move faster. b) The corresponding cluster size distribution displays the same exponent and small changes in the cutoff. c) In the VEC null cases cells detach from the front and invade the space individually following trajectories as the ones Footline Author illustrated. d) The average mean-square displacement of the trajectories indicates an initial ballistic regime followed by a slowing down. i i PNAS Issue Date Volume Issue Number 7 i i 10-410-310-210-1100101s10-410-310-210-1100101102103104105p(s)HeLa (fibrillar)HeLa (soluble)MCF-10Alung-ECsVEC-nullVEC-positiveHeLa (plastic)Simulationi i "fullversion" - 2018/5/10 - 13:21 - page 8 - #8 i i 8 www.pnas.org/cgi/doi/10.1073/pnas.0709640104 Footline Author i i i i i i "fullversion" - 2018/5/10 - 13:21 - page 9 - #9 i i Supplementary information Details about the fitting methods We employ different strate- gies to obtain the best parameters describing the cluster size distributions in simulations and experiments. We first con- sider the log-spaced binned probability density functions and perform a lest-square fitting with the function f (x) = Ax −τ exp(−Cx) , [ 7 ] To obtain a reliable estimate of the cutoff, we first take the logarithm of the distribution and fit it with the logarithm of Eq. 7. The fitting is performed in python using the pyFit- ting function (https://github.com/gdurin/pyFitting) and the results are reported in table ?? and ?? for experiments and simulations, respectively. The second strategy relies on the maximum-likelihood esti- mate. Given the set of measured cluster sizes Si, we consider the log-likelihood function L = −(cid:88) i log(f (Si))), [ 8 ] where the function f (x) is given by Eq. 7. The function L is maximized with respect to the parameters A, τ and α. The results are again reported in table ?? and ??. A comparison of the fitted function and the data can be presented in a way that is independent on binning by plotting the cumulative dis- tribution functions in Fig. S10. The theoretical function in this case is given by CDF = Γ(1 − τ, Cx) Γ(1 − τ, Cxmin) [ 9 ] where Γ(a, b) is the incomplete gamma function. Finally, we collapse together all the experimental probability density function into a unique set. To this end, we first com- (cid:104)S2(cid:105) pute the characteristic cluster size S∗ = 2(cid:104)S(cid:105) from each exper- imental data set. We then rescale the cluster sizes, defining a reduced size s = S/S∗ and a scaling function p(s) = P (S) S∗ (cid:104)S(cid:105) . Notice that p(s) is not a distribution, since its integral is not equal to one. Once data for different experiments are rescaled according to this prescription, they are joined into a single set which is fitted by the least-square method using the func- tion reported in Eq. 7. The result yields A = 0.13 ± 0.01, τ = 1.58 ± 0.02 and C = 0.20 ± 0.03. Footline Author PNAS Issue Date Volume Issue Number 9 i i i i i i "fullversion" - 2018/5/10 - 13:21 - page 10 - #10 i i Fig. S2. The characteristic cluster size S∗ = 2(cid:104)S2(cid:105)/(cid:104)S(cid:105) as a function of the velocity threshold c. a) Hela Cells on different substrates. b) Mouse endothelial cells expressing VE-cadherin or not. c) Comparison between different cell lines d) Simulations of the model for different values of β. Fig. S1. Determination of activity areas from images. Scale bars are 100µm. a) Spatial front velocity maps. Each pixel is colored according to the velocity of the front once it passes through that location. b) Spatio-temporal velocity map. Each pixel corresponds to the velocity (vf (x, t)) at time t of the front at point x c) Activity clusters are obtained by setting threshold c to the data in panel a. Results obtained with different values of c are reported in three panels. d) Cluster size distributions for different values of c. Fig. S3. Scanning electron micrograph of the fibrillar collagen substrate at two different magnifications. From the high magnification image b) it is possible to esti- mate the fiber diameter. 10 www.pnas.org/cgi/doi/10.1073/pnas.0709640104 Footline Author i i i i 36.61013-60-40-200604020(a)(b)76-7638-380xyxt,hv, μm/hv, μm/h(c)(d)c=1.0c=2.0c=3.0110100100010000S,µm210-610-410-2100P(S)c=1c=2c=31230350700HeLa (plastic)HeLa (soluble)HeLa (fibrillar)12305001000VEC-positiveVEC-null12307501500HeLa (soluble)lung-EcSMCF-10AVEC-positive123020004000β=10β=60β=100ccccS, µm2S, µm2S, µm2S, µm2(a)(b)(c)(d)a)b)i i "fullversion" - 2018/5/10 - 13:21 - page 11 - #11 i i Fig. S7. Simulated cluster size distributions a) for different β (the strength of alignment between cells), b) for different ζ (correlation length of the random field). Fig. S4. The average cluster size as a function of the velocity threshold c. a) HeLa Cells on different substrates. b) Mouse endothelial cells expressing VE-cadherin or not. c) Comparison between different cell lines d) Simulations of the model for different values of β. Fig. S8. a) Simulations of the model for different values of β show that fronts are slower when β is increased. b) Simulations of the model for different values of the adhesion strength U1 show that fronts are slower when U1 is increased. Fig. S5. Dependence of the front dynamics on the substrate properties. a) Front position as a function of time indicates that fronts move slowly on plastic substrates. b) The standard deviation of the front (i.e. the front roughness) is larger for cells moving on fibrillar collagen. c) Angular velocity distribution as obtained from PIV. d) Distribution of absolute values of velocities as obtained by PIV. Footline Author PNAS Issue Date Volume Issue Number 11 Fig. S6. (VEC-null). a) Fronts. b) Activity clusters. c) Results from PIV. Front dynamics in mouse endothelial cells with VE-cadherin knock down i i i i 0.511.522.533.5153045HeLa (plastic)HeLa (soluble)HeLa (fibrillar)0.511.522.533.51020304050VEC-positiveVEC-null0.511.522.533.550100150200HeLa (soluble)lung-EcSMCF-10AVEC-positive0.511.522.533.5406080β=10β=60β=100<S>, µm2<S>, µm2<S>, µm2<S>, µm2(a)(b)(c)(d)cccc051015t,h050100h, µm051015t, h203040<(h-<h>)2>0.011100v, µm/h10-610-3100P(v)HeLa (soluble)HeLa (plastic)HeLa (fibrillar)0ππ/23π/2P(θ)0.020.04(a)(b)(c)(d)10010210410-910-610-3100β=10β=60β=10010010210410-910-610-3100ζ=10ζ=20ζ=50ζ=100P(S)S, µm2P(S)S, µm2(a)(b)010203040050100150200250300β=10β=60β=100010203040050100150200250300U1=2U1=4U1=8U1=16h, µmt, hh, µmt, hi i "fullversion" - 2018/5/10 - 13:21 - page 12 - #12 i i Fig. S9. The non-binned cumulative distribution functions (CDF) of the cluster sizes from simulations a) with and without pinning (i. e. no surface particles and no random field) and b) with and without interactions (i.e. β = 0, U0 = U1 = 0). The lack of either pinning or interactions leads to clear deviation from a power law distribution, as also shown by fitting the simulation results with the maximum likeli- hood method. The fit yield scaling exponents τ = 1.2 (panel a, red), τ = 1.38 (panel a, green), τ = 1.51 (panel b, red), τ = 1.56 (panel b, green), but the fitted curves deviate systematically from the data and the scaling regime is extremely small. We also notice that in order to record a front at all in these conditions, one should also reduce the noise parameter σ0. Fig. S10. The non-binned cumulative distribution functions (CDF) of the clus- ter sizes for the different experiments are compared with the results of maximum likelihood estimates. 12 www.pnas.org/cgi/doi/10.1073/pnas.0709640104 Footline Author i i i i 110010000S µm20.011CDF(S)with pinning (σ0=150)no pinning σ0=15no pinning σ0 =7511000S, µm20.0011CDF(S)with interactions (σ0=150)without interactions (σ0=15)without interactions (σ0=75)1101001000100001e+05S, µm20.001110001e+06CDFHeLa (soluble)HeLa (plastic)HeLa (fibrillar)lung-ECsMCF-10AVEC-positiveVEC-nulli i "fullversion" - 2018/5/10 - 13:21 - page 13 - #13 i i Footline Author PNAS Issue Date Volume Issue Number 13 i i i i i i "fullversion" - 2018/5/10 - 13:21 - page 14 - #14 i i Fitting Name HeLa (soluble) HeLa (plastic) HeLa (fibrillar) lung-ECs MCF-10A VEC-positive VEC-null A 0.18 ± 0.02 0.17 ± 0.03 0.19 ± 0.03 0.81 ± 0.36 0.17 ± 0.06 0.33 ± 0.09 0.33 ± 0.04 τ 1.48 ± 0.03 1.47 ± 0.05 1.49 ± 0.04 1.58 ± 0.07 1.58 ± 0.08 1.62 ± 0.05 1.62 ± 0.03 C 0.00054 ± 5.2 × 10−5 0.0006 ± 0.0001 (20 ± 7) × 10−5 (7.2 ± 1.4) × 10−5 (16.7 ± 9.2) × 10−5 (8.6 ± 4.9) × 10−5 0.0001 ± 5 × 10−5 Maximum likelihood τ 1.48 ± 0.10 1.49 ± 0.09 1.34 ± 0.14 1.45 ± 0.13 1.56 ± 0.12 1.55 ± 0.12 1.61 ± 0.13 (3.2 ± 2.43) × 10−4 (3.5 ± 2.8) × 10−4 (2.82 ± 2.0) × 10−4 (1.1 ± 0.6) × 10−4 (1.2 ± 1.0) × 10−4 (1.2 ± 1.0) × 10−4 (1.7 ± 1.4) × 10−4 C xmin 12.8 8.0 28.8 176 43.5 53.8 46.1 β 10 60 100 A 0.34 ± 0.07 0.103 ± 0.007 0.11 ± 0.02 Fitting τ 1.56 ± 0.04 1.37 ± 0.01 1.39 ± 0.04 C (10 ± 1) × 10−5 (3 ± 0.13) × 10−4 (3.8 ± 0.3) × 10−4 Maximum likelihood τ 1.60 ± 0.02 (6.3 ± 1.9) × 10−4 (2.5 ± 0.8) × 10−4 1.36 ± 0.05 (3.9 ± 0.7) × 10−4 1.33 ± 0.03 C xmin 21 16 6 Name Number of fronts 1 px in µm 1 frame in min. HeLa (plastic) HeLa (soluble) HeLa (fibrillar) lung-ECs MCF-10A VEC-positive VEC-null 6 5 8 11 8 4 4 1.2658 1.2658 1.2658 4 1.6 1.6 1.6 10 10 10 5 5 2 2 14 www.pnas.org/cgi/doi/10.1073/pnas.0709640104 Footline Author i i i i i i "fullversion" - 2018/5/10 - 13:21 - page 15 - #15 i i Footline Author PNAS Issue Date Volume Issue Number 15 i i i i i i Supplementary video captions "fullversion" - 2018/5/10 - 13:21 - page 16 - #16 i i Video S1 (Left) Time lapse of a representative sheet of Hela cells invading a fibrillar collagen substrate. The reconstructed front is reported in red. (Right) Evolution of the corre- sponding local velocity map obtained by PIV. Video S2 (Left) Time lapse of a representative sheet of Hela cells invading a soluble collagen substrate. The reconstructed front is reported in red. (Right) Evolution of the corre- sponding local velocity map obtained by PIV. Video S3 (Left) Time lapse of a representative sheet of Hela cells invading a plastic substrate. The reconstructed front is re- ported in red. (Right) Evolution of the corresponding local velocity map obtained by PIV. Video S4 (Left) Time lapse of a representative sheet of lung derived endothelial cells. The reconstructed front is reported in red. (Right) Evolution of the corresponding local velocity map obtained by PIV. null mouse endothelial cell. Video S5 (Left) Time lapse of a representative sheet of MCF10-A cells invading a plastic substrate. The reconstructed front is reported in red. (Right) Evolution of the corresponding local velocity map obtained by PIV. Video S6 (Left) Time lapse of a representative sheet of VEC-null mouse endothelial cells invading a plastic substrate. The reconstructed front is reported in red. (Right) Evolution of the corresponding local velocity map obtained by PIV. Video S7 (Left) Time lapse of a representative sheet of VEC-positive mouse endothelial cells invading a plastic substrate. The reconstructed front is reported in red. (Right) Evolution of the corresponding local velocity map obtained by PIV. Video S8 (Left) Representative simulation results. The recon- structed front is reported in red. (Right) Evolution of the corresponding local velocity map obtained by PIV. Video S9 A representative trajectory (yellow) of an isolated VEC- 16 www.pnas.org/cgi/doi/10.1073/pnas.0709640104 Footline Author i i i i
1809.05306	1	1809	2018-09-14T08:55:44	Single-cell bacterial electrophysiology reveals mechanisms of stress-induced damage	[ "physics.bio-ph" ]	Electrochemical gradient of protons, or proton motive force (PMF), is at the basis of bacterial energetics. It powers vital cellular processes and defines the physiological state of the cell. Here we use an electric circuit analogy of an Escherichia coli cell to mathematically describe the relationship between bacterial PMF, electric properties of the cell membrane and catabolism. We combine the analogy with the use of bacterial flagellar motor as a single-cell "voltmeter" to measure cellular PMF in varied and dynamic external environments, for example, under different stresses. We find that butanol acts as an ionophore, and functionally characterise membrane damage caused by the light of shorter wavelengths. Our approach coalesces non-invasive and fast single-cell voltmeter with a well-defined mathematical framework to enable quantitative bacterial electrophysiology.	physics.bio-ph	physics	Single-cell bacterial electrophysiology reveals mechanisms of stress induced damage Ekaterina Krasnopeeva1, Chien-Jung Lo2, and Teuta Pilizota1 1Centre for Synthetic and Systems Biology, Institute of Cell Biology, School of Biological Sciences, University of Edinburgh, Alexander Crum Brown Road, EH9 3FF, Edinburgh, UK 2Department of Physics and Graduate Institute of Biophysics, National Central University, Jhongli, Taiwan 32001, ROC ABSTRACT Electrochemical gradient of protons, or proton motive force (PMF), is at the basis of bacterial energetics. It powers vital cellular processes and defines the physiological state of the cell. Here we use an electric circuit analogy of an Escherichia coli cell to mathematically describe the relationship between bacterial PMF, electric properties of the cell membrane and catabolism. We combine the analogy with the use of bacterial flagellar motor as a single-cell "voltmeter" to measure cellular PMF in varied and dynamic external environments, for example, under different stresses. We find that butanol acts as an ionophore, and functionally characterise membrane damage caused by the light of shorter wavelengths. Our approach coalesces non-invasive and fast single-cell voltmeter with a well-defined mathematical framework to enable quantitative bac- terial electrophysiology. Keywords: surements, bacterial physiology, indole, butanol, photodamage bacterial energetics, proton motive force, bacterial membrane damage, single-cell mea- INTRODUCTION To stay alive bacteria, like other cells, maintain adequate supplies of free energy, and under various external stresses attempt to stay viable by distributing it to processes essential for coping with the challenge, while simultaneously maintaining core cellular functions. The two main sources of free energy in living cells are adenosine triphosphate (ATP) molecule and proton motive force (PMF). The ATP molecule is the energy "currency" of living organisms used for biosynthesis and transport. The PMF is a direct consequence of the activity of the electron transport chain or substrate level phosphorylation, and serves as the energy source driving numerous cellular processes: ATP production, motility and active membrane transport. The two are interlinked, ordinarily PMF is used to synthesise ATP, but ATP can drive the production of PMF as well (Keis et al., 2006). As early as 1791 Luigi Galvani proposed that life processes generate electricity (Galvani, 1791; Green, 1953). However, it took more than a century for Hugo Fricke to measure the capacitance of biological membrane (Fricke, 1923) and for Peter Mitchell to explain that PMF is an electrochemical gradient of protons across the membrane that powers the production of ATP (Mitchell, 1961). PMF consists of the two components: pH difference between cytoplasm and the external environment (ƊpH = pHin -- pHout), and the electrical potential across the membrane (Vm, we note that the build up of charge occurs at ∼ nm-thin layer close to the biological membrane (Nelson, 2003)). PMF = Vm -- 2.303kT e ƊpH (1) where k is the Boltzmann constant, T is the temperature and e is the elementary charge. Since life generates electricity used to power its processes and cell membrane acts as a capacitor, it is reasonable to represent the rest of the cell components with an electrical circuit analogy (Van Rotterdam et al., 2002; Walter et al., 2007), Fig. 1A. Then, proton fluxes are currents, oxidative or substrate-level phosphorylation can be considered as an imperfect battery with non-zero internal resistance, and the membrane resistance and capacitance are connected in parallel. If external pH equals that of the bacterial cytoplasm, and for Escherichia coli the latter is known (Slonczewski et al., 1981; Zilberstein et al., 1984; Wilks and Slonczewski, 2007), Vm in the circuit equals the PMF. The circuit analogy in Fig. 1A gives a mathematical framework that helps us understand cellular free energy maintenance in a range of different conditions. For example, we can predict changes in Vm when circuit parameters change: a battery depends on the available carbon source and internal resistance Ri increases in presence of electron transport chain inhibitors (such as sodium azide (Noumi et al., 1987)). Furthermore, if we could measure Vm with an equivalent of a "voltmeter" we could predict the mechanism and dynamics of the damage as the cells are exposed to various external stresses, as well as obtain functional dependence between affected circuit parameters and the amplitude of the stress. Here we report the use of bacterial flagellar motor as such a "voltmeter" and reveal the mechanisms of damage caused by chosen stresses. We confirm the behaviour of a known ionophore (indole) (Chimerel et al., 2012), discover that butanol is an ionophore, and quantitatively describe the nature of damage caused by the light of shorter wavelengths. Our approach of combining high-precision PMF (Vm) measurements and the "electrical circuit interpretation" of the cell serves as a powerful tool needed for quantitative bacterial electrophysiology. MATERIALS AND METHODS E. coli strains E. coli EK07 strain is constructed as described in Supplementary Materials. Highly motile E. coli strain MG1655 with an insertion sequence element in the flhD operon (Barker et al., 2004) is modified to have fliC gene replaced by fliCsticky (Kuwajima, 1988), which produces flagellar filaments that stick to glass or polystyrene surfaces. Additionally, pHluorin (Miesenbock et al., 1998; Morimoto et al., 2011) gene under strong constitutive Vibrio harveyi cytochrome C oxidase promoter (Pilizota and Shaevitz, 2012) is placed onto attTn7 site of the chromosome. All the chromosomal alterations are generated using plasmid mediated gene replacement technique (Link and Phillips, 1997). E. coli growth and media EK07 cells are grown in Lysogeny broth (LB: 10 g tryptone, 5 g yeast extract, 10 g NaCl per 1 L) after diluting from the overnight culture as 1:2000, at 37◦C with shaking (220 rpm) to OD=2.0 (Spectronic 200E Spectrophotometer, Thermo Scientific, UK). We found that the yield of the single motor experiment, which is defined as the number of spinning beads in the field of view and likely corresponding to the flagellar motor expression level, was maximised at this OD in agreement with (Amsler et al., 1993). Growth curves of the EK07 and the parent MG1655 strain are given in SI Fig. 1. After growth cells are washed (3 times by centrifugation at 8000 g for 2 min) into MM9 (aqueous solution of 50 mM Na2HPO4, 25 mM NaH2PO4, 8.5 mM NaCl and 18.7 mM NH4Cl with added 0.1 mM CaCl2, 1 mM KCl, 2 mM MgSO4 and 0.3% D-glucose) adjusted to pH=7.5 or PBS (aqueous solution of 154 mM NaCl, 5 mM Na2HPO4 and 1.5 mM KH2PO4) adjusted to pH=7.5. Indole treatment is performed in MM9 and butanol and photodamage experiments in MM9 and PBS. Microscope slides preparation To shorten flagella, cells are "sheared" as described previously (Bai et al., 2010; Rosko et al., 2017) and washed as above. For butanol and indole treatment tunnel-slides are prepared as before (Rosko et al., 2017), see also SI Fig. 2A). For photodamage experiments flow-cells are manufactured by drilling (AcerDent, UK) two 1.8 mm holes on opposite ends of the microscope slide and attaching Tygon R(cid:13) Microbore tubing (Saint Gobain Performance Plastics, France). The flow-cell is then created by attaching gene frame (Fisher Scientific Ltd, USA) to the slide and covering it with a cover glass (SI Fig. 2B). Surface of the cover slide is coated with 0.1% poly-L-lysine (PLL) by flushing PLL through the flow- cell/tunnel-slide for ∼10 s followed by washing it out with the excessive volume of growth medium. Sheared and washed cells are then loaded into the flow-cell/tunnel-slide and incubated for 10 min to allow attachment. Excessive cells are washed out with the growth medium. Subsequently, 0.5 m in diameter polystyrene beads (Polysciences, Inc, USA) are added to the flow-cell/tunnel-slide and incubated for 10 min with consequent washing out of the non-attached beads. 2/26 Microscopy and data collection Back-focal-plane interferometry (Denk and Webb, 1990; Svoboda et al., 1993) is performed as previously described (Rosko et al., 2017). Briefly, heavily attenuated optical trap (855 nm laser) is used to detect the rotation of a polystyrene bead attached to a truncated flagellar filament (Fig. 1B). Time course of the bead rotation is recorded with the position-sensitive detector (PSD Model 2931, New Focus, USA) at 10 kHz, and a 2.5 kHz cutoff anti-aliasing filter applied before processing (Fig. 1B). Bead position (x,y) is calculated from photocurrents I1 -- I4 as (I1 + I2 -- (I3 + I4))/(I1 + I3 + I2 + I4) = 2x/L and (I1 + I3 -- (I2 + I4))/(I1 + I3 + I2 + I4) = 2y/L, where L is the PSD detector side length. Fluorescent images of pH sensitive pHluorin are taken in the same custom-built microscope with iXon Ultra EMCCD camera (Andor, UK). OptoLED Dual (Cairn Research Ltd, UK) independently driving two LEDs is used for the illumination. Narrow spectrum UV LED is used for excitation at 395 nm and Neutral White LED with ET470/40x filter (Chroma Technology, USA) for 475 nm excitation. Emission is taken at 520 nm using ET525/40x filter (Chroma Technology, USA). Exposure time is fixed at 50 ms for butanol and indole treatment experiments and varies from 10 to 200 ms for photodamage experiments. Applying stresses Butanol (1-Butanol for molecular biology, (cid:62)99%, Sigma-Aldrich, USA) and indole (Indole, analytical standard, Sigma-Aldrich, USA) treatment is performed as follows: after recording the motor speed for 2 min, 20 l of MM9 (or PBS) supplemented with a given concentration of butanol or indole is flushed into the tunnel-slide. Flush is done by placing a droplet of liquid on one, and collecting it with a piece of tissue paper on the other side of the tunnel (Buda et al., 2016). Duration of the flush is no longer than 10 s. 10 l droplets of shocking solution are then placed on both side of the tunnel to minimise evaporation. The shock motor speed is recorded for 10 min, followed by a flush back into MM9 (or PBS) medium. Postshock speed is recorded for 5 min. The motor speed recording is uninterrupted for the duration of the experiment (total of 17 min). For pH control experiments fluorescent images are taken every 90 seconds. Control flushes with media containing no indole/butanol are shown in SI Fig. 3. Photodamage experiments are performed as follows: using the flow-cell MM9 or PBS is constantly supplied at 10 l/min rate with a syringe pump (Fusion 400, Chemyx Inc., USA). Cells are sequentially exposed to the light of λ=395 nm and 475 nm. Speed recording starts simultaneously with the light exposure. The camera exposure time (texp,cam) and sampling rate are controlled with a custom written LabView program. texp,cam are set the same for both wavelengths, however hardware adds a different delay, thus effective light exposure times are texp,light =225 ms+texp,cam for 475 nm and 55 ms+texp,cam for 395 nm. We record texp,light and sampling rate throughout the experiment to calculate the effective light power (Peff) as the total energy delivered, divided by the total length of the individual motor speed recording. Total energy delivered is estimated by measuring the illumination power in the sample plane multiplied by the total time of light exposure and divided by the illumination area. We measured the illumination area by photobleaching part of the slide and measuring the diameter of the bleached region (d ≈ 220 m). Control speed traces with no light exposure are shown in SI Fig. 4. Data analysis A flat-top window discrete Fourier transform (window size=16384 data points with dt =0.01 s) is applied to x and y coordinates of bead position to obtain a time series motor speed record. This speed records we refer to as raw speed traces (Fig. 1C, 2A, 3A, 4A, SI Fig. 3, 4, 7, 8). Raw traces are further processed as follows: (a) absolute values are taken, (b) values below 10 Hz are removed and 50 Hz AC frequency values disregarded, (c) remaining data points are median filtered with 201 points moving window. To calculate mean speeds we apply a 10 s moving window on the speed traces processed as above. In addition to above, photodamage traces are normalised. First, 30 s of the trace is split into 60 windows containing 50 points each. The mean of maximum values found within each window is calculated and considered the initial speed value, by which the rest of the trace is normalised. Each normalised trace is fitted with a single parameter exponential: y = e-αx. For Fig. 2D and 3D hyperbolic function fitted is y = 1 Kx+1 and Kx2+1, where K is a fitting parameter. All fittings are performed in Python (SciPy quadratic hyperbolic y = module, curve fit optimization) with maximum number of calls to the optimization function taken as 20 000. 1 3/26 RESULTS PMF measurements via flagellar motor speed can be used to analyse stress-induced damage. The electric circuit analogy (Fig. 1A) gives a mathematical framework needed to understand cellular free energy maintenance in a range of different conditions. For example, under given external stress it allows us to discern the affected component of the cell and predict the mechanism of damage caused by the stress in the following manner. Membrane capacitance is set by the geometry of the lipid bilayer and unlikely to be altered on shorter time scales. Vc is the theoretical maximum potential a cell can generate in a given environment and from a given internalised (carbon) source. Stress can affect Vc only by damaging specific carbon transporters and, thus, is media-dependent. Furthermore, in starvation buffer where E. coli uses internal carbon sources (Nystrom and Gustavsson, 1998) Vc will not be changed by the stress. Ri defines the inefficiency of the catabolism, comprising the drop from Vc as a specific carbon source gets metabolised via a large number of catabolic enzymes. These enzymes are at least partially carbon source specific, thus the stress that targets Ri will be media-dependent. Finally, while the Re value is growth media-dependent, the membrane targeting stresses that influence Re will be media-independent. Once we pin-down the affected component, we employ Kirchoff's laws to express it as a function of stress-induced membrane potential change (Vm/Vm,0), which we measure using bacterial flagellar motor (BFM) as a "voltmeter"(Fig. 1A). BFM is a nano-machine that enables bacterial swimming (Sowa and Berry, 2008) via PMF powered rotation (Manson et al., 1980; Matsuura et al., 1977; Meister et al., 1987; Fung and Berg, 1995). The motor speed (ω), usually reaching couple of hundred Hz (Lowe et al., 1987), varies linearly with PMF (Fung and Berg, 1995; Gabel and Berg, 2003). While BFM can be actively slowed down, e.g. when cell enter stationary phase (Amsler et al., 1993), on shorter time scales the linearity between the motor speed and PMF allows us to use the motor speed as a PMF indicator, and when pHcytoplasm = pHexternal as a Vm indicator as well. Here we consider only the situation where ƊpH ≈ 0, which we set by adjusting the external pH to known internal pH of E. coli (Slonczewski et al., 1981), and in the rest of the text use PMF and Vm interchangeably. In addition, EK07 strain we constructed (see Materials and Methods) carries a chromosomal copy of the gene encoding pHluorin protein, which we use to check that our expectation is correct, i.e. cytoplasmic pH during the experiments stays constant and at the level of external pH (SI Fig. 5). We thus have: ω = ξ· PMF = ξ· Vm ω ω0 = PMF PMF0 = Vm Vm,0 = f(S,t) (2a) (2b) where we assumed that ω changes as a function of stress amplitude and time f(S,t), ξ is a constant and index 0 denotes the variable value prior to stress. We measure ω using back-focal-plane interferometry (Svoboda et al., 1993) and a polystyrene bead attached to a short filament stub (see Materials and Methods and Fig. 1B) (Bai et al., 2010). An example trace of BFM speed is given in Fig. 1C. Using equation (2b) and the circuit analogy we can express each circuit component as a function of stress. To do so, we simplify the electric circuit by estimating the RC constant of the cell membrane. Capacitance and resistance of the bacterial membrane have been reported as C∼1 F/cm2 (Fricke et al., 1956; Hodgkin et al., 1952) and R∼10-1000 Ohm· cm2 (Miyamoto and Thompson, 1967; Chimerel et al., 2012), which gives RC in the range of 10 -- 5 to 10 -- 3 s. Thus, the current through the capacitor is zero prior to the stress application (when the system is in steady-state), as well as post stress application when t > 1 ms (i.e. on the time scales of our experiment). Next we consider ƊG of NADH oxidation only, and compute that respiratory chain can produce Vc ∼-360 mV (Walter et al., 2007). Yet, physiological value of the membrane potential of respiring bacteria is approximately equal to -160 mV (Tran and Unden, 1998), indicating that roughly half of the membrane potential drops at the internal resistance, i.e. Ri,0 ≈ Re,0. Taking the two simplifications into account we arrive to (see Fig. 1A and Supplementary Material for detailed deduction of equations): Vc Vc,0 = f(S,t) (3a) 4/26 Ri Ri,0 Re Re,0 = = 2 f(S,t) -- 1 f(S,t) 2 -- f(S,t) (3b) (3c) Figure 1. (A) Electric circuit equivalent of an E. coli cell. Oxidative (or substrate-level) phosphorylation is shown as a battery Vc with an internal resistance Ri, the membrane with capacitance C and resistance Re, and i1 to i3 are the currents. Bacterial flagellar motor (BFM) is shown as a "voltmeter" that measures membrane potential, Vm. (B) Schematic of the "bead-assay" and back-focal-plane interferometry. A cell is attached to a cover glass with a truncated flagellar filament made "sticky" to polystyrene beads. The bead is brought into a heavily attenuated optical trap and its position measured with position sensitive detector (see Materials and Methods). (C) An example of raw motor speed trace recorded with back- focal-plane interferometry. Positive frequencies correspond to counter-clockwise and negative to the clockwise rotation of the flagellar motor (Bai et al., 2010). In the subsequent figures we show absolute values of the rotational speeds. PMF dynamics analysis confirms indole is an ionophore We test the proposed circuit analogy and applicability of the BFM speed as the voltmeter by applying a known membrane stress. We choose a cell signaling molecule indole that at millimolar concentrations forms a dimer and acts as an ionophore (Chimerel et al., 2012). Ionophores are molecules that carry ions across the lipid bilayer, thus we expect the membrane resistance to decrease (ion conductance increases) when indole is present in the medium. Furthermore, we expect to recover previously demonstrated parabolic dependence of membrane conductance on indole concentration (Chimerel et al., 2012). Fig. 2A shows examples of individual motor speed recordings prior, during and post treatment with a given concentration of indole. Motor speed drops immediately with the addition of indole, and stays at approximately the same level until indole is removed, at which point it recovers to the initial level. The speed change caused by indole is faster than 10 ms (our experimental resolution), confirming the estimate of membrane RC constant, and justifying the assumption that the current through the capacitance in Fig. 1A circuit is negligible. To confirm the dependence of the membrane resistance on indole we find the relative change in motor speed at a given stress concentration. Fig. 2B shows the mean speed traces for different indole concentrations (see Materials and Methods for mean speed calculation) and in Fig. 2C we plot the probability densities of preshock and shock speeds. From the Gaussian fits to preshock and shock speed distributions we obtain mean shock speeds for a given indole concentration, and plot them normalised to the preshock speed, Fig. 2D. We fit the normalised speeds with hyperbolic or quadratic hyperbolic function (see Materials and Methods, both of which yield good quality fits with R2 higher than 0.90). The concentrations of indole at which the quadratic dependence becomes particularly obvious are higher than 2.5 mM (Chimerel et al., 2012), where we use 0-2.5 mM range. Therefore, our result confirms the accuracy of our proposed approach. 5/26 Figure 2. BFM speed drops rapidly and increasingly with increasing indole concentration (A) Examples of raw motor speed traces at 5 different indole concentrations. Indole is delivered into the tunnel-slide 2 min after the recording commences, and removed after 12 min. (B) Mean speeds of n(cid:62)20 motor speeds for each indole concentration are shown against time. Each motor recording is performed on a different cell, thus the number of motors corresponds to the number of different individual cells. Preshock speed is calculated for time interval between 0 and 110 s (indicated in the figure). Shock speed is calculated from the 130 to 660 s of the motor recording. Preshock and shock intervals were chosen to exclude the duration of the flush. Standard errors are given, but not visible (for standard deviations see SI Fig. 6A). (C) Probability density of motor speeds for each indole concentration. Experimental data is fitted with a Gaussian probability density function. (D) Normalised BMF speeds plotted against indole concentration. Error bars represent standard error of the mean, and dotted lines show hyperbolic (black) and quadratic hyperbolic (grey) fit (R2 = 0.97 and R2 = 0.95 respectively). Butanol acts as an ionophore, changing membrane conductance linearly with concen- tration. To determine the mechanism of action of an unknown stress we choose butanol. Previous work indicates that butanol interacts with cell membrane and weakens it, but the exact mechanism of cell damage is unknown (Fletcher et al., 2016). We perform the BFM speed measurements in E. coli cells treated with butanol. The experimental protocol of butanol delivery is the same as for indole. Fig. 3A and 3B show examples of raw traces and mean speed traces prior, during and post butanol shock in MM9. Immediately upon butanol stress motor speed drops, and upon butanol removal it recovers to the initial value, Fig. 3A. Motor speed distributions at a given butanol concentrations remain narrow, and we fit them with Gaussian curves (Fig. 3C). Fig. 3D shows normalised motor speeds, calculated as mean values of the distributions given in Fig. 3C, and plotted against butanol concentration for both MM9 media and PBS. The relative speed drop observed in the presence of butanol is media independent, and alike that observed for indole. The finding suggesting that, on the time scale of our experiment, butanol cause non-permanent membrane damage and acts as an ionophore. The normalised motor speed dependence on butanol concentration is hyperbolic, and we obtain equation (3c) for membrane resistance: 6/26 Figure 3. BFM speed drops sharply and reversibly after butanol treatment. (A) Examples of raw BFM speed traces for 5 different butanol concentrations. Butanol is delivered 2 min into the recording and removed after 12 min. (B) Mean speeds of n(cid:62)20 cells per different butanol concentrations are plotted against time. Preshock and shock speeds are calculated in the 0 to 110 s, and 130 to 660 s time interval, respectively. Standard errors of the mean are given, but not visible. Standard deviations of the same traces are given in SI Fig. 6B. (C) Probability densities of shock speed for each butanol concentration and the preshock speed. (D) Shock speeds obtained from the distributions are normalised by the preshock speed and plotted against butanol concentration. Blue diamonds show cells in MM9 media and red diamonds cells in PBS. Error bars represent standard error of the mean. Hyperbolic fit is given as a black dotted line (R2 = 0.96). Re = Re,0 7.8· cbut + 1 (4) where cbut is a butanol concentration in percents (%) and 7.8 is a value of constant K obtained from the hyperbolic fit (see Materials and Methods). We observe the speed restoration after butanol removal even after multiple treatments of the same cell. SI Fig. 7 shows several consecutive butanol stresses each lasting 60 s (A) or 30 s (B), where after each treatment motor speed is fully restored. Photodamage increases membrane conductance that scales with the light power. As an example of a complex stress we next choose to characterise light induced damage. While previous reports indicate that light causes wavelength dependent damage to bacterial cells (Ashkin et al., 1987; Neuman et al., 1999), they also suggest that the nature of damage is complex. Most likely the cause of the damage is formation of reactive oxygen species (ROS) (de Jager et al., 2017; Lockwood et al., 2005), which have been shown to perturb multiple components of the cell: DNA, RNA, proteins and lipids (Cabiscol et al., 2000; Zhao and Drlica, 2014). To apply light of a certain wavelength and intensity to bacterial cells we use a flow-cell (see Materials and Methods). During the light exposure cells are continuously supplied with fresh media at 10 l/min flow rate. We apply the light of 395 nm and 475 nm wavelengths as the choice allows us to simultaneously measure internal pH of bacteria. 7/26 Fig. 4A shows example BFM speed traces during exposure to light of different effective powers (Peff) delivered to the cells. Peff is calculated as the total energy delivered divided by the total time the light is on (see Materials and Methods). Fig. 4A shows that BFM speed gradually decreases in time during exposure to light and that the decrease rate scales with the Peff, also visible in Fig. 4B showing mean BFM speed traces for the same four effective powers. Figure 4. Rate of the motor speed decay increases with the light power. (A) Examples of raw traces at four different effective powers, Peff = 20, 38.4, 153 or 591 mW/cm2. (B) Mean BFM speed at different illumination powers (21 to 34 cells are recorded per condition). (C) Averaged exponential fits for different illumination powers with standard error. Each individual motor trace is fitted with an exponential function and the mean of fitting parameter α is calculated for each Peff. (D) Exponential fit coefficient α is plotted against illumination power. Blue diamonds show cells in MM9 media and red diamonds cells in PBS. Error bars represent standard error and dotted line the logarithmic fit (R2 = 0.906). The total number of cells in MM9 is 277 and in PBS 116. To identify the functional dependence of the speed decrease rate on Peff we fit individual normalised traces with the simple exponential function: ω/ω0 = e -- αt, with the single fitting parameter α. Mean of the fits with standard errors at corresponding four different powers are shown in Fig. 4C, and Fig. 4D shows fit coefficient α plotted against the light power for both MM9 medium and PBS. The effect of light on Vm is present in PBS and of same functional dependence, thus on the time scales of our experiment light affects primarily the membrane resistance, Re. Together with the fact that the speed decrease rates stay the same at a given Peff, the finding suggests that on the time scale of our experiment there is no active membrane repair. We further confirm this by measuring the motor speed after we expose the cells to light for shorter periods of time. SI Fig. 8 shows that when the illumination ceases after 5 or 15 min the (decreased) BFM speed remains the same with no visible recovery. We also check that light damage is not enhanced by the presence of the fluorescent protein (pHluorin) in the cytoplasm, SI Fig. 9. Fig. 4D enables us to determine functional relationship between effective power and α, which increases as a logarithm of the normalised Peff, i.e. Peff,norm = Peff/(mW· cm -- 2). Thus, for our initial exponential fit we obtain: (5) where a and b are wavelength specific parameters, a = 0.00064 s -- 1 and b = -0.00181 s -- 1 and equation (5) holds for Peff > Peff,0. ω = ω0 · e -- (alnPeff,norm+b)t 8/26 The minimum power required for the damage to occur is defined as Peff,0 = e -- b 395 nm and 475 nm this is ∼ 17 mW/cm2. Re-writing the equation (5) in terms of Peff,0 we get: a mW/cm2, and for (cid:18)Peff,0 (cid:19)at Peff ω = ω0 (6) (7) Finally, applying (6) to equation (3c) we derive Re functional dependence on the effective power: where a is the fit coefficient in Fig. 4D. Re = 2 (cid:17)at Re,0 (cid:16) Peff Peff,0 , -- 1 DISCUSSION Arguably, one of the defining features of life is its ability to avoid thermodynamic equilibrium (death) by achieving a steady state supply of free energy. Chemiosmotic theory explained that the production of life's energy currency, the ATP molecule, proceeds via the generation of trans-membrane electrochemical potential. The ability to measure and control voltage and current across the cellular membrane with the patch-clamp technique had far reaching consequences for our understanding of cells such as neurones, where the electrical inputs govern signal transmission (Hodgkin et al., 1952). In the cases of bacteria, and their small size, we are unable to gain the same level of control over these parameters (Ruthe and Adler, 1985; Martinac et al., 1987), despite the fact that the ability to do so would open a range of currently inaccessible questions that are at the basis of bacterial free energy maintenance, and consequently, survival. Here we demonstrate the use of BFM as a fast voltmeter, enabling quantitative, in vivo studies of electrochemical properties of the bacterial membrane. Alternative methods for measuring Vm in E. coli rely on fluorescent readout (Ehrenberg et al., 1988; Prindle et al., 2015; Kralj et al., 2011). However, Nernstian dyes (Ehrenberg et al., 1988; Prindle et al., 2015) sometimes fail to penetrate E. coli's membrane (Lo et al., 2007), can be a substrate for the outer membrane efflux system TolC (Mancini et al., 2018) and in external conditions where they do equilibrate across the membrane, they do so on the time scales of minutes (Lo et al., 2007; te Winkel et al., 2016). Voltage sensitive membrane proteins that can be used in E. coli require delivery of light of high power (Kralj et al., 2011). BFM, on the other hand, is native to E. coli and expressed in a range of conditions (Cremer et al., 2018). Speed measurements via back-focal-plane interferometry or fast cameras do not rely on fluorescent illumination and offer high time resolution (up to 0.5 ms (Pilizota et al., 2007)). We choose to work with cells grown into late exponential phase in LB to maximise our experimental yield. The approach is, however, more widely applicable as BFM is expressed in a range of other conditions (with the exception of late stationary phase cells (Amsler et al., 1993; Cremer et al., 2018)), where we expect the BFM bead-assay yield to somewhat vary with the condition. It is possible that cells grown to early or mid exponential phase, or cells grown to steady state in different growth media, will have different electrochemical properties, which can be measured with our approach in the future. We choose to work in the conditions that satisfy ƊpH≈0, and thus Vm is the only contribution to the PMF. However, BFM speed measurements can be extended to conditions where ƊpH contribution to the PMF is not negligible, Vm in this case will be calculated from equation (1). Extending the use of BFM as the voltmeter for long term measurements (into hours and days) is possible. We note that on longer time scales motor can be actively slowed down via YcgR protein (Boehm et al., 2010; Paul et al., 2010), and such long term measurements would likely require YcgR deletion background. We base the use of BFM as the cell's voltmeter on the proportionality between motor speed and PMF, measured first more than 20 years ago (Fung and Berg, 1995; Gabel and Berg, 2003). Recent experiments show that BFM also exhibits mechanosensing (Lele et al., 2013; Tipping et al., 2013), where stator unit incorporation depends on the motor torque. These recent findings indicate an intriguing control mechanism, where mechanosensing and the ion flux combined result in the characteristic proportional relationship between the BFM speed and PMF. It will be interesting to fully ascertain the exact molecular mechanism behind the PMF-motor speed relationship, and we think the ability to fine-control the PMF loss can contribute to that understanding. Using the electric circuit analogy for the membrane fluxes, and BFM as the cell's "voltmeter" we demonstrate the effect of three different stresses on the cell's membrane conductance. For the known stress, 9/26 indole, we confirm it acts as an ionophore. For the first unknown stress we applied, butanol, we show its presence decreases membrane resistance, inversely proportional to the butanol concentration. Thus, we conclude that, in the concentration range we tested and on the 15 min time scale, butanol behaves as an ionophore in a manner similar to indole or CCCP (Chimerel et al., 2012). With analysis alike we presented, butanol action can be characterised further, e. g. defining the minimum concentration and incubation time required for the effect to become irreversible. For our last stress, light of short wavelengths, we show that it affects membrane resistance and functionally describe the damage in relation to time and Peff. Light-induced changes in membrane permeability have been reported in artificial planar lipid bilayer systems and liposomes in the presence of photosensitisers (McRae et al., 1985; Pashkovskaya et al., 2010; Kotova et al., 2011; Wong-Ekkabut et al., 2007). The most likely cause of such changes is ROS induced chain-reaction lipid peroxidation (Girotti, 1985, 1990; Halliwell et al., 1993; Heck et al., 2003; Lavi et al., 2010). Presence of peroxidised lipids can change bilayer physical and electrical properties (Dobretsov et al., 1977; Richter, 1987; Birben et al., 2012), e.g. it has been suggested that it induces formation of hydrophobic pre-pores and their later transformation into hydrophilic pores permeable to ions (Kotova et al., 2011; Wong-Ekkabut et al., 2007). Based on the previous work, and our real time, in vivo measurements we propose the following model for the complex nature of the light-induced membrane damage. Exposure to light causes the formation of ROS that induce lipid peroxidation, and thus alter the electric properties of the membrane. In particular, its permeability to ions due to the formation of hydrophilic pores. In contrast to the ionophores that carry ions across the membrane without causing membrane damage, the drop in Vm we observe under light proceeds as a result of slower, multi-step formation of lipid pores that require active repair to be mitigated. Therefore, we do not see any fast recovery after illumination ceases (SI Fig. 8), and the chain-reaction nature of the process results in the exponential-like decay of membrane potential. Living cells have built-in mechanisms of coping with oxidative stress, for example SoxRS/OxyR regulons containing multiple antioxidant-encoding genes, such as sodA (manganese superoxide dismutase) or katG (hydroperoxidase I) (Storz and Imlayt, 1999; Birben et al., 2012). The existence of defence mechanisms explains the occurrence of the minimum power required to cause the damage. Less power, even if it causes ROS formation, will not damage the cells that cope using internal protection enzymes. The value of the minimal damage-causing power we measured can indicate the abundance of internal protective resources available to the cell, as well as define the power range for fluorescence imaging that should be used to ensures no (unaccounted for) damage is inflicted to the cells by the exposure to light. Future applications of our approach include, but are not limited to, studying other damage mechanisms and characterising unknown bacterial membrane properties, e. g. overall resistance in different growth conditions. Lastly, based on our measurements we suggest the use of light for delivery of small molecules, such as antimicrobial peptides or fluorescent dyes, which otherwise fail to penetrate E. coli's membrane (Lo et al., 2007). Author Contributions EK, CJL and TP designed research. EK performed research and analysed data. EK, CJL and TP interpreted results and wrote the paper. Acknowledgements We thank all the members of Pilizota and Lo laboratories, Zaki Leghtas, Jelena Baranovic, Bai Fan, Peter Swain, Ivan Maryshev, Ivan Erofeev, Calin Guet and Munehiro Assaly for useful discussions. EK was supported by the Global Research and Principal's Career Development PhD Scholarships. TP and CJL were supported by the Human Frontier Science Program Grant (RGP0041/2015), and CJL by the Ministry of Science and Technology, Republic of China (MOST-106-2112-M-008-023). REFERENCES Amsler, C. D., Cho, M., and Matsumura, P. (1993). Multiple factors underlying the maximum motility of Escherichia coli as cultures enter post-exponential growth. Journal of Bacteriology, 175(19):6238 -- 6244. Ashkin, A., Dziedzic, J. M., and Yamane, T. (1987). Optical trapping and manipulation of single cells using infrared laser beams. Nature, 330(6150):769 -- 771. 10/26 Bai, F., Branch, R. W., Nicolau, D. V., Pilizota, T., Steel, B. C., Maini, P. K., and Berry, R. M. (2010). Conformational spread as a mechanism for cooperativity in the bacterial flagellar switch. Science, 327(5966):685 -- 689. Barker, C. S., Pruss, B. M., and Matsumura, P. (2004). Increased motility of Escherichia coli by insertion sequence element integration into the regulatory region of the flhD operon. Journal of Bacteriology, 186(22):7529 -- 7537. Berg, H. C. and Turner, L. (1993). Torque generated by the flagellar motor of Escherichia coli. Biophysical Journal, 65(5):2201 -- 16. Birben, E., Murat, U., Md, S., Sackesen, C., Erzurum, S., and Kalayci, O. (2012). Oxidative Stress and Antioxidant Defense. World Allergy Organization Journal, 5(1):9 -- 19. Boehm, A., Kaiser, M., Li, H., Spangler, C., Kasper, C. A., Ackermann, M., Kaever, V., Sourjik, V., Roth, V., and Jenal, U. (2010). Second Messenger-Mediated Adjustment of Bacterial Swimming Velocity. Cell, 141(1):107 -- 116. Buda, R., Liu, Y., Yang, J., Hegde, S., Stevenson, K., Bai, F., and Pilizota, T. (2016). Dynamics of Escherichia coli 's passive response to a sudden decrease in external osmolarity. Proceedings of the National Academy of Sciences of the United States of America, 113(40):E5838 -- E5846. Cabiscol, E., Tamarit, J., and Ros, J. (2000). Oxidative stress in bacteria and protein damage by reactive oxygen species. International Microbiology, 3(1):3 -- 8. Chimerel, C., Field, C. M., Pinero-Fernandez, S., Keyser, U. F., and Summers, D. K. (2012). Indole prevents Escherichia coli cell division by modulating membrane potential. Biochimica et Biophysica Acta - Biomembranes, 1818(7):1590 -- 1594. Cremer, J., Honda, T., Tang, Y., Ng, J., Vergassola, M., and Hwa, T. (2018). On the growth and migration of chemotactic bacterial populations in nutrient-replete environments. In preparation. de Jager, T. L., Cockrell, A. E., and Du Plessis, S. S. (2017). Ultraviolet Light Induced Generation of Reactive Oxygen Species. In Ahmad, S., editor, Ultraviolet Light in Human Health, Diseases and Environment. Advances in Experimental Medicine and Biology, pages 15 -- 23. Springer, Cham. Denk, W. and Webb, W. W. (1990). Optical measurement of picometer displacements of transparent microscopic objects. Applied optics, 29(16):2382 -- 2391. Dobretsov, G. E., Borschevskaya, T. A., Petrov, V. A., and Vladimirov, Y. A. (1977). The increase of phospholipid bilayer rigidity after lipid peroxidation. FEBS Letters, 84(1):125 -- 128. Ehrenberg, B., Montana, V., Wei, M. D., Wuskell, J. P., and Loew, L. M. (1988). Membrane potential can be determined in individual cells from the nernstian distribution of cationic dyes. Biophysical Journal, 53(5):785 -- 794. Fletcher, E., Pilizota, T., Davies, P. R., McVey, A., and French, C. E. (2016). Characterization of the effects of n-butanol on the cell envelope of E. coli. Applied Microbiology and Biotechnology, 100(22):9653 -- 9659. Fricke, H. (1923). The electric capacity of cell suspension. Physical Review, 21:708 -- 709. Fricke, H., Schwan, H. P., Li, K., and Bryson, V. (1956). A dielectric study of the low-conductance surface membrane in E. coli. Nature, 177(4499):134 -- 135. Fung, D. C. and Berg, H. C. (1995). Powering the flagellar motor of Escherichia coli with an external voltage source. Nature, 375(6534):809 -- 812. Gabel, C. V. and Berg, H. C. (2003). The speed of the flagellar rotary motor of Escherichia coli varies linearly with protonmotive force. Proceedings of the National Academy of Sciences of the United States of America, 100(15):8748 -- 8751. Galvani, L. (1791). De viribus electricitatis in motu musculari commentarius. De Bononiensi Scientiarum et Artium Instituto atque Academia Commentarii, tomus septimus., pages 363 -- 418. Girotti, A. W. (1985). Mechanisms of lipid peroxidation. Journal of Free Radicals in Biology and Medicine, 1(2):87 -- 95. photobiology, 51(4):497 -- 509. Girotti, A. W. (1990). Photodynamic lipid peroxidation in biological systems. Photochemistry and Green, R. M. (1953). A Translation of Luigi Galvani's De viribus electricitatis in motu musculari commentarius. Commentary on the Effect of Electricity on Muscular Motion. Journal of the American Medical Association, 153(10):989. Halliwell, B., Chirico, S., Crawford, M. A., Bjerve, K. S., and Gey, K. F. (1993). Lipid peroxida- tion: Its mechanism, measurement, and significance. American Journal of Clinical Nutrition, 57(5 11/26 SUPPL.):715S -- 725S. Heck, D. E., Vetrano, A. M., Mariano, T. M., and Laskin, J. D. (2003). UVB light stimulates production of reactive oxygen species: Unexpected role for catalase. Journal of Biological Chemistry, 278(25):22432 -- 22436. Hodgkin, A. L., Huxley, A. F., and Katz, B. (1952). Measurement of current-voltage relations in the membrane of the giant axon of Loligo. The Journal of Physiology, 116(4):424 -- 448. Keis, S., Stocker, A., Dimroth, P., and Cook, G. M. (2006). Inhibition of ATP hydrolysis by thermoalka- liphilic F1F 0-ATP synthase is controlled by the C terminus of the ε subunit. Journal of Bacteriology, 188(11):3796 -- 3804. Kotova, E. A., Kuzevanov, A. V., Pashkovskaya, A. A., and Antonenko, Y. N. (2011). Selective permeabi- lization of lipid membranes by photodynamic action via formation of hydrophobic defects or pre-pores. Biochimica et Biophysica Acta - Biomembranes, 1808(9):2252 -- 2257. Kralj, J. M., Hochbaum, D. R., Douglass, A. D., and Cohen, A. E. (2011). Electrical spiking in Escherichia coli probed with a fluorescent voltage-indicating protein. Science, 333(6040):345 -- 348. Kuwajima, G. (1988). Construction of a minimum-size functional flagellin of Escherichia coli. Journal of Bacteriology, 170(7):3305 -- 3309. Lavi, R., Shainberg, A., Shneyvays, V., Hochauser, E., Isaac, A., Zinman, T., Friedmann, H., and Lubart, R. (2010). Detailed analysis of reactive oxygen species induced by visible light in various cell types. Lasers in Surgery and Medicine, 42(6):473 -- 480. Lele, P. P., Hosu, B. G., and Berg, H. C. (2013). Dynamics of mechanosensing in the bacterial flag- ellar motor. Proceedings of the National Academy of Sciences of the United States of America, 110(29):11839 -- 11844. Link, A. J. and Phillips, D. (1997). Methods for generating precise deletions and insertions in the genome of wild-type Escherichia coli : application to open reading frame characterization. Journal of Bacteriology, 179(20):6228 -- 6237. Lo, C. J., Leake, M. C., Pilizota, T., and Berry, R. M. (2007). Nonequivalence of membrane voltage and ion-gradient as driving forces for the bacterial flagellar motor at low load. Biophysical Journal, 93(1):294 -- 302. Lockwood, D. B., Wataha, J. C., Lewis, J. B., Tseng, W. Y., Messer, R. L. W., and Hsu, S. D. (2005). Blue light generates reactive oxygen species (ROS) differentially in tumor vs. normal epithelial cells. Dental Materials, 21(7):683 -- 688. Lowe, G., Meister, M., and Berg, H. C. (1987). Rapid rotation of flagellar bundles in swimming bacteria. Mancini, L., Pu, Y., Tian, T., Bai, F., and Pilizota, T. (2018). Membrane voltage dyes: probes or actuators? Nature, 325:637 -- 640. In preparation. Manson, M. D., Tedesco, P. M., and Berg, H. C. (1980). Energetics of flagellar rotation in bacteria. Journal of Molecular Biology, 138(3):541 -- 561. Martinac, B., Buechner, M., Delcour, A. H., Adler, J., and Kung, C. (1987). Pressure-sensitive ion channel in Escherichia coli. Proceedings of the National Academy of Sciences of the United States of America, 84(8):2297 -- 2301. Martinez, K. A., Kitko, R. D., Mershon, J. P., Adcox, H. E., Malek, K. A., Berkmen, M. B., and Slonczewski, J. L. (2012). Cytoplasmic pH response to acid stress in individual cells of Escherichia coli and Bacillus subtilis observed by fluorescence ratio imaging microscopy. Applied and Environmental Microbiology, 78(10):3706 -- 3714. Matsuura, S., Shioi, J. i., and Imae, Y. (1977). Motility in Bacillus subtilis driven by an artificial protonmotive force. FEBS Letters, 82(2):187 -- 190. McRae, D. G., Yamamoto, E., and Towers, G. H. (1985). The mode of action of polyacetylene and thiophene photosensitizers on liposome permeability to glucose. Biochimica et Biophysica Acta - Biomembranes, 821(3):488 -- 496. Meister, M., Lowe, G., and Berg, H. C. (1987). The proton flux through the bacterial flagellar motor. Cell, 49(5):643 -- 650. Merlin, C., Mcateer, S., and Masters, M. (2002). Tools for Characterization of Escherichia coli Genes of Unknown Function. Journal of Bacteriology, 184(16):4573 -- 4581. Miesenbock, G., De Angelis, D. A., and Rothman, J. E. (1998). Visualizing secretion and synaptic transmission with pH-sensitive green fluorescent proteins. Nature, 394(6689):192 -- 195. 12/26 Mitchell, P. (1961). Coupling of phosphorylation to electron and hydrogen transfer by a chemi-osmotic Miyamoto, V. and Thompson, T. (1967). Some electrical properties of lipid bilayer membranes. Journal type of mechanism. Nature, 191(4784):144 -- 148. of Colloid and Interface Science, 25(1):16 -- 25. Morimoto, Y. V., Kojima, S., Namba, K., and Minamino, T. (2011). M153R mutation in a pH-sensitive green fluorescent protein stabilizes its fusion proteins. PLoS ONE, 6(5). Nelson, P. C. (2003). Biological Physics: Energy, Information, Life. Freeman, W.H. Neuman, K. C., Chadd, E. H., Liou, G. F., Bergman, K., and Block, S. M. (1999). Characterization of photodamage to Escherichia coli in optical traps. Biophysical Journal, 77(5):2856 -- 2863. Noumi, T., Maeda, M., and Futai, M. (1987). Mode of inhibition of sodium azide on H+-ATPase of Escherichia coli. FEBS Letters, 213(2):381 -- 384. Nystrom, T. and Gustavsson, N. (1998). Maintenance energy requirement: What is required for stasis survival of Escherichia coli? Biochimica et Biophysica Acta - Bioenergetics, 1365(1-2):225 -- 231. Otsu, N. (1979). A Threshold Selection Method from Gray-Level Histograms. IEEE Transactions on Systems, Man, and Cybernetics, 9(1):62 -- 66. Pashkovskaya, A., Kotova, E., Zorlu, Y., Dumoulin, F., Ahsen, V., Agapov, I., and Antonenko, Y. (2010). Light-triggered liposomal release: Membrane permeabilization by photodynamic action. Langmuir, 26(8):5725 -- 5733. Paul, K., Nieto, V., Carlquist, W. C., Blair, D. F., and Harshey, R. M. (2010). The c-di-GMP Binding Protein YcgR Controls Flagellar Motor Direction and Speed to Affect Chemotaxis by a "Backstop Brake" Mechanism. Molecular Cell, 38(1):128 -- 139. Pilizota, T., Bilyard, T., Bai, F., Futai, M., Hosokawa, H., and Berry, R. M. (2007). A programmable optical angle clamp for rotary molecular motors. Biophysical Journal, 93(1):264 -- 275. Pilizota, T. and Shaevitz, J. W. (2012). Fast, multiphase volume adaptation to hyperosmotic shock by Escherichia coli. PLoS ONE, 7(4):e35205. Prindle, A., Liu, J., Asally, M., Ly, S., Garcia-Ojalvo, J., and Suel, G. M. (2015). Ion channels enable electrical communication in bacterial communities. Nature, 527(7576):59 -- 63. Richter, C. (1987). Biophysical consequences of lipid peroxidation in membranes. Chemistry and Physics of Lipids, 44(2-4):175 -- 189. Rosko, J. (2017). Osmotaxis in Escherichia coli. PhD thesis, The Edinburgh University. Rosko, J., Martinez, V., Poon, W., and Pilizota, T. (2017). Osmotaxis in Escherichia coli through changes in motor speed. Proceedings of the National Academy of Sciences of the United States of America, 114(38):E7969 -- E7976. Ruthe, H. J. and Adler, J. (1985). Fusion of bacterial spheroplasts by electric fields. Biochimica et Biophysica Acta - Biomembranes, 819(1):105 -- 113. Slonczewski, J. L., Rosen, B. P., Alger, J. R., and Macnab, R. M. (1981). pH homeostasis in Es- cherichia coli: measurement by 31P nuclear magnetic resonance of methylphosphonate and phosphate. Proceedings of the National Academy of Sciences of the United States of America, 78(10):6271 -- 6275. Sowa, Y. and Berry, R. M. (2008). Bacterial flagellar motor. Quarterly Reviews of Biophysics, 41(2):103 -- 132. Storz, G. and Imlayt, J. A. (1999). Oxidative stress. Current Opinion in Microbiology, 2(2):188 -- 194. Svoboda, K., Schmidt, C. F., Schnapp, B. J., and Block, S. M. (1993). Direct observation of kinesin stepping by optical trapping interferometry. Nature, 365(6448):721 -- 727. te Winkel, J. D., Gray, D. A., Seistrup, K. H., Hamoen, L. W., and Strahl, H. (2016). Analysis of Antimicrobial-Triggered Membrane Depolarization Using Voltage Sensitive Dyes. Frontiers in Cell and Developmental Biology, 4:29. Tipping, M. J., Delalez, N. J., Lim, R., Berry, R. M., and Armitage, J. P. (2013). Load-dependent assembly of the bacterial flagellar motor. mBio, 4(4). Tran, Q. H. and Unden, G. (1998). Changes in the proton potential and the cellular energetics of Escherichia coli during growth by aerobic and anaerobic respiration or by fermentation. European Journal of Biochemistry, 251(1-2):538 -- 543. Urh, M., Simpson, D., and Zhao, K. (2009). Affinity Chromatography. General Methods. In Methods in Enzymology, volume 463, chapter 26, pages 417 -- 438. Academic Press. Van Rotterdam, B. J., Crielaard, W., Van Stokkum, I. H., Hellingwerf, K. J., and Westerhoff, H. V. (2002). Simplicity in complexity: The photosynthetic reaction center performs as a simple 0.2 V battery. FEBS 13/26 Letters, 510(1-2):105 -- 107. Walter, J. M., Greenfield, D., Bustamante, C., and Liphardt, J. (2007). Light-powering Escherichia coli with proteorhodopsin. Proceedings of the National Academy of Sciences of the United States of America, 104(7):2408 -- 12. Wang, Y. K., Krasnopeeva, E., Bai, F., Pilizota, T., and Lo, C.-J. (2018). Bacterial Immobilization Assays for Single Cell Experiment. In preparation. Wilks, J. C. and Slonczewski, J. L. (2007). pH of the cytoplasm and periplasm of Escherichia coli: Rapid measurement by green fluorescent protein fluorimetry. Journal of Bacteriology, 189(15):5601 -- 5607. Wong-Ekkabut, J., Xu, Z., Triampo, W., Tang, I. M., Tieleman, D. P., and Monticelli, L. (2007). Effect of lipid peroxidation on the properties of lipid bilayers: A molecular dynamics study. Biophysical Journal, 93(12):4225 -- 4236. Zhao, X. and Drlica, K. (2014). Reactive oxygen species and the bacterial response to lethal stress. Current Opinion in Microbiology, 21:1 -- 6. Zilberstein, D., Agmon, V., Schuldiner, S., and Padan, E. (1984). Escherichia coli intracellular pH, membrane potential, and cell growth. Journal of Bacteriology, 158(1):246 -- 252. 14/26 SUPPLEMENTARY MATERIALS Deduction of electric circuit model Let us consider the circuit from Fig. 1A without yet applying simplifications discussed in the main text. When the system is in equilibrium current trough the capacitor i3 = 0. Upon application of a given stress one of the three components of the circuit change (Ri, Vc or Re), and all three currents become non zero. Then, based on Kirchoff's laws we have: Vc = i1Ri + i2Re i1 = i2 + i3 i2Re = Q C (S1a) (S1b) (S1c) We do not consider scenarios that change more than one circuit element at the same time. Let us first consider the case when overall membrane resistance, Re, changes, i.e. membrane has been damaged and Ri, Vc and C are kept fixed. Based on Ohm's law membrane voltage can be expressed as Vm = i2Re. Thus, to get the functional dependence of Re on the stress amplitude we need to find Vm/Vm,0, equation (2b) and (3c) in the main text. Re Re,0 = Vm Vm,0 · i2,0 i2 (S2) where i2,0 is the current through Re,0, i.e. before Re changed. We are now looking for expressions for i2(Re,t) and i2,0. We know that the current through the capacitor is i3 = dQ/dt and thus from (S1c) follows: We express i3 from (S3) and i1 from (S1b) and apply it to (S1a): di2 dt Re + i2 dRe dt = i3 C di2 dt + i2 Ri + Re + CRi CRiRe dRe dt -- Vc CRiRe = 0 When i3 = 0, i2,0 = i1,0 = i0 based on equation (S1b) and by implementing to (S1a) we get: i0 = Vc Ri + Re,0 (S3) (S4) (S5) because Vc and Ri do not change during the application of the stress. Using (2b) in the main text and (S5) we can now express (S2) as: Re Re,0 = f(S,t)· Vc Ri + Re,0 · 1 i2(Re,t) , (S6) where i2(Re,t) is a solution of (S4). Alternatively, if the stress-affected element is Ri or Vc equation (S3) and (S4) become: di2 dt Re = i3 C di2 dt + i2 Ri + Re CRiRe -- Vc CRiRe = 0 (S7) (S8) We now apply simplifications mentioned in the main text: i3 is always zero (RC is in the range of 10 -- 5 to 10 -- 3 s), and Ri,0 ≈ Re,0. Taking the two simplifications into account we arrive to: i = Vc Ri + Re (S9) 15/26 and f(S,t) = Vm Vm,0 = Vc · Re Ri + Re · Ri,0 + Re,0 Vc,0 · Re,0 = Vc · Re Ri + Re · 2 Vc,0 Then, if Re is affected by the stress, Vc = Vc,0 and Ri = Ri,0 = Re,0 Vm Vm,0 = 2· Re Re + Re,0 = f(S,t) If Ri is affected by the stress, Vc = Vc,0 and Re = Re,0 = Ri,0 Vm Vm,0 = 2· Ri,0 Ri + Ri,0 = f(S,t) Finally, if Vc is affected by the stress, Re = Re,0 = Ri = Ri,0 Vm Vm,0 = Vc Vc,0 = f(S,t) (S10) (S11) (S12) (S13) Supplementary Methods:E. coli strain construction All the chromosomal manipulations were performed using plasmid mediated gene replacement method (PMGR) described previously (Link and Phillips, 1997; Merlin et al., 2002). The method is based on RecA-mediated homologous recombination occurring between homologous regions on the chromosome and plasmid. Backbone plasmid pTOF24 (Merlin et al., 2002) was digested with SalI and PstI restriction enzymes. Inserts were amplified with primers listed in SI Table 1 and assembled together in Gibson assembly reaction. MG1655 was transformed with a resulting pTOF-fliCsticky (SI Fig. 10A). The gene replacement was then performed following the PMGR protocol from (Merlin et al., 2002). Resulting strain EK01 was transformed with pTOF-pHluorin plasmid (SI Fig. 10B) and the protocol was repeated. Obtained EK07 strain differs from parental MG1655 strain in having "sticky" flagella and pHluorin gene on the attTn7 site of the chromosome, which is confirmed by sequencing. Supplementary Methods: pHluorin calibration The in vivo calibration of the pH sensor (pHluorin) is performed as described in (Wang et al., 2018). Briefly, MM9 medium is adjusted to a set of pH values in the range between 5.5 and 9 and supplemented with 40 mM potassium benzoate and 40 mM methylamine hydrochloride (Martinez et al., 2012). The medium of a known pH is flushed into the tunnel-slide with cells attached to the surface as described before and incubated for 5 min. Total of ∼100 cells are imaged at 50 ms exposure time. The calibration curve obtained is given in SI Fig. 11A and fitted with a sigmoid function R395/475 = (a1ek(pH -- pH0) + a2)/(ek(pH -- pH0) + 1), where a1,a2,k,pH0 are fitting parameters. The in vitro calibration (SI Fig. 11B) is done using his-tagged pHluorin, purified with affinity chromatography (Urh et al., 2009) diluted into buffer of known pH (and supplemented with indole or butanol when testing pHluorin sensitivity to it). The pHluorin emission intensity is measured for 395 nm and 475 nm excitation in Spark 10M multimode plate reader (Tecan Trading AG, Switzerland). pH of the buffers in presence of indole/butanol is confirmed with pH meter (FE20 FiveEasyTM, Mettler-Toledo International Inc, Switzerland). To account for photobleaching, MM9 medium supplemented with 5% ethanol is supplied (at 10 l/min flow rate) to the cells attached to the surface in a flow-cell. After 5 to 10 min incubation imaging of ethanol treated cells is performed as during the photodamage experiment. The change in pHluorin intensities ratio is determined for ethanol treated cells and used to account for photobleaching by calculating the normalisation coefficients for each time point. Supplementary Methods: pHluorin image analysis To analyse images of pHluorin expressing cells we first identify cells uniformly attached to the surface (so called "flat cells" (Pilizota and Shaevitz, 2012; Buda et al., 2016)), and calculate the background intensity by dividing the whole image into four identical squares. Next we find the minimal intensity in each square and average 3x3 pixel box surrounding the minimal intensity pixel to get the background intensity value for that square. We subtract the background intensity of the corresponding square based on the location of the "flat" cell. We create a cell mask for flat cells using Otsu thresholding method (Otsu, 1979) followed by binary erosion and binary dilation to eliminate noise. We obtain the intensities for 475 nm and 395 nm 16/26 channels by calculating the mean pixel intensity within the identified cell mask, and calculate the ratio of the two to obtain the cytoplasmic pH from the in vivo calibration curve (see above and SI Fig. 11A). 17/26 Supplementary Tables Plasmid Fragment name Template Gibson assembly primers fliA shoulder MG1655 CCGCTTATGTCTATTGCTG- 5'- GTCTCGGTACCCGACCTGCACAAT- GCTTCGTGACGCACCA 5'- AGCAGGTTCTGTCTCTGCT- GCAGGGTTAATCGTTGTAACCTGAT- TAACTGAGACTGA pTOF- fliCsticky fliCsticky pFD313 (Berg and Turner, 1993) 5'- CGTCAGTCTCAGTTAATCAGGT- TACAACGATTAACCCTGCAGCAGA- GACAGAACCTGCT fliD shoulder MG1655 Tn7 Left Shoulder MG1655 pTOF- pHluorin V. harveyi promoter pWR20 (Pilizota and Shaevitz, 2012) pHluorin pkk223-3/pHluorin (Morimoto 2011) et al., 5'- CAACGACTTGCAATATAGGATAAC- GAATCATGGCACAAGTCATTAATAC- CAACAGCCTC GAGGCTGTTGGTATTAAT- 5'- GACTTGTGCCATGATTCGTTATC- CTATATTGCAAGTCGTTG 5'- GCTACAGGGCGCGTCCCATTCGC- CACCGGTCGAAAGTTTAGCG- GTAAACGACGATTG TATGTCTATTGCTGGTCTCG- 5'- GTACCCGACCTGCAATGCCGGT- TATTGTTGTTGCACCGA 5'- TCGAAAGACTGGGCCTTTCGTTT- TATCTGCCCGCTTACGCAGGGCATC- CATTTATTACT AGTGAAAAGTTCTTCTCCTT- 5'- TACTCATATGTATATCTCCTTAACTAG- GTAATTATCAAGC ATGTTTGATTAAAAA- 5'- CATAACAGGAAGAAAAATGCCC- CGCATTTCGACACCTTCGTCCTC GTAATAAATGGATGCCCTGCG- 5'- TAAGCGGGCAGATAAAACGAAAG- GCCCAGTCTTTCGAC TGATAATTACCTAGTTAAGGA- 5'- GATATACATATGAGTAAAGGAGAA- GAACTTTTCACTGGA 18/26 pTOF- pHluorin Tn7 Right Shoulder MG1655 5'- CACTTACCTGAGGACGAAGGT- GTCGAAATGCGGGGCATTTTTCTTC- CTGTTATGTTTTTA 5'- CAGGGCGCGTCCCATTCGCCAC- CGGTCGACAAACACAGAGAAAG- CACTCATCGATAAGG Table 1. List of primers used to generate pTOF24 derivatives for PMGR method. 19/26 Supplementary Figures SI Figure 1. MG1655 and EK07 growth curves in LB diluted from the overnight culture to the initial OD 0.05. Cells are grown in 125 ml flasks at 37◦C while shaken at 220 rpm. Dotted line indicates OD=2, at which cells are taken for the measurements as described in Materials and Methods. SI Figure 2. A cartoon of the tunnel-slide and flow-cell used in the experiments. (A) Top view and side view of the tunnel-slide used for fast flush experiments (indole and butanol treatment): double-sided sticky tape forms a channel and is sandwiched between the microscope slide and the cover glass. Overall volume of the tunnel-slide is 10-15 l (26x5x0.1 mm). (B) Top view and side view of the flow-cell used for photodamage experiments: two 1.8 mm holes are drilled in the microscope slide and connected to the tubing. The experimental chamber is formed by the gene frame and 22x60 mm cover glass. Overall volume of the flow-cell is ∼175-200 l. 20/26 SI Figure 3. Solely exchanging media does not affect the motor speed. (A) Motor speed of a cell in the tunnel-slide is recorded for 2 min, at which point fresh MM9 is flushed into the tunnel (red vertical lines indicate the duration of the flush). (B) Mean trace with standard deviation (shaded) of 9 cells exposed to the MM9 exchange as in (A). SI Figure 4. Laser power used for back-focal-plane interferometry does not cause PMF damage. (A) Motor speed of a cell in a flow-cell with MM9 continuously exchanged at a 10 l/min rate, recorded for 20 min. (B) Mean BFM speed trace of 17 cells monitored as in (A). Standard deviation is shown as shaded grey area. The control experiments were previously done for up to 45 min (Rosko, 2017) 21/26 SI Figure 5. Cytoplasmic pH measurements during stress treatment. (A) 0.5 mM and (B) 2.5 mM indole, (C) 0.5% and (D) 1% butanol, (E) 153 mW/cm2 and (F) 591 mW/cm2 effective light power. Error bars represent standard deviation, and the red line value of the external pH=7.5, which we set according to the previous measurements (Wang et al., 2018). Throughout the measurements ƊpH ∼0, where the cytoplasmic pH before shock in our buffers is slightly higher then previously reported (up to maximum of ƊpH ∼0.5). Difference in 0.5 pH units gives ∼ 30 mV of PMF, which is within the standard deviation of the motor speed measurements (Fig. 2D or 3D). 22/26 SI Figure 6. Average motor speed traces with standard deviations for all the stresses. (A) Version of main text Fig. 2B, (B) Fig. 2D, (C) Fig. 3B, (D) Fig. 3D, (E) Fig. 4B, (F) Fig. 4C, and (G) Fig. 4D, but with standard deviations instead or standard errors. In (D) and (G) red diamonds represent cells in phosphate buffer saline, and blue cells in MM9. Dotted lines in (B), (D) and (G) are fits described in the main text. 23/26 SI Figure 7. Motor speed is reversible during multiple butanol shocks. Butanol is flushed in at the time denoted with red, and out with green arrow. First three treatments are done with 0.5%, and subsequent three with 0.25% butanol. Intervals between butanol and MM9 flushes are of the (A) 1 min, or (B) 30 s duration. SI Figure 8. Raw single motor speed traces during and after light exposure showing that the damage is not reversible on the time scale of our experiment. Shaded blue regions indicate periods of time when the light was on. Cells are kept in MM9 in (A) and (B) and in PBS in (C) and (D). Effective light power is 38 mW/cm2 for (A), 350 mW/cm2 for (B) and 591 mW/cm2 for (C) and (D). 24/26 SI Figure 9. The presence of pHluorin does not alter the response to light induced damage. Averaged and normalised motor speed traces (n(cid:62)20 cells) during exposure to light with 38.4 and 350 mW/cm2 effective power. Red traces are obtained with EK07 strain expressing pHluorin protein, and blue traces with the same strain lacking the pHluorin gene (EK01). Shaded regions represent standard deviations. SI Figure 10. Maps of plasmids used to construct EK07 strain. Plasmid backbone contains chloram- phenicol resistance marker (CmR), gene encoding levansucrase (SacB), which makes host cells sensitive to sucrose, and temperature sensitive repAts replicon. (A) Map of pTOF-fliCsticky. fliCsticky gene flanked by ∼400bp sequence of homology to the surrounding chromosome (here named fliA and fliD shoulders) is cloned into pTOF24 between PstI and SalI cloning sites. Resulting plasmid has a chloramphenicol resistance. (B) Map of pTOF-pHluorin. pHluorin gene under strong constitutive V. harveyi cytochrome C oxidase promoter (Pilizota and Shaevitz, 2012) flanked by ∼350bp sequence of homology to the attTn7 site on the chromosome is cloned into pTOF24 between PstI and SalI cloning sites. 25/26 SI Figure 11. pHluorin is calibrated in vivo and in vitro. (A) In vivo calibration curve is obtained as described in SI Materials and methods. 395/475 nm emission intensity ratio plotted against pH with standard deviation. The experimental points are fitted with a sigmoid function R395/475 = (a1ek(pH -- pH0) + a2)/(ek(pH -- pH0) + 1), where a1 = 1.1, a2 = 0.168, k = 1.7 and pH0 = 6.976 are obtained from the fit. (B) 395/475 nm emission intensity ratio of purified pHluorin in presence of butanol and indole at different pH; error bars are standard deviation. pHluorin readings are influenced by the presence of indole/butanol for pH 9, but not pH 7 (or 5.5) at which we perform our experiments. Inset: Calibration curve of the purified pHluorin. Parameters obtained from fitting the experimental data to the sigmoid function are: a1 = 2.1, a2 = 0.33, k = 1.7 and pH0 = 6.97. 26/26

Subsets and Splits

No saved queries yet

Save your SQL queries to embed, download, and access them later. Queries will appear here once saved.